]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
c54fce6e | 2 | * kernel/workqueue.c - generic async execution with shared worker pool |
1da177e4 | 3 | * |
c54fce6e | 4 | * Copyright (C) 2002 Ingo Molnar |
1da177e4 | 5 | * |
c54fce6e TH |
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> | |
1da177e4 | 11 | * |
c54fce6e | 12 | * Made to use alloc_percpu by Christoph Lameter. |
1da177e4 | 13 | * |
c54fce6e TH |
14 | * Copyright (C) 2010 SUSE Linux Products GmbH |
15 | * Copyright (C) 2010 Tejun Heo <tj@kernel.org> | |
89ada679 | 16 | * |
c54fce6e TH |
17 | * This is the generic async execution mechanism. Work items as are |
18 | * executed in process context. The worker pool is shared and | |
19 | * automatically managed. There is one worker pool for each CPU and | |
20 | * one extra for works which are better served by workers which are | |
21 | * not bound to any specific CPU. | |
22 | * | |
23 | * Please read Documentation/workqueue.txt for details. | |
1da177e4 LT |
24 | */ |
25 | ||
9984de1a | 26 | #include <linux/export.h> |
1da177e4 LT |
27 | #include <linux/kernel.h> |
28 | #include <linux/sched.h> | |
29 | #include <linux/init.h> | |
30 | #include <linux/signal.h> | |
31 | #include <linux/completion.h> | |
32 | #include <linux/workqueue.h> | |
33 | #include <linux/slab.h> | |
34 | #include <linux/cpu.h> | |
35 | #include <linux/notifier.h> | |
36 | #include <linux/kthread.h> | |
1fa44eca | 37 | #include <linux/hardirq.h> |
46934023 | 38 | #include <linux/mempolicy.h> |
341a5958 | 39 | #include <linux/freezer.h> |
d5abe669 PZ |
40 | #include <linux/kallsyms.h> |
41 | #include <linux/debug_locks.h> | |
4e6045f1 | 42 | #include <linux/lockdep.h> |
c34056a3 | 43 | #include <linux/idr.h> |
29c91e99 | 44 | #include <linux/jhash.h> |
42f8570f | 45 | #include <linux/hashtable.h> |
76af4d93 | 46 | #include <linux/rculist.h> |
e22bee78 | 47 | |
ea138446 | 48 | #include "workqueue_internal.h" |
1da177e4 | 49 | |
c8e55f36 | 50 | enum { |
24647570 TH |
51 | /* |
52 | * worker_pool flags | |
bc2ae0f5 | 53 | * |
24647570 | 54 | * A bound pool is either associated or disassociated with its CPU. |
bc2ae0f5 TH |
55 | * While associated (!DISASSOCIATED), all workers are bound to the |
56 | * CPU and none has %WORKER_UNBOUND set and concurrency management | |
57 | * is in effect. | |
58 | * | |
59 | * While DISASSOCIATED, the cpu may be offline and all workers have | |
60 | * %WORKER_UNBOUND set and concurrency management disabled, and may | |
24647570 | 61 | * be executing on any CPU. The pool behaves as an unbound one. |
bc2ae0f5 | 62 | * |
bc3a1afc TH |
63 | * Note that DISASSOCIATED should be flipped only while holding |
64 | * manager_mutex to avoid changing binding state while | |
24647570 | 65 | * create_worker() is in progress. |
bc2ae0f5 | 66 | */ |
11ebea50 | 67 | POOL_MANAGE_WORKERS = 1 << 0, /* need to manage workers */ |
24647570 | 68 | POOL_DISASSOCIATED = 1 << 2, /* cpu can't serve workers */ |
35b6bb63 | 69 | POOL_FREEZING = 1 << 3, /* freeze in progress */ |
db7bccf4 | 70 | |
c8e55f36 TH |
71 | /* worker flags */ |
72 | WORKER_STARTED = 1 << 0, /* started */ | |
73 | WORKER_DIE = 1 << 1, /* die die die */ | |
74 | WORKER_IDLE = 1 << 2, /* is idle */ | |
e22bee78 | 75 | WORKER_PREP = 1 << 3, /* preparing to run works */ |
fb0e7beb | 76 | WORKER_CPU_INTENSIVE = 1 << 6, /* cpu intensive */ |
f3421797 | 77 | WORKER_UNBOUND = 1 << 7, /* worker is unbound */ |
a9ab775b | 78 | WORKER_REBOUND = 1 << 8, /* worker was rebound */ |
e22bee78 | 79 | |
a9ab775b TH |
80 | WORKER_NOT_RUNNING = WORKER_PREP | WORKER_CPU_INTENSIVE | |
81 | WORKER_UNBOUND | WORKER_REBOUND, | |
db7bccf4 | 82 | |
e34cdddb | 83 | NR_STD_WORKER_POOLS = 2, /* # standard pools per cpu */ |
4ce62e9e | 84 | |
29c91e99 | 85 | UNBOUND_POOL_HASH_ORDER = 6, /* hashed by pool->attrs */ |
c8e55f36 | 86 | BUSY_WORKER_HASH_ORDER = 6, /* 64 pointers */ |
db7bccf4 | 87 | |
e22bee78 TH |
88 | MAX_IDLE_WORKERS_RATIO = 4, /* 1/4 of busy can be idle */ |
89 | IDLE_WORKER_TIMEOUT = 300 * HZ, /* keep idle ones for 5 mins */ | |
90 | ||
3233cdbd TH |
91 | MAYDAY_INITIAL_TIMEOUT = HZ / 100 >= 2 ? HZ / 100 : 2, |
92 | /* call for help after 10ms | |
93 | (min two ticks) */ | |
e22bee78 TH |
94 | MAYDAY_INTERVAL = HZ / 10, /* and then every 100ms */ |
95 | CREATE_COOLDOWN = HZ, /* time to breath after fail */ | |
e22bee78 TH |
96 | |
97 | /* | |
98 | * Rescue workers are used only on emergencies and shared by | |
99 | * all cpus. Give -20. | |
100 | */ | |
101 | RESCUER_NICE_LEVEL = -20, | |
3270476a | 102 | HIGHPRI_NICE_LEVEL = -20, |
c8e55f36 | 103 | }; |
1da177e4 LT |
104 | |
105 | /* | |
4690c4ab TH |
106 | * Structure fields follow one of the following exclusion rules. |
107 | * | |
e41e704b TH |
108 | * I: Modifiable by initialization/destruction paths and read-only for |
109 | * everyone else. | |
4690c4ab | 110 | * |
e22bee78 TH |
111 | * P: Preemption protected. Disabling preemption is enough and should |
112 | * only be modified and accessed from the local cpu. | |
113 | * | |
d565ed63 | 114 | * L: pool->lock protected. Access with pool->lock held. |
4690c4ab | 115 | * |
d565ed63 TH |
116 | * X: During normal operation, modification requires pool->lock and should |
117 | * be done only from local cpu. Either disabling preemption on local | |
118 | * cpu or grabbing pool->lock is enough for read access. If | |
119 | * POOL_DISASSOCIATED is set, it's identical to L. | |
e22bee78 | 120 | * |
822d8405 TH |
121 | * MG: pool->manager_mutex and pool->lock protected. Writes require both |
122 | * locks. Reads can happen under either lock. | |
123 | * | |
68e13a67 | 124 | * PL: wq_pool_mutex protected. |
5bcab335 | 125 | * |
68e13a67 | 126 | * PR: wq_pool_mutex protected for writes. Sched-RCU protected for reads. |
76af4d93 | 127 | * |
3c25a55d LJ |
128 | * WQ: wq->mutex protected. |
129 | * | |
b5927605 | 130 | * WR: wq->mutex protected for writes. Sched-RCU protected for reads. |
2e109a28 TH |
131 | * |
132 | * MD: wq_mayday_lock protected. | |
1da177e4 | 133 | */ |
1da177e4 | 134 | |
2eaebdb3 | 135 | /* struct worker is defined in workqueue_internal.h */ |
c34056a3 | 136 | |
bd7bdd43 | 137 | struct worker_pool { |
d565ed63 | 138 | spinlock_t lock; /* the pool lock */ |
d84ff051 | 139 | int cpu; /* I: the associated cpu */ |
9daf9e67 | 140 | int id; /* I: pool ID */ |
11ebea50 | 141 | unsigned int flags; /* X: flags */ |
bd7bdd43 TH |
142 | |
143 | struct list_head worklist; /* L: list of pending works */ | |
144 | int nr_workers; /* L: total number of workers */ | |
ea1abd61 LJ |
145 | |
146 | /* nr_idle includes the ones off idle_list for rebinding */ | |
bd7bdd43 TH |
147 | int nr_idle; /* L: currently idle ones */ |
148 | ||
149 | struct list_head idle_list; /* X: list of idle workers */ | |
150 | struct timer_list idle_timer; /* L: worker idle timeout */ | |
151 | struct timer_list mayday_timer; /* L: SOS timer for workers */ | |
152 | ||
c5aa87bb | 153 | /* a workers is either on busy_hash or idle_list, or the manager */ |
c9e7cf27 TH |
154 | DECLARE_HASHTABLE(busy_hash, BUSY_WORKER_HASH_ORDER); |
155 | /* L: hash of busy workers */ | |
156 | ||
bc3a1afc | 157 | /* see manage_workers() for details on the two manager mutexes */ |
34a06bd6 | 158 | struct mutex manager_arb; /* manager arbitration */ |
bc3a1afc | 159 | struct mutex manager_mutex; /* manager exclusion */ |
822d8405 | 160 | struct idr worker_idr; /* MG: worker IDs and iteration */ |
e19e397a | 161 | |
7a4e344c | 162 | struct workqueue_attrs *attrs; /* I: worker attributes */ |
68e13a67 LJ |
163 | struct hlist_node hash_node; /* PL: unbound_pool_hash node */ |
164 | int refcnt; /* PL: refcnt for unbound pools */ | |
7a4e344c | 165 | |
e19e397a TH |
166 | /* |
167 | * The current concurrency level. As it's likely to be accessed | |
168 | * from other CPUs during try_to_wake_up(), put it in a separate | |
169 | * cacheline. | |
170 | */ | |
171 | atomic_t nr_running ____cacheline_aligned_in_smp; | |
29c91e99 TH |
172 | |
173 | /* | |
174 | * Destruction of pool is sched-RCU protected to allow dereferences | |
175 | * from get_work_pool(). | |
176 | */ | |
177 | struct rcu_head rcu; | |
8b03ae3c TH |
178 | } ____cacheline_aligned_in_smp; |
179 | ||
1da177e4 | 180 | /* |
112202d9 TH |
181 | * The per-pool workqueue. While queued, the lower WORK_STRUCT_FLAG_BITS |
182 | * of work_struct->data are used for flags and the remaining high bits | |
183 | * point to the pwq; thus, pwqs need to be aligned at two's power of the | |
184 | * number of flag bits. | |
1da177e4 | 185 | */ |
112202d9 | 186 | struct pool_workqueue { |
bd7bdd43 | 187 | struct worker_pool *pool; /* I: the associated pool */ |
4690c4ab | 188 | struct workqueue_struct *wq; /* I: the owning workqueue */ |
73f53c4a TH |
189 | int work_color; /* L: current color */ |
190 | int flush_color; /* L: flushing color */ | |
8864b4e5 | 191 | int refcnt; /* L: reference count */ |
73f53c4a TH |
192 | int nr_in_flight[WORK_NR_COLORS]; |
193 | /* L: nr of in_flight works */ | |
1e19ffc6 | 194 | int nr_active; /* L: nr of active works */ |
a0a1a5fd | 195 | int max_active; /* L: max active works */ |
1e19ffc6 | 196 | struct list_head delayed_works; /* L: delayed works */ |
3c25a55d | 197 | struct list_head pwqs_node; /* WR: node on wq->pwqs */ |
2e109a28 | 198 | struct list_head mayday_node; /* MD: node on wq->maydays */ |
8864b4e5 TH |
199 | |
200 | /* | |
201 | * Release of unbound pwq is punted to system_wq. See put_pwq() | |
202 | * and pwq_unbound_release_workfn() for details. pool_workqueue | |
203 | * itself is also sched-RCU protected so that the first pwq can be | |
b09f4fd3 | 204 | * determined without grabbing wq->mutex. |
8864b4e5 TH |
205 | */ |
206 | struct work_struct unbound_release_work; | |
207 | struct rcu_head rcu; | |
e904e6c2 | 208 | } __aligned(1 << WORK_STRUCT_FLAG_BITS); |
1da177e4 | 209 | |
73f53c4a TH |
210 | /* |
211 | * Structure used to wait for workqueue flush. | |
212 | */ | |
213 | struct wq_flusher { | |
3c25a55d LJ |
214 | struct list_head list; /* WQ: list of flushers */ |
215 | int flush_color; /* WQ: flush color waiting for */ | |
73f53c4a TH |
216 | struct completion done; /* flush completion */ |
217 | }; | |
218 | ||
226223ab TH |
219 | struct wq_device; |
220 | ||
1da177e4 | 221 | /* |
c5aa87bb TH |
222 | * The externally visible workqueue. It relays the issued work items to |
223 | * the appropriate worker_pool through its pool_workqueues. | |
1da177e4 LT |
224 | */ |
225 | struct workqueue_struct { | |
87fc741e | 226 | unsigned int flags; /* WQ: WQ_* flags */ |
420c0ddb | 227 | struct pool_workqueue __percpu *cpu_pwqs; /* I: per-cpu pwq's */ |
3c25a55d | 228 | struct list_head pwqs; /* WR: all pwqs of this wq */ |
68e13a67 | 229 | struct list_head list; /* PL: list of all workqueues */ |
73f53c4a | 230 | |
3c25a55d LJ |
231 | struct mutex mutex; /* protects this wq */ |
232 | int work_color; /* WQ: current work color */ | |
233 | int flush_color; /* WQ: current flush color */ | |
112202d9 | 234 | atomic_t nr_pwqs_to_flush; /* flush in progress */ |
3c25a55d LJ |
235 | struct wq_flusher *first_flusher; /* WQ: first flusher */ |
236 | struct list_head flusher_queue; /* WQ: flush waiters */ | |
237 | struct list_head flusher_overflow; /* WQ: flush overflow list */ | |
73f53c4a | 238 | |
2e109a28 | 239 | struct list_head maydays; /* MD: pwqs requesting rescue */ |
e22bee78 TH |
240 | struct worker *rescuer; /* I: rescue worker */ |
241 | ||
87fc741e | 242 | int nr_drainers; /* WQ: drain in progress */ |
a357fc03 | 243 | int saved_max_active; /* WQ: saved pwq max_active */ |
226223ab TH |
244 | |
245 | #ifdef CONFIG_SYSFS | |
246 | struct wq_device *wq_dev; /* I: for sysfs interface */ | |
247 | #endif | |
4e6045f1 | 248 | #ifdef CONFIG_LOCKDEP |
4690c4ab | 249 | struct lockdep_map lockdep_map; |
4e6045f1 | 250 | #endif |
b196be89 | 251 | char name[]; /* I: workqueue name */ |
1da177e4 LT |
252 | }; |
253 | ||
e904e6c2 TH |
254 | static struct kmem_cache *pwq_cache; |
255 | ||
68e13a67 | 256 | static DEFINE_MUTEX(wq_pool_mutex); /* protects pools and workqueues list */ |
2e109a28 | 257 | static DEFINE_SPINLOCK(wq_mayday_lock); /* protects wq->maydays list */ |
5bcab335 | 258 | |
68e13a67 LJ |
259 | static LIST_HEAD(workqueues); /* PL: list of all workqueues */ |
260 | static bool workqueue_freezing; /* PL: have wqs started freezing? */ | |
7d19c5ce TH |
261 | |
262 | /* the per-cpu worker pools */ | |
263 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct worker_pool [NR_STD_WORKER_POOLS], | |
264 | cpu_worker_pools); | |
265 | ||
68e13a67 | 266 | static DEFINE_IDR(worker_pool_idr); /* PR: idr of all pools */ |
7d19c5ce | 267 | |
68e13a67 | 268 | /* PL: hash of all unbound pools keyed by pool->attrs */ |
29c91e99 TH |
269 | static DEFINE_HASHTABLE(unbound_pool_hash, UNBOUND_POOL_HASH_ORDER); |
270 | ||
c5aa87bb | 271 | /* I: attributes used when instantiating standard unbound pools on demand */ |
29c91e99 TH |
272 | static struct workqueue_attrs *unbound_std_wq_attrs[NR_STD_WORKER_POOLS]; |
273 | ||
d320c038 | 274 | struct workqueue_struct *system_wq __read_mostly; |
d320c038 | 275 | EXPORT_SYMBOL_GPL(system_wq); |
044c782c | 276 | struct workqueue_struct *system_highpri_wq __read_mostly; |
1aabe902 | 277 | EXPORT_SYMBOL_GPL(system_highpri_wq); |
044c782c | 278 | struct workqueue_struct *system_long_wq __read_mostly; |
d320c038 | 279 | EXPORT_SYMBOL_GPL(system_long_wq); |
044c782c | 280 | struct workqueue_struct *system_unbound_wq __read_mostly; |
f3421797 | 281 | EXPORT_SYMBOL_GPL(system_unbound_wq); |
044c782c | 282 | struct workqueue_struct *system_freezable_wq __read_mostly; |
24d51add | 283 | EXPORT_SYMBOL_GPL(system_freezable_wq); |
d320c038 | 284 | |
7d19c5ce TH |
285 | static int worker_thread(void *__worker); |
286 | static void copy_workqueue_attrs(struct workqueue_attrs *to, | |
287 | const struct workqueue_attrs *from); | |
288 | ||
97bd2347 TH |
289 | #define CREATE_TRACE_POINTS |
290 | #include <trace/events/workqueue.h> | |
291 | ||
68e13a67 | 292 | #define assert_rcu_or_pool_mutex() \ |
5bcab335 | 293 | rcu_lockdep_assert(rcu_read_lock_sched_held() || \ |
68e13a67 LJ |
294 | lockdep_is_held(&wq_pool_mutex), \ |
295 | "sched RCU or wq_pool_mutex should be held") | |
5bcab335 | 296 | |
b09f4fd3 | 297 | #define assert_rcu_or_wq_mutex(wq) \ |
76af4d93 | 298 | rcu_lockdep_assert(rcu_read_lock_sched_held() || \ |
b5927605 | 299 | lockdep_is_held(&wq->mutex), \ |
b09f4fd3 | 300 | "sched RCU or wq->mutex should be held") |
76af4d93 | 301 | |
822d8405 TH |
302 | #ifdef CONFIG_LOCKDEP |
303 | #define assert_manager_or_pool_lock(pool) \ | |
519e3c11 LJ |
304 | WARN_ONCE(debug_locks && \ |
305 | !lockdep_is_held(&(pool)->manager_mutex) && \ | |
822d8405 TH |
306 | !lockdep_is_held(&(pool)->lock), \ |
307 | "pool->manager_mutex or ->lock should be held") | |
308 | #else | |
309 | #define assert_manager_or_pool_lock(pool) do { } while (0) | |
310 | #endif | |
311 | ||
f02ae73a TH |
312 | #define for_each_cpu_worker_pool(pool, cpu) \ |
313 | for ((pool) = &per_cpu(cpu_worker_pools, cpu)[0]; \ | |
314 | (pool) < &per_cpu(cpu_worker_pools, cpu)[NR_STD_WORKER_POOLS]; \ | |
7a62c2c8 | 315 | (pool)++) |
4ce62e9e | 316 | |
17116969 TH |
317 | /** |
318 | * for_each_pool - iterate through all worker_pools in the system | |
319 | * @pool: iteration cursor | |
611c92a0 | 320 | * @pi: integer used for iteration |
fa1b54e6 | 321 | * |
68e13a67 LJ |
322 | * This must be called either with wq_pool_mutex held or sched RCU read |
323 | * locked. If the pool needs to be used beyond the locking in effect, the | |
324 | * caller is responsible for guaranteeing that the pool stays online. | |
fa1b54e6 TH |
325 | * |
326 | * The if/else clause exists only for the lockdep assertion and can be | |
327 | * ignored. | |
17116969 | 328 | */ |
611c92a0 TH |
329 | #define for_each_pool(pool, pi) \ |
330 | idr_for_each_entry(&worker_pool_idr, pool, pi) \ | |
68e13a67 | 331 | if (({ assert_rcu_or_pool_mutex(); false; })) { } \ |
fa1b54e6 | 332 | else |
17116969 | 333 | |
822d8405 TH |
334 | /** |
335 | * for_each_pool_worker - iterate through all workers of a worker_pool | |
336 | * @worker: iteration cursor | |
337 | * @wi: integer used for iteration | |
338 | * @pool: worker_pool to iterate workers of | |
339 | * | |
340 | * This must be called with either @pool->manager_mutex or ->lock held. | |
341 | * | |
342 | * The if/else clause exists only for the lockdep assertion and can be | |
343 | * ignored. | |
344 | */ | |
345 | #define for_each_pool_worker(worker, wi, pool) \ | |
346 | idr_for_each_entry(&(pool)->worker_idr, (worker), (wi)) \ | |
347 | if (({ assert_manager_or_pool_lock((pool)); false; })) { } \ | |
348 | else | |
349 | ||
49e3cf44 TH |
350 | /** |
351 | * for_each_pwq - iterate through all pool_workqueues of the specified workqueue | |
352 | * @pwq: iteration cursor | |
353 | * @wq: the target workqueue | |
76af4d93 | 354 | * |
b09f4fd3 | 355 | * This must be called either with wq->mutex held or sched RCU read locked. |
794b18bc TH |
356 | * If the pwq needs to be used beyond the locking in effect, the caller is |
357 | * responsible for guaranteeing that the pwq stays online. | |
76af4d93 TH |
358 | * |
359 | * The if/else clause exists only for the lockdep assertion and can be | |
360 | * ignored. | |
49e3cf44 TH |
361 | */ |
362 | #define for_each_pwq(pwq, wq) \ | |
76af4d93 | 363 | list_for_each_entry_rcu((pwq), &(wq)->pwqs, pwqs_node) \ |
b09f4fd3 | 364 | if (({ assert_rcu_or_wq_mutex(wq); false; })) { } \ |
76af4d93 | 365 | else |
f3421797 | 366 | |
dc186ad7 TG |
367 | #ifdef CONFIG_DEBUG_OBJECTS_WORK |
368 | ||
369 | static struct debug_obj_descr work_debug_descr; | |
370 | ||
99777288 SG |
371 | static void *work_debug_hint(void *addr) |
372 | { | |
373 | return ((struct work_struct *) addr)->func; | |
374 | } | |
375 | ||
dc186ad7 TG |
376 | /* |
377 | * fixup_init is called when: | |
378 | * - an active object is initialized | |
379 | */ | |
380 | static int work_fixup_init(void *addr, enum debug_obj_state state) | |
381 | { | |
382 | struct work_struct *work = addr; | |
383 | ||
384 | switch (state) { | |
385 | case ODEBUG_STATE_ACTIVE: | |
386 | cancel_work_sync(work); | |
387 | debug_object_init(work, &work_debug_descr); | |
388 | return 1; | |
389 | default: | |
390 | return 0; | |
391 | } | |
392 | } | |
393 | ||
394 | /* | |
395 | * fixup_activate is called when: | |
396 | * - an active object is activated | |
397 | * - an unknown object is activated (might be a statically initialized object) | |
398 | */ | |
399 | static int work_fixup_activate(void *addr, enum debug_obj_state state) | |
400 | { | |
401 | struct work_struct *work = addr; | |
402 | ||
403 | switch (state) { | |
404 | ||
405 | case ODEBUG_STATE_NOTAVAILABLE: | |
406 | /* | |
407 | * This is not really a fixup. The work struct was | |
408 | * statically initialized. We just make sure that it | |
409 | * is tracked in the object tracker. | |
410 | */ | |
22df02bb | 411 | if (test_bit(WORK_STRUCT_STATIC_BIT, work_data_bits(work))) { |
dc186ad7 TG |
412 | debug_object_init(work, &work_debug_descr); |
413 | debug_object_activate(work, &work_debug_descr); | |
414 | return 0; | |
415 | } | |
416 | WARN_ON_ONCE(1); | |
417 | return 0; | |
418 | ||
419 | case ODEBUG_STATE_ACTIVE: | |
420 | WARN_ON(1); | |
421 | ||
422 | default: | |
423 | return 0; | |
424 | } | |
425 | } | |
426 | ||
427 | /* | |
428 | * fixup_free is called when: | |
429 | * - an active object is freed | |
430 | */ | |
431 | static int work_fixup_free(void *addr, enum debug_obj_state state) | |
432 | { | |
433 | struct work_struct *work = addr; | |
434 | ||
435 | switch (state) { | |
436 | case ODEBUG_STATE_ACTIVE: | |
437 | cancel_work_sync(work); | |
438 | debug_object_free(work, &work_debug_descr); | |
439 | return 1; | |
440 | default: | |
441 | return 0; | |
442 | } | |
443 | } | |
444 | ||
445 | static struct debug_obj_descr work_debug_descr = { | |
446 | .name = "work_struct", | |
99777288 | 447 | .debug_hint = work_debug_hint, |
dc186ad7 TG |
448 | .fixup_init = work_fixup_init, |
449 | .fixup_activate = work_fixup_activate, | |
450 | .fixup_free = work_fixup_free, | |
451 | }; | |
452 | ||
453 | static inline void debug_work_activate(struct work_struct *work) | |
454 | { | |
455 | debug_object_activate(work, &work_debug_descr); | |
456 | } | |
457 | ||
458 | static inline void debug_work_deactivate(struct work_struct *work) | |
459 | { | |
460 | debug_object_deactivate(work, &work_debug_descr); | |
461 | } | |
462 | ||
463 | void __init_work(struct work_struct *work, int onstack) | |
464 | { | |
465 | if (onstack) | |
466 | debug_object_init_on_stack(work, &work_debug_descr); | |
467 | else | |
468 | debug_object_init(work, &work_debug_descr); | |
469 | } | |
470 | EXPORT_SYMBOL_GPL(__init_work); | |
471 | ||
472 | void destroy_work_on_stack(struct work_struct *work) | |
473 | { | |
474 | debug_object_free(work, &work_debug_descr); | |
475 | } | |
476 | EXPORT_SYMBOL_GPL(destroy_work_on_stack); | |
477 | ||
478 | #else | |
479 | static inline void debug_work_activate(struct work_struct *work) { } | |
480 | static inline void debug_work_deactivate(struct work_struct *work) { } | |
481 | #endif | |
482 | ||
9daf9e67 TH |
483 | /* allocate ID and assign it to @pool */ |
484 | static int worker_pool_assign_id(struct worker_pool *pool) | |
485 | { | |
486 | int ret; | |
487 | ||
68e13a67 | 488 | lockdep_assert_held(&wq_pool_mutex); |
5bcab335 | 489 | |
fa1b54e6 TH |
490 | do { |
491 | if (!idr_pre_get(&worker_pool_idr, GFP_KERNEL)) | |
492 | return -ENOMEM; | |
fa1b54e6 | 493 | ret = idr_get_new(&worker_pool_idr, pool, &pool->id); |
fa1b54e6 | 494 | } while (ret == -EAGAIN); |
9daf9e67 | 495 | |
fa1b54e6 | 496 | return ret; |
7c3eed5c TH |
497 | } |
498 | ||
76af4d93 TH |
499 | /** |
500 | * first_pwq - return the first pool_workqueue of the specified workqueue | |
501 | * @wq: the target workqueue | |
502 | * | |
b09f4fd3 | 503 | * This must be called either with wq->mutex held or sched RCU read locked. |
794b18bc TH |
504 | * If the pwq needs to be used beyond the locking in effect, the caller is |
505 | * responsible for guaranteeing that the pwq stays online. | |
76af4d93 | 506 | */ |
7fb98ea7 | 507 | static struct pool_workqueue *first_pwq(struct workqueue_struct *wq) |
b1f4ec17 | 508 | { |
b09f4fd3 | 509 | assert_rcu_or_wq_mutex(wq); |
76af4d93 TH |
510 | return list_first_or_null_rcu(&wq->pwqs, struct pool_workqueue, |
511 | pwqs_node); | |
b1f4ec17 ON |
512 | } |
513 | ||
73f53c4a TH |
514 | static unsigned int work_color_to_flags(int color) |
515 | { | |
516 | return color << WORK_STRUCT_COLOR_SHIFT; | |
517 | } | |
518 | ||
519 | static int get_work_color(struct work_struct *work) | |
520 | { | |
521 | return (*work_data_bits(work) >> WORK_STRUCT_COLOR_SHIFT) & | |
522 | ((1 << WORK_STRUCT_COLOR_BITS) - 1); | |
523 | } | |
524 | ||
525 | static int work_next_color(int color) | |
526 | { | |
527 | return (color + 1) % WORK_NR_COLORS; | |
528 | } | |
1da177e4 | 529 | |
14441960 | 530 | /* |
112202d9 TH |
531 | * While queued, %WORK_STRUCT_PWQ is set and non flag bits of a work's data |
532 | * contain the pointer to the queued pwq. Once execution starts, the flag | |
7c3eed5c | 533 | * is cleared and the high bits contain OFFQ flags and pool ID. |
7a22ad75 | 534 | * |
112202d9 TH |
535 | * set_work_pwq(), set_work_pool_and_clear_pending(), mark_work_canceling() |
