]>
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 | |
b11895c4 L |
19 | * automatically managed. There are two worker pools for each CPU (one for |
20 | * normal work items and the other for high priority ones) and some extra | |
21 | * pools for workqueues which are not bound to any specific CPU - the | |
22 | * number of these backing pools is dynamic. | |
c54fce6e TH |
23 | * |
24 | * Please read Documentation/workqueue.txt for details. | |
1da177e4 LT |
25 | */ |
26 | ||
9984de1a | 27 | #include <linux/export.h> |
1da177e4 LT |
28 | #include <linux/kernel.h> |
29 | #include <linux/sched.h> | |
30 | #include <linux/init.h> | |
31 | #include <linux/signal.h> | |
32 | #include <linux/completion.h> | |
33 | #include <linux/workqueue.h> | |
34 | #include <linux/slab.h> | |
35 | #include <linux/cpu.h> | |
36 | #include <linux/notifier.h> | |
37 | #include <linux/kthread.h> | |
1fa44eca | 38 | #include <linux/hardirq.h> |
46934023 | 39 | #include <linux/mempolicy.h> |
341a5958 | 40 | #include <linux/freezer.h> |
d5abe669 PZ |
41 | #include <linux/kallsyms.h> |
42 | #include <linux/debug_locks.h> | |
4e6045f1 | 43 | #include <linux/lockdep.h> |
c34056a3 | 44 | #include <linux/idr.h> |
29c91e99 | 45 | #include <linux/jhash.h> |
42f8570f | 46 | #include <linux/hashtable.h> |
76af4d93 | 47 | #include <linux/rculist.h> |
bce90380 | 48 | #include <linux/nodemask.h> |
4c16bd32 | 49 | #include <linux/moduleparam.h> |
3d1cb205 | 50 | #include <linux/uaccess.h> |
e22bee78 | 51 | |
ea138446 | 52 | #include "workqueue_internal.h" |
1da177e4 | 53 | |
c8e55f36 | 54 | enum { |
24647570 TH |
55 | /* |
56 | * worker_pool flags | |
bc2ae0f5 | 57 | * |
24647570 | 58 | * A bound pool is either associated or disassociated with its CPU. |
bc2ae0f5 TH |
59 | * While associated (!DISASSOCIATED), all workers are bound to the |
60 | * CPU and none has %WORKER_UNBOUND set and concurrency management | |
61 | * is in effect. | |
62 | * | |
63 | * While DISASSOCIATED, the cpu may be offline and all workers have | |
64 | * %WORKER_UNBOUND set and concurrency management disabled, and may | |
24647570 | 65 | * be executing on any CPU. The pool behaves as an unbound one. |
bc2ae0f5 | 66 | * |
bc3a1afc TH |
67 | * Note that DISASSOCIATED should be flipped only while holding |
68 | * manager_mutex to avoid changing binding state while | |
24647570 | 69 | * create_worker() is in progress. |
bc2ae0f5 | 70 | */ |
11ebea50 | 71 | POOL_MANAGE_WORKERS = 1 << 0, /* need to manage workers */ |
24647570 | 72 | POOL_DISASSOCIATED = 1 << 2, /* cpu can't serve workers */ |
35b6bb63 | 73 | POOL_FREEZING = 1 << 3, /* freeze in progress */ |
db7bccf4 | 74 | |
c8e55f36 TH |
75 | /* worker flags */ |
76 | WORKER_STARTED = 1 << 0, /* started */ | |
77 | WORKER_DIE = 1 << 1, /* die die die */ | |
78 | WORKER_IDLE = 1 << 2, /* is idle */ | |
e22bee78 | 79 | WORKER_PREP = 1 << 3, /* preparing to run works */ |
fb0e7beb | 80 | WORKER_CPU_INTENSIVE = 1 << 6, /* cpu intensive */ |
f3421797 | 81 | WORKER_UNBOUND = 1 << 7, /* worker is unbound */ |
a9ab775b | 82 | WORKER_REBOUND = 1 << 8, /* worker was rebound */ |
e22bee78 | 83 | |
a9ab775b TH |
84 | WORKER_NOT_RUNNING = WORKER_PREP | WORKER_CPU_INTENSIVE | |
85 | WORKER_UNBOUND | WORKER_REBOUND, | |
db7bccf4 | 86 | |
e34cdddb | 87 | NR_STD_WORKER_POOLS = 2, /* # standard pools per cpu */ |
4ce62e9e | 88 | |
29c91e99 | 89 | UNBOUND_POOL_HASH_ORDER = 6, /* hashed by pool->attrs */ |
c8e55f36 | 90 | BUSY_WORKER_HASH_ORDER = 6, /* 64 pointers */ |
db7bccf4 | 91 | |
e22bee78 TH |
92 | MAX_IDLE_WORKERS_RATIO = 4, /* 1/4 of busy can be idle */ |
93 | IDLE_WORKER_TIMEOUT = 300 * HZ, /* keep idle ones for 5 mins */ | |
94 | ||
3233cdbd TH |
95 | MAYDAY_INITIAL_TIMEOUT = HZ / 100 >= 2 ? HZ / 100 : 2, |
96 | /* call for help after 10ms | |
97 | (min two ticks) */ | |
e22bee78 TH |
98 | MAYDAY_INTERVAL = HZ / 10, /* and then every 100ms */ |
99 | CREATE_COOLDOWN = HZ, /* time to breath after fail */ | |
e22bee78 TH |
100 | |
101 | /* | |
102 | * Rescue workers are used only on emergencies and shared by | |
103 | * all cpus. Give -20. | |
104 | */ | |
105 | RESCUER_NICE_LEVEL = -20, | |
3270476a | 106 | HIGHPRI_NICE_LEVEL = -20, |
ecf6881f TH |
107 | |
108 | WQ_NAME_LEN = 24, | |
c8e55f36 | 109 | }; |
1da177e4 LT |
110 | |
111 | /* | |
4690c4ab TH |
112 | * Structure fields follow one of the following exclusion rules. |
113 | * | |
e41e704b TH |
114 | * I: Modifiable by initialization/destruction paths and read-only for |
115 | * everyone else. | |
4690c4ab | 116 | * |
e22bee78 TH |
117 | * P: Preemption protected. Disabling preemption is enough and should |
118 | * only be modified and accessed from the local cpu. | |
119 | * | |
d565ed63 | 120 | * L: pool->lock protected. Access with pool->lock held. |
4690c4ab | 121 | * |
d565ed63 TH |
122 | * X: During normal operation, modification requires pool->lock and should |
123 | * be done only from local cpu. Either disabling preemption on local | |
124 | * cpu or grabbing pool->lock is enough for read access. If | |
125 | * POOL_DISASSOCIATED is set, it's identical to L. | |
e22bee78 | 126 | * |
9625ab17 | 127 | * M: pool->manager_mutex protected. |
822d8405 | 128 | * |
68e13a67 | 129 | * PL: wq_pool_mutex protected. |
5bcab335 | 130 | * |
68e13a67 | 131 | * PR: wq_pool_mutex protected for writes. Sched-RCU protected for reads. |
76af4d93 | 132 | * |
3c25a55d LJ |
133 | * WQ: wq->mutex protected. |
134 | * | |
b5927605 | 135 | * WR: wq->mutex protected for writes. Sched-RCU protected for reads. |
2e109a28 TH |
136 | * |
137 | * MD: wq_mayday_lock protected. | |
1da177e4 | 138 | */ |
1da177e4 | 139 | |
2eaebdb3 | 140 | /* struct worker is defined in workqueue_internal.h */ |
c34056a3 | 141 | |
bd7bdd43 | 142 | struct worker_pool { |
d565ed63 | 143 | spinlock_t lock; /* the pool lock */ |
d84ff051 | 144 | int cpu; /* I: the associated cpu */ |
f3f90ad4 | 145 | int node; /* I: the associated node ID */ |
9daf9e67 | 146 | int id; /* I: pool ID */ |
11ebea50 | 147 | unsigned int flags; /* X: flags */ |
bd7bdd43 TH |
148 | |
149 | struct list_head worklist; /* L: list of pending works */ | |
150 | int nr_workers; /* L: total number of workers */ | |
ea1abd61 LJ |
151 | |
152 | /* nr_idle includes the ones off idle_list for rebinding */ | |
bd7bdd43 TH |
153 | int nr_idle; /* L: currently idle ones */ |
154 | ||
155 | struct list_head idle_list; /* X: list of idle workers */ | |
156 | struct timer_list idle_timer; /* L: worker idle timeout */ | |
157 | struct timer_list mayday_timer; /* L: SOS timer for workers */ | |
158 | ||
c5aa87bb | 159 | /* a workers is either on busy_hash or idle_list, or the manager */ |
c9e7cf27 TH |
160 | DECLARE_HASHTABLE(busy_hash, BUSY_WORKER_HASH_ORDER); |
161 | /* L: hash of busy workers */ | |
162 | ||
bc3a1afc | 163 | /* see manage_workers() for details on the two manager mutexes */ |
34a06bd6 | 164 | struct mutex manager_arb; /* manager arbitration */ |
bc3a1afc | 165 | struct mutex manager_mutex; /* manager exclusion */ |
9625ab17 | 166 | struct idr worker_idr; /* M: worker IDs and iteration */ |
e19e397a | 167 | |
7a4e344c | 168 | struct workqueue_attrs *attrs; /* I: worker attributes */ |
68e13a67 LJ |
169 | struct hlist_node hash_node; /* PL: unbound_pool_hash node */ |
170 | int refcnt; /* PL: refcnt for unbound pools */ | |
7a4e344c | 171 | |
e19e397a TH |
172 | /* |
173 | * The current concurrency level. As it's likely to be accessed | |
174 | * from other CPUs during try_to_wake_up(), put it in a separate | |
175 | * cacheline. | |
176 | */ | |
177 | atomic_t nr_running ____cacheline_aligned_in_smp; | |
29c91e99 TH |
178 | |
179 | /* | |
180 | * Destruction of pool is sched-RCU protected to allow dereferences | |
181 | * from get_work_pool(). | |
182 | */ | |
183 | struct rcu_head rcu; | |
8b03ae3c TH |
184 | } ____cacheline_aligned_in_smp; |
185 | ||
1da177e4 | 186 | /* |
112202d9 TH |
187 | * The per-pool workqueue. While queued, the lower WORK_STRUCT_FLAG_BITS |
188 | * of work_struct->data are used for flags and the remaining high bits | |
189 | * point to the pwq; thus, pwqs need to be aligned at two's power of the | |
190 | * number of flag bits. | |
1da177e4 | 191 | */ |
112202d9 | 192 | struct pool_workqueue { |
bd7bdd43 | 193 | struct worker_pool *pool; /* I: the associated pool */ |
4690c4ab | 194 | struct workqueue_struct *wq; /* I: the owning workqueue */ |
73f53c4a TH |
195 | int work_color; /* L: current color */ |
196 | int flush_color; /* L: flushing color */ | |
8864b4e5 | 197 | int refcnt; /* L: reference count */ |
73f53c4a TH |
198 | int nr_in_flight[WORK_NR_COLORS]; |
199 | /* L: nr of in_flight works */ | |
1e19ffc6 | 200 | int nr_active; /* L: nr of active works */ |
a0a1a5fd | 201 | int max_active; /* L: max active works */ |
1e19ffc6 | 202 | struct list_head delayed_works; /* L: delayed works */ |
3c25a55d | 203 | struct list_head pwqs_node; /* WR: node on wq->pwqs */ |
2e109a28 | 204 | struct list_head mayday_node; /* MD: node on wq->maydays */ |
8864b4e5 TH |
205 | |
206 | /* | |
207 | * Release of unbound pwq is punted to system_wq. See put_pwq() | |
208 | * and pwq_unbound_release_workfn() for details. pool_workqueue | |
209 | * itself is also sched-RCU protected so that the first pwq can be | |
b09f4fd3 | 210 | * determined without grabbing wq->mutex. |
8864b4e5 TH |
211 | */ |
212 | struct work_struct unbound_release_work; | |
213 | struct rcu_head rcu; | |
e904e6c2 | 214 | } __aligned(1 << WORK_STRUCT_FLAG_BITS); |
1da177e4 | 215 | |
73f53c4a TH |
216 | /* |
217 | * Structure used to wait for workqueue flush. | |
218 | */ | |
219 | struct wq_flusher { | |
3c25a55d LJ |
220 | struct list_head list; /* WQ: list of flushers */ |
221 | int flush_color; /* WQ: flush color waiting for */ | |
73f53c4a TH |
222 | struct completion done; /* flush completion */ |
223 | }; | |
224 | ||
226223ab TH |
225 | struct wq_device; |
226 | ||
1da177e4 | 227 | /* |
c5aa87bb TH |
228 | * The externally visible workqueue. It relays the issued work items to |
229 | * the appropriate worker_pool through its pool_workqueues. | |
1da177e4 LT |
230 | */ |
231 | struct workqueue_struct { | |
3c25a55d | 232 | struct list_head pwqs; /* WR: all pwqs of this wq */ |
68e13a67 | 233 | struct list_head list; /* PL: list of all workqueues */ |
73f53c4a | 234 | |
3c25a55d LJ |
235 | struct mutex mutex; /* protects this wq */ |
236 | int work_color; /* WQ: current work color */ | |
237 | int flush_color; /* WQ: current flush color */ | |
112202d9 | 238 | atomic_t nr_pwqs_to_flush; /* flush in progress */ |
3c25a55d LJ |
239 | struct wq_flusher *first_flusher; /* WQ: first flusher */ |
240 | struct list_head flusher_queue; /* WQ: flush waiters */ | |
241 | struct list_head flusher_overflow; /* WQ: flush overflow list */ | |
73f53c4a | 242 | |
2e109a28 | 243 | struct list_head maydays; /* MD: pwqs requesting rescue */ |
e22bee78 TH |
244 | struct worker *rescuer; /* I: rescue worker */ |
245 | ||
87fc741e | 246 | int nr_drainers; /* WQ: drain in progress */ |
a357fc03 | 247 | int saved_max_active; /* WQ: saved pwq max_active */ |
226223ab | 248 | |
6029a918 | 249 | struct workqueue_attrs *unbound_attrs; /* WQ: only for unbound wqs */ |
4c16bd32 | 250 | struct pool_workqueue *dfl_pwq; /* WQ: only for unbound wqs */ |
6029a918 | 251 | |
226223ab TH |
252 | #ifdef CONFIG_SYSFS |
253 | struct wq_device *wq_dev; /* I: for sysfs interface */ | |
254 | #endif | |
4e6045f1 | 255 | #ifdef CONFIG_LOCKDEP |
4690c4ab | 256 | struct lockdep_map lockdep_map; |
4e6045f1 | 257 | #endif |
ecf6881f | 258 | char name[WQ_NAME_LEN]; /* I: workqueue name */ |
2728fd2f TH |
259 | |
260 | /* hot fields used during command issue, aligned to cacheline */ | |
261 | unsigned int flags ____cacheline_aligned; /* WQ: WQ_* flags */ | |
262 | struct pool_workqueue __percpu *cpu_pwqs; /* I: per-cpu pwqs */ | |
df2d5ae4 | 263 | struct pool_workqueue __rcu *numa_pwq_tbl[]; /* FR: unbound pwqs indexed by node */ |
1da177e4 LT |
264 | }; |
265 | ||
e904e6c2 TH |
266 | static struct kmem_cache *pwq_cache; |
267 | ||
bce90380 TH |
268 | static int wq_numa_tbl_len; /* highest possible NUMA node id + 1 */ |
269 | static cpumask_var_t *wq_numa_possible_cpumask; | |
270 | /* possible CPUs of each node */ | |
271 | ||
d55262c4 TH |
272 | static bool wq_disable_numa; |
273 | module_param_named(disable_numa, wq_disable_numa, bool, 0444); | |
274 | ||
cee22a15 VK |
275 | /* see the comment above the definition of WQ_POWER_EFFICIENT */ |
276 | #ifdef CONFIG_WQ_POWER_EFFICIENT_DEFAULT | |
277 | static bool wq_power_efficient = true; | |
278 | #else | |
279 | static bool wq_power_efficient; | |
280 | #endif | |
281 | ||
282 | module_param_named(power_efficient, wq_power_efficient, bool, 0444); | |
283 | ||
bce90380 TH |
284 | static bool wq_numa_enabled; /* unbound NUMA affinity enabled */ |
285 | ||
4c16bd32 TH |
286 | /* buf for wq_update_unbound_numa_attrs(), protected by CPU hotplug exclusion */ |
287 | static struct workqueue_attrs *wq_update_unbound_numa_attrs_buf; | |
288 | ||
68e13a67 | 289 | static DEFINE_MUTEX(wq_pool_mutex); /* protects pools and workqueues list */ |
2e109a28 | 290 | static DEFINE_SPINLOCK(wq_mayday_lock); /* protects wq->maydays list */ |
5bcab335 | 291 | |
68e13a67 LJ |
292 | static LIST_HEAD(workqueues); /* PL: list of all workqueues */ |
293 | static bool workqueue_freezing; /* PL: have wqs started freezing? */ | |
7d19c5ce TH |
294 | |
295 | /* the per-cpu worker pools */ | |
296 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct worker_pool [NR_STD_WORKER_POOLS], | |
297 | cpu_worker_pools); | |
298 | ||
68e13a67 | 299 | static DEFINE_IDR(worker_pool_idr); /* PR: idr of all pools */ |
7d19c5ce | 300 | |
68e13a67 | 301 | /* PL: hash of all unbound pools keyed by pool->attrs */ |
29c91e99 TH |
302 | static DEFINE_HASHTABLE(unbound_pool_hash, UNBOUND_POOL_HASH_ORDER); |
303 | ||
c5aa87bb | 304 | /* I: attributes used when instantiating standard unbound pools on demand */ |
29c91e99 TH |
305 | static struct workqueue_attrs *unbound_std_wq_attrs[NR_STD_WORKER_POOLS]; |
306 | ||
8a2b7538 TH |
307 | /* I: attributes used when instantiating ordered pools on demand */ |
308 | static struct workqueue_attrs *ordered_wq_attrs[NR_STD_WORKER_POOLS]; | |
309 | ||
d320c038 | 310 | struct workqueue_struct *system_wq __read_mostly; |
ad7b1f84 | 311 | EXPORT_SYMBOL(system_wq); |
044c782c | 312 | struct workqueue_struct *system_highpri_wq __read_mostly; |
1aabe902 | 313 | EXPORT_SYMBOL_GPL(system_highpri_wq); |
044c782c | 314 | struct workqueue_struct *system_long_wq __read_mostly; |
d320c038 | 315 | EXPORT_SYMBOL_GPL(system_long_wq); |
044c782c | 316 | struct workqueue_struct *system_unbound_wq __read_mostly; |
f3421797 | 317 | EXPORT_SYMBOL_GPL(system_unbound_wq); |
044c782c | 318 | struct workqueue_struct *system_freezable_wq __read_mostly; |
24d51add | 319 | EXPORT_SYMBOL_GPL(system_freezable_wq); |
0668106c VK |
320 | struct workqueue_struct *system_power_efficient_wq __read_mostly; |
321 | EXPORT_SYMBOL_GPL(system_power_efficient_wq); | |
322 | struct workqueue_struct *system_freezable_power_efficient_wq __read_mostly; | |
323 | EXPORT_SYMBOL_GPL(system_freezable_power_efficient_wq); | |
d320c038 | 324 | |
7d19c5ce TH |
325 | static int worker_thread(void *__worker); |
326 | static void copy_workqueue_attrs(struct workqueue_attrs *to, | |
327 | const struct workqueue_attrs *from); | |
328 | ||
97bd2347 TH |
329 | #define CREATE_TRACE_POINTS |
330 | #include <trace/events/workqueue.h> | |
331 | ||
68e13a67 | 332 | #define assert_rcu_or_pool_mutex() \ |
5bcab335 | 333 | rcu_lockdep_assert(rcu_read_lock_sched_held() || \ |
68e13a67 LJ |
334 | lockdep_is_held(&wq_pool_mutex), \ |
335 | "sched RCU or wq_pool_mutex should be held") | |
5bcab335 | 336 | |
b09f4fd3 | 337 | #define assert_rcu_or_wq_mutex(wq) \ |
76af4d93 | 338 | rcu_lockdep_assert(rcu_read_lock_sched_held() || \ |
b5927605 | 339 | lockdep_is_held(&wq->mutex), \ |
b09f4fd3 | 340 | "sched RCU or wq->mutex should be held") |
76af4d93 | 341 | |
f02ae73a TH |
342 | #define for_each_cpu_worker_pool(pool, cpu) \ |
343 | for ((pool) = &per_cpu(cpu_worker_pools, cpu)[0]; \ | |
344 | (pool) < &per_cpu(cpu_worker_pools, cpu)[NR_STD_WORKER_POOLS]; \ | |
7a62c2c8 | 345 | (pool)++) |
4ce62e9e | 346 | |
17116969 TH |
347 | /** |
348 | * for_each_pool - iterate through all worker_pools in the system | |
349 | * @pool: iteration cursor | |
611c92a0 | 350 | * @pi: integer used for iteration |
fa1b54e6 | 351 | * |
68e13a67 LJ |
352 | * This must be called either with wq_pool_mutex held or sched RCU read |
353 | * locked. If the pool needs to be used beyond the locking in effect, the | |
354 | * caller is responsible for guaranteeing that the pool stays online. | |
fa1b54e6 TH |
355 | * |
356 | * The if/else clause exists only for the lockdep assertion and can be | |
357 | * ignored. | |
17116969 | 358 | */ |
611c92a0 TH |
359 | #define for_each_pool(pool, pi) \ |
360 | idr_for_each_entry(&worker_pool_idr, pool, pi) \ | |
68e13a67 | 361 | if (({ assert_rcu_or_pool_mutex(); false; })) { } \ |
fa1b54e6 | 362 | else |
17116969 | 363 | |
822d8405 TH |
364 | /** |
365 | * for_each_pool_worker - iterate through all workers of a worker_pool | |
366 | * @worker: iteration cursor | |
367 | * @wi: integer used for iteration | |
368 | * @pool: worker_pool to iterate workers of | |
369 | * | |
9625ab17 | 370 | * This must be called with @pool->manager_mutex. |
822d8405 TH |
371 | * |
372 | * The if/else clause exists only for the lockdep assertion and can be | |
373 | * ignored. | |
374 | */ | |
375 | #define for_each_pool_worker(worker, wi, pool) \ | |
376 | idr_for_each_entry(&(pool)->worker_idr, (worker), (wi)) \ | |
9625ab17 | 377 | if (({ lockdep_assert_held(&pool->manager_mutex); false; })) { } \ |
822d8405 TH |
378 | else |
379 | ||
49e3cf44 TH |
380 | /** |
381 | * for_each_pwq - iterate through all pool_workqueues of the specified workqueue | |
382 | * @pwq: iteration cursor | |
383 | * @wq: the target workqueue | |
76af4d93 | 384 | * |
b09f4fd3 | 385 | * This must be called either with wq->mutex held or sched RCU read locked. |
794b18bc TH |
386 | * If the pwq needs to be used beyond the locking in effect, the caller is |
387 | * responsible for guaranteeing that the pwq stays online. | |
76af4d93 TH |
388 | * |
389 | * The if/else clause exists only for the lockdep assertion and can be | |
390 | * ignored. | |
49e3cf44 TH |
391 | */ |
392 | #define for_each_pwq(pwq, wq) \ | |
76af4d93 | 393 | list_for_each_entry_rcu((pwq), &(wq)->pwqs, pwqs_node) \ |
b09f4fd3 | 394 | if (({ assert_rcu_or_wq_mutex(wq); false; })) { } \ |
76af4d93 | 395 | else |
f3421797 | 396 | |
dc186ad7 TG |
397 | #ifdef CONFIG_DEBUG_OBJECTS_WORK |
398 | ||
399 | static struct debug_obj_descr work_debug_descr; | |
400 | ||
99777288 SG |
401 | static void *work_debug_hint(void *addr) |
402 | { | |
403 | return ((struct work_struct *) addr)->func; | |
404 | } | |
405 | ||
dc186ad7 TG |
406 | /* |
407 | * fixup_init is called when: | |
408 | * - an active object is initialized | |
409 | */ | |
410 | static int work_fixup_init(void *addr, enum debug_obj_state state) | |
411 | { | |
412 | struct work_struct *work = addr; | |
413 | ||
414 | switch (state) { | |
415 | case ODEBUG_STATE_ACTIVE: | |
416 | cancel_work_sync(work); | |
417 | debug_object_init(work, &work_debug_descr); | |
418 | return 1; | |
419 | default: | |