536 | * and clear_work_data() can be used to set the pwq, pool or clear | |
bbb68dfa TH |
537 | * work->data. These functions should only be called while the work is |
538 | * owned - ie. while the PENDING bit is set. | |
7a22ad75 | 539 | * |
112202d9 | 540 | * get_work_pool() and get_work_pwq() can be used to obtain the pool or pwq |
7c3eed5c | 541 | * corresponding to a work. Pool is available once the work has been |
112202d9 | 542 | * queued anywhere after initialization until it is sync canceled. pwq is |
7c3eed5c | 543 | * available only while the work item is queued. |
7a22ad75 | 544 | * |
bbb68dfa TH |
545 | * %WORK_OFFQ_CANCELING is used to mark a work item which is being |
546 | * canceled. While being canceled, a work item may have its PENDING set | |
547 | * but stay off timer and worklist for arbitrarily long and nobody should | |
548 | * try to steal the PENDING bit. | |
14441960 | 549 | */ |
7a22ad75 TH |
550 | static inline void set_work_data(struct work_struct *work, unsigned long data, |
551 | unsigned long flags) | |
365970a1 | 552 | { |
6183c009 | 553 | WARN_ON_ONCE(!work_pending(work)); |
7a22ad75 TH |
554 | atomic_long_set(&work->data, data | flags | work_static(work)); |
555 | } | |
365970a1 | 556 | |
112202d9 | 557 | static void set_work_pwq(struct work_struct *work, struct pool_workqueue *pwq, |
7a22ad75 TH |
558 | unsigned long extra_flags) |
559 | { | |
112202d9 TH |
560 | set_work_data(work, (unsigned long)pwq, |
561 | WORK_STRUCT_PENDING | WORK_STRUCT_PWQ | extra_flags); | |
365970a1 DH |
562 | } |
563 | ||
4468a00f LJ |
564 | static void set_work_pool_and_keep_pending(struct work_struct *work, |
565 | int pool_id) | |
566 | { | |
567 | set_work_data(work, (unsigned long)pool_id << WORK_OFFQ_POOL_SHIFT, | |
568 | WORK_STRUCT_PENDING); | |
569 | } | |
570 | ||
7c3eed5c TH |
571 | static void set_work_pool_and_clear_pending(struct work_struct *work, |
572 | int pool_id) | |
7a22ad75 | 573 | { |
23657bb1 TH |
574 | /* |
575 | * The following wmb is paired with the implied mb in | |
576 | * test_and_set_bit(PENDING) and ensures all updates to @work made | |
577 | * here are visible to and precede any updates by the next PENDING | |
578 | * owner. | |
579 | */ | |
580 | smp_wmb(); | |
7c3eed5c | 581 | set_work_data(work, (unsigned long)pool_id << WORK_OFFQ_POOL_SHIFT, 0); |
7a22ad75 | 582 | } |
f756d5e2 | 583 | |
7a22ad75 | 584 | static void clear_work_data(struct work_struct *work) |
1da177e4 | 585 | { |
7c3eed5c TH |
586 | smp_wmb(); /* see set_work_pool_and_clear_pending() */ |
587 | set_work_data(work, WORK_STRUCT_NO_POOL, 0); | |
1da177e4 LT |
588 | } |
589 | ||
112202d9 | 590 | static struct pool_workqueue *get_work_pwq(struct work_struct *work) |
b1f4ec17 | 591 | { |
e120153d | 592 | unsigned long data = atomic_long_read(&work->data); |
7a22ad75 | 593 | |
112202d9 | 594 | if (data & WORK_STRUCT_PWQ) |
e120153d TH |
595 | return (void *)(data & WORK_STRUCT_WQ_DATA_MASK); |
596 | else | |
597 | return NULL; | |
4d707b9f ON |
598 | } |
599 | ||
7c3eed5c TH |
600 | /** |
601 | * get_work_pool - return the worker_pool a given work was associated with | |
602 | * @work: the work item of interest | |
603 | * | |
604 | * Return the worker_pool @work was last associated with. %NULL if none. | |
fa1b54e6 | 605 | * |
68e13a67 LJ |
606 | * Pools are created and destroyed under wq_pool_mutex, and allows read |
607 | * access under sched-RCU read lock. As such, this function should be | |
608 | * called under wq_pool_mutex or with preemption disabled. | |
fa1b54e6 TH |
609 | * |
610 | * All fields of the returned pool are accessible as long as the above | |
611 | * mentioned locking is in effect. If the returned pool needs to be used | |
612 | * beyond the critical section, the caller is responsible for ensuring the | |
613 | * returned pool is and stays online. | |
7c3eed5c TH |
614 | */ |
615 | static struct worker_pool *get_work_pool(struct work_struct *work) | |
365970a1 | 616 | { |
e120153d | 617 | unsigned long data = atomic_long_read(&work->data); |
7c3eed5c | 618 | int pool_id; |
7a22ad75 | 619 | |
68e13a67 | 620 | assert_rcu_or_pool_mutex(); |
fa1b54e6 | 621 | |
112202d9 TH |
622 | if (data & WORK_STRUCT_PWQ) |
623 | return ((struct pool_workqueue *) | |
7c3eed5c | 624 | (data & WORK_STRUCT_WQ_DATA_MASK))->pool; |
7a22ad75 | 625 | |
7c3eed5c TH |
626 | pool_id = data >> WORK_OFFQ_POOL_SHIFT; |
627 | if (pool_id == WORK_OFFQ_POOL_NONE) | |
7a22ad75 TH |
628 | return NULL; |
629 | ||
fa1b54e6 | 630 | return idr_find(&worker_pool_idr, pool_id); |
7c3eed5c TH |
631 | } |
632 | ||
633 | /** | |
634 | * get_work_pool_id - return the worker pool ID a given work is associated with | |
635 | * @work: the work item of interest | |
636 | * | |
637 | * Return the worker_pool ID @work was last associated with. | |
638 | * %WORK_OFFQ_POOL_NONE if none. | |
639 | */ | |
640 | static int get_work_pool_id(struct work_struct *work) | |
641 | { | |
54d5b7d0 LJ |
642 | unsigned long data = atomic_long_read(&work->data); |
643 | ||
112202d9 TH |
644 | if (data & WORK_STRUCT_PWQ) |
645 | return ((struct pool_workqueue *) | |
54d5b7d0 | 646 | (data & WORK_STRUCT_WQ_DATA_MASK))->pool->id; |
7c3eed5c | 647 | |
54d5b7d0 | 648 | return data >> WORK_OFFQ_POOL_SHIFT; |
7c3eed5c TH |
649 | } |
650 | ||
bbb68dfa TH |
651 | static void mark_work_canceling(struct work_struct *work) |
652 | { | |
7c3eed5c | 653 | unsigned long pool_id = get_work_pool_id(work); |
bbb68dfa | 654 | |
7c3eed5c TH |
655 | pool_id <<= WORK_OFFQ_POOL_SHIFT; |
656 | set_work_data(work, pool_id | WORK_OFFQ_CANCELING, WORK_STRUCT_PENDING); | |
bbb68dfa TH |
657 | } |
658 | ||
659 | static bool work_is_canceling(struct work_struct *work) | |
660 | { | |
661 | unsigned long data = atomic_long_read(&work->data); | |
662 | ||
112202d9 | 663 | return !(data & WORK_STRUCT_PWQ) && (data & WORK_OFFQ_CANCELING); |
bbb68dfa TH |
664 | } |
665 | ||
e22bee78 | 666 | /* |
3270476a TH |
667 | * Policy functions. These define the policies on how the global worker |
668 | * pools are managed. Unless noted otherwise, these functions assume that | |
d565ed63 | 669 | * they're being called with pool->lock held. |
e22bee78 TH |
670 | */ |
671 | ||
63d95a91 | 672 | static bool __need_more_worker(struct worker_pool *pool) |
a848e3b6 | 673 | { |
e19e397a | 674 | return !atomic_read(&pool->nr_running); |
a848e3b6 ON |
675 | } |
676 | ||
4594bf15 | 677 | /* |
e22bee78 TH |
678 | * Need to wake up a worker? Called from anything but currently |
679 | * running workers. | |
974271c4 TH |
680 | * |
681 | * Note that, because unbound workers never contribute to nr_running, this | |
706026c2 | 682 | * function will always return %true for unbound pools as long as the |
974271c4 | 683 | * worklist isn't empty. |
4594bf15 | 684 | */ |
63d95a91 | 685 | static bool need_more_worker(struct worker_pool *pool) |
365970a1 | 686 | { |
63d95a91 | 687 | return !list_empty(&pool->worklist) && __need_more_worker(pool); |
e22bee78 | 688 | } |
4594bf15 | 689 | |
e22bee78 | 690 | /* Can I start working? Called from busy but !running workers. */ |
63d95a91 | 691 | static bool may_start_working(struct worker_pool *pool) |
e22bee78 | 692 | { |
63d95a91 | 693 | return pool->nr_idle; |
e22bee78 TH |
694 | } |
695 | ||
696 | /* Do I need to keep working? Called from currently running workers. */ | |
63d95a91 | 697 | static bool keep_working(struct worker_pool *pool) |
e22bee78 | 698 | { |
e19e397a TH |
699 | return !list_empty(&pool->worklist) && |
700 | atomic_read(&pool->nr_running) <= 1; | |
e22bee78 TH |
701 | } |
702 | ||
703 | /* Do we need a new worker? Called from manager. */ | |
63d95a91 | 704 | static bool need_to_create_worker(struct worker_pool *pool) |
e22bee78 | 705 | { |
63d95a91 | 706 | return need_more_worker(pool) && !may_start_working(pool); |
e22bee78 | 707 | } |
365970a1 | 708 | |
e22bee78 | 709 | /* Do I need to be the manager? */ |
63d95a91 | 710 | static bool need_to_manage_workers(struct worker_pool *pool) |
e22bee78 | 711 | { |
63d95a91 | 712 | return need_to_create_worker(pool) || |
11ebea50 | 713 | (pool->flags & POOL_MANAGE_WORKERS); |
e22bee78 TH |
714 | } |
715 | ||
716 | /* Do we have too many workers and should some go away? */ | |
63d95a91 | 717 | static bool too_many_workers(struct worker_pool *pool) |
e22bee78 | 718 | { |
34a06bd6 | 719 | bool managing = mutex_is_locked(&pool->manager_arb); |
63d95a91 TH |
720 | int nr_idle = pool->nr_idle + managing; /* manager is considered idle */ |
721 | int nr_busy = pool->nr_workers - nr_idle; | |
e22bee78 | 722 | |
ea1abd61 LJ |
723 | /* |
724 | * nr_idle and idle_list may disagree if idle rebinding is in | |
725 | * progress. Never return %true if idle_list is empty. | |
726 | */ | |
727 | if (list_empty(&pool->idle_list)) | |
728 | return false; | |
729 | ||
e22bee78 | 730 | return nr_idle > 2 && (nr_idle - 2) * MAX_IDLE_WORKERS_RATIO >= nr_busy; |
365970a1 DH |
731 | } |
732 | ||
4d707b9f | 733 | /* |
e22bee78 TH |
734 | * Wake up functions. |
735 | */ | |
736 | ||
7e11629d | 737 | /* Return the first worker. Safe with preemption disabled */ |
63d95a91 | 738 | static struct worker *first_worker(struct worker_pool *pool) |
7e11629d | 739 | { |
63d95a91 | 740 | if (unlikely(list_empty(&pool->idle_list))) |
7e11629d TH |
741 | return NULL; |
742 | ||
63d95a91 | 743 | return list_first_entry(&pool->idle_list, struct worker, entry); |
7e11629d TH |
744 | } |
745 | ||
746 | /** | |
747 | * wake_up_worker - wake up an idle worker | |
63d95a91 | 748 | * @pool: worker pool to wake worker from |
7e11629d | 749 | * |
63d95a91 | 750 | * Wake up the first idle worker of @pool. |
7e11629d TH |
751 | * |
752 | * CONTEXT: | |
d565ed63 | 753 | * spin_lock_irq(pool->lock). |
7e11629d | 754 | */ |
63d95a91 | 755 | static void wake_up_worker(struct worker_pool *pool) |
7e11629d | 756 | { |
63d95a91 | 757 | struct worker *worker = first_worker(pool); |
7e11629d TH |
758 | |
759 | if (likely(worker)) | |
760 | wake_up_process(worker->task); | |
761 | } | |
762 | ||
d302f017 | 763 | /** |
e22bee78 TH |
764 | * wq_worker_waking_up - a worker is waking up |
765 | * @task: task waking up | |
766 | * @cpu: CPU @task is waking up to | |
767 | * | |
768 | * This function is called during try_to_wake_up() when a worker is | |
769 | * being awoken. | |
770 | * | |
771 | * CONTEXT: | |
772 | * spin_lock_irq(rq->lock) | |
773 | */ | |
d84ff051 | 774 | void wq_worker_waking_up(struct task_struct *task, int cpu) |
e22bee78 TH |
775 | { |
776 | struct worker *worker = kthread_data(task); | |
777 | ||
36576000 | 778 | if (!(worker->flags & WORKER_NOT_RUNNING)) { |
ec22ca5e | 779 | WARN_ON_ONCE(worker->pool->cpu != cpu); |
e19e397a | 780 | atomic_inc(&worker->pool->nr_running); |
36576000 | 781 | } |
e22bee78 TH |
782 | } |
783 | ||
784 | /** | |
785 | * wq_worker_sleeping - a worker is going to sleep | |
786 | * @task: task going to sleep | |
787 | * @cpu: CPU in question, must be the current CPU number | |
788 | * | |
789 | * This function is called during schedule() when a busy worker is | |
790 | * going to sleep. Worker on the same cpu can be woken up by | |
791 | * returning pointer to its task. | |
792 | * | |
793 | * CONTEXT: | |
794 | * spin_lock_irq(rq->lock) | |
795 | * | |
796 | * RETURNS: | |
797 | * Worker task on @cpu to wake up, %NULL if none. | |
798 | */ | |
d84ff051 | 799 | struct task_struct *wq_worker_sleeping(struct task_struct *task, int cpu) |
e22bee78 TH |
800 | { |
801 | struct worker *worker = kthread_data(task), *to_wakeup = NULL; | |
111c225a | 802 | struct worker_pool *pool; |
e22bee78 | 803 | |
111c225a TH |
804 | /* |
805 | * Rescuers, which may not have all the fields set up like normal | |
806 | * workers, also reach here, let's not access anything before | |
807 | * checking NOT_RUNNING. | |
808 | */ | |
2d64672e | 809 | if (worker->flags & WORKER_NOT_RUNNING) |
e22bee78 TH |
810 | return NULL; |
811 | ||
111c225a | 812 | pool = worker->pool; |
111c225a | 813 | |
e22bee78 | 814 | /* this can only happen on the local cpu */ |
6183c009 TH |
815 | if (WARN_ON_ONCE(cpu != raw_smp_processor_id())) |
816 | return NULL; | |
e22bee78 TH |
817 | |
818 | /* | |
819 | * The counterpart of the following dec_and_test, implied mb, | |
820 | * worklist not empty test sequence is in insert_work(). | |
821 | * Please read comment there. | |
822 | * | |
628c78e7 TH |
823 | * NOT_RUNNING is clear. This means that we're bound to and |
824 | * running on the local cpu w/ rq lock held and preemption | |
825 | * disabled, which in turn means that none else could be | |
d565ed63 | 826 | * manipulating idle_list, so dereferencing idle_list without pool |
628c78e7 | 827 | * lock is safe. |
e22bee78 | 828 | */ |
e19e397a TH |
829 | if (atomic_dec_and_test(&pool->nr_running) && |
830 | !list_empty(&pool->worklist)) | |
63d95a91 | 831 | to_wakeup = first_worker(pool); |
e22bee78 TH |
832 | return to_wakeup ? to_wakeup->task : NULL; |
833 | } | |
834 | ||
835 | /** | |
836 | * worker_set_flags - set worker flags and adjust nr_running accordingly | |
cb444766 | 837 | * @worker: self |
d302f017 TH |
838 | * @flags: flags to set |
839 | * @wakeup: wakeup an idle worker if necessary | |
840 | * | |
e22bee78 TH |
841 | * Set @flags in @worker->flags and adjust nr_running accordingly. If |
842 | * nr_running becomes zero and @wakeup is %true, an idle worker is | |
843 | * woken up. | |
d302f017 | 844 | * |
cb444766 | 845 | * CONTEXT: |
d565ed63 | 846 | * spin_lock_irq(pool->lock) |
d302f017 TH |
847 | */ |
848 | static inline void worker_set_flags(struct worker *worker, unsigned int flags, | |
849 | bool wakeup) | |
850 | { | |
bd7bdd43 | 851 | struct worker_pool *pool = worker->pool; |
e22bee78 | 852 | |
cb444766 TH |
853 | WARN_ON_ONCE(worker->task != current); |
854 | ||
e22bee78 TH |
855 | /* |
856 | * If transitioning into NOT_RUNNING, adjust nr_running and | |
857 | * wake up an idle worker as necessary if requested by | |
858 | * @wakeup. | |
859 | */ | |
860 | if ((flags & WORKER_NOT_RUNNING) && | |
861 | !(worker->flags & WORKER_NOT_RUNNING)) { | |
e22bee78 | 862 | if (wakeup) { |
e19e397a | 863 | if (atomic_dec_and_test(&pool->nr_running) && |
bd7bdd43 | 864 | !list_empty(&pool->worklist)) |
63d95a91 | 865 | wake_up_worker(pool); |
e22bee78 | 866 | } else |
e19e397a | 867 | atomic_dec(&pool->nr_running); |
e22bee78 TH |
868 | } |
869 | ||
d302f017 TH |
870 | worker->flags |= flags; |
871 | } | |
872 | ||
873 | /** | |
e22bee78 | 874 | * worker_clr_flags - clear worker flags and adjust nr_running accordingly |
cb444766 | 875 | * @worker: self |
d302f017 TH |
876 | * @flags: flags to clear |
877 | * | |
e22bee78 | 878 | * Clear @flags in @worker->flags and adjust nr_running accordingly. |
d302f017 | 879 | * |
cb444766 | 880 | * CONTEXT: |
d565ed63 | 881 | * spin_lock_irq(pool->lock) |
d302f017 TH |
882 | */ |
883 | static inline void worker_clr_flags(struct worker *worker, unsigned int flags) | |
884 | { | |
63d95a91 | 885 | struct worker_pool *pool = worker->pool; |
e22bee78 TH |
886 | unsigned int oflags = worker->flags; |
887 | ||
cb444766 TH |
888 | WARN_ON_ONCE(worker->task != current); |
889 | ||
d302f017 | 890 | worker->flags &= ~flags; |
e22bee78 | 891 | |
42c025f3 TH |
892 | /* |
893 | * If transitioning out of NOT_RUNNING, increment nr_running. Note | |
894 | * that the nested NOT_RUNNING is not a noop. NOT_RUNNING is mask | |
895 | * of multiple flags, not a single flag. | |
896 | */ | |
e22bee78 TH |
897 | if ((flags & WORKER_NOT_RUNNING) && (oflags & WORKER_NOT_RUNNING)) |
898 | if (!(worker->flags & WORKER_NOT_RUNNING)) | |
e19e397a | 899 | atomic_inc(&pool->nr_running); |
d302f017 TH |
900 | } |
901 | ||
8cca0eea TH |
902 | /** |
903 | * find_worker_executing_work - find worker which is executing a work | |
c9e7cf27 | 904 | * @pool: pool of interest |
8cca0eea TH |
905 | * @work: work to find worker for |
906 | * | |
c9e7cf27 TH |
907 | * Find a worker which is executing @work on @pool by searching |
908 | * @pool->busy_hash which is keyed by the address of @work. For a worker | |
a2c1c57b TH |
909 | * to match, its current execution should match the address of @work and |
910 | * its work function. This is to avoid unwanted dependency between | |
911 | * unrelated work executions through a work item being recycled while still | |
912 | * being executed. | |
913 | * | |
914 | * This is a bit tricky. A work item may be freed once its execution | |
915 | * starts and nothing prevents the freed area from being recycled for | |
916 | * another work item. If the same work item address ends up being reused | |
917 | * before the original execution finishes, workqueue will identify the | |
918 | * recycled work item as currently executing and make it wait until the | |
919 | * current execution finishes, introducing an unwanted dependency. | |
920 | * | |
c5aa87bb TH |
921 | * This function checks the work item address and work function to avoid |
922 | * false positives. Note that this isn't complete as one may construct a | |
923 | * work function which can introduce dependency onto itself through a | |
924 | * recycled work item. Well, if somebody wants to shoot oneself in the | |
925 | * foot that badly, there's only so much we can do, and if such deadlock | |
926 | * actually occurs, it should be easy to locate the culprit work function. | |
8cca0eea TH |
927 | * |
928 | * CONTEXT: | |
d565ed63 | 929 | * spin_lock_irq(pool->lock). |
8cca0eea TH |
930 | * |
931 | * RETURNS: | |
932 | * Pointer to worker which is executing @work if found, NULL | |
933 | * otherwise. | |
4d707b9f | 934 | */ |
c9e7cf27 | 935 | static struct worker *find_worker_executing_work(struct worker_pool *pool, |
8cca0eea | 936 | struct work_struct *work) |
4d707b9f | 937 | { |
42f8570f | 938 | struct worker *worker; |
42f8570f | 939 | |
b67bfe0d | 940 | hash_for_each_possible(pool->busy_hash, worker, hentry, |
a2c1c57b TH |
941 | (unsigned long)work) |
942 | if (worker->current_work == work && | |
943 | worker->current_func == work->func) | |
42f8570f SL |
944 | return worker; |
945 | ||
946 | return NULL; | |
4d707b9f ON |
947 | } |
948 | ||
bf4ede01 TH |
949 | /** |
950 | * move_linked_works - move linked works to a list | |
951 | * @work: start of series of works to be scheduled | |
952 | * @head: target list to append @work to | |
953 | * @nextp: out paramter for nested worklist walking | |
954 | * | |
955 | * Schedule linked works starting from @work to @head. Work series to | |
956 | * be scheduled starts at @work and includes any consecutive work with | |
957 | * WORK_STRUCT_LINKED set in its predecessor. | |
958 | * | |
959 | * If @nextp is not NULL, it's updated to point to the next work of | |
960 | * the last scheduled work. This allows move_linked_works() to be | |
961 | * nested inside outer list_for_each_entry_safe(). | |
962 | * | |
963 | * CONTEXT: | |
d565ed63 | 964 | * spin_lock_irq(pool->lock). |
bf4ede01 TH |
965 | */ |
966 | static void move_linked_works(struct work_struct *work, struct list_head *head, | |
967 | struct work_struct **nextp) | |
968 | { | |
969 | struct work_struct *n; | |
970 | ||
971 | /* | |
972 | * Linked worklist will always end before the end of the list, | |
973 | * use NULL for list head. | |
974 | */ | |
975 | list_for_each_entry_safe_from(work, n, NULL, entry) { | |
976 | list_move_tail(&work->entry, head); | |
977 | if (!(*work_data_bits(work) & WORK_STRUCT_LINKED)) | |
978 | break; | |
979 | } | |
980 | ||
981 | /* | |
982 | * If we're already inside safe list traversal and have moved | |
983 | * multiple works to the scheduled queue, the next position | |
984 | * needs to be updated. | |
985 | */ | |
986 | if (nextp) | |
987 | *nextp = n; | |
988 | } | |
989 | ||
8864b4e5 TH |
990 | /** |
991 | * get_pwq - get an extra reference on the specified pool_workqueue | |
992 | * @pwq: pool_workqueue to get | |
993 | * | |
994 | * Obtain an extra reference on @pwq. The caller should guarantee that | |
995 | * @pwq has positive refcnt and be holding the matching pool->lock. | |
996 | */ | |
997 | static void get_pwq(struct pool_workqueue *pwq) | |
998 | { | |
999 | lockdep_assert_held(&pwq->pool->lock); | |
1000 | WARN_ON_ONCE(pwq->refcnt <= 0); | |
1001 | pwq->refcnt++; | |
1002 | } | |
1003 | ||
1004 | /** | |
1005 | * put_pwq - put a pool_workqueue reference | |
1006 | * @pwq: pool_workqueue to put | |
1007 | * | |
1008 | * Drop a reference of @pwq. If its refcnt reaches zero, schedule its | |
1009 | * destruction. The caller should be holding the matching pool->lock. | |
1010 | */ | |
1011 | static void put_pwq(struct pool_workqueue *pwq) | |
1012 | { | |
1013 | lockdep_assert_held(&pwq->pool->lock); | |
1014 | if (likely(--pwq->refcnt)) | |
1015 | return; | |
1016 | if (WARN_ON_ONCE(!(pwq->wq->flags & WQ_UNBOUND))) | |
1017 | return; | |
1018 | /* | |
1019 | * @pwq can't be released under pool->lock, bounce to | |
1020 | * pwq_unbound_release_workfn(). This never recurses on the same | |
1021 | * pool->lock as this path is taken only for unbound workqueues and | |
1022 | * the release work item is scheduled on a per-cpu workqueue. To | |
1023 | * avoid lockdep warning, unbound pool->locks are given lockdep | |
1024 | * subclass of 1 in get_unbound_pool(). | |
1025 | */ | |
1026 | schedule_work(&pwq->unbound_release_work); | |
1027 | } | |
1028 | ||
112202d9 | 1029 | static void pwq_activate_delayed_work(struct work_struct *work) |
bf4ede01 | 1030 | { |
112202d9 | 1031 | struct pool_workqueue *pwq = get_work_pwq(work); |
bf4ede01 TH |
1032 | |
1033 | trace_workqueue_activate_work(work); | |
112202d9 | 1034 | move_linked_works(work, &pwq->pool->worklist, NULL); |
bf4ede01 | 1035 | __clear_bit(WORK_STRUCT_DELAYED_BIT, work_data_bits(work)); |
112202d9 | 1036 | pwq->nr_active++; |
bf4ede01 TH |
1037 | } |
1038 | ||
112202d9 | 1039 | static void pwq_activate_first_delayed(struct pool_workqueue *pwq) |
3aa62497 | 1040 | { |
112202d9 | 1041 | struct work_struct *work = list_first_entry(&pwq->delayed_works, |
3aa62497 LJ |
1042 | struct work_struct, entry); |
1043 | ||
112202d9 | 1044 | pwq_activate_delayed_work(work); |
3aa62497 LJ |
1045 | } |
1046 | ||
bf4ede01 | 1047 | /** |
112202d9 TH |
1048 | * pwq_dec_nr_in_flight - decrement pwq's nr_in_flight |
1049 | * @pwq: pwq of interest | |
bf4ede01 | 1050 | * @color: color of work which left the queue |
bf4ede01 TH |
1051 | * |
1052 | * A work either has completed or is removed from pending queue, | |
112202d9 | 1053 | * decrement nr_in_flight of its pwq and handle workqueue flushing. |
bf4ede01 TH |
1054 | * |
1055 | * CONTEXT: | |
d565ed63 | 1056 | * spin_lock_irq(pool->lock). |
bf4ede01 | 1057 | */ |
112202d9 | 1058 | static void pwq_dec_nr_in_flight(struct pool_workqueue *pwq, int color) |
bf4ede01 | 1059 | { |
8864b4e5 | 1060 | /* uncolored work items don't participate in flushing or nr_active */ |
bf4ede01 | 1061 | if (color == WORK_NO_COLOR) |
8864b4e5 | 1062 | goto out_put; |
bf4ede01 | 1063 | |
112202d9 | 1064 | pwq->nr_in_flight[color]--; |
bf4ede01 | 1065 | |
112202d9 TH |
1066 | pwq->nr_active--; |
1067 | if (!list_empty(&pwq->delayed_works)) { | |
b3f9f405 | 1068 | /* one down, submit a delayed one */ |
112202d9 TH |
1069 | if (pwq->nr_active < pwq->max_active) |
1070 | pwq_activate_first_delayed(pwq); | |
bf4ede01 TH |
1071 | } |
1072 | ||
1073 | /* is flush in progress and are we at the flushing tip? */ | |