420 | return 0; | |
421 | } | |
422 | } | |
423 | ||
424 | /* | |
425 | * fixup_activate is called when: | |
426 | * - an active object is activated | |
427 | * - an unknown object is activated (might be a statically initialized object) | |
428 | */ | |
429 | static int work_fixup_activate(void *addr, enum debug_obj_state state) | |
430 | { | |
431 | struct work_struct *work = addr; | |
432 | ||
433 | switch (state) { | |
434 | ||
435 | case ODEBUG_STATE_NOTAVAILABLE: | |
436 | /* | |
437 | * This is not really a fixup. The work struct was | |
438 | * statically initialized. We just make sure that it | |
439 | * is tracked in the object tracker. | |
440 | */ | |
22df02bb | 441 | if (test_bit(WORK_STRUCT_STATIC_BIT, work_data_bits(work))) { |
dc186ad7 TG |
442 | debug_object_init(work, &work_debug_descr); |
443 | debug_object_activate(work, &work_debug_descr); | |
444 | return 0; | |
445 | } | |
446 | WARN_ON_ONCE(1); | |
447 | return 0; | |
448 | ||
449 | case ODEBUG_STATE_ACTIVE: | |
450 | WARN_ON(1); | |
451 | ||
452 | default: | |
453 | return 0; | |
454 | } | |
455 | } | |
456 | ||
457 | /* | |
458 | * fixup_free is called when: | |
459 | * - an active object is freed | |
460 | */ | |
461 | static int work_fixup_free(void *addr, enum debug_obj_state state) | |
462 | { | |
463 | struct work_struct *work = addr; | |
464 | ||
465 | switch (state) { | |
466 | case ODEBUG_STATE_ACTIVE: | |
467 | cancel_work_sync(work); | |
468 | debug_object_free(work, &work_debug_descr); | |
469 | return 1; | |
470 | default: | |
471 | return 0; | |
472 | } | |
473 | } | |
474 | ||
475 | static struct debug_obj_descr work_debug_descr = { | |
476 | .name = "work_struct", | |
99777288 | 477 | .debug_hint = work_debug_hint, |
dc186ad7 TG |
478 | .fixup_init = work_fixup_init, |
479 | .fixup_activate = work_fixup_activate, | |
480 | .fixup_free = work_fixup_free, | |
481 | }; | |
482 | ||
483 | static inline void debug_work_activate(struct work_struct *work) | |
484 | { | |
485 | debug_object_activate(work, &work_debug_descr); | |
486 | } | |
487 | ||
488 | static inline void debug_work_deactivate(struct work_struct *work) | |
489 | { | |
490 | debug_object_deactivate(work, &work_debug_descr); | |
491 | } | |
492 | ||
493 | void __init_work(struct work_struct *work, int onstack) | |
494 | { | |
495 | if (onstack) | |
496 | debug_object_init_on_stack(work, &work_debug_descr); | |
497 | else | |
498 | debug_object_init(work, &work_debug_descr); | |
499 | } | |
500 | EXPORT_SYMBOL_GPL(__init_work); | |
501 | ||
502 | void destroy_work_on_stack(struct work_struct *work) | |
503 | { | |
504 | debug_object_free(work, &work_debug_descr); | |
505 | } | |
506 | EXPORT_SYMBOL_GPL(destroy_work_on_stack); | |
507 | ||
ea2e64f2 TG |
508 | void destroy_delayed_work_on_stack(struct delayed_work *work) |
509 | { | |
510 | destroy_timer_on_stack(&work->timer); | |
511 | debug_object_free(&work->work, &work_debug_descr); | |
512 | } | |
513 | EXPORT_SYMBOL_GPL(destroy_delayed_work_on_stack); | |
514 | ||
dc186ad7 TG |
515 | #else |
516 | static inline void debug_work_activate(struct work_struct *work) { } | |
517 | static inline void debug_work_deactivate(struct work_struct *work) { } | |
518 | #endif | |
519 | ||
4e8b22bd LB |
520 | /** |
521 | * worker_pool_assign_id - allocate ID and assing it to @pool | |
522 | * @pool: the pool pointer of interest | |
523 | * | |
524 | * Returns 0 if ID in [0, WORK_OFFQ_POOL_NONE) is allocated and assigned | |
525 | * successfully, -errno on failure. | |
526 | */ | |
9daf9e67 TH |
527 | static int worker_pool_assign_id(struct worker_pool *pool) |
528 | { | |
529 | int ret; | |
530 | ||
68e13a67 | 531 | lockdep_assert_held(&wq_pool_mutex); |
5bcab335 | 532 | |
4e8b22bd LB |
533 | ret = idr_alloc(&worker_pool_idr, pool, 0, WORK_OFFQ_POOL_NONE, |
534 | GFP_KERNEL); | |
229641a6 | 535 | if (ret >= 0) { |
e68035fb | 536 | pool->id = ret; |
229641a6 TH |
537 | return 0; |
538 | } | |
fa1b54e6 | 539 | return ret; |
7c3eed5c TH |
540 | } |
541 | ||
df2d5ae4 TH |
542 | /** |
543 | * unbound_pwq_by_node - return the unbound pool_workqueue for the given node | |
544 | * @wq: the target workqueue | |
545 | * @node: the node ID | |
546 | * | |
547 | * This must be called either with pwq_lock held or sched RCU read locked. | |
548 | * If the pwq needs to be used beyond the locking in effect, the caller is | |
549 | * responsible for guaranteeing that the pwq stays online. | |
d185af30 YB |
550 | * |
551 | * Return: The unbound pool_workqueue for @node. | |
df2d5ae4 TH |
552 | */ |
553 | static struct pool_workqueue *unbound_pwq_by_node(struct workqueue_struct *wq, | |
554 | int node) | |
555 | { | |
556 | assert_rcu_or_wq_mutex(wq); | |
557 | return rcu_dereference_raw(wq->numa_pwq_tbl[node]); | |
558 | } | |
559 | ||
73f53c4a TH |
560 | static unsigned int work_color_to_flags(int color) |
561 | { | |
562 | return color << WORK_STRUCT_COLOR_SHIFT; | |
563 | } | |
564 | ||
565 | static int get_work_color(struct work_struct *work) | |
566 | { | |
567 | return (*work_data_bits(work) >> WORK_STRUCT_COLOR_SHIFT) & | |
568 | ((1 << WORK_STRUCT_COLOR_BITS) - 1); | |
569 | } | |
570 | ||
571 | static int work_next_color(int color) | |
572 | { | |
573 | return (color + 1) % WORK_NR_COLORS; | |
574 | } | |
1da177e4 | 575 | |
14441960 | 576 | /* |
112202d9 TH |
577 | * While queued, %WORK_STRUCT_PWQ is set and non flag bits of a work's data |
578 | * contain the pointer to the queued pwq. Once execution starts, the flag | |
7c3eed5c | 579 | * is cleared and the high bits contain OFFQ flags and pool ID. |
7a22ad75 | 580 | * |
112202d9 TH |
581 | * set_work_pwq(), set_work_pool_and_clear_pending(), mark_work_canceling() |
582 | * and clear_work_data() can be used to set the pwq, pool or clear | |
bbb68dfa TH |
583 | * work->data. These functions should only be called while the work is |
584 | * owned - ie. while the PENDING bit is set. | |
7a22ad75 | 585 | * |
112202d9 | 586 | * get_work_pool() and get_work_pwq() can be used to obtain the pool or pwq |
7c3eed5c | 587 | * corresponding to a work. Pool is available once the work has been |
112202d9 | 588 | * queued anywhere after initialization until it is sync canceled. pwq is |
7c3eed5c | 589 | * available only while the work item is queued. |
7a22ad75 | 590 | * |
bbb68dfa TH |
591 | * %WORK_OFFQ_CANCELING is used to mark a work item which is being |
592 | * canceled. While being canceled, a work item may have its PENDING set | |
593 | * but stay off timer and worklist for arbitrarily long and nobody should | |
594 | * try to steal the PENDING bit. | |
14441960 | 595 | */ |
7a22ad75 TH |
596 | static inline void set_work_data(struct work_struct *work, unsigned long data, |
597 | unsigned long flags) | |
365970a1 | 598 | { |
6183c009 | 599 | WARN_ON_ONCE(!work_pending(work)); |
7a22ad75 TH |
600 | atomic_long_set(&work->data, data | flags | work_static(work)); |
601 | } | |
365970a1 | 602 | |
112202d9 | 603 | static void set_work_pwq(struct work_struct *work, struct pool_workqueue *pwq, |
7a22ad75 TH |
604 | unsigned long extra_flags) |
605 | { | |
112202d9 TH |
606 | set_work_data(work, (unsigned long)pwq, |
607 | WORK_STRUCT_PENDING | WORK_STRUCT_PWQ | extra_flags); | |
365970a1 DH |
608 | } |
609 | ||
4468a00f LJ |
610 | static void set_work_pool_and_keep_pending(struct work_struct *work, |
611 | int pool_id) | |
612 | { | |
613 | set_work_data(work, (unsigned long)pool_id << WORK_OFFQ_POOL_SHIFT, | |
614 | WORK_STRUCT_PENDING); | |
615 | } | |
616 | ||
7c3eed5c TH |
617 | static void set_work_pool_and_clear_pending(struct work_struct *work, |
618 | int pool_id) | |
7a22ad75 | 619 | { |
23657bb1 TH |
620 | /* |
621 | * The following wmb is paired with the implied mb in | |
622 | * test_and_set_bit(PENDING) and ensures all updates to @work made | |
623 | * here are visible to and precede any updates by the next PENDING | |
624 | * owner. | |
625 | */ | |
626 | smp_wmb(); | |
7c3eed5c | 627 | set_work_data(work, (unsigned long)pool_id << WORK_OFFQ_POOL_SHIFT, 0); |
7a22ad75 | 628 | } |
f756d5e2 | 629 | |
7a22ad75 | 630 | static void clear_work_data(struct work_struct *work) |
1da177e4 | 631 | { |
7c3eed5c TH |
632 | smp_wmb(); /* see set_work_pool_and_clear_pending() */ |
633 | set_work_data(work, WORK_STRUCT_NO_POOL, 0); | |
1da177e4 LT |
634 | } |
635 | ||
112202d9 | 636 | static struct pool_workqueue *get_work_pwq(struct work_struct *work) |
b1f4ec17 | 637 | { |
e120153d | 638 | unsigned long data = atomic_long_read(&work->data); |
7a22ad75 | 639 | |
112202d9 | 640 | if (data & WORK_STRUCT_PWQ) |
e120153d TH |
641 | return (void *)(data & WORK_STRUCT_WQ_DATA_MASK); |
642 | else | |
643 | return NULL; | |
4d707b9f ON |
644 | } |
645 | ||
7c3eed5c TH |
646 | /** |
647 | * get_work_pool - return the worker_pool a given work was associated with | |
648 | * @work: the work item of interest | |
649 | * | |
68e13a67 LJ |
650 | * Pools are created and destroyed under wq_pool_mutex, and allows read |
651 | * access under sched-RCU read lock. As such, this function should be | |
652 | * called under wq_pool_mutex or with preemption disabled. | |
fa1b54e6 TH |
653 | * |
654 | * All fields of the returned pool are accessible as long as the above | |
655 | * mentioned locking is in effect. If the returned pool needs to be used | |
656 | * beyond the critical section, the caller is responsible for ensuring the | |
657 | * returned pool is and stays online. | |
d185af30 YB |
658 | * |
659 | * Return: The worker_pool @work was last associated with. %NULL if none. | |
7c3eed5c TH |
660 | */ |
661 | static struct worker_pool *get_work_pool(struct work_struct *work) | |
365970a1 | 662 | { |