112202d9 | 1074 | if (likely(pwq->flush_color != color)) |
8864b4e5 | 1075 | goto out_put; |
bf4ede01 TH |
1076 | |
1077 | /* are there still in-flight works? */ | |
112202d9 | 1078 | if (pwq->nr_in_flight[color]) |
8864b4e5 | 1079 | goto out_put; |
bf4ede01 | 1080 | |
112202d9 TH |
1081 | /* this pwq is done, clear flush_color */ |
1082 | pwq->flush_color = -1; | |
bf4ede01 TH |
1083 | |
1084 | /* | |
112202d9 | 1085 | * If this was the last pwq, wake up the first flusher. It |
bf4ede01 TH |
1086 | * will handle the rest. |
1087 | */ | |
112202d9 TH |
1088 | if (atomic_dec_and_test(&pwq->wq->nr_pwqs_to_flush)) |
1089 | complete(&pwq->wq->first_flusher->done); | |
8864b4e5 TH |
1090 | out_put: |
1091 | put_pwq(pwq); | |
bf4ede01 TH |
1092 | } |
1093 | ||
36e227d2 | 1094 | /** |
bbb68dfa | 1095 | * try_to_grab_pending - steal work item from worklist and disable irq |
36e227d2 TH |
1096 | * @work: work item to steal |
1097 | * @is_dwork: @work is a delayed_work | |
bbb68dfa | 1098 | * @flags: place to store irq state |
36e227d2 TH |
1099 | * |
1100 | * Try to grab PENDING bit of @work. This function can handle @work in any | |
1101 | * stable state - idle, on timer or on worklist. Return values are | |
1102 | * | |
1103 | * 1 if @work was pending and we successfully stole PENDING | |
1104 | * 0 if @work was idle and we claimed PENDING | |
1105 | * -EAGAIN if PENDING couldn't be grabbed at the moment, safe to busy-retry | |
bbb68dfa TH |
1106 | * -ENOENT if someone else is canceling @work, this state may persist |
1107 | * for arbitrarily long | |
36e227d2 | 1108 | * |
bbb68dfa | 1109 | * On >= 0 return, the caller owns @work's PENDING bit. To avoid getting |
e0aecdd8 TH |
1110 | * interrupted while holding PENDING and @work off queue, irq must be |
1111 | * disabled on entry. This, combined with delayed_work->timer being | |
1112 | * irqsafe, ensures that we return -EAGAIN for finite short period of time. | |
bbb68dfa TH |
1113 | * |
1114 | * On successful return, >= 0, irq is disabled and the caller is | |
1115 | * responsible for releasing it using local_irq_restore(*@flags). | |
1116 | * | |
e0aecdd8 | 1117 | * This function is safe to call from any context including IRQ handler. |
bf4ede01 | 1118 | */ |
bbb68dfa TH |
1119 | static int try_to_grab_pending(struct work_struct *work, bool is_dwork, |
1120 | unsigned long *flags) | |
bf4ede01 | 1121 | { |
d565ed63 | 1122 | struct worker_pool *pool; |
112202d9 | 1123 | struct pool_workqueue *pwq; |
bf4ede01 | 1124 | |
bbb68dfa TH |
1125 | local_irq_save(*flags); |
1126 | ||
36e227d2 TH |
1127 | /* try to steal the timer if it exists */ |
1128 | if (is_dwork) { | |
1129 | struct delayed_work *dwork = to_delayed_work(work); | |
1130 | ||
e0aecdd8 TH |
1131 | /* |
1132 | * dwork->timer is irqsafe. If del_timer() fails, it's | |
1133 | * guaranteed that the timer is not queued anywhere and not | |
1134 | * running on the local CPU. | |
1135 | */ | |
36e227d2 TH |
1136 | if (likely(del_timer(&dwork->timer))) |
1137 | return 1; | |
1138 | } | |
1139 | ||
1140 | /* try to claim PENDING the normal way */ | |
bf4ede01 TH |
1141 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) |
1142 | return 0; | |
1143 | ||
1144 | /* | |
1145 | * The queueing is in progress, or it is already queued. Try to | |
1146 | * steal it from ->worklist without clearing WORK_STRUCT_PENDING. | |
1147 | */ | |
d565ed63 TH |
1148 | pool = get_work_pool(work); |
1149 | if (!pool) | |
bbb68dfa | 1150 | goto fail; |
bf4ede01 | 1151 | |
d565ed63 | 1152 | spin_lock(&pool->lock); |
0b3dae68 | 1153 | /* |
112202d9 TH |
1154 | * work->data is guaranteed to point to pwq only while the work |
1155 | * item is queued on pwq->wq, and both updating work->data to point | |
1156 | * to pwq on queueing and to pool on dequeueing are done under | |
1157 | * pwq->pool->lock. This in turn guarantees that, if work->data | |
1158 | * points to pwq which is associated with a locked pool, the work | |
0b3dae68 LJ |
1159 | * item is currently queued on that pool. |
1160 | */ | |
112202d9 TH |
1161 | pwq = get_work_pwq(work); |
1162 | if (pwq && pwq->pool == pool) { | |
16062836 TH |
1163 | debug_work_deactivate(work); |
1164 | ||
1165 | /* | |
1166 | * A delayed work item cannot be grabbed directly because | |
1167 | * it might have linked NO_COLOR work items which, if left | |
112202d9 | 1168 | * on the delayed_list, will confuse pwq->nr_active |
16062836 TH |
1169 | * management later on and cause stall. Make sure the work |
1170 | * item is activated before grabbing. | |
1171 | */ | |
1172 | if (*work_data_bits(work) & WORK_STRUCT_DELAYED) | |
112202d9 | 1173 | pwq_activate_delayed_work(work); |
16062836 TH |
1174 | |
1175 | list_del_init(&work->entry); | |
112202d9 | 1176 | pwq_dec_nr_in_flight(get_work_pwq(work), get_work_color(work)); |
16062836 | 1177 | |
112202d9 | 1178 | /* work->data points to pwq iff queued, point to pool */ |
16062836 TH |
1179 | set_work_pool_and_keep_pending(work, pool->id); |
1180 | ||
1181 | spin_unlock(&pool->lock); | |
1182 | return 1; | |
bf4ede01 | 1183 | } |
d565ed63 | 1184 | spin_unlock(&pool->lock); |
bbb68dfa TH |
1185 | fail: |
1186 | local_irq_restore(*flags); | |
1187 | if (work_is_canceling(work)) | |
1188 | return -ENOENT; | |
1189 | cpu_relax(); | |
36e227d2 | 1190 | return -EAGAIN; |
bf4ede01 TH |
1191 | } |
1192 | ||
4690c4ab | 1193 | /** |
706026c2 | 1194 | * insert_work - insert a work into a pool |
112202d9 | 1195 | * @pwq: pwq @work belongs to |
4690c4ab TH |
1196 | * @work: work to insert |
1197 | * @head: insertion point | |
1198 | * @extra_flags: extra WORK_STRUCT_* flags to set | |
1199 | * | |
112202d9 | 1200 | * Insert @work which belongs to @pwq after @head. @extra_flags is or'd to |
706026c2 | 1201 | * work_struct flags. |
4690c4ab TH |
1202 | * |
1203 | * CONTEXT: | |
d565ed63 | 1204 | * spin_lock_irq(pool->lock). |
4690c4ab | 1205 | */ |
112202d9 TH |
1206 | static void insert_work(struct pool_workqueue *pwq, struct work_struct *work, |
1207 | struct list_head *head, unsigned int extra_flags) | |
b89deed3 | 1208 | { |
112202d9 | 1209 | struct worker_pool *pool = pwq->pool; |
e22bee78 | 1210 | |
4690c4ab | 1211 | /* we own @work, set data and link */ |
112202d9 | 1212 | set_work_pwq(work, pwq, extra_flags); |
1a4d9b0a | 1213 | list_add_tail(&work->entry, head); |
8864b4e5 | 1214 | get_pwq(pwq); |
e22bee78 TH |
1215 | |
1216 | /* | |
c5aa87bb TH |
1217 | * Ensure either wq_worker_sleeping() sees the above |
1218 | * list_add_tail() or we see zero nr_running to avoid workers lying | |
1219 | * around lazily while there are works to be processed. | |
e22bee78 TH |
1220 | */ |
1221 | smp_mb(); | |
1222 | ||
63d95a91 TH |
1223 | if (__need_more_worker(pool)) |
1224 | wake_up_worker(pool); | |
b89deed3 ON |
1225 | } |
1226 | ||
c8efcc25 TH |
1227 | /* |
1228 | * Test whether @work is being queued from another work executing on the | |
8d03ecfe | 1229 | * same workqueue. |
c8efcc25 TH |
1230 | */ |
1231 | static bool is_chained_work(struct workqueue_struct *wq) | |
1232 | { | |
8d03ecfe TH |
1233 | struct worker *worker; |
1234 | ||
1235 | worker = current_wq_worker(); | |
1236 | /* | |
1237 | * Return %true iff I'm a worker execuing a work item on @wq. If | |
1238 | * I'm @worker, it's safe to dereference it without locking. | |
1239 | */ | |
112202d9 | 1240 | return worker && worker->current_pwq->wq == wq; |
c8efcc25 TH |
1241 | } |
1242 | ||
d84ff051 | 1243 | static void __queue_work(int cpu, struct workqueue_struct *wq, |
1da177e4 LT |
1244 | struct work_struct *work) |
1245 | { | |
112202d9 | 1246 | struct pool_workqueue *pwq; |
c9178087 | 1247 | struct worker_pool *last_pool; |
1e19ffc6 | 1248 | struct list_head *worklist; |
8a2e8e5d | 1249 | unsigned int work_flags; |
b75cac93 | 1250 | unsigned int req_cpu = cpu; |
8930caba TH |
1251 | |
1252 | /* | |
1253 | * While a work item is PENDING && off queue, a task trying to | |
1254 | * steal the PENDING will busy-loop waiting for it to either get | |
1255 | * queued or lose PENDING. Grabbing PENDING and queueing should | |
1256 | * happen with IRQ disabled. | |
1257 | */ | |
1258 | WARN_ON_ONCE(!irqs_disabled()); | |
1da177e4 | 1259 | |
dc186ad7 | 1260 | debug_work_activate(work); |
1e19ffc6 | 1261 | |
c8efcc25 | 1262 | /* if dying, only works from the same workqueue are allowed */ |
618b01eb | 1263 | if (unlikely(wq->flags & __WQ_DRAINING) && |
c8efcc25 | 1264 | WARN_ON_ONCE(!is_chained_work(wq))) |
e41e704b | 1265 | return; |
9e8cd2f5 | 1266 | retry: |
c9178087 | 1267 | /* pwq which will be used unless @work is executing elsewhere */ |
c7fc77f7 | 1268 | if (!(wq->flags & WQ_UNBOUND)) { |
57469821 | 1269 | if (cpu == WORK_CPU_UNBOUND) |
c7fc77f7 | 1270 | cpu = raw_smp_processor_id(); |
7fb98ea7 | 1271 | pwq = per_cpu_ptr(wq->cpu_pwqs, cpu); |
c9178087 TH |
1272 | } else { |
1273 | pwq = first_pwq(wq); | |
1274 | } | |
dbf2576e | 1275 | |
c9178087 TH |
1276 | /* |
1277 | * If @work was previously on a different pool, it might still be | |
1278 | * running there, in which case the work needs to be queued on that | |
1279 | * pool to guarantee non-reentrancy. | |
1280 | */ | |
1281 | last_pool = get_work_pool(work); | |
1282 | if (last_pool && last_pool != pwq->pool) { | |
1283 | struct worker *worker; | |
18aa9eff | 1284 | |
c9178087 | 1285 | spin_lock(&last_pool->lock); |
18aa9eff | 1286 | |
c9178087 | 1287 | worker = find_worker_executing_work(last_pool, work); |
18aa9eff | 1288 | |
c9178087 TH |
1289 | if (worker && worker->current_pwq->wq == wq) { |
1290 | pwq = worker->current_pwq; | |
8930caba | 1291 | } else { |
c9178087 TH |
1292 | /* meh... not running there, queue here */ |
1293 | spin_unlock(&last_pool->lock); | |
112202d9 | 1294 | spin_lock(&pwq->pool->lock); |
8930caba | 1295 | } |
f3421797 | 1296 | } else { |
112202d9 | 1297 | spin_lock(&pwq->pool->lock); |
502ca9d8 TH |
1298 | } |
1299 | ||
9e8cd2f5 TH |
1300 | /* |
1301 | * pwq is determined and locked. For unbound pools, we could have | |
1302 | * raced with pwq release and it could already be dead. If its | |
1303 | * refcnt is zero, repeat pwq selection. Note that pwqs never die | |
1304 | * without another pwq replacing it as the first pwq or while a | |
1305 | * work item is executing on it, so the retying is guaranteed to | |
1306 | * make forward-progress. | |
1307 | */ | |
1308 | if (unlikely(!pwq->refcnt)) { | |
1309 | if (wq->flags & WQ_UNBOUND) { | |
1310 | spin_unlock(&pwq->pool->lock); | |
1311 | cpu_relax(); | |
1312 | goto retry; | |
1313 | } | |
1314 | /* oops */ | |
1315 | WARN_ONCE(true, "workqueue: per-cpu pwq for %s on cpu%d has 0 refcnt", | |
1316 | wq->name, cpu); | |
1317 | } | |
1318 | ||
112202d9 TH |
1319 | /* pwq determined, queue */ |
1320 | trace_workqueue_queue_work(req_cpu, pwq, work); | |
502ca9d8 | 1321 | |
f5b2552b | 1322 | if (WARN_ON(!list_empty(&work->entry))) { |
112202d9 | 1323 | spin_unlock(&pwq->pool->lock); |
f5b2552b DC |
1324 | return; |
1325 | } | |
1e19ffc6 | 1326 | |
112202d9 TH |
1327 | pwq->nr_in_flight[pwq->work_color]++; |
1328 | work_flags = work_color_to_flags(pwq->work_color); | |
1e19ffc6 | 1329 | |
112202d9 | 1330 | if (likely(pwq->nr_active < pwq->max_active)) { |
cdadf009 | 1331 | trace_workqueue_activate_work(work); |
112202d9 TH |
1332 | pwq->nr_active++; |
1333 | worklist = &pwq->pool->worklist; | |
8a2e8e5d TH |
1334 | } else { |
1335 | work_flags |= WORK_STRUCT_DELAYED; | |
112202d9 | 1336 | worklist = &pwq->delayed_works; |
8a2e8e5d | 1337 | } |
1e19ffc6 | 1338 | |
112202d9 | 1339 | insert_work(pwq, work, worklist, work_flags); |
1e19ffc6 | 1340 | |
112202d9 | 1341 | spin_unlock(&pwq->pool->lock); |
1da177e4 LT |
1342 | } |
1343 | ||
0fcb78c2 | 1344 | /** |
c1a220e7 ZR |
1345 | * queue_work_on - queue work on specific cpu |
1346 | * @cpu: CPU number to execute work on | |
0fcb78c2 REB |
1347 | * @wq: workqueue to use |
1348 | * @work: work to queue | |
1349 | * | |
d4283e93 | 1350 | * Returns %false if @work was already on a queue, %true otherwise. |
1da177e4 | 1351 | * |
c1a220e7 ZR |
1352 | * We queue the work to a specific CPU, the caller must ensure it |
1353 | * can't go away. | |
1da177e4 | 1354 | */ |
d4283e93 TH |
1355 | bool queue_work_on(int cpu, struct workqueue_struct *wq, |
1356 | struct work_struct *work) | |
1da177e4 | 1357 | { |
d4283e93 | 1358 | bool ret = false; |
8930caba | 1359 | unsigned long flags; |
ef1ca236 | 1360 | |
8930caba | 1361 | local_irq_save(flags); |
c1a220e7 | 1362 | |
22df02bb | 1363 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
4690c4ab | 1364 | __queue_work(cpu, wq, work); |
d4283e93 | 1365 | ret = true; |
c1a220e7 | 1366 | } |
ef1ca236 | 1367 | |
8930caba | 1368 | local_irq_restore(flags); |
1da177e4 LT |
1369 | return ret; |
1370 | } | |
c1a220e7 | 1371 | EXPORT_SYMBOL_GPL(queue_work_on); |
1da177e4 | 1372 | |
d8e794df | 1373 | void delayed_work_timer_fn(unsigned long __data) |
1da177e4 | 1374 | { |
52bad64d | 1375 | struct delayed_work *dwork = (struct delayed_work *)__data; |
1da177e4 | 1376 | |
e0aecdd8 | 1377 | /* should have been called from irqsafe timer with irq already off */ |
60c057bc | 1378 | __queue_work(dwork->cpu, dwork->wq, &dwork->work); |
1da177e4 | 1379 | } |
1438ade5 | 1380 | EXPORT_SYMBOL(delayed_work_timer_fn); |
1da177e4 | 1381 | |
7beb2edf TH |
1382 | static void __queue_delayed_work(int cpu, struct workqueue_struct *wq, |
1383 | struct delayed_work *dwork, unsigned long delay) | |
1da177e4 | 1384 | { |
7beb2edf TH |
1385 | struct timer_list *timer = &dwork->timer; |
1386 | struct work_struct *work = &dwork->work; | |
7beb2edf TH |
1387 | |
1388 | WARN_ON_ONCE(timer->function != delayed_work_timer_fn || | |
1389 | timer->data != (unsigned long)dwork); | |
fc4b514f TH |
1390 | WARN_ON_ONCE(timer_pending(timer)); |
1391 | WARN_ON_ONCE(!list_empty(&work->entry)); | |
7beb2edf | 1392 | |
8852aac2 TH |
1393 | /* |
1394 | * If @delay is 0, queue @dwork->work immediately. This is for | |
1395 | * both optimization and correctness. The earliest @timer can | |
1396 | * expire is on the closest next tick and delayed_work users depend | |
1397 | * on that there's no such delay when @delay is 0. | |
1398 | */ | |
1399 | if (!delay) { | |
1400 | __queue_work(cpu, wq, &dwork->work); | |
1401 | return; | |
1402 | } | |
1403 | ||
7beb2edf | 1404 | timer_stats_timer_set_start_info(&dwork->timer); |
1da177e4 | 1405 | |
60c057bc | 1406 | dwork->wq = wq; |
1265057f | 1407 | dwork->cpu = cpu; |
7beb2edf TH |
1408 | timer->expires = jiffies + delay; |
1409 | ||
1410 | if (unlikely(cpu != WORK_CPU_UNBOUND)) | |
1411 | add_timer_on(timer, cpu); | |
1412 | else | |
1413 | add_timer(timer); | |
1da177e4 LT |
1414 | } |
1415 | ||
0fcb78c2 REB |
1416 | /** |
1417 | * queue_delayed_work_on - queue work on specific CPU after delay | |
1418 | * @cpu: CPU number to execute work on | |
1419 | * @wq: workqueue to use | |
af9997e4 | 1420 | * @dwork: work to queue |
0fcb78c2 REB |
1421 | * @delay: number of jiffies to wait before queueing |
1422 | * | |
715f1300 TH |
1423 | * Returns %false if @work was already on a queue, %true otherwise. If |
1424 | * @delay is zero and @dwork is idle, it will be scheduled for immediate | |
1425 | * execution. | |
0fcb78c2 | 1426 | */ |
d4283e93 TH |
1427 | bool queue_delayed_work_on(int cpu, struct workqueue_struct *wq, |
1428 | struct delayed_work *dwork, unsigned long delay) | |
7a6bc1cd | 1429 | { |
52bad64d | 1430 | struct work_struct *work = &dwork->work; |
d4283e93 | 1431 | bool ret = false; |
8930caba | 1432 | unsigned long flags; |
7a6bc1cd | 1433 | |
8930caba TH |
1434 | /* read the comment in __queue_work() */ |
1435 | local_irq_save(flags); | |
7a6bc1cd | 1436 | |
22df02bb | 1437 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
7beb2edf | 1438 | __queue_delayed_work(cpu, wq, dwork, delay); |
d4283e93 | 1439 | ret = true; |
7a6bc1cd | 1440 | } |
8a3e77cc | 1441 | |
8930caba | 1442 | local_irq_restore(flags); |
7a6bc1cd VP |
1443 | return ret; |
1444 | } | |
ae90dd5d | 1445 | EXPORT_SYMBOL_GPL(queue_delayed_work_on); |
c7fc77f7 | 1446 | |
8376fe22 TH |
1447 | /** |
1448 | * mod_delayed_work_on - modify delay of or queue a delayed work on specific CPU | |
1449 | * @cpu: CPU number to execute work on | |
1450 | * @wq: workqueue to use | |
1451 | * @dwork: work to queue | |
1452 | * @delay: number of jiffies to wait before queueing | |
1453 | * | |
1454 | * If @dwork is idle, equivalent to queue_delayed_work_on(); otherwise, | |
1455 | * modify @dwork's timer so that it expires after @delay. If @delay is | |
1456 | * zero, @work is guaranteed to be scheduled immediately regardless of its | |
1457 | * current state. | |
1458 | * | |
1459 | * Returns %false if @dwork was idle and queued, %true if @dwork was | |
1460 | * pending and its timer was modified. | |
1461 | * | |
e0aecdd8 | 1462 | * This function is safe to call from any context including IRQ handler. |
8376fe22 TH |
1463 | * See try_to_grab_pending() for details. |
1464 | */ | |
1465 | bool mod_delayed_work_on(int cpu, struct workqueue_struct *wq, | |
1466 | struct delayed_work *dwork, unsigned long delay) | |
1467 | { | |
1468 | unsigned long flags; | |
1469 | int ret; | |
c7fc77f7 | 1470 | |
8376fe22 TH |
1471 | do { |
1472 | ret = try_to_grab_pending(&dwork->work, true, &flags); | |
1473 | } while (unlikely(ret == -EAGAIN)); | |
63bc0362 | 1474 | |
8376fe22 TH |
1475 | if (likely(ret >= 0)) { |
1476 | __queue_delayed_work(cpu, wq, dwork, delay); | |
1477 | local_irq_restore(flags); | |
7a6bc1cd | 1478 | } |
8376fe22 TH |
1479 | |
1480 | /* -ENOENT from try_to_grab_pending() becomes %true */ | |
7a6bc1cd VP |
1481 | return ret; |
1482 | } | |
8376fe22 TH |
1483 | EXPORT_SYMBOL_GPL(mod_delayed_work_on); |
1484 | ||
c8e55f36 TH |
1485 | /** |
1486 | * worker_enter_idle - enter idle state | |
1487 | * @worker: worker which is entering idle state | |
1488 | * | |
1489 | * @worker is entering idle state. Update stats and idle timer if | |
1490 | * necessary. | |
1491 | * | |
1492 | * LOCKING: | |
d565ed63 | 1493 | * spin_lock_irq(pool->lock). |
c8e55f36 TH |
1494 | */ |
1495 | static void worker_enter_idle(struct worker *worker) | |
1da177e4 | 1496 | { |
bd7bdd43 | 1497 | struct worker_pool *pool = worker->pool; |
c8e55f36 | 1498 | |
6183c009 TH |
1499 | if (WARN_ON_ONCE(worker->flags & WORKER_IDLE) || |
1500 | WARN_ON_ONCE(!list_empty(&worker->entry) && | |
1501 | (worker->hentry.next || worker->hentry.pprev))) | |
1502 | return; | |
c8e55f36 | 1503 | |
cb444766 TH |
1504 | /* can't use worker_set_flags(), also called from start_worker() */ |
1505 | worker->flags |= WORKER_IDLE; | |
bd7bdd43 | 1506 | pool->nr_idle++; |
e22bee78 | 1507 | worker->last_active = jiffies; |
c8e55f36 TH |
1508 | |
1509 | /* idle_list is LIFO */ | |
bd7bdd43 | 1510 | list_add(&worker->entry, &pool->idle_list); |
db7bccf4 | 1511 | |
628c78e7 TH |
1512 | if (too_many_workers(pool) && !timer_pending(&pool->idle_timer)) |
1513 | mod_timer(&pool->idle_timer, jiffies + IDLE_WORKER_TIMEOUT); | |
cb444766 | 1514 | |
544ecf31 | 1515 | /* |
706026c2 | 1516 | * Sanity check nr_running. Because wq_unbind_fn() releases |
d565ed63 | 1517 | * pool->lock between setting %WORKER_UNBOUND and zapping |
628c78e7 TH |
1518 | * nr_running, the warning may trigger spuriously. Check iff |
1519 | * unbind is not in progress. | |
544ecf31 | 1520 | */ |
24647570 | 1521 | WARN_ON_ONCE(!(pool->flags & POOL_DISASSOCIATED) && |
bd7bdd43 | 1522 | pool->nr_workers == pool->nr_idle && |
e19e397a | 1523 | atomic_read(&pool->nr_running)); |
c8e55f36 TH |
1524 | } |
1525 | ||
1526 | /** | |
1527 | * worker_leave_idle - leave idle state | |
1528 | * @worker: worker which is leaving idle state | |
1529 | * | |
1530 | * @worker is leaving idle state. Update stats. | |
1531 | * | |
1532 | * LOCKING: | |
d565ed63 | 1533 | * spin_lock_irq(pool->lock). |
c8e55f36 TH |
1534 | */ |
1535 | static void worker_leave_idle(struct worker *worker) | |
1536 | { | |
bd7bdd43 | 1537 | struct worker_pool *pool = worker->pool; |
c8e55f36 | 1538 | |
6183c009 TH |
1539 | if (WARN_ON_ONCE(!(worker->flags & WORKER_IDLE))) |
1540 | return; | |
d302f017 | 1541 | worker_clr_flags(worker, WORKER_IDLE); |
bd7bdd43 | 1542 | pool->nr_idle--; |
c8e55f36 TH |
1543 | list_del_init(&worker->entry); |
1544 | } | |
1545 | ||
e22bee78 | 1546 | /** |
f36dc67b LJ |
1547 | * worker_maybe_bind_and_lock - try to bind %current to worker_pool and lock it |
1548 | * @pool: target worker_pool | |
1549 | * | |
1550 | * Bind %current to the cpu of @pool if it is associated and lock @pool. | |
e22bee78 TH |
1551 | * |
1552 | * Works which are scheduled while the cpu is online must at least be | |
1553 | * scheduled to a worker which is bound to the cpu so that if they are | |
1554 | * flushed from cpu callbacks while cpu is going down, they are | |
1555 | * guaranteed to execute on the cpu. | |
1556 | * | |
f5faa077 | 1557 | * This function is to be used by unbound workers and rescuers to bind |
e22bee78 TH |
1558 | * themselves to the target cpu and may race with cpu going down or |
1559 | * coming online. kthread_bind() can't be used because it may put the | |
1560 | * worker to already dead cpu and set_cpus_allowed_ptr() can't be used | |
706026c2 | 1561 | * verbatim as it's best effort and blocking and pool may be |
e22bee78 TH |
1562 | * [dis]associated in the meantime. |
1563 | * | |
706026c2 | 1564 | * This function tries set_cpus_allowed() and locks pool and verifies the |
24647570 | 1565 | * binding against %POOL_DISASSOCIATED which is set during |
f2d5a0ee TH |
1566 | * %CPU_DOWN_PREPARE and cleared during %CPU_ONLINE, so if the worker |
1567 | * enters idle state or fetches works without dropping lock, it can | |
1568 | * guarantee the scheduling requirement described in the first paragraph. | |
e22bee78 TH |
1569 | * |
1570 | * CONTEXT: | |
d565ed63 | 1571 | * Might sleep. Called without any lock but returns with pool->lock |
e22bee78 TH |
1572 | * held. |
1573 | * | |
1574 | * RETURNS: | |
706026c2 | 1575 | * %true if the associated pool is online (@worker is successfully |
e22bee78 TH |
1576 | * bound), %false if offline. |
1577 | */ | |
f36dc67b | 1578 | static bool worker_maybe_bind_and_lock(struct worker_pool *pool) |
d565ed63 | 1579 | __acquires(&pool->lock) |
e22bee78 | 1580 | { |
e22bee78 | 1581 | while (true) { |
4e6045f1 | 1582 | /* |
e22bee78 TH |
1583 | * The following call may fail, succeed or succeed |
1584 | * without actually migrating the task to the cpu if | |
1585 | * it races with cpu hotunplug operation. Verify | |
24647570 | 1586 | * against POOL_DISASSOCIATED. |
4e6045f1 | 1587 | */ |
24647570 | 1588 | if (!(pool->flags & POOL_DISASSOCIATED)) |
7a4e344c | 1589 | set_cpus_allowed_ptr(current, pool->attrs->cpumask); |
e22bee78 | 1590 | |
d565ed63 | 1591 | spin_lock_irq(&pool->lock); |
24647570 | 1592 | if (pool->flags & POOL_DISASSOCIATED) |
e22bee78 | 1593 | return false; |
f5faa077 | 1594 | if (task_cpu(current) == pool->cpu && |
7a4e344c | 1595 | cpumask_equal(¤t->cpus_allowed, pool->attrs->cpumask)) |
e22bee78 | 1596 | return true; |
d565ed63 | 1597 | spin_unlock_irq(&pool->lock); |
e22bee78 | 1598 | |
5035b20f TH |
1599 | /* |
1600 | * We've raced with CPU hot[un]plug. Give it a breather | |
1601 | * and retry migration. cond_resched() is required here; | |
1602 | * otherwise, we might deadlock against cpu_stop trying to | |
1603 | * bring down the CPU on non-preemptive kernel. | |
1604 | */ | |