e120153d | 663 | unsigned long data = atomic_long_read(&work->data); |
7c3eed5c | 664 | int pool_id; |
7a22ad75 | 665 | |
68e13a67 | 666 | assert_rcu_or_pool_mutex(); |
fa1b54e6 | 667 | |
112202d9 TH |
668 | if (data & WORK_STRUCT_PWQ) |
669 | return ((struct pool_workqueue *) | |
7c3eed5c | 670 | (data & WORK_STRUCT_WQ_DATA_MASK))->pool; |
7a22ad75 | 671 | |
7c3eed5c TH |
672 | pool_id = data >> WORK_OFFQ_POOL_SHIFT; |
673 | if (pool_id == WORK_OFFQ_POOL_NONE) | |
7a22ad75 TH |
674 | return NULL; |
675 | ||
fa1b54e6 | 676 | return idr_find(&worker_pool_idr, pool_id); |
7c3eed5c TH |
677 | } |
678 | ||
679 | /** | |
680 | * get_work_pool_id - return the worker pool ID a given work is associated with | |
681 | * @work: the work item of interest | |
682 | * | |
d185af30 | 683 | * Return: The worker_pool ID @work was last associated with. |
7c3eed5c TH |
684 | * %WORK_OFFQ_POOL_NONE if none. |
685 | */ | |
686 | static int get_work_pool_id(struct work_struct *work) | |
687 | { | |
54d5b7d0 LJ |
688 | unsigned long data = atomic_long_read(&work->data); |
689 | ||
112202d9 TH |
690 | if (data & WORK_STRUCT_PWQ) |
691 | return ((struct pool_workqueue *) | |
54d5b7d0 | 692 | (data & WORK_STRUCT_WQ_DATA_MASK))->pool->id; |
7c3eed5c | 693 | |
54d5b7d0 | 694 | return data >> WORK_OFFQ_POOL_SHIFT; |
7c3eed5c TH |
695 | } |
696 | ||
bbb68dfa TH |
697 | static void mark_work_canceling(struct work_struct *work) |
698 | { | |
7c3eed5c | 699 | unsigned long pool_id = get_work_pool_id(work); |
bbb68dfa | 700 | |
7c3eed5c TH |
701 | pool_id <<= WORK_OFFQ_POOL_SHIFT; |
702 | set_work_data(work, pool_id | WORK_OFFQ_CANCELING, WORK_STRUCT_PENDING); | |
bbb68dfa TH |
703 | } |
704 | ||
705 | static bool work_is_canceling(struct work_struct *work) | |
706 | { | |
707 | unsigned long data = atomic_long_read(&work->data); | |
708 | ||
112202d9 | 709 | return !(data & WORK_STRUCT_PWQ) && (data & WORK_OFFQ_CANCELING); |
bbb68dfa TH |
710 | } |
711 | ||
e22bee78 | 712 | /* |
3270476a TH |
713 | * Policy functions. These define the policies on how the global worker |
714 | * pools are managed. Unless noted otherwise, these functions assume that | |
d565ed63 | 715 | * they're being called with pool->lock held. |
e22bee78 TH |
716 | */ |
717 | ||
63d95a91 | 718 | static bool __need_more_worker(struct worker_pool *pool) |
a848e3b6 | 719 | { |
e19e397a | 720 | return !atomic_read(&pool->nr_running); |
a848e3b6 ON |
721 | } |
722 | ||
4594bf15 | 723 | /* |
e22bee78 TH |
724 | * Need to wake up a worker? Called from anything but currently |
725 | * running workers. | |
974271c4 TH |
726 | * |
727 | * Note that, because unbound workers never contribute to nr_running, this | |
706026c2 | 728 | * function will always return %true for unbound pools as long as the |
974271c4 | 729 | * worklist isn't empty. |
4594bf15 | 730 | */ |
63d95a91 | 731 | static bool need_more_worker(struct worker_pool *pool) |
365970a1 | 732 | { |
63d95a91 | 733 | return !list_empty(&pool->worklist) && __need_more_worker(pool); |
e22bee78 | 734 | } |
4594bf15 | 735 | |
e22bee78 | 736 | /* Can I start working? Called from busy but !running workers. */ |
63d95a91 | 737 | static bool may_start_working(struct worker_pool *pool) |
e22bee78 | 738 | { |
63d95a91 | 739 | return pool->nr_idle; |
e22bee78 TH |
740 | } |
741 | ||
742 | /* Do I need to keep working? Called from currently running workers. */ | |
63d95a91 | 743 | static bool keep_working(struct worker_pool *pool) |
e22bee78 | 744 | { |
e19e397a TH |
745 | return !list_empty(&pool->worklist) && |
746 | atomic_read(&pool->nr_running) <= 1; | |
e22bee78 TH |
747 | } |
748 | ||
749 | /* Do we need a new worker? Called from manager. */ | |
63d95a91 | 750 | static bool need_to_create_worker(struct worker_pool *pool) |
e22bee78 | 751 | { |
63d95a91 | 752 | return need_more_worker(pool) && !may_start_working(pool); |
e22bee78 | 753 | } |
365970a1 | 754 | |
e22bee78 | 755 | /* Do I need to be the manager? */ |
63d95a91 | 756 | static bool need_to_manage_workers(struct worker_pool *pool) |
e22bee78 | 757 | { |
63d95a91 | 758 | return need_to_create_worker(pool) || |
11ebea50 | 759 | (pool->flags & POOL_MANAGE_WORKERS); |
e22bee78 TH |
760 | } |
761 | ||
762 | /* Do we have too many workers and should some go away? */ | |
63d95a91 | 763 | static bool too_many_workers(struct worker_pool *pool) |
e22bee78 | 764 | { |
34a06bd6 | 765 | bool managing = mutex_is_locked(&pool->manager_arb); |
63d95a91 TH |
766 | int nr_idle = pool->nr_idle + managing; /* manager is considered idle */ |
767 | int nr_busy = pool->nr_workers - nr_idle; | |
e22bee78 | 768 | |
ea1abd61 LJ |
769 | /* |
770 | * nr_idle and idle_list may disagree if idle rebinding is in | |
771 | * progress. Never return %true if idle_list is empty. | |
772 | */ | |
773 | if (list_empty(&pool->idle_list)) | |
774 | return false; | |
775 | ||
e22bee78 | 776 | return nr_idle > 2 && (nr_idle - 2) * MAX_IDLE_WORKERS_RATIO >= nr_busy; |
365970a1 DH |
777 | } |
778 | ||
4d707b9f | 779 | /* |
e22bee78 TH |
780 | * Wake up functions. |
781 | */ | |
782 | ||
7e11629d | 783 | /* Return the first worker. Safe with preemption disabled */ |
63d95a91 | 784 | static struct worker *first_worker(struct worker_pool *pool) |
7e11629d | 785 | { |
63d95a91 | 786 | if (unlikely(list_empty(&pool->idle_list))) |
7e11629d TH |
787 | return NULL; |
788 | ||
63d95a91 | 789 | return list_first_entry(&pool->idle_list, struct worker, entry); |
7e11629d TH |
790 | } |
791 | ||
792 | /** | |
793 | * wake_up_worker - wake up an idle worker | |
63d95a91 | 794 | * @pool: worker pool to wake worker from |
7e11629d | 795 | * |
63d95a91 | 796 | * Wake up the first idle worker of @pool. |
7e11629d TH |
797 | * |
798 | * CONTEXT: | |
d565ed63 | 799 | * spin_lock_irq(pool->lock). |
7e11629d | 800 | */ |
63d95a91 | 801 | static void wake_up_worker(struct worker_pool *pool) |
7e11629d | 802 | { |
63d95a91 | 803 | struct worker *worker = first_worker(pool); |
7e11629d TH |
804 | |
805 | if (likely(worker)) | |
806 | wake_up_process(worker->task); | |
807 | } | |
808 | ||
d302f017 | 809 | /** |
e22bee78 TH |
810 | * wq_worker_waking_up - a worker is waking up |
811 | * @task: task waking up | |
812 | * @cpu: CPU @task is waking up to | |
813 | * | |
814 | * This function is called during try_to_wake_up() when a worker is | |
815 | * being awoken. | |
816 | * | |
817 | * CONTEXT: | |
818 | * spin_lock_irq(rq->lock) | |
819 | */ | |
d84ff051 | 820 | void wq_worker_waking_up(struct task_struct *task, int cpu) |
e22bee78 TH |
821 | { |
822 | struct worker *worker = kthread_data(task); | |
823 | ||
36576000 | 824 | if (!(worker->flags & WORKER_NOT_RUNNING)) { |
ec22ca5e | 825 | WARN_ON_ONCE(worker->pool->cpu != cpu); |
e19e397a | 826 | atomic_inc(&worker->pool->nr_running); |
36576000 | 827 | } |
e22bee78 TH |
828 | } |
829 | ||
830 | /** | |
831 | * wq_worker_sleeping - a worker is going to sleep | |
832 | * @task: task going to sleep | |
833 | * @cpu: CPU in question, must be the current CPU number | |
834 | * | |
835 | * This function is called during schedule() when a busy worker is | |
836 | * going to sleep. Worker on the same cpu can be woken up by | |
837 | * returning pointer to its task. | |
838 | * | |
839 | * CONTEXT: | |
840 | * spin_lock_irq(rq->lock) | |
841 | * | |
d185af30 | 842 | * Return: |
e22bee78 TH |
843 | * Worker task on @cpu to wake up, %NULL if none. |
844 | */ | |
d84ff051 | 845 | struct task_struct *wq_worker_sleeping(struct task_struct *task, int cpu) |
e22bee78 TH |
846 | { |
847 | struct worker *worker = kthread_data(task), *to_wakeup = NULL; | |
111c225a | 848 | struct worker_pool *pool; |
e22bee78 | 849 | |
111c225a TH |
850 | /* |
851 | * Rescuers, which may not have all the fields set up like normal | |
852 | * workers, also reach here, let's not access anything before | |
853 | * checking NOT_RUNNING. | |
854 | */ | |
2d64672e | 855 | if (worker->flags & WORKER_NOT_RUNNING) |
e22bee78 TH |
856 | return NULL; |
857 | ||
111c225a | 858 | pool = worker->pool; |
111c225a | 859 | |
e22bee78 | 860 | /* this can only happen on the local cpu */ |
6183c009 TH |
861 | if (WARN_ON_ONCE(cpu != raw_smp_processor_id())) |
862 | return NULL; | |
e22bee78 TH |
863 | |
864 | /* | |
865 | * The counterpart of the following dec_and_test, implied mb, | |
866 | * worklist not empty test sequence is in insert_work(). | |
867 | * Please read comment there. | |
868 | * | |
628c78e7 TH |
869 | * NOT_RUNNING is clear. This means that we're bound to and |
870 | * running on the local cpu w/ rq lock held and preemption | |
871 | * disabled, which in turn means that none else could be | |
d565ed63 | 872 | * manipulating idle_list, so dereferencing idle_list without pool |
628c78e7 | 873 | * lock is safe. |
e22bee78 | 874 | */ |
e19e397a TH |
875 | if (atomic_dec_and_test(&pool->nr_running) && |
876 | !list_empty(&pool->worklist)) | |
63d95a91 | 877 | to_wakeup = first_worker(pool); |
e22bee78 TH |
878 | return to_wakeup ? to_wakeup->task : NULL; |
879 | } | |
880 | ||
881 | /** | |
882 | * worker_set_flags - set worker flags and adjust nr_running accordingly | |
cb444766 | 883 | * @worker: self |
d302f017 TH |
884 | * @flags: flags to set |
885 | * @wakeup: wakeup an idle worker if necessary | |
886 | * | |
e22bee78 TH |
887 | * Set @flags in @worker->flags and adjust nr_running accordingly. If |
888 | * nr_running becomes zero and @wakeup is %true, an idle worker is | |
889 | * woken up. | |
d302f017 | 890 | * |
cb444766 | 891 | * CONTEXT: |
d565ed63 | 892 | * spin_lock_irq(pool->lock) |
d302f017 TH |
893 | */ |
894 | static inline void worker_set_flags(struct worker *worker, unsigned int flags, | |
895 | bool wakeup) | |
896 | { | |