e22bee78 | 1605 | cpu_relax(); |
5035b20f | 1606 | cond_resched(); |
e22bee78 TH |
1607 | } |
1608 | } | |
1609 | ||
c34056a3 TH |
1610 | static struct worker *alloc_worker(void) |
1611 | { | |
1612 | struct worker *worker; | |
1613 | ||
1614 | worker = kzalloc(sizeof(*worker), GFP_KERNEL); | |
c8e55f36 TH |
1615 | if (worker) { |
1616 | INIT_LIST_HEAD(&worker->entry); | |
affee4b2 | 1617 | INIT_LIST_HEAD(&worker->scheduled); |
e22bee78 TH |
1618 | /* on creation a worker is in !idle && prep state */ |
1619 | worker->flags = WORKER_PREP; | |
c8e55f36 | 1620 | } |
c34056a3 TH |
1621 | return worker; |
1622 | } | |
1623 | ||
1624 | /** | |
1625 | * create_worker - create a new workqueue worker | |
63d95a91 | 1626 | * @pool: pool the new worker will belong to |
c34056a3 | 1627 | * |
63d95a91 | 1628 | * Create a new worker which is bound to @pool. The returned worker |
c34056a3 TH |
1629 | * can be started by calling start_worker() or destroyed using |
1630 | * destroy_worker(). | |
1631 | * | |
1632 | * CONTEXT: | |
1633 | * Might sleep. Does GFP_KERNEL allocations. | |
1634 | * | |
1635 | * RETURNS: | |
1636 | * Pointer to the newly created worker. | |
1637 | */ | |
bc2ae0f5 | 1638 | static struct worker *create_worker(struct worker_pool *pool) |
c34056a3 | 1639 | { |
7a4e344c | 1640 | const char *pri = pool->attrs->nice < 0 ? "H" : ""; |
c34056a3 | 1641 | struct worker *worker = NULL; |
f3421797 | 1642 | int id = -1; |
c34056a3 | 1643 | |
cd549687 TH |
1644 | lockdep_assert_held(&pool->manager_mutex); |
1645 | ||
822d8405 TH |
1646 | /* |
1647 | * ID is needed to determine kthread name. Allocate ID first | |
1648 | * without installing the pointer. | |
1649 | */ | |
1650 | idr_preload(GFP_KERNEL); | |
d565ed63 | 1651 | spin_lock_irq(&pool->lock); |
822d8405 TH |
1652 | |
1653 | id = idr_alloc(&pool->worker_idr, NULL, 0, 0, GFP_NOWAIT); | |
1654 | ||
d565ed63 | 1655 | spin_unlock_irq(&pool->lock); |
822d8405 TH |
1656 | idr_preload_end(); |
1657 | if (id < 0) | |
1658 | goto fail; | |
c34056a3 TH |
1659 | |
1660 | worker = alloc_worker(); | |
1661 | if (!worker) | |
1662 | goto fail; | |
1663 | ||
bd7bdd43 | 1664 | worker->pool = pool; |
c34056a3 TH |
1665 | worker->id = id; |
1666 | ||
29c91e99 | 1667 | if (pool->cpu >= 0) |
94dcf29a | 1668 | worker->task = kthread_create_on_node(worker_thread, |
ec22ca5e | 1669 | worker, cpu_to_node(pool->cpu), |
d84ff051 | 1670 | "kworker/%d:%d%s", pool->cpu, id, pri); |
f3421797 TH |
1671 | else |
1672 | worker->task = kthread_create(worker_thread, worker, | |
ac6104cd TH |
1673 | "kworker/u%d:%d%s", |
1674 | pool->id, id, pri); | |
c34056a3 TH |
1675 | if (IS_ERR(worker->task)) |
1676 | goto fail; | |
1677 | ||
c5aa87bb TH |
1678 | /* |
1679 | * set_cpus_allowed_ptr() will fail if the cpumask doesn't have any | |
1680 | * online CPUs. It'll be re-applied when any of the CPUs come up. | |
1681 | */ | |
7a4e344c TH |
1682 | set_user_nice(worker->task, pool->attrs->nice); |
1683 | set_cpus_allowed_ptr(worker->task, pool->attrs->cpumask); | |
3270476a | 1684 | |
14a40ffc TH |
1685 | /* prevent userland from meddling with cpumask of workqueue workers */ |
1686 | worker->task->flags |= PF_NO_SETAFFINITY; | |
7a4e344c TH |
1687 | |
1688 | /* | |
1689 | * The caller is responsible for ensuring %POOL_DISASSOCIATED | |
1690 | * remains stable across this function. See the comments above the | |
1691 | * flag definition for details. | |
1692 | */ | |
1693 | if (pool->flags & POOL_DISASSOCIATED) | |
bc2ae0f5 | 1694 | worker->flags |= WORKER_UNBOUND; |
c34056a3 | 1695 | |
822d8405 TH |
1696 | /* successful, commit the pointer to idr */ |
1697 | spin_lock_irq(&pool->lock); | |
1698 | idr_replace(&pool->worker_idr, worker, worker->id); | |
1699 | spin_unlock_irq(&pool->lock); | |
1700 | ||
c34056a3 | 1701 | return worker; |
822d8405 | 1702 | |
c34056a3 TH |
1703 | fail: |
1704 | if (id >= 0) { | |
d565ed63 | 1705 | spin_lock_irq(&pool->lock); |
822d8405 | 1706 | idr_remove(&pool->worker_idr, id); |
d565ed63 | 1707 | spin_unlock_irq(&pool->lock); |
c34056a3 TH |
1708 | } |
1709 | kfree(worker); | |
1710 | return NULL; | |
1711 | } | |
1712 | ||
1713 | /** | |
1714 | * start_worker - start a newly created worker | |
1715 | * @worker: worker to start | |
1716 | * | |
706026c2 | 1717 | * Make the pool aware of @worker and start it. |
c34056a3 TH |
1718 | * |
1719 | * CONTEXT: | |
d565ed63 | 1720 | * spin_lock_irq(pool->lock). |
c34056a3 TH |
1721 | */ |
1722 | static void start_worker(struct worker *worker) | |
1723 | { | |
cb444766 | 1724 | worker->flags |= WORKER_STARTED; |
bd7bdd43 | 1725 | worker->pool->nr_workers++; |
c8e55f36 | 1726 | worker_enter_idle(worker); |
c34056a3 TH |
1727 | wake_up_process(worker->task); |
1728 | } | |
1729 | ||
ebf44d16 TH |
1730 | /** |
1731 | * create_and_start_worker - create and start a worker for a pool | |
1732 | * @pool: the target pool | |
1733 | * | |
cd549687 | 1734 | * Grab the managership of @pool and create and start a new worker for it. |
ebf44d16 TH |
1735 | */ |
1736 | static int create_and_start_worker(struct worker_pool *pool) | |
1737 | { | |
1738 | struct worker *worker; | |
1739 | ||
cd549687 TH |
1740 | mutex_lock(&pool->manager_mutex); |
1741 | ||
ebf44d16 TH |
1742 | worker = create_worker(pool); |
1743 | if (worker) { | |
1744 | spin_lock_irq(&pool->lock); | |
1745 | start_worker(worker); | |
1746 | spin_unlock_irq(&pool->lock); | |
1747 | } | |
1748 | ||
cd549687 TH |
1749 | mutex_unlock(&pool->manager_mutex); |
1750 | ||
ebf44d16 TH |
1751 | return worker ? 0 : -ENOMEM; |
1752 | } | |
1753 | ||
c34056a3 TH |
1754 | /** |
1755 | * destroy_worker - destroy a workqueue worker | |
1756 | * @worker: worker to be destroyed | |
1757 | * | |
706026c2 | 1758 | * Destroy @worker and adjust @pool stats accordingly. |
c8e55f36 TH |
1759 | * |
1760 | * CONTEXT: | |
d565ed63 | 1761 | * spin_lock_irq(pool->lock) which is released and regrabbed. |
c34056a3 TH |
1762 | */ |
1763 | static void destroy_worker(struct worker *worker) | |
1764 | { | |
bd7bdd43 | 1765 | struct worker_pool *pool = worker->pool; |
c34056a3 | 1766 | |
cd549687 TH |
1767 | lockdep_assert_held(&pool->manager_mutex); |
1768 | lockdep_assert_held(&pool->lock); | |
1769 | ||
c34056a3 | 1770 | /* sanity check frenzy */ |
6183c009 TH |
1771 | if (WARN_ON(worker->current_work) || |
1772 | WARN_ON(!list_empty(&worker->scheduled))) | |
1773 | return; | |
c34056a3 | 1774 | |
c8e55f36 | 1775 | if (worker->flags & WORKER_STARTED) |
bd7bdd43 | 1776 | pool->nr_workers--; |
c8e55f36 | 1777 | if (worker->flags & WORKER_IDLE) |
bd7bdd43 | 1778 | pool->nr_idle--; |
c8e55f36 TH |
1779 | |
1780 | list_del_init(&worker->entry); | |
cb444766 | 1781 | worker->flags |= WORKER_DIE; |
c8e55f36 | 1782 | |
822d8405 TH |
1783 | idr_remove(&pool->worker_idr, worker->id); |
1784 | ||
d565ed63 | 1785 | spin_unlock_irq(&pool->lock); |
c8e55f36 | 1786 | |
c34056a3 TH |
1787 | kthread_stop(worker->task); |
1788 | kfree(worker); | |
1789 | ||
d565ed63 | 1790 | spin_lock_irq(&pool->lock); |
c34056a3 TH |
1791 | } |
1792 | ||
63d95a91 | 1793 | static void idle_worker_timeout(unsigned long __pool) |
e22bee78 | 1794 | { |
63d95a91 | 1795 | struct worker_pool *pool = (void *)__pool; |
e22bee78 | 1796 | |
d565ed63 | 1797 | spin_lock_irq(&pool->lock); |
e22bee78 | 1798 | |
63d95a91 | 1799 | if (too_many_workers(pool)) { |
e22bee78 TH |
1800 | struct worker *worker; |
1801 | unsigned long expires; | |
1802 | ||
1803 | /* idle_list is kept in LIFO order, check the last one */ | |
63d95a91 | 1804 | worker = list_entry(pool->idle_list.prev, struct worker, entry); |
e22bee78 TH |
1805 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; |
1806 | ||
1807 | if (time_before(jiffies, expires)) | |
63d95a91 | 1808 | mod_timer(&pool->idle_timer, expires); |
e22bee78 TH |
1809 | else { |
1810 | /* it's been idle for too long, wake up manager */ | |
11ebea50 | 1811 | pool->flags |= POOL_MANAGE_WORKERS; |
63d95a91 | 1812 | wake_up_worker(pool); |
d5abe669 | 1813 | } |
e22bee78 TH |
1814 | } |
1815 | ||
d565ed63 | 1816 | spin_unlock_irq(&pool->lock); |
e22bee78 | 1817 | } |
d5abe669 | 1818 | |
493a1724 | 1819 | static void send_mayday(struct work_struct *work) |
e22bee78 | 1820 | { |
112202d9 TH |
1821 | struct pool_workqueue *pwq = get_work_pwq(work); |
1822 | struct workqueue_struct *wq = pwq->wq; | |
493a1724 | 1823 | |
2e109a28 | 1824 | lockdep_assert_held(&wq_mayday_lock); |
e22bee78 | 1825 | |
493008a8 | 1826 | if (!wq->rescuer) |
493a1724 | 1827 | return; |
e22bee78 TH |
1828 | |
1829 | /* mayday mayday mayday */ | |
493a1724 TH |
1830 | if (list_empty(&pwq->mayday_node)) { |
1831 | list_add_tail(&pwq->mayday_node, &wq->maydays); | |
e22bee78 | 1832 | wake_up_process(wq->rescuer->task); |
493a1724 | 1833 | } |
e22bee78 TH |
1834 | } |
1835 | ||
706026c2 | 1836 | static void pool_mayday_timeout(unsigned long __pool) |
e22bee78 | 1837 | { |
63d95a91 | 1838 | struct worker_pool *pool = (void *)__pool; |
e22bee78 TH |
1839 | struct work_struct *work; |
1840 | ||
2e109a28 | 1841 | spin_lock_irq(&wq_mayday_lock); /* for wq->maydays */ |
493a1724 | 1842 | spin_lock(&pool->lock); |
e22bee78 | 1843 | |
63d95a91 | 1844 | if (need_to_create_worker(pool)) { |
e22bee78 TH |
1845 | /* |
1846 | * We've been trying to create a new worker but | |
1847 | * haven't been successful. We might be hitting an | |
1848 | * allocation deadlock. Send distress signals to | |
1849 | * rescuers. | |
1850 | */ | |
63d95a91 | 1851 | list_for_each_entry(work, &pool->worklist, entry) |
e22bee78 | 1852 | send_mayday(work); |
1da177e4 | 1853 | } |
e22bee78 | 1854 | |
493a1724 | 1855 | spin_unlock(&pool->lock); |
2e109a28 | 1856 | spin_unlock_irq(&wq_mayday_lock); |
e22bee78 | 1857 | |
63d95a91 | 1858 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INTERVAL); |
1da177e4 LT |
1859 | } |
1860 | ||
e22bee78 TH |
1861 | /** |
1862 | * maybe_create_worker - create a new worker if necessary | |
63d95a91 | 1863 | * @pool: pool to create a new worker for |
e22bee78 | 1864 | * |
63d95a91 | 1865 | * Create a new worker for @pool if necessary. @pool is guaranteed to |
e22bee78 TH |
1866 | * have at least one idle worker on return from this function. If |
1867 | * creating a new worker takes longer than MAYDAY_INTERVAL, mayday is | |
63d95a91 | 1868 | * sent to all rescuers with works scheduled on @pool to resolve |
e22bee78 TH |
1869 | * possible allocation deadlock. |
1870 | * | |
c5aa87bb TH |
1871 | * On return, need_to_create_worker() is guaranteed to be %false and |
1872 | * may_start_working() %true. | |
e22bee78 TH |
1873 | * |
1874 | * LOCKING: | |
d565ed63 | 1875 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
e22bee78 TH |
1876 | * multiple times. Does GFP_KERNEL allocations. Called only from |
1877 | * manager. | |
1878 | * | |
1879 | * RETURNS: | |
c5aa87bb | 1880 | * %false if no action was taken and pool->lock stayed locked, %true |
e22bee78 TH |
1881 | * otherwise. |
1882 | */ | |
63d95a91 | 1883 | static bool maybe_create_worker(struct worker_pool *pool) |
d565ed63 TH |
1884 | __releases(&pool->lock) |
1885 | __acquires(&pool->lock) | |
1da177e4 | 1886 | { |
63d95a91 | 1887 | if (!need_to_create_worker(pool)) |
e22bee78 TH |
1888 | return false; |
1889 | restart: | |
d565ed63 | 1890 | spin_unlock_irq(&pool->lock); |
9f9c2364 | 1891 | |
e22bee78 | 1892 | /* if we don't make progress in MAYDAY_INITIAL_TIMEOUT, call for help */ |
63d95a91 | 1893 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INITIAL_TIMEOUT); |
e22bee78 TH |
1894 | |
1895 | while (true) { | |
1896 | struct worker *worker; | |
1897 | ||
bc2ae0f5 | 1898 | worker = create_worker(pool); |
e22bee78 | 1899 | if (worker) { |
63d95a91 | 1900 | del_timer_sync(&pool->mayday_timer); |
d565ed63 | 1901 | spin_lock_irq(&pool->lock); |
e22bee78 | 1902 | start_worker(worker); |
6183c009 TH |
1903 | if (WARN_ON_ONCE(need_to_create_worker(pool))) |
1904 | goto restart; | |
e22bee78 TH |
1905 | return true; |
1906 | } | |
1907 | ||
63d95a91 | 1908 | if (!need_to_create_worker(pool)) |
e22bee78 | 1909 | break; |
1da177e4 | 1910 | |
e22bee78 TH |
1911 | __set_current_state(TASK_INTERRUPTIBLE); |
1912 | schedule_timeout(CREATE_COOLDOWN); | |
9f9c2364 | 1913 | |
63d95a91 | 1914 | if (!need_to_create_worker(pool)) |
e22bee78 TH |
1915 | break; |
1916 | } | |
1917 | ||
63d95a91 | 1918 | del_timer_sync(&pool->mayday_timer); |
d565ed63 | 1919 | spin_lock_irq(&pool->lock); |
63d95a91 | 1920 | if (need_to_create_worker(pool)) |
e22bee78 TH |
1921 | goto restart; |
1922 | return true; | |
1923 | } | |
1924 | ||
1925 | /** | |
1926 | * maybe_destroy_worker - destroy workers which have been idle for a while | |
63d95a91 | 1927 | * @pool: pool to destroy workers for |
e22bee78 | 1928 | * |
63d95a91 | 1929 | * Destroy @pool workers which have been idle for longer than |
e22bee78 TH |
1930 | * IDLE_WORKER_TIMEOUT. |
1931 | * | |
1932 | * LOCKING: | |
d565ed63 | 1933 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
e22bee78 TH |
1934 | * multiple times. Called only from manager. |
1935 | * | |
1936 | * RETURNS: | |
c5aa87bb | 1937 | * %false if no action was taken and pool->lock stayed locked, %true |
e22bee78 TH |
1938 | * otherwise. |
1939 | */ | |
63d95a91 | 1940 | static bool maybe_destroy_workers(struct worker_pool *pool) |
e22bee78 TH |
1941 | { |
1942 | bool ret = false; | |
1da177e4 | 1943 | |
63d95a91 | 1944 | while (too_many_workers(pool)) { |
e22bee78 TH |
1945 | struct worker *worker; |
1946 | unsigned long expires; | |
3af24433 | 1947 | |
63d95a91 | 1948 | worker = list_entry(pool->idle_list.prev, struct worker, entry); |
e22bee78 | 1949 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; |
85f4186a | 1950 | |
e22bee78 | 1951 | if (time_before(jiffies, expires)) { |
63d95a91 | 1952 | mod_timer(&pool->idle_timer, expires); |
3af24433 | 1953 | break; |
e22bee78 | 1954 | } |
1da177e4 | 1955 | |
e22bee78 TH |
1956 | destroy_worker(worker); |
1957 | ret = true; | |
1da177e4 | 1958 | } |
1e19ffc6 | 1959 | |
e22bee78 | 1960 | return ret; |
1e19ffc6 TH |
1961 | } |
1962 | ||
73f53c4a | 1963 | /** |
e22bee78 TH |
1964 | * manage_workers - manage worker pool |
1965 | * @worker: self | |
73f53c4a | 1966 | * |
706026c2 | 1967 | * Assume the manager role and manage the worker pool @worker belongs |
e22bee78 | 1968 | * to. At any given time, there can be only zero or one manager per |
706026c2 | 1969 | * pool. The exclusion is handled automatically by this function. |
e22bee78 TH |
1970 | * |
1971 | * The caller can safely start processing works on false return. On | |
1972 | * true return, it's guaranteed that need_to_create_worker() is false | |
1973 | * and may_start_working() is true. | |
73f53c4a TH |
1974 | * |
1975 | * CONTEXT: | |
d565ed63 | 1976 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
e22bee78 TH |
1977 | * multiple times. Does GFP_KERNEL allocations. |
1978 | * | |
1979 | * RETURNS: | |
d565ed63 TH |
1980 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
1981 | * multiple times. Does GFP_KERNEL allocations. | |
73f53c4a | 1982 | */ |
e22bee78 | 1983 | static bool manage_workers(struct worker *worker) |
73f53c4a | 1984 | { |
63d95a91 | 1985 | struct worker_pool *pool = worker->pool; |
e22bee78 | 1986 | bool ret = false; |
73f53c4a | 1987 | |
bc3a1afc TH |
1988 | /* |
1989 | * Managership is governed by two mutexes - manager_arb and | |
1990 | * manager_mutex. manager_arb handles arbitration of manager role. | |
1991 | * Anyone who successfully grabs manager_arb wins the arbitration | |
1992 | * and becomes the manager. mutex_trylock() on pool->manager_arb | |
1993 | * failure while holding pool->lock reliably indicates that someone | |
1994 | * else is managing the pool and the worker which failed trylock | |
1995 | * can proceed to executing work items. This means that anyone | |
1996 | * grabbing manager_arb is responsible for actually performing | |
1997 | * manager duties. If manager_arb is grabbed and released without | |
1998 | * actual management, the pool may stall indefinitely. | |
1999 | * | |
2000 | * manager_mutex is used for exclusion of actual management | |
2001 | * operations. The holder of manager_mutex can be sure that none | |
2002 | * of management operations, including creation and destruction of | |
2003 | * workers, won't take place until the mutex is released. Because | |
2004 | * manager_mutex doesn't interfere with manager role arbitration, | |
2005 | * it is guaranteed that the pool's management, while may be | |
2006 | * delayed, won't be disturbed by someone else grabbing | |
2007 | * manager_mutex. | |
2008 | */ | |
34a06bd6 | 2009 | if (!mutex_trylock(&pool->manager_arb)) |
e22bee78 | 2010 | return ret; |
1e19ffc6 | 2011 | |
ee378aa4 | 2012 | /* |
bc3a1afc TH |
2013 | * With manager arbitration won, manager_mutex would be free in |
2014 | * most cases. trylock first without dropping @pool->lock. | |
ee378aa4 | 2015 | */ |
bc3a1afc | 2016 | if (unlikely(!mutex_trylock(&pool->manager_mutex))) { |
d565ed63 | 2017 | spin_unlock_irq(&pool->lock); |
bc3a1afc | 2018 | mutex_lock(&pool->manager_mutex); |
ee378aa4 LJ |
2019 | ret = true; |
2020 | } | |
73f53c4a | 2021 | |
11ebea50 | 2022 | pool->flags &= ~POOL_MANAGE_WORKERS; |
73f53c4a TH |
2023 | |
2024 | /* | |
e22bee78 TH |
2025 | * Destroy and then create so that may_start_working() is true |
2026 | * on return. | |
73f53c4a | 2027 | */ |
63d95a91 TH |
2028 | ret |= maybe_destroy_workers(pool); |
2029 | ret |= maybe_create_worker(pool); | |
e22bee78 | 2030 | |
bc3a1afc | 2031 | mutex_unlock(&pool->manager_mutex); |
34a06bd6 | 2032 | mutex_unlock(&pool->manager_arb); |
e22bee78 | 2033 | return ret; |
73f53c4a TH |
2034 | } |
2035 | ||
a62428c0 TH |
2036 | /** |
2037 | * process_one_work - process single work | |
c34056a3 | 2038 | * @worker: self |
a62428c0 TH |
2039 | * @work: work to process |
2040 | * | |
2041 | * Process @work. This function contains all the logics necessary to | |
2042 | * process a single work including synchronization against and | |
2043 | * interaction with other workers on the same cpu, queueing and | |
2044 | * flushing. As long as context requirement is met, any worker can | |
2045 | * call this function to process a work. | |
2046 | * | |
2047 | * CONTEXT: | |
d565ed63 | 2048 | * spin_lock_irq(pool->lock) which is released and regrabbed. |
a62428c0 | 2049 | */ |
c34056a3 | 2050 | static void process_one_work(struct worker *worker, struct work_struct *work) |
d565ed63 TH |
2051 | __releases(&pool->lock) |
2052 | __acquires(&pool->lock) | |
a62428c0 | 2053 | { |
112202d9 | 2054 | struct pool_workqueue *pwq = get_work_pwq(work); |
bd7bdd43 | 2055 | struct worker_pool *pool = worker->pool; |
112202d9 | 2056 | bool cpu_intensive = pwq->wq->flags & WQ_CPU_INTENSIVE; |
73f53c4a | 2057 | int work_color; |
7e11629d | 2058 | struct worker *collision; |
a62428c0 TH |
2059 | #ifdef CONFIG_LOCKDEP |
2060 | /* | |
2061 | * It is permissible to free the struct work_struct from | |
2062 | * inside the function that is called from it, this we need to | |
2063 | * take into account for lockdep too. To avoid bogus "held | |
2064 | * lock freed" warnings as well as problems when looking into | |
2065 | * work->lockdep_map, make a copy and use that here. | |
2066 | */ | |
4d82a1de PZ |
2067 | struct lockdep_map lockdep_map; |
2068 | ||
2069 | lockdep_copy_map(&lockdep_map, &work->lockdep_map); | |
a62428c0 | 2070 | #endif |
6fec10a1 TH |
2071 | /* |
2072 | * Ensure we're on the correct CPU. DISASSOCIATED test is | |
2073 | * necessary to avoid spurious warnings from rescuers servicing the | |
24647570 | 2074 | * unbound or a disassociated pool. |
6fec10a1 | 2075 | */ |
5f7dabfd | 2076 | WARN_ON_ONCE(!(worker->flags & WORKER_UNBOUND) && |
24647570 | 2077 | !(pool->flags & POOL_DISASSOCIATED) && |
ec22ca5e | 2078 | raw_smp_processor_id() != pool->cpu); |
25511a47 | 2079 | |
7e11629d TH |
2080 | /* |
2081 | * A single work shouldn't be executed concurrently by | |
2082 | * multiple workers on a single cpu. Check whether anyone is | |
2083 | * already processing the work. If so, defer the work to the | |
2084 | * currently executing one. | |
2085 | */ | |
c9e7cf27 | 2086 | collision = find_worker_executing_work(pool, work); |
7e11629d TH |
2087 | if (unlikely(collision)) { |
2088 | move_linked_works(work, &collision->scheduled, NULL); | |
2089 | return; | |
2090 | } | |
2091 | ||
8930caba | 2092 | /* claim and dequeue */ |
a62428c0 | 2093 | debug_work_deactivate(work); |
c9e7cf27 | 2094 | hash_add(pool->busy_hash, &worker->hentry, (unsigned long)work); |
c34056a3 | 2095 | worker->current_work = work; |
a2c1c57b | 2096 | worker->current_func = work->func; |
112202d9 | 2097 | worker->current_pwq = pwq; |
73f53c4a | 2098 | work_color = get_work_color(work); |
7a22ad75 | 2099 | |
a62428c0 TH |
2100 | list_del_init(&work->entry); |
2101 | ||
fb0e7beb TH |
2102 | /* |
2103 | * CPU intensive works don't participate in concurrency | |
2104 | * management. They're the scheduler's responsibility. | |
2105 | */ | |
2106 | if (unlikely(cpu_intensive)) | |
2107 | worker_set_flags(worker, WORKER_CPU_INTENSIVE, true); | |
2108 | ||
974271c4 | 2109 | /* |
d565ed63 | 2110 | * Unbound pool isn't concurrency managed and work items should be |
974271c4 TH |
2111 | * executed ASAP. Wake up another worker if necessary. |
2112 | */ | |
63d95a91 TH |
2113 | if ((worker->flags & WORKER_UNBOUND) && need_more_worker(pool)) |
2114 | wake_up_worker(pool); | |
974271c4 | 2115 | |
8930caba | 2116 | /* |
7c3eed5c | 2117 | * Record the last pool and clear PENDING which should be the last |
d565ed63 | 2118 | * update to @work. Also, do this inside @pool->lock so that |
23657bb1 TH |
2119 | * PENDING and queued state changes happen together while IRQ is |
2120 | * disabled. | |
8930caba | 2121 | */ |
7c3eed5c | 2122 | set_work_pool_and_clear_pending(work, pool->id); |
a62428c0 | 2123 | |
d565ed63 | 2124 | spin_unlock_irq(&pool->lock); |
a62428c0 | 2125 | |
112202d9 | 2126 | lock_map_acquire_read(&pwq->wq->lockdep_map); |
a62428c0 | 2127 | lock_map_acquire(&lockdep_map); |
e36c886a | 2128 | trace_workqueue_execute_start(work); |
a2c1c57b | 2129 | worker->current_func(work); |
e36c886a AV |
2130 | /* |
2131 | * While we must be careful to not use "work" after this, the trace | |
2132 | * point will only record its address. | |
2133 | */ | |
2134 | trace_workqueue_execute_end(work); | |
a62428c0 | 2135 | lock_map_release(&lockdep_map); |
112202d9 | 2136 | lock_map_release(&pwq->wq->lockdep_map); |
a62428c0 TH |
2137 | |
2138 | if (unlikely(in_atomic() || lockdep_depth(current) > 0)) { | |
044c782c VI |
2139 | pr_err("BUG: workqueue leaked lock or atomic: %s/0x%08x/%d\n" |
2140 | " last function: %pf\n", | |
a2c1c57b TH |
2141 | current->comm, preempt_count(), task_pid_nr(current), |
2142 | worker->current_func); | |
a62428c0 TH |
2143 | debug_show_held_locks(current); |
2144 | dump_stack(); | |
2145 | } | |
2146 | ||
d565ed63 | 2147 | spin_lock_irq(&pool->lock); |
a62428c0 | 2148 | |
fb0e7beb TH |
2149 | /* clear cpu intensive status */ |
2150 | if (unlikely(cpu_intensive)) | |
2151 | worker_clr_flags(worker, WORKER_CPU_INTENSIVE); | |
2152 | ||
a62428c0 | 2153 | /* we're done with it, release */ |
42f8570f | 2154 | hash_del(&worker->hentry); |
c34056a3 | 2155 | worker->current_work = NULL; |
a2c1c57b | 2156 | worker->current_func = NULL; |
112202d9 TH |
2157 | worker->current_pwq = NULL; |
2158 | pwq_dec_nr_in_flight(pwq, work_color); | |
a62428c0 TH |
2159 | } |
2160 | ||
affee4b2 TH |
2161 | /** |
2162 | * process_scheduled_works - process scheduled works | |
2163 | * @worker: self | |
2164 | * | |
2165 | * Process all scheduled works. Please note that the scheduled list | |
2166 | * may change while processing a work, so this function repeatedly | |
2167 | * fetches a work from the top and executes it. | |
2168 | * | |
2169 | * CONTEXT: | |
d565ed63 | 2170 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
affee4b2 TH |
2171 | * multiple times. |
2172 | */ | |
2173 | static void process_scheduled_works(struct worker *worker) | |
1da177e4 | 2174 | { |
affee4b2 TH |
2175 | while (!list_empty(&worker->scheduled)) { |
2176 | struct work_struct *work = list_first_entry(&worker->scheduled, | |
1da177e4 | 2177 | struct work_struct, entry); |
c34056a3 | 2178 | process_one_work(worker, work); |
1da177e4 | 2179 | } |
1da177e4 LT |
2180 | } |
2181 | ||
4690c4ab TH |
2182 | /** |
2183 | * worker_thread - the worker thread function | |
c34056a3 | 2184 | * @__worker: self |
4690c4ab | 2185 | * |
c5aa87bb TH |
2186 | * The worker thread function. All workers belong to a worker_pool - |
2187 | * either a per-cpu one or dynamic unbound one. These workers process all | |
2188 | * work items regardless of their specific target workqueue. The only | |
2189 | * exception is work items which belong to workqueues with a rescuer which | |
2190 | * will be explained in rescuer_thread(). | |
4690c4ab | 2191 | */ |
c34056a3 | 2192 | static int worker_thread(void *__worker) |
1da177e4 | 2193 | { |
c34056a3 | 2194 | struct worker *worker = __worker; |
bd7bdd43 | 2195 | struct worker_pool *pool = worker->pool; |
1da177e4 | 2196 | |
e22bee78 TH |
2197 | /* tell the scheduler that this is a workqueue worker */ |
2198 | worker->task->flags |= PF_WQ_WORKER; | |
c8e55f36 | 2199 | woke_up: |
d565ed63 | 2200 | spin_lock_irq(&pool->lock); |
1da177e4 | 2201 | |
a9ab775b TH |
2202 | /* am I supposed to die? */ |
2203 | if (unlikely(worker->flags & WORKER_DIE)) { | |
d565ed63 | 2204 | spin_unlock_irq(&pool->lock); |
a9ab775b TH |
2205 | WARN_ON_ONCE(!list_empty(&worker->entry)); |
2206 | worker->task->flags &= ~PF_WQ_WORKER; | |
2207 | return 0; | |
c8e55f36 | 2208 | } |
affee4b2 | 2209 | |
c8e55f36 | 2210 | worker_leave_idle(worker); |
db7bccf4 | 2211 | recheck: |
e22bee78 | 2212 | /* no more worker necessary? */ |
63d95a91 | 2213 | if (!need_more_worker(pool)) |
e22bee78 TH |
2214 | goto sleep; |
2215 | ||
2216 | /* do we need to manage? */ | |
63d95a91 | 2217 | if (unlikely(!may_start_working(pool)) && manage_workers(worker)) |
e22bee78 TH |
2218 | goto recheck; |
2219 | ||
c8e55f36 TH |
2220 | /* |
2221 | * ->scheduled list can only be filled while a worker is | |
2222 | * preparing to process a work or actually processing it. | |
2223 | * Make sure nobody diddled with it while I was sleeping. | |
2224 | */ | |
6183c009 | 2225 | WARN_ON_ONCE(!list_empty(&worker->scheduled)); |
c8e55f36 | 2226 | |
e22bee78 | 2227 | /* |
a9ab775b TH |
2228 | * Finish PREP stage. We're guaranteed to have at least one idle |
2229 | * worker or that someone else has already assumed the manager | |
2230 | * role. This is where @worker starts participating in concurrency | |
2231 | * management if applicable and concurrency management is restored | |
2232 | * after being rebound. See rebind_workers() for details. | |
e22bee78 | 2233 | */ |
a9ab775b | 2234 | worker_clr_flags(worker, WORKER_PREP | WORKER_REBOUND); |
e22bee78 TH |
2235 | |
2236 | do { | |
c8e55f36 | 2237 | struct work_struct *work = |
bd7bdd43 | 2238 | list_first_entry(&pool->worklist, |
c8e55f36 TH |
2239 | struct work_struct, entry); |
2240 | ||
2241 | if (likely(!(*work_data_bits(work) & WORK_STRUCT_LINKED))) { | |
2242 | /* optimization path, not strictly necessary */ | |
2243 | process_one_work(worker, work); | |
2244 | if (unlikely(!list_empty(&worker->scheduled))) | |
affee4b2 | 2245 | process_scheduled_works(worker); |
c8e55f36 TH |
2246 | } else { |
2247 | move_linked_works(work, &worker->scheduled, NULL); | |
2248 | process_scheduled_works(worker); | |
affee4b2 | 2249 | } |
63d95a91 | 2250 | } while (keep_working(pool)); |
e22bee78 TH |
2251 | |
2252 | worker_set_flags(worker, WORKER_PREP, false); | |
d313dd85 | 2253 | sleep: |
63d95a91 | 2254 | if (unlikely(need_to_manage_workers(pool)) && manage_workers(worker)) |
e22bee78 | 2255 | goto recheck; |
d313dd85 | 2256 | |
c8e55f36 | 2257 | /* |
d565ed63 TH |
2258 | * pool->lock is held and there's no work to process and no need to |
2259 | * manage, sleep. Workers are woken up only while holding | |
2260 | * pool->lock or from local cpu, so setting the current state | |
2261 | * before releasing pool->lock is enough to prevent losing any | |
2262 | * event. | |
c8e55f36 TH |
2263 | */ |
2264 | worker_enter_idle(worker); | |
2265 | __set_current_state(TASK_INTERRUPTIBLE); | |
d565ed63 | 2266 | spin_unlock_irq(&pool->lock); |
c8e55f36 TH |
2267 | schedule(); |
2268 | goto woke_up; | |
1da177e4 LT |
2269 | } |
2270 | ||
e22bee78 TH |
2271 | /** |
2272 | * rescuer_thread - the rescuer thread function | |
111c225a | 2273 | * @__rescuer: self |
e22bee78 TH |
2274 | * |
2275 | * Workqueue rescuer thread function. There's one rescuer for each | |
493008a8 | 2276 | * workqueue which has WQ_MEM_RECLAIM set. |
e22bee78 | 2277 | * |
706026c2 | 2278 | * Regular work processing on a pool may block trying to create a new |
e22bee78 TH |
2279 | * worker which uses GFP_KERNEL allocation which has slight chance of |
2280 | * developing into deadlock if some works currently on the same queue | |
2281 | * need to be processed to satisfy the GFP_KERNEL allocation. This is | |
2282 | * the problem rescuer solves. | |
2283 | * | |
706026c2 TH |
2284 | * When such condition is possible, the pool summons rescuers of all |
2285 | * workqueues which have works queued on the pool and let them process | |
e22bee78 TH |
2286 | * those works so that forward progress can be guaranteed. |
2287 | * | |
2288 | * This should happen rarely. | |
2289 | */ | |
111c225a | 2290 | static int rescuer_thread(void *__rescuer) |
e22bee78 | 2291 | { |
111c225a TH |
2292 | struct worker *rescuer = __rescuer; |
2293 | struct workqueue_struct *wq = rescuer->rescue_wq; | |
e22bee78 | 2294 | struct list_head *scheduled = &rescuer->scheduled; |
e22bee78 TH |
2295 | |
2296 | set_user_nice(current, RESCUER_NICE_LEVEL); | |
111c225a TH |
2297 | |
2298 | /* | |
2299 | * Mark rescuer as worker too. As WORKER_PREP is never cleared, it | |
2300 | * doesn't participate in concurrency management. | |
2301 | */ | |
2302 | rescuer->task->flags |= PF_WQ_WORKER; | |
e22bee78 TH |
2303 | repeat: |
2304 | set_current_state(TASK_INTERRUPTIBLE); | |
2305 | ||
412d32e6 MG |
2306 | if (kthread_should_stop()) { |
2307 | __set_current_state(TASK_RUNNING); | |
111c225a | 2308 | rescuer->task->flags &= ~PF_WQ_WORKER; |
e22bee78 | 2309 | return 0; |
412d32e6 | 2310 | } |
e22bee78 | 2311 | |
493a1724 | 2312 | /* see whether any pwq is asking for help */ |
2e109a28 | 2313 | spin_lock_irq(&wq_mayday_lock); |
493a1724 TH |
2314 | |
2315 | while (!list_empty(&wq->maydays)) { | |
2316 | struct pool_workqueue *pwq = list_first_entry(&wq->maydays, | |
2317 | struct pool_workqueue, mayday_node); | |
112202d9 | 2318 | struct worker_pool *pool = pwq->pool; |
e22bee78 TH |
2319 | struct work_struct *work, *n; |
2320 | ||
2321 | __set_current_state(TASK_RUNNING); | |
493a1724 TH |
2322 | list_del_init(&pwq->mayday_node); |
2323 | ||
2e109a28 | 2324 | spin_unlock_irq(&wq_mayday_lock); |
e22bee78 TH |
2325 | |
2326 | /* migrate to the target cpu if possible */ | |
f36dc67b | 2327 | worker_maybe_bind_and_lock(pool); |
b3104104 | 2328 | rescuer->pool = pool; |
e22bee78 TH |
2329 | |
2330 | /* | |
2331 | * Slurp in all works issued via this workqueue and | |
2332 | * process'em. | |
2333 | */ | |
6183c009 | 2334 | WARN_ON_ONCE(!list_empty(&rescuer->scheduled)); |
bd7bdd43 | 2335 | list_for_each_entry_safe(work, n, &pool->worklist, entry) |
112202d9 | 2336 | if (get_work_pwq(work) == pwq) |
e22bee78 TH |
2337 | move_linked_works(work, scheduled, &n); |
2338 | ||
2339 | process_scheduled_works(rescuer); | |
7576958a TH |
2340 | |
2341 | /* | |
d565ed63 | 2342 | * Leave this pool. If keep_working() is %true, notify a |
7576958a TH |
2343 | * regular worker; otherwise, we end up with 0 concurrency |
2344 | * and stalling the execution. | |
2345 | */ | |
63d95a91 TH |
2346 | if (keep_working(pool)) |
2347 | wake_up_worker(pool); | |
7576958a | 2348 | |
b3104104 | 2349 | rescuer->pool = NULL; |
493a1724 | 2350 | spin_unlock(&pool->lock); |
2e109a28 | 2351 | spin_lock(&wq_mayday_lock); |
e22bee78 TH |
2352 | } |
2353 | ||
2e109a28 | 2354 | spin_unlock_irq(&wq_mayday_lock); |
493a1724 | 2355 | |
111c225a TH |
2356 | /* rescuers should never participate in concurrency management */ |
2357 | WARN_ON_ONCE(!(rescuer->flags & WORKER_NOT_RUNNING)); | |
e22bee78 TH |
2358 | schedule(); |
2359 | goto repeat; | |
1da177e4 LT |
2360 | } |
2361 | ||
fc2e4d70 ON |
2362 | struct wq_barrier { |
2363 | struct work_struct work; | |
2364 | struct completion done; | |
2365 | }; | |
2366 | ||
2367 | static void wq_barrier_func(struct work_struct *work) | |
2368 | { | |
2369 | struct wq_barrier *barr = container_of(work, struct wq_barrier, work); | |
2370 | complete(&barr->done); | |
2371 | } | |
2372 | ||
4690c4ab TH |
2373 | /** |
2374 | * insert_wq_barrier - insert a barrier work | |
112202d9 | 2375 | * @pwq: pwq to insert barrier into |
4690c4ab | 2376 | * @barr: wq_barrier to insert |
affee4b2 TH |
2377 | * @target: target work to attach @barr to |
2378 | * @worker: worker currently executing @target, NULL if @target is not executing | |
4690c4ab | 2379 | * |
affee4b2 TH |
2380 | * @barr is linked to @target such that @barr is completed only after |
2381 | * @target finishes execution. Please note that the ordering | |
2382 | * guarantee is observed only with respect to @target and on the local | |
2383 | * cpu. | |
2384 | * | |
2385 | * Currently, a queued barrier can't be canceled. This is because | |
2386 | * try_to_grab_pending() can't determine whether the work to be | |
2387 | * grabbed is at the head of the queue and thus can't clear LINKED | |
2388 | * flag of the previous work while there must be a valid next work | |
2389 | * after a work with LINKED flag set. | |
2390 | * | |
2391 | * Note that when @worker is non-NULL, @target may be modified | |
112202d9 | 2392 | * underneath us, so we can't reliably determine pwq from @target. |
4690c4ab TH |
2393 | * |
2394 | * CONTEXT: | |
d565ed63 | 2395 | * spin_lock_irq(pool->lock). |
4690c4ab | 2396 | */ |
112202d9 | 2397 | static void insert_wq_barrier(struct pool_workqueue *pwq, |
affee4b2 TH |
2398 | struct wq_barrier *barr, |
2399 | struct work_struct *target, struct worker *worker) | |
fc2e4d70 | 2400 | { |
affee4b2 TH |
2401 | struct list_head *head; |
2402 | unsigned int linked = 0; | |
2403 | ||
dc186ad7 | 2404 | /* |
d565ed63 | 2405 | * debugobject calls are safe here even with pool->lock locked |
dc186ad7 TG |
2406 | * as we know for sure that this will not trigger any of the |
2407 | * checks and call back into the fixup functions where we | |
2408 | * might deadlock. | |
2409 | */ | |
ca1cab37 | 2410 | INIT_WORK_ONSTACK(&barr->work, wq_barrier_func); |
22df02bb | 2411 | __set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(&barr->work)); |
fc2e4d70 | 2412 | init_completion(&barr->done); |
83c22520 | 2413 | |
affee4b2 TH |
2414 | /* |
2415 | * If @target is currently being executed, schedule the | |
2416 | * barrier to the worker; otherwise, put it after @target. | |
2417 | */ | |
2418 | if (worker) | |
2419 | head = worker->scheduled.next; | |
2420 | else { | |
2421 | unsigned long *bits = work_data_bits(target); | |
2422 | ||
2423 | head = target->entry.next; | |
2424 | /* there can already be other linked works, inherit and set */ | |
2425 | linked = *bits & WORK_STRUCT_LINKED; | |
2426 | __set_bit(WORK_STRUCT_LINKED_BIT, bits); | |
2427 | } | |
2428 | ||
dc186ad7 | 2429 | debug_work_activate(&barr->work); |
112202d9 | 2430 | insert_work(pwq, &barr->work, head, |
affee4b2 | 2431 | work_color_to_flags(WORK_NO_COLOR) | linked); |
fc2e4d70 ON |
2432 | } |
2433 | ||
73f53c4a | 2434 | /** |
112202d9 | 2435 | * flush_workqueue_prep_pwqs - prepare pwqs for workqueue flushing |
73f53c4a TH |
2436 | * @wq: workqueue being flushed |
2437 | * @flush_color: new flush color, < 0 for no-op | |
2438 | * @work_color: new work color, < 0 for no-op | |
2439 | * | |
112202d9 | 2440 | * Prepare pwqs for workqueue flushing. |
73f53c4a | 2441 | * |
112202d9 TH |
2442 | * If @flush_color is non-negative, flush_color on all pwqs should be |
2443 | * -1. If no pwq has in-flight commands at the specified color, all | |
2444 | * pwq->flush_color's stay at -1 and %false is returned. If any pwq | |
2445 | * has in flight commands, its pwq->flush_color is set to | |
2446 | * @flush_color, @wq->nr_pwqs_to_flush is updated accordingly, pwq | |
73f53c4a TH |
2447 | * wakeup logic is armed and %true is returned. |
2448 | * | |
2449 | * The caller should have initialized @wq->first_flusher prior to | |
2450 | * calling this function with non-negative @flush_color. If | |
2451 | * @flush_color is negative, no flush color update is done and %false | |
2452 | * is returned. | |
2453 | * | |
112202d9 | 2454 | * If @work_color is non-negative, all pwqs should have the same |
73f53c4a TH |
2455 | * work_color which is previous to @work_color and all will be |
2456 | * advanced to @work_color. | |
2457 | * | |
2458 | * CONTEXT: | |
3c25a55d | 2459 | * mutex_lock(wq->mutex). |
73f53c4a TH |
2460 | * |
2461 | * RETURNS: | |
2462 | * %true if @flush_color >= 0 and there's something to flush. %false | |
2463 | * otherwise. | |
2464 | */ | |
112202d9 | 2465 | static bool flush_workqueue_prep_pwqs(struct workqueue_struct *wq, |
73f53c4a | 2466 | int flush_color, int work_color) |
1da177e4 | 2467 | { |
73f53c4a | 2468 | bool wait = false; |
49e3cf44 | 2469 | struct pool_workqueue *pwq; |
1da177e4 | 2470 | |
73f53c4a | 2471 | if (flush_color >= 0) { |
6183c009 | 2472 | WARN_ON_ONCE(atomic_read(&wq->nr_pwqs_to_flush)); |
112202d9 | 2473 | atomic_set(&wq->nr_pwqs_to_flush, 1); |
1da177e4 | 2474 | } |
2355b70f | 2475 | |
49e3cf44 | 2476 | for_each_pwq(pwq, wq) { |
112202d9 | 2477 | struct worker_pool *pool = pwq->pool; |
fc2e4d70 | 2478 | |
b09f4fd3 | 2479 | spin_lock_irq(&pool->lock); |
83c22520 | 2480 | |
73f53c4a | 2481 | if (flush_color >= 0) { |
6183c009 | 2482 | WARN_ON_ONCE(pwq->flush_color != -1); |
fc2e4d70 | 2483 | |
112202d9 TH |
2484 | if (pwq->nr_in_flight[flush_color]) { |
2485 | pwq->flush_color = flush_color; | |
2486 | atomic_inc(&wq->nr_pwqs_to_flush); | |
73f53c4a TH |
2487 | wait = true; |
2488 | } | |
2489 | } | |
1da177e4 | 2490 | |
73f53c4a | 2491 | if (work_color >= 0) { |
6183c009 | 2492 | WARN_ON_ONCE(work_color != work_next_color(pwq->work_color)); |
112202d9 | 2493 | pwq->work_color = work_color; |
73f53c4a | 2494 | } |
1da177e4 | 2495 | |
b09f4fd3 | 2496 | spin_unlock_irq(&pool->lock); |
1da177e4 | 2497 | } |
2355b70f | 2498 | |
112202d9 | 2499 | if (flush_color >= 0 && atomic_dec_and_test(&wq->nr_pwqs_to_flush)) |
73f53c4a | 2500 | complete(&wq->first_flusher->done); |
14441960 | 2501 | |
73f53c4a | 2502 | return wait; |
1da177e4 LT |
2503 | } |
2504 | ||
0fcb78c2 | 2505 | /** |
1da177e4 | 2506 | * flush_workqueue - ensure that any scheduled work has run to completion. |
0fcb78c2 | 2507 | * @wq: workqueue to flush |
1da177e4 | 2508 | * |
c5aa87bb TH |
2509 | * This function sleeps until all work items which were queued on entry |
2510 | * have finished execution, but it is not livelocked by new incoming ones. | |
1da177e4 | 2511 | */ |
7ad5b3a5 | 2512 | void flush_workqueue(struct workqueue_struct *wq) |
1da177e4 | 2513 | { |
73f53c4a TH |
2514 | struct wq_flusher this_flusher = { |
2515 | .list = LIST_HEAD_INIT(this_flusher.list), | |
2516 | .flush_color = -1, | |
2517 | .done = COMPLETION_INITIALIZER_ONSTACK(this_flusher.done), | |
2518 | }; | |
2519 | int next_color; | |
1da177e4 | 2520 | |
3295f0ef IM |
2521 | lock_map_acquire(&wq->lockdep_map); |
2522 | lock_map_release(&wq->lockdep_map); | |
73f53c4a | 2523 | |
3c25a55d | 2524 | mutex_lock(&wq->mutex); |
73f53c4a TH |
2525 | |
2526 | /* | |
2527 | * Start-to-wait phase | |
2528 | */ | |
2529 | next_color = work_next_color(wq->work_color); | |
2530 | ||
2531 | if (next_color != wq->flush_color) { | |
2532 | /* | |
2533 | * Color space is not full. The current work_color | |
2534 | * becomes our flush_color and work_color is advanced | |
2535 | * by one. | |
2536 | */ | |
6183c009 | 2537 | WARN_ON_ONCE(!list_empty(&wq->flusher_overflow)); |
73f53c4a TH |
2538 | this_flusher.flush_color = wq->work_color; |
2539 | wq->work_color = next_color; | |
2540 | ||
2541 | if (!wq->first_flusher) { | |
2542 | /* no flush in progress, become the first flusher */ | |
6183c009 | 2543 | WARN_ON_ONCE(wq->flush_color != this_flusher.flush_color); |
73f53c4a TH |
2544 | |
2545 | wq->first_flusher = &this_flusher; | |
2546 | ||
112202d9 | 2547 | if (!flush_workqueue_prep_pwqs(wq, wq->flush_color, |
73f53c4a TH |
2548 | wq->work_color)) { |
2549 | /* nothing to flush, done */ | |
2550 | wq->flush_color = next_color; | |
2551 | wq->first_flusher = NULL; | |
2552 | goto out_unlock; | |
2553 | } | |
2554 | } else { | |
2555 | /* wait in queue */ | |
6183c009 | 2556 | WARN_ON_ONCE(wq->flush_color == this_flusher.flush_color); |
73f53c4a | 2557 | list_add_tail(&this_flusher.list, &wq->flusher_queue); |
112202d9 | 2558 | flush_workqueue_prep_pwqs(wq, -1, wq->work_color); |
73f53c4a TH |
2559 | } |
2560 | } else { | |
2561 | /* | |
2562 | * Oops, color space is full, wait on overflow queue. | |
2563 | * The next flush completion will assign us | |
2564 | * flush_color and transfer to flusher_queue. | |
2565 | */ | |
2566 | list_add_tail(&this_flusher.list, &wq->flusher_overflow); | |
2567 | } | |
2568 | ||
3c25a55d | 2569 | mutex_unlock(&wq->mutex); |
73f53c4a TH |
2570 | |
2571 | wait_for_completion(&this_flusher.done); | |
2572 | ||
2573 | /* | |
2574 | * Wake-up-and-cascade phase | |
2575 | * | |
2576 | * First flushers are responsible for cascading flushes and | |
2577 | * handling overflow. Non-first flushers can simply return. | |
2578 | */ | |
2579 | if (wq->first_flusher != &this_flusher) | |
2580 | return; | |
2581 | ||
3c25a55d | 2582 | mutex_lock(&wq->mutex); |
73f53c4a | 2583 | |
4ce48b37 TH |
2584 | /* we might have raced, check again with mutex held */ |
2585 | if (wq->first_flusher != &this_flusher) | |
2586 | goto out_unlock; | |
2587 | ||
73f53c4a TH |
2588 | wq->first_flusher = NULL; |
2589 | ||
6183c009 TH |
2590 | WARN_ON_ONCE(!list_empty(&this_flusher.list)); |
2591 | WARN_ON_ONCE(wq->flush_color != this_flusher.flush_color); | |
73f53c4a TH |
2592 | |
2593 | while (true) { | |
2594 | struct wq_flusher *next, *tmp; | |
2595 | ||
2596 | /* complete all the flushers sharing the current flush color */ | |
2597 | list_for_each_entry_safe(next, tmp, &wq->flusher_queue, list) { | |
2598 | if (next->flush_color != wq->flush_color) | |
2599 | break; | |
2600 | list_del_init(&next->list); | |
2601 | complete(&next->done); | |
2602 | } | |
2603 | ||
6183c009 TH |
2604 | WARN_ON_ONCE(!list_empty(&wq->flusher_overflow) && |
2605 | wq->flush_color != work_next_color(wq->work_color)); | |
73f53c4a TH |
2606 | |
2607 | /* this flush_color is finished, advance by one */ | |
2608 | wq->flush_color = work_next_color(wq->flush_color); | |
2609 | ||
2610 | /* one color has been freed, handle overflow queue */ | |
2611 | if (!list_empty(&wq->flusher_overflow)) { | |
2612 | /* | |
2613 | * Assign the same color to all overflowed | |
2614 | * flushers, advance work_color and append to | |
2615 | * flusher_queue. This is the start-to-wait | |
2616 | * phase for these overflowed flushers. | |
2617 | */ | |
2618 | list_for_each_entry(tmp, &wq->flusher_overflow, list) | |
2619 | tmp->flush_color = wq->work_color; | |
2620 | ||
2621 | wq->work_color = work_next_color(wq->work_color); | |
2622 | ||
2623 | list_splice_tail_init(&wq->flusher_overflow, | |
2624 | &wq->flusher_queue); | |
112202d9 | 2625 | flush_workqueue_prep_pwqs(wq, -1, wq->work_color); |
73f53c4a TH |
2626 | } |
2627 | ||
2628 | if (list_empty(&wq->flusher_queue)) { | |
6183c009 | 2629 | WARN_ON_ONCE(wq->flush_color != wq->work_color); |
73f53c4a TH |
2630 | break; |
2631 | } | |
2632 | ||
2633 | /* | |
2634 | * Need to flush more colors. Make the next flusher | |
112202d9 | 2635 | * the new first flusher and arm pwqs. |
73f53c4a | 2636 | */ |
6183c009 TH |
2637 | WARN_ON_ONCE(wq->flush_color == wq->work_color); |
2638 | WARN_ON_ONCE(wq->flush_color != next->flush_color); | |
73f53c4a TH |
2639 | |
2640 | list_del_init(&next->list); | |
2641 | wq->first_flusher = next; | |
2642 | ||
112202d9 | 2643 | if (flush_workqueue_prep_pwqs(wq, wq->flush_color, -1)) |
73f53c4a TH |
2644 | break; |
2645 | ||
2646 | /* | |
2647 | * Meh... this color is already done, clear first | |
2648 | * flusher and repeat cascading. | |
2649 | */ | |
2650 | wq->first_flusher = NULL; | |
2651 | } | |
2652 | ||
2653 | out_unlock: | |
3c25a55d | 2654 | mutex_unlock(&wq->mutex); |
1da177e4 | 2655 | } |
ae90dd5d | 2656 | EXPORT_SYMBOL_GPL(flush_workqueue); |
1da177e4 | 2657 | |
9c5a2ba7 TH |
2658 | /** |
2659 | * drain_workqueue - drain a workqueue | |
2660 | * @wq: workqueue to drain | |
2661 | * | |
2662 | * Wait until the workqueue becomes empty. While draining is in progress, | |
2663 | * only chain queueing is allowed. IOW, only currently pending or running | |
2664 | * work items on @wq can queue further work items on it. @wq is flushed | |
2665 | * repeatedly until it becomes empty. The number of flushing is detemined | |
2666 | * by the depth of chaining and should be relatively short. Whine if it | |
2667 | * takes too long. | |
2668 | */ | |
2669 | void drain_workqueue(struct workqueue_struct *wq) | |
2670 | { | |
2671 | unsigned int flush_cnt = 0; | |
49e3cf44 | 2672 | struct pool_workqueue *pwq; |
9c5a2ba7 TH |
2673 | |
2674 | /* | |
2675 | * __queue_work() needs to test whether there are drainers, is much | |
2676 | * hotter than drain_workqueue() and already looks at @wq->flags. | |
618b01eb | 2677 | * Use __WQ_DRAINING so that queue doesn't have to check nr_drainers. |
9c5a2ba7 | 2678 | */ |
87fc741e | 2679 | mutex_lock(&wq->mutex); |
9c5a2ba7 | 2680 | if (!wq->nr_drainers++) |
618b01eb | 2681 | wq->flags |= __WQ_DRAINING; |
87fc741e | 2682 | mutex_unlock(&wq->mutex); |
9c5a2ba7 TH |
2683 | reflush: |
2684 | flush_workqueue(wq); | |
2685 | ||
b09f4fd3 | 2686 | mutex_lock(&wq->mutex); |
76af4d93 | 2687 | |
49e3cf44 | 2688 | for_each_pwq(pwq, wq) { |
fa2563e4 | 2689 | bool drained; |
9c5a2ba7 | 2690 | |
b09f4fd3 | 2691 | spin_lock_irq(&pwq->pool->lock); |
112202d9 | 2692 | drained = !pwq->nr_active && list_empty(&pwq->delayed_works); |
b09f4fd3 | 2693 | spin_unlock_irq(&pwq->pool->lock); |
fa2563e4 TT |
2694 | |
2695 | if (drained) | |
9c5a2ba7 TH |
2696 | continue; |
2697 | ||
2698 | if (++flush_cnt == 10 || | |
2699 | (flush_cnt % 100 == 0 && flush_cnt <= 1000)) | |
c5aa87bb | 2700 | pr_warn("workqueue %s: drain_workqueue() isn't complete after %u tries\n", |
044c782c | 2701 | wq->name, flush_cnt); |
76af4d93 | 2702 | |