bd7bdd43 | 897 | struct worker_pool *pool = worker->pool; |
e22bee78 | 898 | |
cb444766 TH |
899 | WARN_ON_ONCE(worker->task != current); |
900 | ||
e22bee78 TH |
901 | /* |
902 | * If transitioning into NOT_RUNNING, adjust nr_running and | |
903 | * wake up an idle worker as necessary if requested by | |
904 | * @wakeup. | |
905 | */ | |
906 | if ((flags & WORKER_NOT_RUNNING) && | |
907 | !(worker->flags & WORKER_NOT_RUNNING)) { | |
e22bee78 | 908 | if (wakeup) { |
e19e397a | 909 | if (atomic_dec_and_test(&pool->nr_running) && |
bd7bdd43 | 910 | !list_empty(&pool->worklist)) |
63d95a91 | 911 | wake_up_worker(pool); |
e22bee78 | 912 | } else |
e19e397a | 913 | atomic_dec(&pool->nr_running); |
e22bee78 TH |
914 | } |
915 | ||
d302f017 TH |
916 | worker->flags |= flags; |
917 | } | |
918 | ||
919 | /** | |
e22bee78 | 920 | * worker_clr_flags - clear worker flags and adjust nr_running accordingly |
cb444766 | 921 | * @worker: self |
d302f017 TH |
922 | * @flags: flags to clear |
923 | * | |
e22bee78 | 924 | * Clear @flags in @worker->flags and adjust nr_running accordingly. |
d302f017 | 925 | * |
cb444766 | 926 | * CONTEXT: |
d565ed63 | 927 | * spin_lock_irq(pool->lock) |
d302f017 TH |
928 | */ |
929 | static inline void worker_clr_flags(struct worker *worker, unsigned int flags) | |
930 | { | |
63d95a91 | 931 | struct worker_pool *pool = worker->pool; |
e22bee78 TH |
932 | unsigned int oflags = worker->flags; |
933 | ||
cb444766 TH |
934 | WARN_ON_ONCE(worker->task != current); |
935 | ||
d302f017 | 936 | worker->flags &= ~flags; |
e22bee78 | 937 | |
42c025f3 TH |
938 | /* |
939 | * If transitioning out of NOT_RUNNING, increment nr_running. Note | |
940 | * that the nested NOT_RUNNING is not a noop. NOT_RUNNING is mask | |
941 | * of multiple flags, not a single flag. | |
942 | */ | |
e22bee78 TH |
943 | if ((flags & WORKER_NOT_RUNNING) && (oflags & WORKER_NOT_RUNNING)) |
944 | if (!(worker->flags & WORKER_NOT_RUNNING)) | |
e19e397a | 945 | atomic_inc(&pool->nr_running); |
d302f017 TH |
946 | } |
947 | ||
8cca0eea TH |
948 | /** |
949 | * find_worker_executing_work - find worker which is executing a work | |
c9e7cf27 | 950 | * @pool: pool of interest |
8cca0eea TH |
951 | * @work: work to find worker for |
952 | * | |
c9e7cf27 TH |
953 | * Find a worker which is executing @work on @pool by searching |
954 | * @pool->busy_hash which is keyed by the address of @work. For a worker | |
a2c1c57b TH |
955 | * to match, its current execution should match the address of @work and |
956 | * its work function. This is to avoid unwanted dependency between | |
957 | * unrelated work executions through a work item being recycled while still | |
958 | * being executed. | |
959 | * | |
960 | * This is a bit tricky. A work item may be freed once its execution | |
961 | * starts and nothing prevents the freed area from being recycled for | |
962 | * another work item. If the same work item address ends up being reused | |
963 | * before the original execution finishes, workqueue will identify the | |
964 | * recycled work item as currently executing and make it wait until the | |
965 | * current execution finishes, introducing an unwanted dependency. | |
966 | * | |
c5aa87bb TH |
967 | * This function checks the work item address and work function to avoid |
968 | * false positives. Note that this isn't complete as one may construct a | |
969 | * work function which can introduce dependency onto itself through a | |
970 | * recycled work item. Well, if somebody wants to shoot oneself in the | |
971 | * foot that badly, there's only so much we can do, and if such deadlock | |
972 | * actually occurs, it should be easy to locate the culprit work function. | |
8cca0eea TH |
973 | * |
974 | * CONTEXT: | |
d565ed63 | 975 | * spin_lock_irq(pool->lock). |
8cca0eea | 976 | * |
d185af30 YB |
977 | * Return: |
978 | * Pointer to worker which is executing @work if found, %NULL | |
8cca0eea | 979 | * otherwise. |
4d707b9f | 980 | */ |
c9e7cf27 | 981 | static struct worker *find_worker_executing_work(struct worker_pool *pool, |
8cca0eea | 982 | struct work_struct *work) |
4d707b9f | 983 | { |
42f8570f | 984 | struct worker *worker; |
42f8570f | 985 | |
b67bfe0d | 986 | hash_for_each_possible(pool->busy_hash, worker, hentry, |
a2c1c57b TH |
987 | (unsigned long)work) |
988 | if (worker->current_work == work && | |
989 | worker->current_func == work->func) | |
42f8570f SL |
990 | return worker; |
991 | ||
992 | return NULL; | |
4d707b9f ON |
993 | } |
994 | ||
bf4ede01 TH |
995 | /** |
996 | * move_linked_works - move linked works to a list | |
997 | * @work: start of series of works to be scheduled | |
998 | * @head: target list to append @work to | |
999 | * @nextp: out paramter for nested worklist walking | |
1000 | * | |
1001 | * Schedule linked works starting from @work to @head. Work series to | |
1002 | * be scheduled starts at @work and includes any consecutive work with | |
1003 | * WORK_STRUCT_LINKED set in its predecessor. | |
1004 | * | |
1005 | * If @nextp is not NULL, it's updated to point to the next work of | |
1006 | * the last scheduled work. This allows move_linked_works() to be | |
1007 | * nested inside outer list_for_each_entry_safe(). | |
1008 | * | |
1009 | * CONTEXT: | |
d565ed63 | 1010 | * spin_lock_irq(pool->lock). |
bf4ede01 TH |
1011 | */ |
1012 | static void move_linked_works(struct work_struct *work, struct list_head *head, | |
1013 | struct work_struct **nextp) | |
1014 | { | |
1015 | struct work_struct *n; | |
1016 | ||
1017 | /* | |
1018 | * Linked worklist will always end before the end of the list, | |
1019 | * use NULL for list head. | |
1020 | */ | |
1021 | list_for_each_entry_safe_from(work, n, NULL, entry) { | |
1022 | list_move_tail(&work->entry, head); | |
1023 | if (!(*work_data_bits(work) & WORK_STRUCT_LINKED)) | |
1024 | break; | |
1025 | } | |
1026 | ||
1027 | /* | |
1028 | * If we're already inside safe list traversal and have moved | |
1029 | * multiple works to the scheduled queue, the next position | |
1030 | * needs to be updated. | |
1031 | */ | |
1032 | if (nextp) | |
1033 | *nextp = n; | |
1034 | } | |
1035 | ||
8864b4e5 TH |
1036 | /** |
1037 | * get_pwq - get an extra reference on the specified pool_workqueue | |
1038 | * @pwq: pool_workqueue to get | |
1039 | * | |
1040 | * Obtain an extra reference on @pwq. The caller should guarantee that | |
1041 | * @pwq has positive refcnt and be holding the matching pool->lock. | |
1042 | */ | |
1043 | static void get_pwq(struct pool_workqueue *pwq) | |
1044 | { | |
1045 | lockdep_assert_held(&pwq->pool->lock); | |
1046 | WARN_ON_ONCE(pwq->refcnt <= 0); | |
1047 | pwq->refcnt++; | |
1048 | } | |
1049 | ||
1050 | /** | |
1051 | * put_pwq - put a pool_workqueue reference | |
1052 | * @pwq: pool_workqueue to put | |
1053 | * | |
1054 | * Drop a reference of @pwq. If its refcnt reaches zero, schedule its | |
1055 | * destruction. The caller should be holding the matching pool->lock. | |
1056 | */ | |
1057 | static void put_pwq(struct pool_workqueue *pwq) | |
1058 | { | |
1059 | lockdep_assert_held(&pwq->pool->lock); | |
1060 | if (likely(--pwq->refcnt)) | |
1061 | return; | |
1062 | if (WARN_ON_ONCE(!(pwq->wq->flags & WQ_UNBOUND))) | |
1063 | return; | |
1064 | /* | |
1065 | * @pwq can't be released under pool->lock, bounce to | |
1066 | * pwq_unbound_release_workfn(). This never recurses on the same | |
1067 | * pool->lock as this path is taken only for unbound workqueues and | |
1068 | * the release work item is scheduled on a per-cpu workqueue. To | |
1069 | * avoid lockdep warning, unbound pool->locks are given lockdep | |
1070 | * subclass of 1 in get_unbound_pool(). | |
1071 | */ | |
1072 | schedule_work(&pwq->unbound_release_work); | |
1073 | } | |
1074 | ||
dce90d47 TH |
1075 | /** |
1076 | * put_pwq_unlocked - put_pwq() with surrounding pool lock/unlock | |
1077 | * @pwq: pool_workqueue to put (can be %NULL) | |
1078 | * | |
1079 | * put_pwq() with locking. This function also allows %NULL @pwq. | |
1080 | */ | |
1081 | static void put_pwq_unlocked(struct pool_workqueue *pwq) | |
1082 | { | |
1083 | if (pwq) { | |
1084 | /* | |
1085 | * As both pwqs and pools are sched-RCU protected, the | |
1086 | * following lock operations are safe. | |
1087 | */ | |
1088 | spin_lock_irq(&pwq->pool->lock); | |
1089 | put_pwq(pwq); | |
1090 | spin_unlock_irq(&pwq->pool->lock); | |
1091 | } | |
1092 | } | |
1093 | ||
112202d9 | 1094 | static void pwq_activate_delayed_work(struct work_struct *work) |
bf4ede01 | 1095 | { |
112202d9 | 1096 | struct pool_workqueue *pwq = get_work_pwq(work); |
bf4ede01 TH |
1097 | |
1098 | trace_workqueue_activate_work(work); | |
112202d9 | 1099 | move_linked_works(work, &pwq->pool->worklist, NULL); |
bf4ede01 | 1100 | __clear_bit(WORK_STRUCT_DELAYED_BIT, work_data_bits(work)); |
112202d9 | 1101 | pwq->nr_active++; |
bf4ede01 TH |
1102 | } |
1103 | ||
112202d9 | 1104 | static void pwq_activate_first_delayed(struct pool_workqueue *pwq) |
3aa62497 | 1105 | { |
112202d9 | 1106 | struct work_struct *work = list_first_entry(&pwq->delayed_works, |
3aa62497 LJ |
1107 | struct work_struct, entry); |
1108 | ||
112202d9 | 1109 | pwq_activate_delayed_work(work); |
3aa62497 LJ |
1110 | } |
1111 | ||
bf4ede01 | 1112 | /** |
112202d9 TH |
1113 | * pwq_dec_nr_in_flight - decrement pwq's nr_in_flight |
1114 | * @pwq: pwq of interest | |
bf4ede01 | 1115 | * @color: color of work which left the queue |
bf4ede01 TH |
1116 | * |
1117 | * A work either has completed or is removed from pending queue, | |
112202d9 | 1118 | * decrement nr_in_flight of its pwq and handle workqueue flushing. |
bf4ede01 TH |
1119 | * |
1120 | * CONTEXT: | |
d565ed63 | 1121 | * spin_lock_irq(pool->lock). |