b09f4fd3 | 2703 | mutex_unlock(&wq->mutex); |
9c5a2ba7 TH |
2704 | goto reflush; |
2705 | } | |
2706 | ||
9c5a2ba7 | 2707 | if (!--wq->nr_drainers) |
618b01eb | 2708 | wq->flags &= ~__WQ_DRAINING; |
87fc741e | 2709 | mutex_unlock(&wq->mutex); |
9c5a2ba7 TH |
2710 | } |
2711 | EXPORT_SYMBOL_GPL(drain_workqueue); | |
2712 | ||
606a5020 | 2713 | static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr) |
db700897 | 2714 | { |
affee4b2 | 2715 | struct worker *worker = NULL; |
c9e7cf27 | 2716 | struct worker_pool *pool; |
112202d9 | 2717 | struct pool_workqueue *pwq; |
db700897 ON |
2718 | |
2719 | might_sleep(); | |
fa1b54e6 TH |
2720 | |
2721 | local_irq_disable(); | |
c9e7cf27 | 2722 | pool = get_work_pool(work); |
fa1b54e6 TH |
2723 | if (!pool) { |
2724 | local_irq_enable(); | |
baf59022 | 2725 | return false; |
fa1b54e6 | 2726 | } |
db700897 | 2727 | |
fa1b54e6 | 2728 | spin_lock(&pool->lock); |
0b3dae68 | 2729 | /* see the comment in try_to_grab_pending() with the same code */ |
112202d9 TH |
2730 | pwq = get_work_pwq(work); |
2731 | if (pwq) { | |
2732 | if (unlikely(pwq->pool != pool)) | |
4690c4ab | 2733 | goto already_gone; |
606a5020 | 2734 | } else { |
c9e7cf27 | 2735 | worker = find_worker_executing_work(pool, work); |
affee4b2 | 2736 | if (!worker) |
4690c4ab | 2737 | goto already_gone; |
112202d9 | 2738 | pwq = worker->current_pwq; |
606a5020 | 2739 | } |
db700897 | 2740 | |
112202d9 | 2741 | insert_wq_barrier(pwq, barr, work, worker); |
d565ed63 | 2742 | spin_unlock_irq(&pool->lock); |
7a22ad75 | 2743 | |
e159489b TH |
2744 | /* |
2745 | * If @max_active is 1 or rescuer is in use, flushing another work | |
2746 | * item on the same workqueue may lead to deadlock. Make sure the | |
2747 | * flusher is not running on the same workqueue by verifying write | |
2748 | * access. | |
2749 | */ | |
493008a8 | 2750 | if (pwq->wq->saved_max_active == 1 || pwq->wq->rescuer) |
112202d9 | 2751 | lock_map_acquire(&pwq->wq->lockdep_map); |
e159489b | 2752 | else |
112202d9 TH |
2753 | lock_map_acquire_read(&pwq->wq->lockdep_map); |
2754 | lock_map_release(&pwq->wq->lockdep_map); | |
e159489b | 2755 | |
401a8d04 | 2756 | return true; |
4690c4ab | 2757 | already_gone: |
d565ed63 | 2758 | spin_unlock_irq(&pool->lock); |
401a8d04 | 2759 | return false; |
db700897 | 2760 | } |
baf59022 TH |
2761 | |
2762 | /** | |
2763 | * flush_work - wait for a work to finish executing the last queueing instance | |
2764 | * @work: the work to flush | |
2765 | * | |
606a5020 TH |
2766 | * Wait until @work has finished execution. @work is guaranteed to be idle |
2767 | * on return if it hasn't been requeued since flush started. | |
baf59022 TH |
2768 | * |
2769 | * RETURNS: | |
2770 | * %true if flush_work() waited for the work to finish execution, | |
2771 | * %false if it was already idle. | |
2772 | */ | |
2773 | bool flush_work(struct work_struct *work) | |
2774 | { | |
2775 | struct wq_barrier barr; | |
2776 | ||
0976dfc1 SB |
2777 | lock_map_acquire(&work->lockdep_map); |
2778 | lock_map_release(&work->lockdep_map); | |
2779 | ||
606a5020 | 2780 | if (start_flush_work(work, &barr)) { |
401a8d04 TH |
2781 | wait_for_completion(&barr.done); |
2782 | destroy_work_on_stack(&barr.work); | |
2783 | return true; | |
606a5020 | 2784 | } else { |
401a8d04 | 2785 | return false; |
6e84d644 | 2786 | } |
6e84d644 | 2787 | } |
606a5020 | 2788 | EXPORT_SYMBOL_GPL(flush_work); |
6e84d644 | 2789 | |
36e227d2 | 2790 | static bool __cancel_work_timer(struct work_struct *work, bool is_dwork) |
1f1f642e | 2791 | { |
bbb68dfa | 2792 | unsigned long flags; |
1f1f642e ON |
2793 | int ret; |
2794 | ||
2795 | do { | |
bbb68dfa TH |
2796 | ret = try_to_grab_pending(work, is_dwork, &flags); |
2797 | /* | |
2798 | * If someone else is canceling, wait for the same event it | |
2799 | * would be waiting for before retrying. | |
2800 | */ | |
2801 | if (unlikely(ret == -ENOENT)) | |
606a5020 | 2802 | flush_work(work); |
1f1f642e ON |
2803 | } while (unlikely(ret < 0)); |
2804 | ||
bbb68dfa TH |
2805 | /* tell other tasks trying to grab @work to back off */ |
2806 | mark_work_canceling(work); | |
2807 | local_irq_restore(flags); | |
2808 | ||
606a5020 | 2809 | flush_work(work); |
7a22ad75 | 2810 | clear_work_data(work); |
1f1f642e ON |
2811 | return ret; |
2812 | } | |
2813 | ||
6e84d644 | 2814 | /** |
401a8d04 TH |
2815 | * cancel_work_sync - cancel a work and wait for it to finish |
2816 | * @work: the work to cancel | |
6e84d644 | 2817 | * |
401a8d04 TH |
2818 | * Cancel @work and wait for its execution to finish. This function |
2819 | * can be used even if the work re-queues itself or migrates to | |
2820 | * another workqueue. On return from this function, @work is | |
2821 | * guaranteed to be not pending or executing on any CPU. | |
1f1f642e | 2822 | * |
401a8d04 TH |
2823 | * cancel_work_sync(&delayed_work->work) must not be used for |
2824 | * delayed_work's. Use cancel_delayed_work_sync() instead. | |
6e84d644 | 2825 | * |
401a8d04 | 2826 | * The caller must ensure that the workqueue on which @work was last |
6e84d644 | 2827 | * queued can't be destroyed before this function returns. |
401a8d04 TH |
2828 | * |
2829 | * RETURNS: | |
2830 | * %true if @work was pending, %false otherwise. | |
6e84d644 | 2831 | */ |
401a8d04 | 2832 | bool cancel_work_sync(struct work_struct *work) |
6e84d644 | 2833 | { |
36e227d2 | 2834 | return __cancel_work_timer(work, false); |
b89deed3 | 2835 | } |
28e53bdd | 2836 | EXPORT_SYMBOL_GPL(cancel_work_sync); |
b89deed3 | 2837 | |
6e84d644 | 2838 | /** |
401a8d04 TH |
2839 | * flush_delayed_work - wait for a dwork to finish executing the last queueing |
2840 | * @dwork: the delayed work to flush | |
6e84d644 | 2841 | * |
401a8d04 TH |
2842 | * Delayed timer is cancelled and the pending work is queued for |
2843 | * immediate execution. Like flush_work(), this function only | |
2844 | * considers the last queueing instance of @dwork. | |
1f1f642e | 2845 | * |
401a8d04 TH |
2846 | * RETURNS: |
2847 | * %true if flush_work() waited for the work to finish execution, | |
2848 | * %false if it was already idle. | |
6e84d644 | 2849 | */ |
401a8d04 TH |
2850 | bool flush_delayed_work(struct delayed_work *dwork) |
2851 | { | |
8930caba | 2852 | local_irq_disable(); |
401a8d04 | 2853 | if (del_timer_sync(&dwork->timer)) |
60c057bc | 2854 | __queue_work(dwork->cpu, dwork->wq, &dwork->work); |
8930caba | 2855 | local_irq_enable(); |
401a8d04 TH |
2856 | return flush_work(&dwork->work); |
2857 | } | |
2858 | EXPORT_SYMBOL(flush_delayed_work); | |
2859 | ||
09383498 | 2860 | /** |
57b30ae7 TH |
2861 | * cancel_delayed_work - cancel a delayed work |
2862 | * @dwork: delayed_work to cancel | |
09383498 | 2863 | * |
57b30ae7 TH |
2864 | * Kill off a pending delayed_work. Returns %true if @dwork was pending |
2865 | * and canceled; %false if wasn't pending. Note that the work callback | |
2866 | * function may still be running on return, unless it returns %true and the | |
2867 | * work doesn't re-arm itself. Explicitly flush or use | |
2868 | * cancel_delayed_work_sync() to wait on it. | |
09383498 | 2869 | * |
57b30ae7 | 2870 | * This function is safe to call from any context including IRQ handler. |
09383498 | 2871 | */ |
57b30ae7 | 2872 | bool cancel_delayed_work(struct delayed_work *dwork) |
09383498 | 2873 | { |
57b30ae7 TH |
2874 | unsigned long flags; |
2875 | int ret; | |
2876 | ||
2877 | do { | |
2878 | ret = try_to_grab_pending(&dwork->work, true, &flags); | |
2879 | } while (unlikely(ret == -EAGAIN)); | |
2880 | ||
2881 | if (unlikely(ret < 0)) | |
2882 | return false; | |
2883 | ||
7c3eed5c TH |
2884 | set_work_pool_and_clear_pending(&dwork->work, |
2885 | get_work_pool_id(&dwork->work)); | |
57b30ae7 | 2886 | local_irq_restore(flags); |
c0158ca6 | 2887 | return ret; |
09383498 | 2888 | } |
57b30ae7 | 2889 | EXPORT_SYMBOL(cancel_delayed_work); |
09383498 | 2890 | |
401a8d04 TH |
2891 | /** |
2892 | * cancel_delayed_work_sync - cancel a delayed work and wait for it to finish | |
2893 | * @dwork: the delayed work cancel | |
2894 | * | |
2895 | * This is cancel_work_sync() for delayed works. | |
2896 | * | |
2897 | * RETURNS: | |
2898 | * %true if @dwork was pending, %false otherwise. | |
2899 | */ | |
2900 | bool cancel_delayed_work_sync(struct delayed_work *dwork) | |
6e84d644 | 2901 | { |
36e227d2 | 2902 | return __cancel_work_timer(&dwork->work, true); |
6e84d644 | 2903 | } |
f5a421a4 | 2904 | EXPORT_SYMBOL(cancel_delayed_work_sync); |
1da177e4 | 2905 | |
b6136773 | 2906 | /** |
31ddd871 | 2907 | * schedule_on_each_cpu - execute a function synchronously on each online CPU |
b6136773 | 2908 | * @func: the function to call |
b6136773 | 2909 | * |
31ddd871 TH |
2910 | * schedule_on_each_cpu() executes @func on each online CPU using the |
2911 | * system workqueue and blocks until all CPUs have completed. | |
b6136773 | 2912 | * schedule_on_each_cpu() is very slow. |
31ddd871 TH |
2913 | * |
2914 | * RETURNS: | |
2915 | * 0 on success, -errno on failure. | |
b6136773 | 2916 | */ |
65f27f38 | 2917 | int schedule_on_each_cpu(work_func_t func) |
15316ba8 CL |
2918 | { |
2919 | int cpu; | |
38f51568 | 2920 | struct work_struct __percpu *works; |
15316ba8 | 2921 | |
b6136773 AM |
2922 | works = alloc_percpu(struct work_struct); |
2923 | if (!works) | |
15316ba8 | 2924 | return -ENOMEM; |
b6136773 | 2925 | |
93981800 TH |
2926 | get_online_cpus(); |
2927 | ||
15316ba8 | 2928 | for_each_online_cpu(cpu) { |
9bfb1839 IM |
2929 | struct work_struct *work = per_cpu_ptr(works, cpu); |
2930 | ||
2931 | INIT_WORK(work, func); | |
b71ab8c2 | 2932 | schedule_work_on(cpu, work); |
65a64464 | 2933 | } |
93981800 TH |
2934 | |
2935 | for_each_online_cpu(cpu) | |
2936 | flush_work(per_cpu_ptr(works, cpu)); | |
2937 | ||
95402b38 | 2938 | put_online_cpus(); |
b6136773 | 2939 | free_percpu(works); |
15316ba8 CL |
2940 | return 0; |
2941 | } | |
2942 | ||
eef6a7d5 AS |
2943 | /** |
2944 | * flush_scheduled_work - ensure that any scheduled work has run to completion. | |
2945 | * | |
2946 | * Forces execution of the kernel-global workqueue and blocks until its | |
2947 | * completion. | |
2948 | * | |
2949 | * Think twice before calling this function! It's very easy to get into | |
2950 | * trouble if you don't take great care. Either of the following situations | |
2951 | * will lead to deadlock: | |
2952 | * | |
2953 | * One of the work items currently on the workqueue needs to acquire | |
2954 | * a lock held by your code or its caller. | |
2955 | * | |
2956 | * Your code is running in the context of a work routine. | |
2957 | * | |
2958 | * They will be detected by lockdep when they occur, but the first might not | |
2959 | * occur very often. It depends on what work items are on the workqueue and | |
2960 | * what locks they need, which you have no control over. | |
2961 | * | |
2962 | * In most situations flushing the entire workqueue is overkill; you merely | |
2963 | * need to know that a particular work item isn't queued and isn't running. | |
2964 | * In such cases you should use cancel_delayed_work_sync() or | |
2965 | * cancel_work_sync() instead. | |
2966 | */ | |
1da177e4 LT |
2967 | void flush_scheduled_work(void) |
2968 | { | |
d320c038 | 2969 | flush_workqueue(system_wq); |
1da177e4 | 2970 | } |
ae90dd5d | 2971 | EXPORT_SYMBOL(flush_scheduled_work); |
1da177e4 | 2972 | |
1fa44eca JB |
2973 | /** |
2974 | * execute_in_process_context - reliably execute the routine with user context | |
2975 | * @fn: the function to execute | |
1fa44eca JB |
2976 | * @ew: guaranteed storage for the execute work structure (must |
2977 | * be available when the work executes) | |
2978 | * | |
2979 | * Executes the function immediately if process context is available, | |
2980 | * otherwise schedules the function for delayed execution. | |
2981 | * | |
2982 | * Returns: 0 - function was executed | |
2983 | * 1 - function was scheduled for execution | |
2984 | */ | |
65f27f38 | 2985 | int execute_in_process_context(work_func_t fn, struct execute_work *ew) |
1fa44eca JB |
2986 | { |
2987 | if (!in_interrupt()) { | |
65f27f38 | 2988 | fn(&ew->work); |
1fa44eca JB |
2989 | return 0; |
2990 | } | |
2991 | ||
65f27f38 | 2992 | INIT_WORK(&ew->work, fn); |
1fa44eca JB |
2993 | schedule_work(&ew->work); |
2994 | ||
2995 | return 1; | |
2996 | } | |
2997 | EXPORT_SYMBOL_GPL(execute_in_process_context); | |
2998 | ||
226223ab TH |
2999 | #ifdef CONFIG_SYSFS |
3000 | /* | |
3001 | * Workqueues with WQ_SYSFS flag set is visible to userland via | |
3002 | * /sys/bus/workqueue/devices/WQ_NAME. All visible workqueues have the | |
3003 | * following attributes. | |
3004 | * | |
3005 | * per_cpu RO bool : whether the workqueue is per-cpu or unbound | |
3006 | * max_active RW int : maximum number of in-flight work items | |
3007 | * | |
3008 | * Unbound workqueues have the following extra attributes. | |
3009 | * | |
3010 | * id RO int : the associated pool ID | |
3011 | * nice RW int : nice value of the workers | |
3012 | * cpumask RW mask : bitmask of allowed CPUs for the workers | |
3013 | */ | |
3014 | struct wq_device { | |
3015 | struct workqueue_struct *wq; | |
3016 | struct device dev; | |
3017 | }; | |
3018 | ||
3019 | static struct workqueue_struct *dev_to_wq(struct device *dev) | |
3020 | { | |
3021 | struct wq_device *wq_dev = container_of(dev, struct wq_device, dev); | |
3022 | ||
3023 | return wq_dev->wq; | |
3024 | } | |
3025 | ||
3026 | static ssize_t wq_per_cpu_show(struct device *dev, | |
3027 | struct device_attribute *attr, char *buf) | |
3028 | { | |
3029 | struct workqueue_struct *wq = dev_to_wq(dev); | |
3030 | ||
3031 | return scnprintf(buf, PAGE_SIZE, "%d\n", (bool)!(wq->flags & WQ_UNBOUND)); | |
3032 | } | |
3033 | ||
3034 | static ssize_t wq_max_active_show(struct device *dev, | |
3035 | struct device_attribute *attr, char *buf) | |
3036 | { | |
3037 | struct workqueue_struct *wq = dev_to_wq(dev); | |
3038 | ||
3039 | return scnprintf(buf, PAGE_SIZE, "%d\n", wq->saved_max_active); | |
3040 | } | |
3041 | ||
3042 | static ssize_t wq_max_active_store(struct device *dev, | |
3043 | struct device_attribute *attr, | |
3044 | const char *buf, size_t count) | |
3045 | { | |
3046 | struct workqueue_struct *wq = dev_to_wq(dev); | |
3047 | int val; | |
3048 | ||
3049 | if (sscanf(buf, "%d", &val) != 1 || val <= 0) | |
3050 | return -EINVAL; | |
3051 | ||
3052 | workqueue_set_max_active(wq, val); | |
3053 | return count; | |
3054 | } | |
3055 | ||
3056 | static struct device_attribute wq_sysfs_attrs[] = { | |
3057 | __ATTR(per_cpu, 0444, wq_per_cpu_show, NULL), | |
3058 | __ATTR(max_active, 0644, wq_max_active_show, wq_max_active_store), | |
3059 | __ATTR_NULL, | |
3060 | }; | |
3061 | ||
3062 | static ssize_t wq_pool_id_show(struct device *dev, | |
3063 | struct device_attribute *attr, char *buf) | |
3064 | { | |
3065 | struct workqueue_struct *wq = dev_to_wq(dev); | |
3066 | struct worker_pool *pool; | |
3067 | int written; | |
3068 | ||
3069 | rcu_read_lock_sched(); | |
3070 | pool = first_pwq(wq)->pool; | |
3071 | written = scnprintf(buf, PAGE_SIZE, "%d\n", pool->id); | |
3072 | rcu_read_unlock_sched(); | |
3073 | ||
3074 | return written; | |
3075 | } | |
3076 | ||
3077 | static ssize_t wq_nice_show(struct device *dev, struct device_attribute *attr, | |
3078 | char *buf) | |
3079 | { | |
3080 | struct workqueue_struct *wq = dev_to_wq(dev); | |
3081 | int written; | |
3082 | ||
3083 | rcu_read_lock_sched(); | |
3084 | written = scnprintf(buf, PAGE_SIZE, "%d\n", | |
3085 | first_pwq(wq)->pool->attrs->nice); | |
3086 | rcu_read_unlock_sched(); | |
3087 | ||
3088 | return written; | |
3089 | } | |
3090 | ||
3091 | /* prepare workqueue_attrs for sysfs store operations */ | |
3092 | static struct workqueue_attrs *wq_sysfs_prep_attrs(struct workqueue_struct *wq) | |
3093 | { | |
3094 | struct workqueue_attrs *attrs; | |
3095 | ||
3096 | attrs = alloc_workqueue_attrs(GFP_KERNEL); | |
3097 | if (!attrs) | |
3098 | return NULL; | |
3099 | ||
3100 | rcu_read_lock_sched(); | |
3101 | copy_workqueue_attrs(attrs, first_pwq(wq)->pool->attrs); | |
3102 | rcu_read_unlock_sched(); | |
3103 | return attrs; | |
3104 | } | |
3105 | ||
3106 | static ssize_t wq_nice_store(struct device *dev, struct device_attribute *attr, | |
3107 | const char *buf, size_t count) | |
3108 | { | |
3109 | struct workqueue_struct *wq = dev_to_wq(dev); | |
3110 | struct workqueue_attrs *attrs; | |
3111 | int ret; | |
3112 | ||
3113 | attrs = wq_sysfs_prep_attrs(wq); | |
3114 | if (!attrs) | |
3115 | return -ENOMEM; | |
3116 | ||
3117 | if (sscanf(buf, "%d", &attrs->nice) == 1 && | |
3118 | attrs->nice >= -20 && attrs->nice <= 19) | |
3119 | ret = apply_workqueue_attrs(wq, attrs); | |
3120 | else | |
3121 | ret = -EINVAL; | |
3122 | ||
3123 | free_workqueue_attrs(attrs); | |
3124 | return ret ?: count; | |
3125 | } | |
3126 | ||
3127 | static ssize_t wq_cpumask_show(struct device *dev, | |
3128 | struct device_attribute *attr, char *buf) | |
3129 | { | |
3130 | struct workqueue_struct *wq = dev_to_wq(dev); | |
3131 | int written; | |
3132 | ||
3133 | rcu_read_lock_sched(); | |
3134 | written = cpumask_scnprintf(buf, PAGE_SIZE, | |
3135 | first_pwq(wq)->pool->attrs->cpumask); | |
3136 | rcu_read_unlock_sched(); | |
3137 | ||
3138 | written += scnprintf(buf + written, PAGE_SIZE - written, "\n"); | |
3139 | return written; | |
3140 | } | |
3141 | ||
3142 | static ssize_t wq_cpumask_store(struct device *dev, | |
3143 | struct device_attribute *attr, | |
3144 | const char *buf, size_t count) | |
3145 | { | |
3146 | struct workqueue_struct *wq = dev_to_wq(dev); | |
3147 | struct workqueue_attrs *attrs; | |
3148 | int ret; | |
3149 | ||
3150 | attrs = wq_sysfs_prep_attrs(wq); | |
3151 | if (!attrs) | |
3152 | return -ENOMEM; | |
3153 | ||
3154 | ret = cpumask_parse(buf, attrs->cpumask); | |
3155 | if (!ret) | |
3156 | ret = apply_workqueue_attrs(wq, attrs); | |
3157 | ||
3158 | free_workqueue_attrs(attrs); | |
3159 | return ret ?: count; | |
3160 | } | |
3161 | ||
3162 | static struct device_attribute wq_sysfs_unbound_attrs[] = { | |
3163 | __ATTR(pool_id, 0444, wq_pool_id_show, NULL), | |
3164 | __ATTR(nice, 0644, wq_nice_show, wq_nice_store), | |
3165 | __ATTR(cpumask, 0644, wq_cpumask_show, wq_cpumask_store), | |
3166 | __ATTR_NULL, | |
3167 | }; | |
3168 | ||
3169 | static struct bus_type wq_subsys = { | |
3170 | .name = "workqueue", | |
3171 | .dev_attrs = wq_sysfs_attrs, | |
3172 | }; | |
3173 | ||
3174 | static int __init wq_sysfs_init(void) | |
3175 | { | |
3176 | return subsys_virtual_register(&wq_subsys, NULL); | |
3177 | } | |
3178 | core_initcall(wq_sysfs_init); | |
3179 | ||
3180 | static void wq_device_release(struct device *dev) | |
3181 | { | |
3182 | struct wq_device *wq_dev = container_of(dev, struct wq_device, dev); | |
3183 | ||
3184 | kfree(wq_dev); | |
3185 | } | |
3186 | ||
3187 | /** | |
3188 | * workqueue_sysfs_register - make a workqueue visible in sysfs | |
3189 | * @wq: the workqueue to register | |
3190 | * | |
3191 | * Expose @wq in sysfs under /sys/bus/workqueue/devices. | |
3192 | * alloc_workqueue*() automatically calls this function if WQ_SYSFS is set | |
3193 | * which is the preferred method. | |
3194 | * | |
3195 | * Workqueue user should use this function directly iff it wants to apply | |
3196 | * workqueue_attrs before making the workqueue visible in sysfs; otherwise, | |
3197 | * apply_workqueue_attrs() may race against userland updating the | |
3198 | * attributes. | |
3199 | * | |
3200 | * Returns 0 on success, -errno on failure. | |
3201 | */ | |
3202 | int workqueue_sysfs_register(struct workqueue_struct *wq) | |
3203 | { | |
3204 | struct wq_device *wq_dev; | |
3205 | int ret; | |
3206 | ||
3207 | /* | |
3208 | * Adjusting max_active or creating new pwqs by applyting | |
3209 | * attributes breaks ordering guarantee. Disallow exposing ordered | |
3210 | * workqueues. | |
3211 | */ | |
3212 | if (WARN_ON(wq->flags & __WQ_ORDERED)) | |
3213 | return -EINVAL; | |
3214 | ||
3215 | wq->wq_dev = wq_dev = kzalloc(sizeof(*wq_dev), GFP_KERNEL); | |
3216 | if (!wq_dev) | |
3217 | return -ENOMEM; | |
3218 | ||
3219 | wq_dev->wq = wq; | |
3220 | wq_dev->dev.bus = &wq_subsys; | |
3221 | wq_dev->dev.init_name = wq->name; | |
3222 | wq_dev->dev.release = wq_device_release; | |
3223 | ||
3224 | /* | |
3225 | * unbound_attrs are created separately. Suppress uevent until | |
3226 | * everything is ready. | |
3227 | */ | |
3228 | dev_set_uevent_suppress(&wq_dev->dev, true); | |
3229 | ||
3230 | ret = device_register(&wq_dev->dev); | |
3231 | if (ret) { | |
3232 | kfree(wq_dev); | |
3233 | wq->wq_dev = NULL; | |
3234 | return ret; | |
3235 | } | |
3236 | ||
3237 | if (wq->flags & WQ_UNBOUND) { | |
3238 | struct device_attribute *attr; | |
3239 | ||
3240 | for (attr = wq_sysfs_unbound_attrs; attr->attr.name; attr++) { | |
3241 | ret = device_create_file(&wq_dev->dev, attr); | |
3242 | if (ret) { | |
3243 | device_unregister(&wq_dev->dev); | |
3244 | wq->wq_dev = NULL; | |
3245 | return ret; | |
3246 | } | |
3247 | } | |
3248 | } | |
3249 | ||
3250 | kobject_uevent(&wq_dev->dev.kobj, KOBJ_ADD); | |
3251 | return 0; | |
3252 | } | |
3253 | ||
3254 | /** | |
3255 | * workqueue_sysfs_unregister - undo workqueue_sysfs_register() | |
3256 | * @wq: the workqueue to unregister | |
3257 | * | |
3258 | * If @wq is registered to sysfs by workqueue_sysfs_register(), unregister. | |
3259 | */ | |
3260 | static void workqueue_sysfs_unregister(struct workqueue_struct *wq) | |
3261 | { | |
3262 | struct wq_device *wq_dev = wq->wq_dev; | |
3263 | ||
3264 | if (!wq->wq_dev) | |
3265 | return; | |
3266 | ||
3267 | wq->wq_dev = NULL; | |
3268 | device_unregister(&wq_dev->dev); | |
3269 | } | |
3270 | #else /* CONFIG_SYSFS */ | |
3271 | static void workqueue_sysfs_unregister(struct workqueue_struct *wq) { } | |
3272 | #endif /* CONFIG_SYSFS */ | |
3273 | ||
7a4e344c TH |
3274 | /** |
3275 | * free_workqueue_attrs - free a workqueue_attrs | |
3276 | * @attrs: workqueue_attrs to free | |
3277 | * | |
3278 | * Undo alloc_workqueue_attrs(). | |
3279 | */ | |
3280 | void free_workqueue_attrs(struct workqueue_attrs *attrs) | |
3281 | { | |
3282 | if (attrs) { | |
3283 | free_cpumask_var(attrs->cpumask); | |
3284 | kfree(attrs); | |
3285 | } | |
3286 | } | |
3287 | ||
3288 | /** | |
3289 | * alloc_workqueue_attrs - allocate a workqueue_attrs | |
3290 | * @gfp_mask: allocation mask to use | |
3291 | * | |
3292 | * Allocate a new workqueue_attrs, initialize with default settings and | |
3293 | * return it. Returns NULL on failure. | |
3294 | */ | |
3295 | struct workqueue_attrs *alloc_workqueue_attrs(gfp_t gfp_mask) | |
3296 | { | |
3297 | struct workqueue_attrs *attrs; | |
3298 | ||
3299 | attrs = kzalloc(sizeof(*attrs), gfp_mask); | |
3300 | if (!attrs) | |
3301 | goto fail; | |
3302 | if (!alloc_cpumask_var(&attrs->cpumask, gfp_mask)) | |
3303 | goto fail; | |
3304 | ||
13e2e556 | 3305 | cpumask_copy(attrs->cpumask, cpu_possible_mask); |
7a4e344c TH |
3306 | return attrs; |
3307 | fail: | |
3308 | free_workqueue_attrs(attrs); | |
3309 | return NULL; | |
3310 | } | |
3311 | ||
29c91e99 TH |
3312 | static void copy_workqueue_attrs(struct workqueue_attrs *to, |
3313 | const struct workqueue_attrs *from) | |
3314 | { | |
3315 | to->nice = from->nice; | |
3316 | cpumask_copy(to->cpumask, from->cpumask); | |
3317 | } | |
3318 | ||
29c91e99 TH |
3319 | /* hash value of the content of @attr */ |
3320 | static u32 wqattrs_hash(const struct workqueue_attrs *attrs) | |
3321 | { | |
3322 | u32 hash = 0; | |
3323 | ||
3324 | hash = jhash_1word(attrs->nice, hash); | |
13e2e556 TH |
3325 | hash = jhash(cpumask_bits(attrs->cpumask), |
3326 | BITS_TO_LONGS(nr_cpumask_bits) * sizeof(long), hash); | |
29c91e99 TH |
3327 | return hash; |
3328 | } | |
3329 | ||
3330 | /* content equality test */ | |
3331 | static bool wqattrs_equal(const struct workqueue_attrs *a, | |
3332 | const struct workqueue_attrs *b) | |
3333 | { | |
3334 | if (a->nice != b->nice) | |
3335 | return false; | |
3336 | if (!cpumask_equal(a->cpumask, b->cpumask)) | |
3337 | return false; | |
3338 | return true; | |
3339 | } | |
3340 | ||
7a4e344c TH |
3341 | /** |
3342 | * init_worker_pool - initialize a newly zalloc'd worker_pool | |
3343 | * @pool: worker_pool to initialize | |
3344 | * | |
3345 | * Initiailize a newly zalloc'd @pool. It also allocates @pool->attrs. | |
29c91e99 TH |
3346 | * Returns 0 on success, -errno on failure. Even on failure, all fields |
3347 | * inside @pool proper are initialized and put_unbound_pool() can be called | |
3348 | * on @pool safely to release it. | |
7a4e344c TH |
3349 | */ |
3350 | static int init_worker_pool(struct worker_pool *pool) | |
4e1a1f9a TH |
3351 | { |
3352 | spin_lock_init(&pool->lock); | |
29c91e99 TH |
3353 | pool->id = -1; |
3354 | pool->cpu = -1; | |
4e1a1f9a TH |
3355 | pool->flags |= POOL_DISASSOCIATED; |
3356 | INIT_LIST_HEAD(&pool->worklist); | |
3357 | INIT_LIST_HEAD(&pool->idle_list); | |
3358 | hash_init(pool->busy_hash); | |
3359 | ||
3360 | init_timer_deferrable(&pool->idle_timer); | |
3361 | pool->idle_timer.function = idle_worker_timeout; | |
3362 | pool->idle_timer.data = (unsigned long)pool; | |
3363 | ||
3364 | setup_timer(&pool->mayday_timer, pool_mayday_timeout, | |
3365 | (unsigned long)pool); | |
3366 | ||
3367 | mutex_init(&pool->manager_arb); | |
bc3a1afc | 3368 | mutex_init(&pool->manager_mutex); |
822d8405 | 3369 | idr_init(&pool->worker_idr); |
7a4e344c | 3370 | |
29c91e99 TH |
3371 | INIT_HLIST_NODE(&pool->hash_node); |
3372 | pool->refcnt = 1; | |
3373 | ||
3374 | /* shouldn't fail above this point */ | |
7a4e344c TH |
3375 | pool->attrs = alloc_workqueue_attrs(GFP_KERNEL); |
3376 | if (!pool->attrs) | |
3377 | return -ENOMEM; | |
3378 | return 0; | |
4e1a1f9a TH |
3379 | } |
3380 | ||
29c91e99 TH |
3381 | static void rcu_free_pool(struct rcu_head *rcu) |
3382 | { | |
3383 | struct worker_pool *pool = container_of(rcu, struct worker_pool, rcu); | |
3384 | ||
822d8405 | 3385 | idr_destroy(&pool->worker_idr); |
29c91e99 TH |
3386 | free_workqueue_attrs(pool->attrs); |
3387 | kfree(pool); | |
3388 | } | |
3389 | ||
3390 | /** | |
3391 | * put_unbound_pool - put a worker_pool | |
3392 | * @pool: worker_pool to put | |
3393 | * | |
3394 | * Put @pool. If its refcnt reaches zero, it gets destroyed in sched-RCU | |
c5aa87bb TH |
3395 | * safe manner. get_unbound_pool() calls this function on its failure path |
3396 | * and this function should be able to release pools which went through, | |
3397 | * successfully or not, init_worker_pool(). | |
a892cacc TH |
3398 | * |
3399 | * Should be called with wq_pool_mutex held. | |
29c91e99 TH |
3400 | */ |
3401 | static void put_unbound_pool(struct worker_pool *pool) | |
3402 | { | |
3403 | struct worker *worker; | |
3404 | ||
a892cacc TH |
3405 | lockdep_assert_held(&wq_pool_mutex); |
3406 | ||
3407 | if (--pool->refcnt) | |
29c91e99 | 3408 | return; |
29c91e99 TH |
3409 | |
3410 | /* sanity checks */ | |
3411 | if (WARN_ON(!(pool->flags & POOL_DISASSOCIATED)) || | |
a892cacc | 3412 | WARN_ON(!list_empty(&pool->worklist))) |
29c91e99 | 3413 | return; |
29c91e99 TH |
3414 | |
3415 | /* release id and unhash */ | |
3416 | if (pool->id >= 0) | |
3417 | idr_remove(&worker_pool_idr, pool->id); | |
3418 | hash_del(&pool->hash_node); | |
3419 | ||
c5aa87bb TH |
3420 | /* |
3421 | * Become the manager and destroy all workers. Grabbing | |
3422 | * manager_arb prevents @pool's workers from blocking on | |
3423 | * manager_mutex. | |
3424 | */ | |
29c91e99 | 3425 | mutex_lock(&pool->manager_arb); |
cd549687 | 3426 | mutex_lock(&pool->manager_mutex); |
29c91e99 TH |
3427 | spin_lock_irq(&pool->lock); |
3428 | ||
3429 | while ((worker = first_worker(pool))) | |
3430 | destroy_worker(worker); | |
3431 | WARN_ON(pool->nr_workers || pool->nr_idle); | |
3432 | ||
3433 | spin_unlock_irq(&pool->lock); | |
cd549687 | 3434 | mutex_unlock(&pool->manager_mutex); |
29c91e99 TH |
3435 | mutex_unlock(&pool->manager_arb); |
3436 | ||
3437 | /* shut down the timers */ | |
3438 | del_timer_sync(&pool->idle_timer); | |
3439 | del_timer_sync(&pool->mayday_timer); | |
3440 | ||
3441 | /* sched-RCU protected to allow dereferences from get_work_pool() */ | |
3442 | call_rcu_sched(&pool->rcu, rcu_free_pool); | |
3443 | } | |
3444 | ||
3445 | /** | |
3446 | * get_unbound_pool - get a worker_pool with the specified attributes | |
3447 | * @attrs: the attributes of the worker_pool to get | |
3448 | * | |
3449 | * Obtain a worker_pool which has the same attributes as @attrs, bump the | |
3450 | * reference count and return it. If there already is a matching | |
3451 | * worker_pool, it will be used; otherwise, this function attempts to | |
3452 | * create a new one. On failure, returns NULL. | |
a892cacc TH |
3453 | * |
3454 | * Should be called with wq_pool_mutex held. | |
29c91e99 TH |
3455 | */ |
3456 | static struct worker_pool *get_unbound_pool(const struct workqueue_attrs *attrs) | |
3457 | { | |
29c91e99 TH |
3458 | u32 hash = wqattrs_hash(attrs); |
3459 | struct worker_pool *pool; | |
29c91e99 | 3460 | |
a892cacc | 3461 | lockdep_assert_held(&wq_pool_mutex); |
29c91e99 TH |
3462 | |
3463 | /* do we already have a matching pool? */ | |
29c91e99 TH |
3464 | hash_for_each_possible(unbound_pool_hash, pool, hash_node, hash) { |
3465 | if (wqattrs_equal(pool->attrs, attrs)) { | |
3466 | pool->refcnt++; | |
3467 | goto out_unlock; | |
3468 | } | |
3469 | } | |
29c91e99 TH |
3470 | |
3471 | /* nope, create a new one */ | |
3472 | pool = kzalloc(sizeof(*pool), GFP_KERNEL); | |
3473 | if (!pool || init_worker_pool(pool) < 0) | |
3474 | goto fail; | |
3475 | ||
12ee4fc6 LJ |
3476 | if (workqueue_freezing) |
3477 | pool->flags |= POOL_FREEZING; | |
3478 | ||
8864b4e5 | 3479 | lockdep_set_subclass(&pool->lock, 1); /* see put_pwq() */ |
29c91e99 TH |
3480 | copy_workqueue_attrs(pool->attrs, attrs); |
3481 | ||
3482 | if (worker_pool_assign_id(pool) < 0) | |
3483 | goto fail; | |
3484 | ||
3485 | /* create and start the initial worker */ | |
ebf44d16 | 3486 | if (create_and_start_worker(pool) < 0) |
29c91e99 TH |
3487 | goto fail; |
3488 | ||
29c91e99 | 3489 | /* install */ |
29c91e99 TH |
3490 | hash_add(unbound_pool_hash, &pool->hash_node, hash); |
3491 | out_unlock: | |
29c91e99 TH |
3492 | return pool; |
3493 | fail: | |
29c91e99 TH |
3494 | if (pool) |
3495 | put_unbound_pool(pool); | |
3496 | return NULL; | |
3497 | } | |
3498 | ||
8864b4e5 TH |
3499 | static void rcu_free_pwq(struct rcu_head *rcu) |
3500 | { | |
3501 | kmem_cache_free(pwq_cache, | |
3502 | container_of(rcu, struct pool_workqueue, rcu)); | |
3503 | } | |
3504 | ||
3505 | /* | |
3506 | * Scheduled on system_wq by put_pwq() when an unbound pwq hits zero refcnt | |
3507 | * and needs to be destroyed. | |
3508 | */ | |
3509 | static void pwq_unbound_release_workfn(struct work_struct *work) | |
3510 | { | |
3511 | struct pool_workqueue *pwq = container_of(work, struct pool_workqueue, | |
3512 | unbound_release_work); | |
3513 | struct workqueue_struct *wq = pwq->wq; | |
3514 | struct worker_pool *pool = pwq->pool; | |
bc0caf09 | 3515 | bool is_last; |
8864b4e5 TH |
3516 | |
3517 | if (WARN_ON_ONCE(!(wq->flags & WQ_UNBOUND))) | |
3518 | return; | |
3519 | ||
75ccf595 | 3520 | /* |
3c25a55d | 3521 | * Unlink @pwq. Synchronization against wq->mutex isn't strictly |
75ccf595 TH |
3522 | * necessary on release but do it anyway. It's easier to verify |
3523 | * and consistent with the linking path. | |
3524 | */ | |
3c25a55d | 3525 | mutex_lock(&wq->mutex); |
8864b4e5 | 3526 | list_del_rcu(&pwq->pwqs_node); |
bc0caf09 | 3527 | is_last = list_empty(&wq->pwqs); |
3c25a55d | 3528 | mutex_unlock(&wq->mutex); |
8864b4e5 | 3529 | |
a892cacc | 3530 | mutex_lock(&wq_pool_mutex); |
8864b4e5 | 3531 | put_unbound_pool(pool); |
a892cacc TH |
3532 | mutex_unlock(&wq_pool_mutex); |
3533 | ||
8864b4e5 TH |
3534 | call_rcu_sched(&pwq->rcu, rcu_free_pwq); |
3535 | ||
3536 | /* | |
3537 | * If we're the last pwq going away, @wq is already dead and no one | |
3538 | * is gonna access it anymore. Free it. | |
3539 | */ | |
bc0caf09 | 3540 | if (is_last) |
8864b4e5 TH |
3541 | kfree(wq); |
3542 | } | |
3543 | ||
0fbd95aa | 3544 | /** |
699ce097 | 3545 | * pwq_adjust_max_active - update a pwq's max_active to the current setting |
0fbd95aa | 3546 | * @pwq: target pool_workqueue |
0fbd95aa | 3547 | * |
699ce097 TH |
3548 | * If @pwq isn't freezing, set @pwq->max_active to the associated |
3549 | * workqueue's saved_max_active and activate delayed work items | |
3550 | * accordingly. If @pwq is freezing, clear @pwq->max_active to zero. | |
0fbd95aa | 3551 | */ |
699ce097 | 3552 | static void pwq_adjust_max_active(struct pool_workqueue *pwq) |
0fbd95aa | 3553 | { |
699ce097 TH |
3554 | struct workqueue_struct *wq = pwq->wq; |
3555 | bool freezable = wq->flags & WQ_FREEZABLE; | |
3556 | ||
3557 | /* for @wq->saved_max_active */ | |
a357fc03 | 3558 | lockdep_assert_held(&wq->mutex); |
699ce097 TH |
3559 | |
3560 | /* fast exit for non-freezable wqs */ | |
3561 | if (!freezable && pwq->max_active == wq->saved_max_active) | |
3562 | return; | |
3563 | ||
a357fc03 | 3564 | spin_lock_irq(&pwq->pool->lock); |
699ce097 TH |
3565 | |
3566 | if (!freezable || !(pwq->pool->flags & POOL_FREEZING)) { | |
3567 | pwq->max_active = wq->saved_max_active; | |
0fbd95aa | 3568 | |
699ce097 TH |
3569 | while (!list_empty(&pwq->delayed_works) && |
3570 | pwq->nr_active < pwq->max_active) | |
3571 | pwq_activate_first_delayed(pwq); | |
951a078a LJ |
3572 | |
3573 | /* | |
3574 | * Need to kick a worker after thawed or an unbound wq's | |
3575 | * max_active is bumped. It's a slow path. Do it always. | |
3576 | */ | |
3577 | wake_up_worker(pwq->pool); | |
699ce097 TH |
3578 | } else { |
3579 | pwq->max_active = 0; | |
3580 | } | |
3581 | ||
a357fc03 | 3582 | spin_unlock_irq(&pwq->pool->lock); |
0fbd95aa TH |
3583 | } |
3584 | ||
d2c1d404 TH |
3585 | static void init_and_link_pwq(struct pool_workqueue *pwq, |
3586 | struct workqueue_struct *wq, | |
9e8cd2f5 TH |
3587 | struct worker_pool *pool, |
3588 | struct pool_workqueue **p_last_pwq) | |
d2c1d404 TH |
3589 | { |
3590 | BUG_ON((unsigned long)pwq & WORK_STRUCT_FLAG_MASK); | |
3591 | ||
3592 | pwq->pool = pool; | |
3593 | pwq->wq = wq; | |
3594 | pwq->flush_color = -1; | |
8864b4e5 | 3595 | pwq->refcnt = 1; |
d2c1d404 TH |
3596 | INIT_LIST_HEAD(&pwq->delayed_works); |
3597 | INIT_LIST_HEAD(&pwq->mayday_node); | |
8864b4e5 | 3598 | INIT_WORK(&pwq->unbound_release_work, pwq_unbound_release_workfn); |
d2c1d404 | 3599 | |
3c25a55d | 3600 | mutex_lock(&wq->mutex); |
75ccf595 | 3601 | |
983ca25e TH |
3602 | /* |
3603 | * Set the matching work_color. This is synchronized with | |
3c25a55d | 3604 | * wq->mutex to avoid confusing flush_workqueue(). |
983ca25e | 3605 | */ |
9e8cd2f5 TH |
3606 | if (p_last_pwq) |
3607 | *p_last_pwq = first_pwq(wq); | |
75ccf595 | 3608 | pwq->work_color = wq->work_color; |
983ca25e TH |
3609 | |
3610 | /* sync max_active to the current setting */ | |
3611 | pwq_adjust_max_active(pwq); | |
3612 | ||
3613 | /* link in @pwq */ | |
9e8cd2f5 | 3614 | list_add_rcu(&pwq->pwqs_node, &wq->pwqs); |
a357fc03 | 3615 | |
3c25a55d | 3616 | mutex_unlock(&wq->mutex); |
d2c1d404 TH |
3617 | } |
3618 | ||
9e8cd2f5 TH |
3619 | /** |
3620 | * apply_workqueue_attrs - apply new workqueue_attrs to an unbound workqueue | |
3621 | * @wq: the target workqueue | |
3622 | * @attrs: the workqueue_attrs to apply, allocated with alloc_workqueue_attrs() | |
3623 | * | |
3624 | * Apply @attrs to an unbound workqueue @wq. If @attrs doesn't match the | |
3625 | * current attributes, a new pwq is created and made the first pwq which | |
3626 | * will serve all new work items. Older pwqs are released as in-flight | |
3627 | * work items finish. Note that a work item which repeatedly requeues | |
3628 | * itself back-to-back will stay on its current pwq. | |
3629 | * | |
3630 | * Performs GFP_KERNEL allocations. Returns 0 on success and -errno on | |
3631 | * failure. | |
3632 | */ | |
3633 | int apply_workqueue_attrs(struct workqueue_struct *wq, | |
3634 | const struct workqueue_attrs *attrs) | |
3635 | { | |
13e2e556 TH |
3636 | struct workqueue_attrs *new_attrs; |
3637 | struct pool_workqueue *pwq = NULL, *last_pwq; | |
9e8cd2f5 | 3638 | struct worker_pool *pool; |
4862125b | 3639 | int ret; |
9e8cd2f5 | 3640 | |
8719dcea | 3641 | /* only unbound workqueues can change attributes */ |
9e8cd2f5 TH |
3642 | if (WARN_ON(!(wq->flags & WQ_UNBOUND))) |
3643 | return -EINVAL; | |
3644 | ||
8719dcea TH |
3645 | /* creating multiple pwqs breaks ordering guarantee */ |
3646 | if (WARN_ON((wq->flags & __WQ_ORDERED) && !list_empty(&wq->pwqs))) | |
3647 | return -EINVAL; | |
3648 | ||
13e2e556 TH |
3649 | /* make a copy of @attrs and sanitize it */ |
3650 | new_attrs = alloc_workqueue_attrs(GFP_KERNEL); | |
3651 | if (!new_attrs) | |
3652 | goto enomem; | |
3653 | ||
3654 | copy_workqueue_attrs(new_attrs, attrs); | |
3655 | cpumask_and(new_attrs->cpumask, new_attrs->cpumask, cpu_possible_mask); | |
3656 | ||
a892cacc TH |
3657 | mutex_lock(&wq_pool_mutex); |
3658 | ||
9e8cd2f5 | 3659 | pwq = kmem_cache_zalloc(pwq_cache, GFP_KERNEL); |
a892cacc TH |
3660 | if (!pwq) { |
3661 | mutex_unlock(&wq_pool_mutex); | |
13e2e556 | 3662 | goto enomem; |
a892cacc | 3663 | } |
9e8cd2f5 | 3664 | |
13e2e556 | 3665 | pool = get_unbound_pool(new_attrs); |
a892cacc TH |
3666 | if (!pool) { |
3667 | mutex_unlock(&wq_pool_mutex); | |
13e2e556 | 3668 | goto enomem; |
a892cacc TH |
3669 | } |
3670 | ||
3671 | mutex_unlock(&wq_pool_mutex); | |
9e8cd2f5 TH |
3672 | |
3673 | init_and_link_pwq(pwq, wq, pool, &last_pwq); | |
3674 | if (last_pwq) { | |
3675 | spin_lock_irq(&last_pwq->pool->lock); | |
3676 | put_pwq(last_pwq); | |
3677 | spin_unlock_irq(&last_pwq->pool->lock); | |
3678 | } | |
3679 | ||
4862125b TH |
3680 | ret = 0; |
3681 | /* fall through */ | |
3682 | out_free: | |
3683 | free_workqueue_attrs(new_attrs); | |
3684 | return ret; | |
13e2e556 TH |
3685 | |
3686 | enomem: | |
3687 | kmem_cache_free(pwq_cache, pwq); | |
4862125b TH |
3688 | ret = -ENOMEM; |
3689 | goto out_free; | |
9e8cd2f5 TH |
3690 | } |
3691 | ||
30cdf249 | 3692 | static int alloc_and_link_pwqs(struct workqueue_struct *wq) |
0f900049 | 3693 | { |
49e3cf44 | 3694 | bool highpri = wq->flags & WQ_HIGHPRI; |
30cdf249 TH |
3695 | int cpu; |
3696 | ||
3697 | if (!(wq->flags & WQ_UNBOUND)) { | |
420c0ddb TH |
3698 | wq->cpu_pwqs = alloc_percpu(struct pool_workqueue); |
3699 | if (!wq->cpu_pwqs) | |
30cdf249 TH |
3700 | return -ENOMEM; |
3701 | ||
3702 | for_each_possible_cpu(cpu) { | |
7fb98ea7 TH |
3703 | struct pool_workqueue *pwq = |
3704 | per_cpu_ptr(wq->cpu_pwqs, cpu); | |
7a62c2c8 | 3705 | struct worker_pool *cpu_pools = |
f02ae73a | 3706 | per_cpu(cpu_worker_pools, cpu); |
f3421797 | 3707 | |
9e8cd2f5 | 3708 | init_and_link_pwq(pwq, wq, &cpu_pools[highpri], NULL); |
30cdf249 | 3709 | } |
9e8cd2f5 | 3710 | return 0; |
30cdf249 | 3711 | } else { |
9e8cd2f5 | 3712 | return apply_workqueue_attrs(wq, unbound_std_wq_attrs[highpri]); |
30cdf249 | 3713 | } |
0f900049 TH |
3714 | } |
3715 | ||
f3421797 TH |
3716 | static int wq_clamp_max_active(int max_active, unsigned int flags, |
3717 | const char *name) | |
b71ab8c2 | 3718 | { |
f3421797 TH |
3719 | int lim = flags & WQ_UNBOUND ? WQ_UNBOUND_MAX_ACTIVE : WQ_MAX_ACTIVE; |
3720 | ||
3721 | if (max_active < 1 || max_active > lim) | |
044c782c VI |
3722 | pr_warn("workqueue: max_active %d requested for %s is out of range, clamping between %d and %d\n", |
3723 | max_active, name, 1, lim); | |
b71ab8c2 | 3724 | |
f3421797 | 3725 | return clamp_val(max_active, 1, lim); |
b71ab8c2 TH |
3726 | } |
3727 | ||
b196be89 | 3728 | struct workqueue_struct *__alloc_workqueue_key(const char *fmt, |
d320c038 TH |
3729 | unsigned int flags, |
3730 | int max_active, | |
3731 | struct lock_class_key *key, | |
b196be89 | 3732 | const char *lock_name, ...) |
1da177e4 | 3733 | { |
b196be89 | 3734 | va_list args, args1; |
1da177e4 | 3735 | struct workqueue_struct *wq; |
49e3cf44 | 3736 | struct pool_workqueue *pwq; |
b196be89 TH |
3737 | size_t namelen; |
3738 | ||
3739 | /* determine namelen, allocate wq and format name */ | |
3740 | va_start(args, lock_name); | |
3741 | va_copy(args1, args); | |
3742 | namelen = vsnprintf(NULL, 0, fmt, args) + 1; | |
3743 | ||
3744 | wq = kzalloc(sizeof(*wq) + namelen, GFP_KERNEL); | |
3745 | if (!wq) | |
d2c1d404 | 3746 | return NULL; |
b196be89 TH |
3747 | |
3748 | vsnprintf(wq->name, namelen, fmt, args1); | |
3749 | va_end(args); | |
3750 | va_end(args1); | |
1da177e4 | 3751 | |
d320c038 | 3752 | max_active = max_active ?: WQ_DFL_ACTIVE; |
b196be89 | 3753 | max_active = wq_clamp_max_active(max_active, flags, wq->name); |
3af24433 | 3754 | |
b196be89 | 3755 | /* init wq */ |
97e37d7b | 3756 | wq->flags = flags; |
a0a1a5fd | 3757 | wq->saved_max_active = max_active; |
3c25a55d | 3758 | mutex_init(&wq->mutex); |
112202d9 | 3759 | atomic_set(&wq->nr_pwqs_to_flush, 0); |
30cdf249 | 3760 | INIT_LIST_HEAD(&wq->pwqs); |
73f53c4a TH |
3761 | INIT_LIST_HEAD(&wq->flusher_queue); |
3762 | INIT_LIST_HEAD(&wq->flusher_overflow); | |
493a1724 | 3763 | INIT_LIST_HEAD(&wq->maydays); |
502ca9d8 | 3764 | |
eb13ba87 | 3765 | lockdep_init_map(&wq->lockdep_map, lock_name, key, 0); |
cce1a165 | 3766 | INIT_LIST_HEAD(&wq->list); |
3af24433 | 3767 | |
30cdf249 | 3768 | if (alloc_and_link_pwqs(wq) < 0) |
d2c1d404 | 3769 | goto err_free_wq; |
1537663f | 3770 | |
493008a8 TH |
3771 | /* |
3772 | * Workqueues which may be used during memory reclaim should | |
3773 | * have a rescuer to guarantee forward progress. | |
3774 | */ | |
3775 | if (flags & WQ_MEM_RECLAIM) { | |
e22bee78 TH |
3776 | struct worker *rescuer; |
3777 | ||
d2c1d404 | 3778 | rescuer = alloc_worker(); |
e22bee78 | 3779 | if (!rescuer) |
d2c1d404 | 3780 | goto err_destroy; |
e22bee78 | 3781 | |
111c225a TH |
3782 | rescuer->rescue_wq = wq; |
3783 | rescuer->task = kthread_create(rescuer_thread, rescuer, "%s", | |
b196be89 | 3784 | wq->name); |
d2c1d404 TH |
3785 | if (IS_ERR(rescuer->task)) { |
3786 | kfree(rescuer); | |
3787 | goto err_destroy; | |
3788 | } | |
e22bee78 | 3789 | |
d2c1d404 | 3790 | wq->rescuer = rescuer; |
14a40ffc | 3791 | rescuer->task->flags |= PF_NO_SETAFFINITY; |
e22bee78 | 3792 | wake_up_process(rescuer->task); |
3af24433 ON |
3793 | } |
3794 | ||
226223ab TH |
3795 | if ((wq->flags & WQ_SYSFS) && workqueue_sysfs_register(wq)) |
3796 | goto err_destroy; | |
3797 | ||
a0a1a5fd | 3798 | /* |
68e13a67 LJ |
3799 | * wq_pool_mutex protects global freeze state and workqueues list. |
3800 | * Grab it, adjust max_active and add the new @wq to workqueues | |
3801 | * list. | |
a0a1a5fd | 3802 | */ |
68e13a67 | 3803 | mutex_lock(&wq_pool_mutex); |
a0a1a5fd | 3804 | |
a357fc03 | 3805 | mutex_lock(&wq->mutex); |
699ce097 TH |
3806 | for_each_pwq(pwq, wq) |
3807 | pwq_adjust_max_active(pwq); | |
a357fc03 | 3808 | mutex_unlock(&wq->mutex); |
a0a1a5fd | 3809 | |
1537663f | 3810 | list_add(&wq->list, &workqueues); |
a0a1a5fd | 3811 | |
68e13a67 | 3812 | mutex_unlock(&wq_pool_mutex); |
1537663f | 3813 | |
3af24433 | 3814 | return wq; |
d2c1d404 TH |
3815 | |
3816 | err_free_wq: | |
3817 | kfree(wq); | |
3818 | return NULL; | |
3819 | err_destroy: | |
3820 | destroy_workqueue(wq); | |
4690c4ab | 3821 | return NULL; |
3af24433 | 3822 | } |
d320c038 | 3823 | EXPORT_SYMBOL_GPL(__alloc_workqueue_key); |
1da177e4 | 3824 | |
3af24433 ON |
3825 | /** |
3826 | * destroy_workqueue - safely terminate a workqueue | |
3827 | * @wq: target workqueue | |
3828 | * | |
3829 | * Safely destroy a workqueue. All work currently pending will be done first. | |
3830 | */ | |
3831 | void destroy_workqueue(struct workqueue_struct *wq) | |
3832 | { | |
49e3cf44 | 3833 | struct pool_workqueue *pwq; |
3af24433 | 3834 | |
9c5a2ba7 TH |
3835 | /* drain it before proceeding with destruction */ |
3836 | drain_workqueue(wq); | |
c8efcc25 | 3837 | |
6183c009 | 3838 | /* sanity checks */ |
b09f4fd3 | 3839 | mutex_lock(&wq->mutex); |
49e3cf44 | 3840 | for_each_pwq(pwq, wq) { |
6183c009 TH |
3841 | int i; |
3842 | ||
76af4d93 TH |
3843 | for (i = 0; i < WORK_NR_COLORS; i++) { |
3844 | if (WARN_ON(pwq->nr_in_flight[i])) { | |
b09f4fd3 | 3845 | mutex_unlock(&wq->mutex); |
6183c009 | 3846 | return; |
76af4d93 TH |
3847 | } |
3848 | } | |
3849 | ||
8864b4e5 TH |
3850 | if (WARN_ON(pwq->refcnt > 1) || |
3851 | WARN_ON(pwq->nr_active) || | |
76af4d93 | 3852 | WARN_ON(!list_empty(&pwq->delayed_works))) { |
b09f4fd3 | 3853 | mutex_unlock(&wq->mutex); |
6183c009 | 3854 | return; |
76af4d93 | 3855 | } |
6183c009 | 3856 | } |
b09f4fd3 | 3857 | mutex_unlock(&wq->mutex); |
6183c009 | 3858 | |
a0a1a5fd TH |
3859 | /* |
3860 | * wq list is used to freeze wq, remove from list after | |
3861 | * flushing is complete in case freeze races us. | |
3862 | */ | |
68e13a67 | 3863 | mutex_lock(&wq_pool_mutex); |
d2c1d404 | 3864 | list_del_init(&wq->list); |
68e13a67 | 3865 | mutex_unlock(&wq_pool_mutex); |
3af24433 | 3866 | |
226223ab TH |
3867 | workqueue_sysfs_unregister(wq); |
3868 | ||
493008a8 | 3869 | if (wq->rescuer) { |
e22bee78 | 3870 | kthread_stop(wq->rescuer->task); |
8d9df9f0 | 3871 | kfree(wq->rescuer); |
493008a8 | 3872 | wq->rescuer = NULL; |
e22bee78 TH |
3873 | } |
3874 | ||
8864b4e5 TH |
3875 | if (!(wq->flags & WQ_UNBOUND)) { |
3876 | /* | |
3877 | * The base ref is never dropped on per-cpu pwqs. Directly | |
3878 | * free the pwqs and wq. | |
3879 | */ | |
3880 | free_percpu(wq->cpu_pwqs); | |
3881 | kfree(wq); | |
3882 | } else { | |
3883 | /* | |
3884 | * We're the sole accessor of @wq at this point. Directly | |
3885 | * access the first pwq and put the base ref. As both pwqs | |
3886 | * and pools are sched-RCU protected, the lock operations | |
3887 | * are safe. @wq will be freed when the last pwq is | |
3888 | * released. | |
3889 | */ | |
29c91e99 TH |
3890 | pwq = list_first_entry(&wq->pwqs, struct pool_workqueue, |
3891 | pwqs_node); | |
8864b4e5 TH |
3892 | spin_lock_irq(&pwq->pool->lock); |
3893 | put_pwq(pwq); | |
3894 | spin_unlock_irq(&pwq->pool->lock); | |
29c91e99 | 3895 | } |
3af24433 ON |
3896 | } |
3897 | EXPORT_SYMBOL_GPL(destroy_workqueue); | |
3898 | ||
dcd989cb TH |
3899 | /** |
3900 | * workqueue_set_max_active - adjust max_active of a workqueue | |
3901 | * @wq: target workqueue | |
3902 | * @max_active: new max_active value. | |
3903 | * | |
3904 | * Set max_active of @wq to @max_active. | |
3905 | * | |
3906 | * CONTEXT: | |
3907 | * Don't call from IRQ context. | |
3908 | */ | |
3909 | void workqueue_set_max_active(struct workqueue_struct *wq, int max_active) | |
3910 | { | |
49e3cf44 | 3911 | struct pool_workqueue *pwq; |
dcd989cb | 3912 | |
8719dcea TH |
3913 | /* disallow meddling with max_active for ordered workqueues */ |
3914 | if (WARN_ON(wq->flags & __WQ_ORDERED)) | |