bf4ede01 | 1122 | */ |
112202d9 | 1123 | static void pwq_dec_nr_in_flight(struct pool_workqueue *pwq, int color) |
bf4ede01 | 1124 | { |
8864b4e5 | 1125 | /* uncolored work items don't participate in flushing or nr_active */ |
bf4ede01 | 1126 | if (color == WORK_NO_COLOR) |
8864b4e5 | 1127 | goto out_put; |
bf4ede01 | 1128 | |
112202d9 | 1129 | pwq->nr_in_flight[color]--; |
bf4ede01 | 1130 | |
112202d9 TH |
1131 | pwq->nr_active--; |
1132 | if (!list_empty(&pwq->delayed_works)) { | |
b3f9f405 | 1133 | /* one down, submit a delayed one */ |
112202d9 TH |
1134 | if (pwq->nr_active < pwq->max_active) |
1135 | pwq_activate_first_delayed(pwq); | |
bf4ede01 TH |
1136 | } |
1137 | ||
1138 | /* is flush in progress and are we at the flushing tip? */ | |
112202d9 | 1139 | if (likely(pwq->flush_color != color)) |
8864b4e5 | 1140 | goto out_put; |
bf4ede01 TH |
1141 | |
1142 | /* are there still in-flight works? */ | |
112202d9 | 1143 | if (pwq->nr_in_flight[color]) |
8864b4e5 | 1144 | goto out_put; |
bf4ede01 | 1145 | |
112202d9 TH |
1146 | /* this pwq is done, clear flush_color */ |
1147 | pwq->flush_color = -1; | |
bf4ede01 TH |
1148 | |
1149 | /* | |
112202d9 | 1150 | * If this was the last pwq, wake up the first flusher. It |
bf4ede01 TH |
1151 | * will handle the rest. |
1152 | */ | |
112202d9 TH |
1153 | if (atomic_dec_and_test(&pwq->wq->nr_pwqs_to_flush)) |
1154 | complete(&pwq->wq->first_flusher->done); | |
8864b4e5 TH |
1155 | out_put: |
1156 | put_pwq(pwq); | |
bf4ede01 TH |
1157 | } |
1158 | ||
36e227d2 | 1159 | /** |
bbb68dfa | 1160 | * try_to_grab_pending - steal work item from worklist and disable irq |
36e227d2 TH |
1161 | * @work: work item to steal |
1162 | * @is_dwork: @work is a delayed_work | |
bbb68dfa | 1163 | * @flags: place to store irq state |
36e227d2 TH |
1164 | * |
1165 | * Try to grab PENDING bit of @work. This function can handle @work in any | |
d185af30 | 1166 | * stable state - idle, on timer or on worklist. |
36e227d2 | 1167 | * |
d185af30 | 1168 | * Return: |
36e227d2 TH |
1169 | * 1 if @work was pending and we successfully stole PENDING |
1170 | * 0 if @work was idle and we claimed PENDING | |
1171 | * -EAGAIN if PENDING couldn't be grabbed at the moment, safe to busy-retry | |
bbb68dfa TH |
1172 | * -ENOENT if someone else is canceling @work, this state may persist |
1173 | * for arbitrarily long | |
36e227d2 | 1174 | * |
d185af30 | 1175 | * Note: |
bbb68dfa | 1176 | * On >= 0 return, the caller owns @work's PENDING bit. To avoid getting |
e0aecdd8 TH |
1177 | * interrupted while holding PENDING and @work off queue, irq must be |
1178 | * disabled on entry. This, combined with delayed_work->timer being | |
1179 | * irqsafe, ensures that we return -EAGAIN for finite short period of time. | |
bbb68dfa TH |
1180 | * |
1181 | * On successful return, >= 0, irq is disabled and the caller is | |
1182 | * responsible for releasing it using local_irq_restore(*@flags). | |
1183 | * | |
e0aecdd8 | 1184 | * This function is safe to call from any context including IRQ handler. |
bf4ede01 | 1185 | */ |
bbb68dfa TH |
1186 | static int try_to_grab_pending(struct work_struct *work, bool is_dwork, |
1187 | unsigned long *flags) | |
bf4ede01 | 1188 | { |
d565ed63 | 1189 | struct worker_pool *pool; |
112202d9 | 1190 | struct pool_workqueue *pwq; |
bf4ede01 | 1191 | |
bbb68dfa TH |
1192 | local_irq_save(*flags); |
1193 | ||
36e227d2 TH |
1194 | /* try to steal the timer if it exists */ |
1195 | if (is_dwork) { | |
1196 | struct delayed_work *dwork = to_delayed_work(work); | |
1197 | ||
e0aecdd8 TH |
1198 | /* |
1199 | * dwork->timer is irqsafe. If del_timer() fails, it's | |
1200 | * guaranteed that the timer is not queued anywhere and not | |
1201 | * running on the local CPU. | |
1202 | */ | |
36e227d2 TH |
1203 | if (likely(del_timer(&dwork->timer))) |
1204 | return 1; | |
1205 | } | |
1206 | ||
1207 | /* try to claim PENDING the normal way */ | |
bf4ede01 TH |
1208 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) |
1209 | return 0; | |
1210 | ||
1211 | /* | |
1212 | * The queueing is in progress, or it is already queued. Try to | |
1213 | * steal it from ->worklist without clearing WORK_STRUCT_PENDING. | |
1214 | */ | |
d565ed63 TH |
1215 | pool = get_work_pool(work); |
1216 | if (!pool) | |
bbb68dfa | 1217 | goto fail; |
bf4ede01 | 1218 | |
d565ed63 | 1219 | spin_lock(&pool->lock); |
0b3dae68 | 1220 | /* |
112202d9 TH |
1221 | * work->data is guaranteed to point to pwq only while the work |
1222 | * item is queued on pwq->wq, and both updating work->data to point | |
1223 | * to pwq on queueing and to pool on dequeueing are done under | |
1224 | * pwq->pool->lock. This in turn guarantees that, if work->data | |
1225 | * points to pwq which is associated with a locked pool, the work | |
0b3dae68 LJ |
1226 | * item is currently queued on that pool. |
1227 | */ | |
112202d9 TH |
1228 | pwq = get_work_pwq(work); |
1229 | if (pwq && pwq->pool == pool) { | |
16062836 TH |
1230 | debug_work_deactivate(work); |
1231 | ||
1232 | /* | |
1233 | * A delayed work item cannot be grabbed directly because | |
1234 | * it might have linked NO_COLOR work items which, if left | |
112202d9 | 1235 | * on the delayed_list, will confuse pwq->nr_active |
16062836 TH |
1236 | * management later on and cause stall. Make sure the work |
1237 | * item is activated before grabbing. | |
1238 | */ | |
1239 | if (*work_data_bits(work) & WORK_STRUCT_DELAYED) | |
112202d9 | 1240 | pwq_activate_delayed_work(work); |
16062836 TH |
1241 | |
1242 | list_del_init(&work->entry); | |
112202d9 | 1243 | pwq_dec_nr_in_flight(get_work_pwq(work), get_work_color(work)); |
16062836 | 1244 | |
112202d9 | 1245 | /* work->data points to pwq iff queued, point to pool */ |
16062836 TH |
1246 | set_work_pool_and_keep_pending(work, pool->id); |
1247 | ||
1248 | spin_unlock(&pool->lock); | |
1249 | return 1; | |
bf4ede01 | 1250 | } |
d565ed63 | 1251 | spin_unlock(&pool->lock); |
bbb68dfa TH |
1252 | fail: |
1253 | local_irq_restore(*flags); | |
1254 | if (work_is_canceling(work)) | |
1255 | return -ENOENT; | |
1256 | cpu_relax(); | |
36e227d2 | 1257 | return -EAGAIN; |
bf4ede01 TH |
1258 | } |
1259 | ||
4690c4ab | 1260 | /** |
706026c2 | 1261 | * insert_work - insert a work into a pool |
112202d9 | 1262 | * @pwq: pwq @work belongs to |
4690c4ab TH |
1263 | * @work: work to insert |
1264 | * @head: insertion point | |
1265 | * @extra_flags: extra WORK_STRUCT_* flags to set | |
1266 | * | |
112202d9 | 1267 | * Insert @work which belongs to @pwq after @head. @extra_flags is or'd to |
706026c2 | 1268 | * work_struct flags. |
4690c4ab TH |
1269 | * |
1270 | * CONTEXT: | |
d565ed63 | 1271 | * spin_lock_irq(pool->lock). |
4690c4ab | 1272 | */ |
112202d9 TH |
1273 | static void insert_work(struct pool_workqueue *pwq, struct work_struct *work, |
1274 | struct list_head *head, unsigned int extra_flags) | |
b89deed3 | 1275 | { |
112202d9 | 1276 | struct worker_pool *pool = pwq->pool; |
e22bee78 | 1277 | |
4690c4ab | 1278 | /* we own @work, set data and link */ |
112202d9 | 1279 | set_work_pwq(work, pwq, extra_flags); |
1a4d9b0a | 1280 | list_add_tail(&work->entry, head); |
8864b4e5 | 1281 | get_pwq(pwq); |
e22bee78 TH |
1282 | |
1283 | /* | |
c5aa87bb TH |
1284 | * Ensure either wq_worker_sleeping() sees the above |
1285 | * list_add_tail() or we see zero nr_running to avoid workers lying | |
1286 | * around lazily while there are works to be processed. | |
e22bee78 TH |
1287 | */ |
1288 | smp_mb(); | |
1289 | ||
63d95a91 TH |
1290 | if (__need_more_worker(pool)) |
1291 | wake_up_worker(pool); | |
b89deed3 ON |
1292 | } |
1293 | ||
c8efcc25 TH |
1294 | /* |
1295 | * Test whether @work is being queued from another work executing on the | |
8d03ecfe | 1296 | * same workqueue. |
c8efcc25 TH |
1297 | */ |
1298 | static bool is_chained_work(struct workqueue_struct *wq) | |
1299 | { | |
8d03ecfe TH |
1300 | struct worker *worker; |
1301 | ||
1302 | worker = current_wq_worker(); | |
1303 | /* | |
1304 | * Return %true iff I'm a worker execuing a work item on @wq. If | |
1305 | * I'm @worker, it's safe to dereference it without locking. | |
1306 | */ | |
112202d9 | 1307 | return worker && worker->current_pwq->wq == wq; |
c8efcc25 TH |
1308 | } |
1309 | ||
d84ff051 | 1310 | static void __queue_work(int cpu, struct workqueue_struct *wq, |
1da177e4 LT |
1311 | struct work_struct *work) |
1312 | { | |
112202d9 | 1313 | struct pool_workqueue *pwq; |
c9178087 | 1314 | struct worker_pool *last_pool; |
1e19ffc6 | 1315 | struct list_head *worklist; |
8a2e8e5d | 1316 | unsigned int work_flags; |
b75cac93 | 1317 | unsigned int req_cpu = cpu; |
8930caba TH |
1318 | |
1319 | /* | |
1320 | * While a work item is PENDING && off queue, a task trying to | |
1321 | * steal the PENDING will busy-loop waiting for it to either get | |
1322 | * queued or lose PENDING. Grabbing PENDING and queueing should | |
1323 | * happen with IRQ disabled. | |
1324 | */ | |
1325 | WARN_ON_ONCE(!irqs_disabled()); | |
1da177e4 | 1326 | |
dc186ad7 | 1327 | debug_work_activate(work); |
1e19ffc6 | 1328 | |
9ef28a73 | 1329 | /* if draining, only works from the same workqueue are allowed */ |
618b01eb | 1330 | if (unlikely(wq->flags & __WQ_DRAINING) && |
c8efcc25 | 1331 | WARN_ON_ONCE(!is_chained_work(wq))) |
e41e704b | 1332 | return; |
9e8cd2f5 | 1333 | retry: |
df2d5ae4 TH |
1334 | if (req_cpu == WORK_CPU_UNBOUND) |
1335 | cpu = raw_smp_processor_id(); | |
1336 | ||
c9178087 | 1337 | /* pwq which will be used unless @work is executing elsewhere */ |
df2d5ae4 | 1338 | if (!(wq->flags & WQ_UNBOUND)) |
7fb98ea7 | 1339 | pwq = per_cpu_ptr(wq->cpu_pwqs, cpu); |
df2d5ae4 TH |
1340 | else |
1341 | pwq = unbound_pwq_by_node(wq, cpu_to_node(cpu)); | |