3915 | return; | |
3916 | ||
f3421797 | 3917 | max_active = wq_clamp_max_active(max_active, wq->flags, wq->name); |
dcd989cb | 3918 | |
a357fc03 | 3919 | mutex_lock(&wq->mutex); |
dcd989cb TH |
3920 | |
3921 | wq->saved_max_active = max_active; | |
3922 | ||
699ce097 TH |
3923 | for_each_pwq(pwq, wq) |
3924 | pwq_adjust_max_active(pwq); | |
93981800 | 3925 | |
a357fc03 | 3926 | mutex_unlock(&wq->mutex); |
15316ba8 | 3927 | } |
dcd989cb | 3928 | EXPORT_SYMBOL_GPL(workqueue_set_max_active); |
15316ba8 | 3929 | |
e6267616 TH |
3930 | /** |
3931 | * current_is_workqueue_rescuer - is %current workqueue rescuer? | |
3932 | * | |
3933 | * Determine whether %current is a workqueue rescuer. Can be used from | |
3934 | * work functions to determine whether it's being run off the rescuer task. | |
3935 | */ | |
3936 | bool current_is_workqueue_rescuer(void) | |
3937 | { | |
3938 | struct worker *worker = current_wq_worker(); | |
3939 | ||
6a092dfd | 3940 | return worker && worker->rescue_wq; |
e6267616 TH |
3941 | } |
3942 | ||
eef6a7d5 | 3943 | /** |
dcd989cb TH |
3944 | * workqueue_congested - test whether a workqueue is congested |
3945 | * @cpu: CPU in question | |
3946 | * @wq: target workqueue | |
eef6a7d5 | 3947 | * |
dcd989cb TH |
3948 | * Test whether @wq's cpu workqueue for @cpu is congested. There is |
3949 | * no synchronization around this function and the test result is | |
3950 | * unreliable and only useful as advisory hints or for debugging. | |
eef6a7d5 | 3951 | * |
dcd989cb TH |
3952 | * RETURNS: |
3953 | * %true if congested, %false otherwise. | |
eef6a7d5 | 3954 | */ |
d84ff051 | 3955 | bool workqueue_congested(int cpu, struct workqueue_struct *wq) |
1da177e4 | 3956 | { |
7fb98ea7 | 3957 | struct pool_workqueue *pwq; |
76af4d93 TH |
3958 | bool ret; |
3959 | ||
88109453 | 3960 | rcu_read_lock_sched(); |
7fb98ea7 TH |
3961 | |
3962 | if (!(wq->flags & WQ_UNBOUND)) | |
3963 | pwq = per_cpu_ptr(wq->cpu_pwqs, cpu); | |
3964 | else | |
3965 | pwq = first_pwq(wq); | |
dcd989cb | 3966 | |
76af4d93 | 3967 | ret = !list_empty(&pwq->delayed_works); |
88109453 | 3968 | rcu_read_unlock_sched(); |
76af4d93 TH |
3969 | |
3970 | return ret; | |
1da177e4 | 3971 | } |
dcd989cb | 3972 | EXPORT_SYMBOL_GPL(workqueue_congested); |
1da177e4 | 3973 | |
dcd989cb TH |
3974 | /** |
3975 | * work_busy - test whether a work is currently pending or running | |
3976 | * @work: the work to be tested | |
3977 | * | |
3978 | * Test whether @work is currently pending or running. There is no | |
3979 | * synchronization around this function and the test result is | |
3980 | * unreliable and only useful as advisory hints or for debugging. | |
dcd989cb TH |
3981 | * |
3982 | * RETURNS: | |
3983 | * OR'd bitmask of WORK_BUSY_* bits. | |
3984 | */ | |
3985 | unsigned int work_busy(struct work_struct *work) | |
1da177e4 | 3986 | { |
fa1b54e6 | 3987 | struct worker_pool *pool; |
dcd989cb TH |
3988 | unsigned long flags; |
3989 | unsigned int ret = 0; | |
1da177e4 | 3990 | |
dcd989cb TH |
3991 | if (work_pending(work)) |
3992 | ret |= WORK_BUSY_PENDING; | |
1da177e4 | 3993 | |
fa1b54e6 TH |
3994 | local_irq_save(flags); |
3995 | pool = get_work_pool(work); | |
038366c5 | 3996 | if (pool) { |
fa1b54e6 | 3997 | spin_lock(&pool->lock); |
038366c5 LJ |
3998 | if (find_worker_executing_work(pool, work)) |
3999 | ret |= WORK_BUSY_RUNNING; | |
fa1b54e6 | 4000 | spin_unlock(&pool->lock); |
038366c5 | 4001 | } |
fa1b54e6 | 4002 | local_irq_restore(flags); |
1da177e4 | 4003 | |
dcd989cb | 4004 | return ret; |
1da177e4 | 4005 | } |
dcd989cb | 4006 | EXPORT_SYMBOL_GPL(work_busy); |
1da177e4 | 4007 | |
db7bccf4 TH |
4008 | /* |
4009 | * CPU hotplug. | |
4010 | * | |
e22bee78 | 4011 | * There are two challenges in supporting CPU hotplug. Firstly, there |
112202d9 | 4012 | * are a lot of assumptions on strong associations among work, pwq and |
706026c2 | 4013 | * pool which make migrating pending and scheduled works very |
e22bee78 | 4014 | * difficult to implement without impacting hot paths. Secondly, |
94cf58bb | 4015 | * worker pools serve mix of short, long and very long running works making |
e22bee78 TH |
4016 | * blocked draining impractical. |
4017 | * | |
24647570 | 4018 | * This is solved by allowing the pools to be disassociated from the CPU |
628c78e7 TH |
4019 | * running as an unbound one and allowing it to be reattached later if the |
4020 | * cpu comes back online. | |
db7bccf4 | 4021 | */ |
1da177e4 | 4022 | |
706026c2 | 4023 | static void wq_unbind_fn(struct work_struct *work) |
3af24433 | 4024 | { |
38db41d9 | 4025 | int cpu = smp_processor_id(); |
4ce62e9e | 4026 | struct worker_pool *pool; |
db7bccf4 | 4027 | struct worker *worker; |
a9ab775b | 4028 | int wi; |
3af24433 | 4029 | |
f02ae73a | 4030 | for_each_cpu_worker_pool(pool, cpu) { |
6183c009 | 4031 | WARN_ON_ONCE(cpu != smp_processor_id()); |
db7bccf4 | 4032 | |
bc3a1afc | 4033 | mutex_lock(&pool->manager_mutex); |
94cf58bb | 4034 | spin_lock_irq(&pool->lock); |
3af24433 | 4035 | |
94cf58bb | 4036 | /* |
bc3a1afc | 4037 | * We've blocked all manager operations. Make all workers |
94cf58bb TH |
4038 | * unbound and set DISASSOCIATED. Before this, all workers |
4039 | * except for the ones which are still executing works from | |
4040 | * before the last CPU down must be on the cpu. After | |
4041 | * this, they may become diasporas. | |
4042 | */ | |
a9ab775b | 4043 | for_each_pool_worker(worker, wi, pool) |
c9e7cf27 | 4044 | worker->flags |= WORKER_UNBOUND; |
06ba38a9 | 4045 | |
24647570 | 4046 | pool->flags |= POOL_DISASSOCIATED; |
f2d5a0ee | 4047 | |
94cf58bb | 4048 | spin_unlock_irq(&pool->lock); |
bc3a1afc | 4049 | mutex_unlock(&pool->manager_mutex); |
94cf58bb | 4050 | } |
628c78e7 | 4051 | |
e22bee78 | 4052 | /* |
403c821d | 4053 | * Call schedule() so that we cross rq->lock and thus can guarantee |
628c78e7 TH |
4054 | * sched callbacks see the %WORKER_UNBOUND flag. This is necessary |
4055 | * as scheduler callbacks may be invoked from other cpus. | |
e22bee78 | 4056 | */ |
e22bee78 | 4057 | schedule(); |
06ba38a9 | 4058 | |
e22bee78 | 4059 | /* |
628c78e7 TH |
4060 | * Sched callbacks are disabled now. Zap nr_running. After this, |
4061 | * nr_running stays zero and need_more_worker() and keep_working() | |
38db41d9 TH |
4062 | * are always true as long as the worklist is not empty. Pools on |
4063 | * @cpu now behave as unbound (in terms of concurrency management) | |
4064 | * pools which are served by workers tied to the CPU. | |
628c78e7 TH |
4065 | * |
4066 | * On return from this function, the current worker would trigger | |
4067 | * unbound chain execution of pending work items if other workers | |
4068 | * didn't already. | |
e22bee78 | 4069 | */ |
f02ae73a | 4070 | for_each_cpu_worker_pool(pool, cpu) |
e19e397a | 4071 | atomic_set(&pool->nr_running, 0); |
3af24433 | 4072 | } |
3af24433 | 4073 | |
bd7c089e TH |
4074 | /** |
4075 | * rebind_workers - rebind all workers of a pool to the associated CPU | |
4076 | * @pool: pool of interest | |
4077 | * | |
a9ab775b | 4078 | * @pool->cpu is coming online. Rebind all workers to the CPU. |
bd7c089e TH |
4079 | */ |
4080 | static void rebind_workers(struct worker_pool *pool) | |
4081 | { | |
a9ab775b TH |
4082 | struct worker *worker; |
4083 | int wi; | |
bd7c089e TH |
4084 | |
4085 | lockdep_assert_held(&pool->manager_mutex); | |
bd7c089e | 4086 | |
a9ab775b TH |
4087 | /* |
4088 | * Restore CPU affinity of all workers. As all idle workers should | |
4089 | * be on the run-queue of the associated CPU before any local | |
4090 | * wake-ups for concurrency management happen, restore CPU affinty | |
4091 | * of all workers first and then clear UNBOUND. As we're called | |
4092 | * from CPU_ONLINE, the following shouldn't fail. | |
4093 | */ | |
4094 | for_each_pool_worker(worker, wi, pool) | |
4095 | WARN_ON_ONCE(set_cpus_allowed_ptr(worker->task, | |
4096 | pool->attrs->cpumask) < 0); | |
bd7c089e | 4097 | |
a9ab775b | 4098 | spin_lock_irq(&pool->lock); |
bd7c089e | 4099 | |
a9ab775b TH |
4100 | for_each_pool_worker(worker, wi, pool) { |
4101 | unsigned int worker_flags = worker->flags; | |
bd7c089e TH |
4102 | |
4103 | /* | |
a9ab775b TH |
4104 | * A bound idle worker should actually be on the runqueue |
4105 | * of the associated CPU for local wake-ups targeting it to | |
4106 | * work. Kick all idle workers so that they migrate to the | |
4107 | * associated CPU. Doing this in the same loop as | |
4108 | * replacing UNBOUND with REBOUND is safe as no worker will | |
4109 | * be bound before @pool->lock is released. | |
bd7c089e | 4110 | */ |
a9ab775b TH |
4111 | if (worker_flags & WORKER_IDLE) |
4112 | wake_up_process(worker->task); | |
bd7c089e | 4113 | |
a9ab775b TH |
4114 | /* |
4115 | * We want to clear UNBOUND but can't directly call | |
4116 | * worker_clr_flags() or adjust nr_running. Atomically | |
4117 | * replace UNBOUND with another NOT_RUNNING flag REBOUND. | |
4118 | * @worker will clear REBOUND using worker_clr_flags() when | |
4119 | * it initiates the next execution cycle thus restoring | |
4120 | * concurrency management. Note that when or whether | |
4121 | * @worker clears REBOUND doesn't affect correctness. | |
4122 | * | |
4123 | * ACCESS_ONCE() is necessary because @worker->flags may be | |
4124 | * tested without holding any lock in | |
4125 | * wq_worker_waking_up(). Without it, NOT_RUNNING test may | |
4126 | * fail incorrectly leading to premature concurrency | |
4127 | * management operations. | |
4128 | */ | |
4129 | WARN_ON_ONCE(!(worker_flags & WORKER_UNBOUND)); | |
4130 | worker_flags |= WORKER_REBOUND; | |
4131 | worker_flags &= ~WORKER_UNBOUND; | |
4132 | ACCESS_ONCE(worker->flags) = worker_flags; | |
bd7c089e | 4133 | } |
a9ab775b TH |
4134 | |
4135 | spin_unlock_irq(&pool->lock); | |
bd7c089e TH |
4136 | } |
4137 | ||
7dbc725e TH |
4138 | /** |
4139 | * restore_unbound_workers_cpumask - restore cpumask of unbound workers | |
4140 | * @pool: unbound pool of interest | |
4141 | * @cpu: the CPU which is coming up | |
4142 | * | |
4143 | * An unbound pool may end up with a cpumask which doesn't have any online | |
4144 | * CPUs. When a worker of such pool get scheduled, the scheduler resets | |
4145 | * its cpus_allowed. If @cpu is in @pool's cpumask which didn't have any | |
4146 | * online CPU before, cpus_allowed of all its workers should be restored. | |
4147 | */ | |
4148 | static void restore_unbound_workers_cpumask(struct worker_pool *pool, int cpu) | |
4149 | { | |
4150 | static cpumask_t cpumask; | |
4151 | struct worker *worker; | |
4152 | int wi; | |
4153 | ||
4154 | lockdep_assert_held(&pool->manager_mutex); | |
4155 | ||
4156 | /* is @cpu allowed for @pool? */ | |
4157 | if (!cpumask_test_cpu(cpu, pool->attrs->cpumask)) | |
4158 | return; | |
4159 | ||
4160 | /* is @cpu the only online CPU? */ | |
4161 | cpumask_and(&cpumask, pool->attrs->cpumask, cpu_online_mask); | |
4162 | if (cpumask_weight(&cpumask) != 1) | |
4163 | return; | |
4164 | ||
4165 | /* as we're called from CPU_ONLINE, the following shouldn't fail */ | |
4166 | for_each_pool_worker(worker, wi, pool) | |
4167 | WARN_ON_ONCE(set_cpus_allowed_ptr(worker->task, | |
4168 | pool->attrs->cpumask) < 0); | |
4169 | } | |
4170 | ||
8db25e78 TH |
4171 | /* |
4172 | * Workqueues should be brought up before normal priority CPU notifiers. | |
4173 | * This will be registered high priority CPU notifier. | |
4174 | */ | |
9fdf9b73 | 4175 | static int __cpuinit workqueue_cpu_up_callback(struct notifier_block *nfb, |
8db25e78 TH |
4176 | unsigned long action, |
4177 | void *hcpu) | |
3af24433 | 4178 | { |
d84ff051 | 4179 | int cpu = (unsigned long)hcpu; |
4ce62e9e | 4180 | struct worker_pool *pool; |
7dbc725e | 4181 | int pi; |
3ce63377 | 4182 | |
8db25e78 | 4183 | switch (action & ~CPU_TASKS_FROZEN) { |
3af24433 | 4184 | case CPU_UP_PREPARE: |
f02ae73a | 4185 | for_each_cpu_worker_pool(pool, cpu) { |
3ce63377 TH |
4186 | if (pool->nr_workers) |
4187 | continue; | |
ebf44d16 | 4188 | if (create_and_start_worker(pool) < 0) |
3ce63377 | 4189 | return NOTIFY_BAD; |
3af24433 | 4190 | } |
8db25e78 | 4191 | break; |
3af24433 | 4192 | |
db7bccf4 TH |
4193 | case CPU_DOWN_FAILED: |
4194 | case CPU_ONLINE: | |
68e13a67 | 4195 | mutex_lock(&wq_pool_mutex); |
7dbc725e TH |
4196 | |
4197 | for_each_pool(pool, pi) { | |
bc3a1afc | 4198 | mutex_lock(&pool->manager_mutex); |
94cf58bb | 4199 | |
7dbc725e TH |
4200 | if (pool->cpu == cpu) { |
4201 | spin_lock_irq(&pool->lock); | |
4202 | pool->flags &= ~POOL_DISASSOCIATED; | |
4203 | spin_unlock_irq(&pool->lock); | |
a9ab775b | 4204 | |
7dbc725e TH |
4205 | rebind_workers(pool); |
4206 | } else if (pool->cpu < 0) { | |
4207 | restore_unbound_workers_cpumask(pool, cpu); | |
4208 | } | |
94cf58bb | 4209 | |
bc3a1afc | 4210 | mutex_unlock(&pool->manager_mutex); |
94cf58bb | 4211 | } |
7dbc725e | 4212 | |
68e13a67 | 4213 | mutex_unlock(&wq_pool_mutex); |
db7bccf4 | 4214 | break; |
00dfcaf7 | 4215 | } |
65758202 TH |
4216 | return NOTIFY_OK; |
4217 | } | |
4218 | ||
4219 | /* | |
4220 | * Workqueues should be brought down after normal priority CPU notifiers. | |
4221 | * This will be registered as low priority CPU notifier. | |
4222 | */ | |
9fdf9b73 | 4223 | static int __cpuinit workqueue_cpu_down_callback(struct notifier_block *nfb, |
65758202 TH |
4224 | unsigned long action, |
4225 | void *hcpu) | |
4226 | { | |
d84ff051 | 4227 | int cpu = (unsigned long)hcpu; |
8db25e78 TH |
4228 | struct work_struct unbind_work; |
4229 | ||
65758202 TH |
4230 | switch (action & ~CPU_TASKS_FROZEN) { |
4231 | case CPU_DOWN_PREPARE: | |
8db25e78 | 4232 | /* unbinding should happen on the local CPU */ |
706026c2 | 4233 | INIT_WORK_ONSTACK(&unbind_work, wq_unbind_fn); |
7635d2fd | 4234 | queue_work_on(cpu, system_highpri_wq, &unbind_work); |
8db25e78 TH |
4235 | flush_work(&unbind_work); |
4236 | break; | |
65758202 TH |
4237 | } |
4238 | return NOTIFY_OK; | |
4239 | } | |
4240 | ||
2d3854a3 | 4241 | #ifdef CONFIG_SMP |
8ccad40d | 4242 | |
2d3854a3 | 4243 | struct work_for_cpu { |
ed48ece2 | 4244 | struct work_struct work; |
2d3854a3 RR |
4245 | long (*fn)(void *); |
4246 | void *arg; | |
4247 | long ret; | |
4248 | }; | |
4249 | ||
ed48ece2 | 4250 | static void work_for_cpu_fn(struct work_struct *work) |
2d3854a3 | 4251 | { |
ed48ece2 TH |
4252 | struct work_for_cpu *wfc = container_of(work, struct work_for_cpu, work); |
4253 | ||
2d3854a3 RR |
4254 | wfc->ret = wfc->fn(wfc->arg); |
4255 | } | |
4256 | ||
4257 | /** | |
4258 | * work_on_cpu - run a function in user context on a particular cpu | |
4259 | * @cpu: the cpu to run on | |
4260 | * @fn: the function to run | |
4261 | * @arg: the function arg | |
4262 | * | |
31ad9081 RR |
4263 | * This will return the value @fn returns. |
4264 | * It is up to the caller to ensure that the cpu doesn't go offline. | |
6b44003e | 4265 | * The caller must not hold any locks which would prevent @fn from completing. |
2d3854a3 | 4266 | */ |
d84ff051 | 4267 | long work_on_cpu(int cpu, long (*fn)(void *), void *arg) |
2d3854a3 | 4268 | { |
ed48ece2 | 4269 | struct work_for_cpu wfc = { .fn = fn, .arg = arg }; |
6b44003e | 4270 | |
ed48ece2 TH |
4271 | INIT_WORK_ONSTACK(&wfc.work, work_for_cpu_fn); |
4272 | schedule_work_on(cpu, &wfc.work); | |
4273 | flush_work(&wfc.work); | |
2d3854a3 RR |
4274 | return wfc.ret; |
4275 | } | |
4276 | EXPORT_SYMBOL_GPL(work_on_cpu); | |
4277 | #endif /* CONFIG_SMP */ | |
4278 | ||
a0a1a5fd TH |
4279 | #ifdef CONFIG_FREEZER |
4280 | ||
4281 | /** | |
4282 | * freeze_workqueues_begin - begin freezing workqueues | |
4283 | * | |
58a69cb4 | 4284 | * Start freezing workqueues. After this function returns, all freezable |
c5aa87bb | 4285 | * workqueues will queue new works to their delayed_works list instead of |
706026c2 | 4286 | * pool->worklist. |
a0a1a5fd TH |
4287 | * |
4288 | * CONTEXT: | |
a357fc03 | 4289 | * Grabs and releases wq_pool_mutex, wq->mutex and pool->lock's. |
a0a1a5fd TH |
4290 | */ |
4291 | void freeze_workqueues_begin(void) | |
4292 | { | |
17116969 | 4293 | struct worker_pool *pool; |
24b8a847 TH |
4294 | struct workqueue_struct *wq; |
4295 | struct pool_workqueue *pwq; | |
611c92a0 | 4296 | int pi; |
a0a1a5fd | 4297 | |
68e13a67 | 4298 | mutex_lock(&wq_pool_mutex); |
a0a1a5fd | 4299 | |
6183c009 | 4300 | WARN_ON_ONCE(workqueue_freezing); |
a0a1a5fd TH |
4301 | workqueue_freezing = true; |
4302 | ||
24b8a847 | 4303 | /* set FREEZING */ |
611c92a0 | 4304 | for_each_pool(pool, pi) { |
5bcab335 | 4305 | spin_lock_irq(&pool->lock); |
17116969 TH |
4306 | WARN_ON_ONCE(pool->flags & POOL_FREEZING); |
4307 | pool->flags |= POOL_FREEZING; | |
5bcab335 | 4308 | spin_unlock_irq(&pool->lock); |
24b8a847 | 4309 | } |
a0a1a5fd | 4310 | |
24b8a847 | 4311 | list_for_each_entry(wq, &workqueues, list) { |
a357fc03 | 4312 | mutex_lock(&wq->mutex); |
699ce097 TH |
4313 | for_each_pwq(pwq, wq) |
4314 | pwq_adjust_max_active(pwq); | |
a357fc03 | 4315 | mutex_unlock(&wq->mutex); |
a0a1a5fd | 4316 | } |
5bcab335 | 4317 | |
68e13a67 | 4318 | mutex_unlock(&wq_pool_mutex); |
a0a1a5fd TH |
4319 | } |
4320 | ||
4321 | /** | |
58a69cb4 | 4322 | * freeze_workqueues_busy - are freezable workqueues still busy? |
a0a1a5fd TH |
4323 | * |
4324 | * Check whether freezing is complete. This function must be called | |
4325 | * between freeze_workqueues_begin() and thaw_workqueues(). | |
4326 | * | |
4327 | * CONTEXT: | |
68e13a67 | 4328 | * Grabs and releases wq_pool_mutex. |
a0a1a5fd TH |
4329 | * |
4330 | * RETURNS: | |
58a69cb4 TH |
4331 | * %true if some freezable workqueues are still busy. %false if freezing |
4332 | * is complete. | |
a0a1a5fd TH |
4333 | */ |
4334 | bool freeze_workqueues_busy(void) | |
4335 | { | |
a0a1a5fd | 4336 | bool busy = false; |
24b8a847 TH |
4337 | struct workqueue_struct *wq; |
4338 | struct pool_workqueue *pwq; | |
a0a1a5fd | 4339 | |
68e13a67 | 4340 | mutex_lock(&wq_pool_mutex); |
a0a1a5fd | 4341 | |
6183c009 | 4342 | WARN_ON_ONCE(!workqueue_freezing); |
a0a1a5fd | 4343 | |
24b8a847 TH |
4344 | list_for_each_entry(wq, &workqueues, list) { |
4345 | if (!(wq->flags & WQ_FREEZABLE)) | |
4346 | continue; | |
a0a1a5fd TH |
4347 | /* |
4348 | * nr_active is monotonically decreasing. It's safe | |
4349 | * to peek without lock. | |
4350 | */ | |
88109453 | 4351 | rcu_read_lock_sched(); |
24b8a847 | 4352 | for_each_pwq(pwq, wq) { |
6183c009 | 4353 | WARN_ON_ONCE(pwq->nr_active < 0); |
112202d9 | 4354 | if (pwq->nr_active) { |
a0a1a5fd | 4355 | busy = true; |
88109453 | 4356 | rcu_read_unlock_sched(); |
a0a1a5fd TH |
4357 | goto out_unlock; |
4358 | } | |
4359 | } | |
88109453 | 4360 | rcu_read_unlock_sched(); |
a0a1a5fd TH |
4361 | } |
4362 | out_unlock: | |
68e13a67 | 4363 | mutex_unlock(&wq_pool_mutex); |
a0a1a5fd TH |
4364 | return busy; |
4365 | } | |
4366 | ||
4367 | /** | |
4368 | * thaw_workqueues - thaw workqueues | |
4369 | * | |
4370 | * Thaw workqueues. Normal queueing is restored and all collected | |
706026c2 | 4371 | * frozen works are transferred to their respective pool worklists. |
a0a1a5fd TH |
4372 | * |
4373 | * CONTEXT: | |
a357fc03 | 4374 | * Grabs and releases wq_pool_mutex, wq->mutex and pool->lock's. |
a0a1a5fd TH |
4375 | */ |
4376 | void thaw_workqueues(void) | |
4377 | { | |
24b8a847 TH |
4378 | struct workqueue_struct *wq; |
4379 | struct pool_workqueue *pwq; | |
4380 | struct worker_pool *pool; | |
611c92a0 | 4381 | int pi; |
a0a1a5fd | 4382 | |
68e13a67 | 4383 | mutex_lock(&wq_pool_mutex); |
a0a1a5fd TH |
4384 | |
4385 | if (!workqueue_freezing) | |
4386 | goto out_unlock; | |
4387 | ||
24b8a847 | 4388 | /* clear FREEZING */ |
611c92a0 | 4389 | for_each_pool(pool, pi) { |
5bcab335 | 4390 | spin_lock_irq(&pool->lock); |
24b8a847 TH |
4391 | WARN_ON_ONCE(!(pool->flags & POOL_FREEZING)); |
4392 | pool->flags &= ~POOL_FREEZING; | |
5bcab335 | 4393 | spin_unlock_irq(&pool->lock); |
24b8a847 | 4394 | } |
8b03ae3c | 4395 | |
24b8a847 TH |
4396 | /* restore max_active and repopulate worklist */ |
4397 | list_for_each_entry(wq, &workqueues, list) { | |
a357fc03 | 4398 | mutex_lock(&wq->mutex); |
699ce097 TH |
4399 | for_each_pwq(pwq, wq) |
4400 | pwq_adjust_max_active(pwq); | |
a357fc03 | 4401 | mutex_unlock(&wq->mutex); |
a0a1a5fd TH |
4402 | } |
4403 | ||
4404 | workqueue_freezing = false; | |
4405 | out_unlock: | |
68e13a67 | 4406 | mutex_unlock(&wq_pool_mutex); |
a0a1a5fd TH |
4407 | } |
4408 | #endif /* CONFIG_FREEZER */ | |
4409 | ||
6ee0578b | 4410 | static int __init init_workqueues(void) |
1da177e4 | 4411 | { |
7a4e344c TH |
4412 | int std_nice[NR_STD_WORKER_POOLS] = { 0, HIGHPRI_NICE_LEVEL }; |
4413 | int i, cpu; | |
c34056a3 | 4414 | |
7c3eed5c TH |
4415 | /* make sure we have enough bits for OFFQ pool ID */ |
4416 | BUILD_BUG_ON((1LU << (BITS_PER_LONG - WORK_OFFQ_POOL_SHIFT)) < | |
6be19588 | 4417 | WORK_CPU_END * NR_STD_WORKER_POOLS); |
b5490077 | 4418 | |
e904e6c2 TH |
4419 | WARN_ON(__alignof__(struct pool_workqueue) < __alignof__(long long)); |
4420 | ||
4421 | pwq_cache = KMEM_CACHE(pool_workqueue, SLAB_PANIC); | |
4422 | ||
65758202 | 4423 | cpu_notifier(workqueue_cpu_up_callback, CPU_PRI_WORKQUEUE_UP); |
a5b4e57d | 4424 | hotcpu_notifier(workqueue_cpu_down_callback, CPU_PRI_WORKQUEUE_DOWN); |
8b03ae3c | 4425 | |
706026c2 | 4426 | /* initialize CPU pools */ |
29c91e99 | 4427 | for_each_possible_cpu(cpu) { |
4ce62e9e | 4428 | struct worker_pool *pool; |
8b03ae3c | 4429 | |
7a4e344c | 4430 | i = 0; |
f02ae73a | 4431 | for_each_cpu_worker_pool(pool, cpu) { |
7a4e344c | 4432 | BUG_ON(init_worker_pool(pool)); |
ec22ca5e | 4433 | pool->cpu = cpu; |
29c91e99 | 4434 | cpumask_copy(pool->attrs->cpumask, cpumask_of(cpu)); |
7a4e344c TH |
4435 | pool->attrs->nice = std_nice[i++]; |
4436 | ||
9daf9e67 | 4437 | /* alloc pool ID */ |
68e13a67 | 4438 | mutex_lock(&wq_pool_mutex); |
9daf9e67 | 4439 | BUG_ON(worker_pool_assign_id(pool)); |
68e13a67 | 4440 | mutex_unlock(&wq_pool_mutex); |
4ce62e9e | 4441 | } |
8b03ae3c TH |
4442 | } |
4443 | ||
e22bee78 | 4444 | /* create the initial worker */ |
29c91e99 | 4445 | for_each_online_cpu(cpu) { |
4ce62e9e | 4446 | struct worker_pool *pool; |
e22bee78 | 4447 | |
f02ae73a | 4448 | for_each_cpu_worker_pool(pool, cpu) { |
29c91e99 | 4449 | pool->flags &= ~POOL_DISASSOCIATED; |
ebf44d16 | 4450 | BUG_ON(create_and_start_worker(pool) < 0); |
4ce62e9e | 4451 | } |
e22bee78 TH |
4452 | } |
4453 | ||
29c91e99 TH |
4454 | /* create default unbound wq attrs */ |
4455 | for (i = 0; i < NR_STD_WORKER_POOLS; i++) { | |
4456 | struct workqueue_attrs *attrs; | |
4457 | ||
4458 | BUG_ON(!(attrs = alloc_workqueue_attrs(GFP_KERNEL))); | |
29c91e99 | 4459 | attrs->nice = std_nice[i]; |
29c91e99 TH |
4460 | unbound_std_wq_attrs[i] = attrs; |
4461 | } | |
4462 | ||
d320c038 | 4463 | system_wq = alloc_workqueue("events", 0, 0); |
1aabe902 | 4464 | system_highpri_wq = alloc_workqueue("events_highpri", WQ_HIGHPRI, 0); |
d320c038 | 4465 | system_long_wq = alloc_workqueue("events_long", 0, 0); |
f3421797 TH |
4466 | system_unbound_wq = alloc_workqueue("events_unbound", WQ_UNBOUND, |
4467 | WQ_UNBOUND_MAX_ACTIVE); | |
24d51add TH |
4468 | system_freezable_wq = alloc_workqueue("events_freezable", |
4469 | WQ_FREEZABLE, 0); | |
1aabe902 | 4470 | BUG_ON(!system_wq || !system_highpri_wq || !system_long_wq || |
ae930e0f | 4471 | !system_unbound_wq || !system_freezable_wq); |
6ee0578b | 4472 | return 0; |
1da177e4 | 4473 | } |
6ee0578b | 4474 | early_initcall(init_workqueues); |