dbf2576e | 1342 | |
c9178087 TH |
1343 | /* |
1344 | * If @work was previously on a different pool, it might still be | |
1345 | * running there, in which case the work needs to be queued on that | |
1346 | * pool to guarantee non-reentrancy. | |
1347 | */ | |
1348 | last_pool = get_work_pool(work); | |
1349 | if (last_pool && last_pool != pwq->pool) { | |
1350 | struct worker *worker; | |
18aa9eff | 1351 | |
c9178087 | 1352 | spin_lock(&last_pool->lock); |
18aa9eff | 1353 | |
c9178087 | 1354 | worker = find_worker_executing_work(last_pool, work); |
18aa9eff | 1355 | |
c9178087 TH |
1356 | if (worker && worker->current_pwq->wq == wq) { |
1357 | pwq = worker->current_pwq; | |
8930caba | 1358 | } else { |
c9178087 TH |
1359 | /* meh... not running there, queue here */ |
1360 | spin_unlock(&last_pool->lock); | |
112202d9 | 1361 | spin_lock(&pwq->pool->lock); |
8930caba | 1362 | } |
f3421797 | 1363 | } else { |
112202d9 | 1364 | spin_lock(&pwq->pool->lock); |
502ca9d8 TH |
1365 | } |
1366 | ||
9e8cd2f5 TH |
1367 | /* |
1368 | * pwq is determined and locked. For unbound pools, we could have | |
1369 | * raced with pwq release and it could already be dead. If its | |
1370 | * refcnt is zero, repeat pwq selection. Note that pwqs never die | |
df2d5ae4 TH |
1371 | * without another pwq replacing it in the numa_pwq_tbl or while |
1372 | * work items are executing on it, so the retrying is guaranteed to | |
9e8cd2f5 TH |
1373 | * make forward-progress. |
1374 | */ | |
1375 | if (unlikely(!pwq->refcnt)) { | |
1376 | if (wq->flags & WQ_UNBOUND) { | |
1377 | spin_unlock(&pwq->pool->lock); | |
1378 | cpu_relax(); | |
1379 | goto retry; | |
1380 | } | |
1381 | /* oops */ | |
1382 | WARN_ONCE(true, "workqueue: per-cpu pwq for %s on cpu%d has 0 refcnt", | |
1383 | wq->name, cpu); | |
1384 | } | |
1385 | ||
112202d9 TH |
1386 | /* pwq determined, queue */ |
1387 | trace_workqueue_queue_work(req_cpu, pwq, work); | |
502ca9d8 | 1388 | |
f5b2552b | 1389 | if (WARN_ON(!list_empty(&work->entry))) { |
112202d9 | 1390 | spin_unlock(&pwq->pool->lock); |
f5b2552b DC |
1391 | return; |
1392 | } | |
1e19ffc6 | 1393 | |
112202d9 TH |
1394 | pwq->nr_in_flight[pwq->work_color]++; |
1395 | work_flags = work_color_to_flags(pwq->work_color); | |
1e19ffc6 | 1396 | |
112202d9 | 1397 | if (likely(pwq->nr_active < pwq->max_active)) { |
cdadf009 | 1398 | trace_workqueue_activate_work(work); |
112202d9 TH |
1399 | pwq->nr_active++; |
1400 | worklist = &pwq->pool->worklist; | |
8a2e8e5d TH |
1401 | } else { |
1402 | work_flags |= WORK_STRUCT_DELAYED; | |
112202d9 | 1403 | worklist = &pwq->delayed_works; |
8a2e8e5d | 1404 | } |
1e19ffc6 | 1405 | |
112202d9 | 1406 | insert_work(pwq, work, worklist, work_flags); |
1e19ffc6 | 1407 | |
112202d9 | 1408 | spin_unlock(&pwq->pool->lock); |
1da177e4 LT |
1409 | } |
1410 | ||
0fcb78c2 | 1411 | /** |
c1a220e7 ZR |
1412 | * queue_work_on - queue work on specific cpu |
1413 | * @cpu: CPU number to execute work on | |
0fcb78c2 REB |
1414 | * @wq: workqueue to use |
1415 | * @work: work to queue | |
1416 | * | |
c1a220e7 ZR |
1417 | * We queue the work to a specific CPU, the caller must ensure it |
1418 | * can't go away. | |
d185af30 YB |
1419 | * |
1420 | * Return: %false if @work was already on a queue, %true otherwise. | |
1da177e4 | 1421 | */ |
d4283e93 TH |
1422 | bool queue_work_on(int cpu, struct workqueue_struct *wq, |
1423 | struct work_struct *work) | |
1da177e4 | 1424 | { |
d4283e93 | 1425 | bool ret = false; |
8930caba | 1426 | unsigned long flags; |
ef1ca236 | 1427 | |
8930caba | 1428 | local_irq_save(flags); |
c1a220e7 | 1429 | |
22df02bb | 1430 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
4690c4ab | 1431 | __queue_work(cpu, wq, work); |
d4283e93 | 1432 | ret = true; |
c1a220e7 | 1433 | } |
ef1ca236 | 1434 | |
8930caba | 1435 | local_irq_restore(flags); |
1da177e4 LT |
1436 | return ret; |
1437 | } | |
ad7b1f84 | 1438 | EXPORT_SYMBOL(queue_work_on); |
1da177e4 | 1439 | |
d8e794df | 1440 | void delayed_work_timer_fn(unsigned long __data) |
1da177e4 | 1441 | { |
52bad64d | 1442 | struct delayed_work *dwork = (struct delayed_work *)__data; |
1da177e4 | 1443 | |
e0aecdd8 | 1444 | /* should have been called from irqsafe timer with irq already off */ |
60c057bc | 1445 | __queue_work(dwork->cpu, dwork->wq, &dwork->work); |
1da177e4 | 1446 | } |
1438ade5 | 1447 | EXPORT_SYMBOL(delayed_work_timer_fn); |
1da177e4 | 1448 | |
7beb2edf TH |
1449 | static void __queue_delayed_work(int cpu, struct workqueue_struct *wq, |
1450 | struct delayed_work *dwork, unsigned long delay) | |
1da177e4 | 1451 | { |
7beb2edf TH |
1452 | struct timer_list *timer = &dwork->timer; |
1453 | struct work_struct *work = &dwork->work; | |
7beb2edf TH |
1454 | |
1455 | WARN_ON_ONCE(timer->function != delayed_work_timer_fn || | |
1456 | timer->data != (unsigned long)dwork); | |
fc4b514f TH |
1457 | WARN_ON_ONCE(timer_pending(timer)); |
1458 | WARN_ON_ONCE(!list_empty(&work->entry)); | |
7beb2edf | 1459 | |
8852aac2 TH |
1460 | /* |
1461 | * If @delay is 0, queue @dwork->work immediately. This is for | |
1462 | * both optimization and correctness. The earliest @timer can | |
1463 | * expire is on the closest next tick and delayed_work users depend | |
1464 | * on that there's no such delay when @delay is 0. | |
1465 | */ | |
1466 | if (!delay) { | |
1467 | __queue_work(cpu, wq, &dwork->work); | |
1468 | return; | |
1469 | } | |
1470 | ||
7beb2edf | 1471 | timer_stats_timer_set_start_info(&dwork->timer); |
1da177e4 | 1472 | |
60c057bc | 1473 | dwork->wq = wq; |
1265057f | 1474 | dwork->cpu = cpu; |
7beb2edf TH |
1475 | timer->expires = jiffies + delay; |
1476 | ||
1477 | if (unlikely(cpu != WORK_CPU_UNBOUND)) | |
1478 | add_timer_on(timer, cpu); | |
1479 | else | |
1480 | add_timer(timer); | |
1da177e4 LT |
1481 | } |
1482 | ||
0fcb78c2 REB |
1483 | /** |
1484 | * queue_delayed_work_on - queue work on specific CPU after delay | |
1485 | * @cpu: CPU number to execute work on | |
1486 | * @wq: workqueue to use | |
af9997e4 | 1487 | * @dwork: work to queue |
0fcb78c2 REB |
1488 | * @delay: number of jiffies to wait before queueing |
1489 | * | |
d185af30 | 1490 | * Return: %false if @work was already on a queue, %true otherwise. If |
715f1300 TH |
1491 | * @delay is zero and @dwork is idle, it will be scheduled for immediate |
1492 | * execution. | |
0fcb78c2 | 1493 | */ |
d4283e93 TH |
1494 | bool queue_delayed_work_on(int cpu, struct workqueue_struct *wq, |
1495 | struct delayed_work *dwork, unsigned long delay) | |
7a6bc1cd | 1496 | { |
52bad64d | 1497 | struct work_struct *work = &dwork->work; |
d4283e93 | 1498 | bool ret = false; |
8930caba | 1499 | unsigned long flags; |
7a6bc1cd | 1500 | |
8930caba TH |
1501 | /* read the comment in __queue_work() */ |
1502 | local_irq_save(flags); | |
7a6bc1cd | 1503 | |
22df02bb | 1504 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
7beb2edf | 1505 | __queue_delayed_work(cpu, wq, dwork, delay); |
d4283e93 | 1506 | ret = true; |
7a6bc1cd | 1507 | } |
8a3e77cc | 1508 | |
8930caba | 1509 | local_irq_restore(flags); |
7a6bc1cd VP |
1510 | return ret; |
1511 | } | |
ad7b1f84 | 1512 | EXPORT_SYMBOL(queue_delayed_work_on); |
c7fc77f7 | 1513 | |
8376fe22 TH |
1514 | /** |
1515 | * mod_delayed_work_on - modify delay of or queue a delayed work on specific CPU | |
1516 | * @cpu: CPU number to execute work on | |
1517 | * @wq: workqueue to use | |
1518 | * @dwork: work to queue | |
1519 | * @delay: number of jiffies to wait before queueing | |
1520 | * | |
1521 | * If @dwork is idle, equivalent to queue_delayed_work_on(); otherwise, | |
1522 | * modify @dwork's timer so that it expires after @delay. If @delay is | |
1523 | * zero, @work is guaranteed to be scheduled immediately regardless of its | |
1524 | * current state. | |
1525 | * | |
d185af30 | 1526 | * Return: %false if @dwork was idle and queued, %true if @dwork was |
8376fe22 TH |
1527 | * pending and its timer was modified. |
1528 | * | |
e0aecdd8 | 1529 | * This function is safe to call from any context including IRQ handler. |
8376fe22 TH |
1530 | * See try_to_grab_pending() for details. |
1531 | */ | |
1532 | bool mod_delayed_work_on(int cpu, struct workqueue_struct *wq, | |
1533 | struct delayed_work *dwork, unsigned long delay) | |
1534 | { | |
1535 | unsigned long flags; | |
1536 | int ret; | |
c7fc77f7 | 1537 | |
8376fe22 TH |
1538 | do { |
1539 | ret = try_to_grab_pending(&dwork->work, true, &flags); | |
1540 | } while (unlikely(ret == -EAGAIN)); | |
63bc0362 | 1541 | |
8376fe22 TH |
1542 | if (likely(ret >= 0)) { |
1543 | __queue_delayed_work(cpu, wq, dwork, delay); | |
1544 | local_irq_restore(flags); | |
7a6bc1cd | 1545 | } |
8376fe22 TH |
1546 | |
1547 | /* -ENOENT from try_to_grab_pending() becomes %true */ | |
7a6bc1cd VP |
1548 | return ret; |
1549 | } | |
8376fe22 TH |
1550 | EXPORT_SYMBOL_GPL(mod_delayed_work_on); |
1551 | ||
c8e55f36 TH |
1552 | /** |
1553 | * worker_enter_idle - enter idle state | |
1554 | * @worker: worker which is entering idle state | |
1555 | * | |
1556 | * @worker is entering idle state. Update stats and idle timer if | |
1557 | * necessary. | |
1558 | * | |
1559 | * LOCKING: | |
d565ed63 | 1560 | * spin_lock_irq(pool->lock). |
c8e55f36 TH |
1561 | */ |
1562 | static void worker_enter_idle(struct worker *worker) | |
1da177e4 | 1563 | { |
bd7bdd43 | 1564 | struct worker_pool *pool = worker->pool; |
c8e55f36 | 1565 | |
6183c009 TH |
1566 | if (WARN_ON_ONCE(worker->flags & WORKER_IDLE) || |
1567 | WARN_ON_ONCE(!list_empty(&worker->entry) && | |
1568 | (worker->hentry.next || worker->hentry.pprev))) | |
1569 | return; | |
c8e55f36 | 1570 | |
cb444766 TH |
1571 | /* can't use worker_set_flags(), also called from start_worker() */ |
1572 | worker->flags |= WORKER_IDLE; | |
bd7bdd43 | 1573 | pool->nr_idle++; |
e22bee78 | 1574 | worker->last_active = jiffies; |
c8e55f36 TH |
1575 | |
1576 | /* idle_list is LIFO */ | |
bd7bdd43 | 1577 | list_add(&worker->entry, &pool->idle_list); |
db7bccf4 | 1578 | |
628c78e7 TH |
1579 | if (too_many_workers(pool) && !timer_pending(&pool->idle_timer)) |
1580 | mod_timer(&pool->idle_timer, jiffies + IDLE_WORKER_TIMEOUT); | |
cb444766 | 1581 | |
544ecf31 | 1582 | /* |
706026c2 | 1583 | * Sanity check nr_running. Because wq_unbind_fn() releases |
d565ed63 | 1584 | * pool->lock between setting %WORKER_UNBOUND and zapping |
628c78e7 TH |
1585 | * nr_running, the warning may trigger spuriously. Check iff |
1586 | * unbind is not in progress. | |
544ecf31 | 1587 | */ |
24647570 | 1588 | WARN_ON_ONCE(!(pool->flags & POOL_DISASSOCIATED) && |
bd7bdd43 | 1589 | pool->nr_workers == pool->nr_idle && |
e19e397a | 1590 | atomic_read(&pool->nr_running)); |
c8e55f36 TH |
1591 | } |
1592 | ||
1593 | /** | |
1594 | * worker_leave_idle - leave idle state | |
1595 | * @worker: worker which is leaving idle state | |
1596 | * | |
1597 | * @worker is leaving idle state. Update stats. | |
1598 | * | |
1599 | * LOCKING: | |
d565ed63 | 1600 | * spin_lock_irq(pool->lock). |
c8e55f36 TH |
1601 | */ |
1602 | static void worker_leave_idle(struct worker *worker) | |
1603 | { | |
bd7bdd43 | 1604 | struct worker_pool *pool = worker->pool; |
c8e55f36 | 1605 | |
6183c009 TH |
1606 | if (WARN_ON_ONCE(!(worker->flags & WORKER_IDLE))) |
1607 | return; | |
d302f017 | 1608 | worker_clr_flags(worker, WORKER_IDLE); |
bd7bdd43 | 1609 | pool->nr_idle--; |
c8e55f36 TH |
1610 | list_del_init(&worker->entry); |
1611 | } | |
1612 | ||
e22bee78 | 1613 | /** |
f36dc67b LJ |
1614 | * worker_maybe_bind_and_lock - try to bind %current to worker_pool and lock it |
1615 | * @pool: target worker_pool | |
1616 | * | |
1617 | * Bind %current to the cpu of @pool if it is associated and lock @pool. | |
e22bee78 TH |
1618 | * |
1619 | * Works which are scheduled while the cpu is online must at least be | |
1620 | * scheduled to a worker which is bound to the cpu so that if they are | |
1621 | * flushed from cpu callbacks while cpu is going down, they are | |
1622 | * guaranteed to execute on the cpu. | |
1623 | * | |
f5faa077 | 1624 | * This function is to be used by unbound workers and rescuers to bind |
e22bee78 TH |
1625 | * themselves to the target cpu and may race with cpu going down or |
1626 | * coming online. kthread_bind() can't be used because it may put the | |
1627 | * worker to already dead cpu and set_cpus_allowed_ptr() can't be used | |
706026c2 | 1628 | * verbatim as it's best effort and blocking and pool may be |
e22bee78 TH |
1629 | * [dis]associated in the meantime. |
1630 | * | |
706026c2 | 1631 | * This function tries set_cpus_allowed() and locks pool and verifies the |
24647570 | 1632 | * binding against %POOL_DISASSOCIATED which is set during |
f2d5a0ee TH |
1633 | * %CPU_DOWN_PREPARE and cleared during %CPU_ONLINE, so if the worker |
1634 | * enters idle state or fetches works without dropping lock, it can | |
1635 | * guarantee the scheduling requirement described in the first paragraph. | |
e22bee78 TH |
1636 | * |
1637 | * CONTEXT: | |
d565ed63 | 1638 | * Might sleep. Called without any lock but returns with pool->lock |
e22bee78 TH |
1639 | * held. |
1640 | * | |
d185af30 | 1641 | * Return: |
706026c2 | 1642 | * %true if the associated pool is online (@worker is successfully |
e22bee78 TH |
1643 | * bound), %false if offline. |
1644 | */ | |
f36dc67b | 1645 | static bool worker_maybe_bind_and_lock(struct worker_pool *pool) |
d565ed63 | 1646 | __acquires(&pool->lock) |
e22bee78 | 1647 | { |
e22bee78 | 1648 | while (true) { |
4e6045f1 | 1649 | /* |
e22bee78 TH |
1650 | * The following call may fail, succeed or succeed |
1651 | * without actually migrating the task to the cpu if | |
1652 | * it races with cpu hotunplug operation. Verify | |
24647570 | 1653 | * against POOL_DISASSOCIATED. |
4e6045f1 | 1654 | */ |
24647570 | 1655 | if (!(pool->flags & POOL_DISASSOCIATED)) |
7a4e344c | 1656 | set_cpus_allowed_ptr(current, pool->attrs->cpumask); |
e22bee78 | 1657 | |
d565ed63 | 1658 | spin_lock_irq(&pool->lock); |
24647570 | 1659 | if (pool->flags & POOL_DISASSOCIATED) |
e22bee78 | 1660 | return false; |
f5faa077 | 1661 | if (task_cpu(current) == pool->cpu && |
7a4e344c | 1662 | cpumask_equal(¤t->cpus_allowed, pool->attrs->cpumask)) |
e22bee78 | 1663 | return true; |
d565ed63 | 1664 | spin_unlock_irq(&pool->lock); |
e22bee78 | 1665 | |
5035b20f TH |
1666 | /* |
1667 | * We've raced with CPU hot[un]plug. Give it a breather | |
1668 | * and retry migration. cond_resched() is required here; | |
1669 | * otherwise, we might deadlock against cpu_stop trying to | |
1670 | * bring down the CPU on non-preemptive kernel. | |
1671 | */ | |
e22bee78 | 1672 | cpu_relax(); |
5035b20f | 1673 | cond_resched(); |
e22bee78 TH |
1674 | } |
1675 | } | |
1676 | ||
c34056a3 TH |
1677 | static struct worker *alloc_worker(void) |
1678 | { | |
1679 | struct worker *worker; | |
1680 | ||
1681 | worker = kzalloc(sizeof(*worker), GFP_KERNEL); | |
c8e55f36 TH |
1682 | if (worker) { |
1683 | INIT_LIST_HEAD(&worker->entry); | |
affee4b2 | 1684 | INIT_LIST_HEAD(&worker->scheduled); |
e22bee78 TH |
1685 | /* on creation a worker is in !idle && prep state */ |
1686 | worker->flags = WORKER_PREP; | |
c8e55f36 | 1687 | } |
c34056a3 TH |
1688 | return worker; |
1689 | } | |
1690 | ||
1691 | /** | |
1692 | * create_worker - create a new workqueue worker | |
63d95a91 | 1693 | * @pool: pool the new worker will belong to |
c34056a3 | 1694 | * |
63d95a91 | 1695 | * Create a new worker which is bound to @pool. The returned worker |
c34056a3 TH |
1696 | * can be started by calling start_worker() or destroyed using |
1697 | * destroy_worker(). | |
1698 | * | |
1699 | * CONTEXT: | |
1700 | * Might sleep. Does GFP_KERNEL allocations. | |
1701 | * | |
d185af30 | 1702 | * Return: |
c34056a3 TH |
1703 | * Pointer to the newly created worker. |
1704 | */ | |
bc2ae0f5 | 1705 | static struct worker *create_worker(struct worker_pool *pool) |
c34056a3 | 1706 | { |
c34056a3 | 1707 | struct worker *worker = NULL; |
f3421797 | 1708 | int id = -1; |
e3c916a4 | 1709 | char id_buf[16]; |
c34056a3 | 1710 | |
cd549687 TH |
1711 | lockdep_assert_held(&pool->manager_mutex); |
1712 | ||
822d8405 TH |
1713 | /* |
1714 | * ID is needed to determine kthread name. Allocate ID first | |
1715 | * without installing the pointer. | |
1716 | */ | |
9625ab17 | 1717 | id = idr_alloc(&pool->worker_idr, NULL, 0, 0, GFP_KERNEL); |
822d8405 TH |
1718 | if (id < 0) |
1719 | goto fail; | |
c34056a3 TH |
1720 | |
1721 | worker = alloc_worker(); | |
1722 | if (!worker) | |
1723 | goto fail; | |
1724 | ||
bd7bdd43 | 1725 | worker->pool = pool; |
c34056a3 TH |
1726 | worker->id = id; |
1727 | ||
29c91e99 | 1728 | if (pool->cpu >= 0) |
e3c916a4 TH |
1729 | snprintf(id_buf, sizeof(id_buf), "%d:%d%s", pool->cpu, id, |
1730 | pool->attrs->nice < 0 ? "H" : ""); | |
f3421797 | 1731 | else |
e3c916a4 TH |
1732 | snprintf(id_buf, sizeof(id_buf), "u%d:%d", pool->id, id); |
1733 | ||
f3f90ad4 | 1734 | worker->task = kthread_create_on_node(worker_thread, worker, pool->node, |
e3c916a4 | 1735 | "kworker/%s", id_buf); |
c34056a3 TH |
1736 | if (IS_ERR(worker->task)) |
1737 | goto fail; | |
1738 | ||
91151228 ON |
1739 | set_user_nice(worker->task, pool->attrs->nice); |
1740 | ||
1741 | /* prevent userland from meddling with cpumask of workqueue workers */ | |
1742 | worker->task->flags |= PF_NO_SETAFFINITY; | |
1743 | ||
c5aa87bb TH |
1744 | /* |
1745 | * set_cpus_allowed_ptr() will fail if the cpumask doesn't have any | |
1746 | * online CPUs. It'll be re-applied when any of the CPUs come up. | |
1747 | */ | |
7a4e344c | 1748 | set_cpus_allowed_ptr(worker->task, pool->attrs->cpumask); |
3270476a | 1749 | |
7a4e344c TH |
1750 | /* |
1751 | * The caller is responsible for ensuring %POOL_DISASSOCIATED | |
1752 | * remains stable across this function. See the comments above the | |
1753 | * flag definition for details. | |
1754 | */ | |
1755 | if (pool->flags & POOL_DISASSOCIATED) | |
bc2ae0f5 | 1756 | worker->flags |= WORKER_UNBOUND; |
c34056a3 | 1757 | |
822d8405 | 1758 | /* successful, commit the pointer to idr */ |
822d8405 | 1759 | idr_replace(&pool->worker_idr, worker, worker->id); |
822d8405 | 1760 | |
c34056a3 | 1761 | return worker; |
822d8405 | 1762 | |
c34056a3 | 1763 | fail: |
9625ab17 | 1764 | if (id >= 0) |
822d8405 | 1765 | idr_remove(&pool->worker_idr, id); |
c34056a3 TH |
1766 | kfree(worker); |
1767 | return NULL; | |
1768 | } | |
1769 | ||
1770 | /** | |
1771 | * start_worker - start a newly created worker | |
1772 | * @worker: worker to start | |
1773 | * | |
706026c2 | 1774 | * Make the pool aware of @worker and start it. |
c34056a3 TH |
1775 | * |
1776 | * CONTEXT: | |
d565ed63 | 1777 | * spin_lock_irq(pool->lock). |
c34056a3 TH |
1778 | */ |
1779 | static void start_worker(struct worker *worker) | |
1780 | { | |
cb444766 | 1781 | worker->flags |= WORKER_STARTED; |
bd7bdd43 | 1782 | worker->pool->nr_workers++; |
c8e55f36 | 1783 | worker_enter_idle(worker); |
c34056a3 TH |
1784 | wake_up_process(worker->task); |
1785 | } | |
1786 | ||
ebf44d16 TH |
1787 | /** |
1788 | * create_and_start_worker - create and start a worker for a pool | |
1789 | * @pool: the target pool | |
1790 | * | |
cd549687 | 1791 | * Grab the managership of @pool and create and start a new worker for it. |
d185af30 YB |
1792 | * |
1793 | * Return: 0 on success. A negative error code otherwise. | |
ebf44d16 TH |
1794 | */ |
1795 | static int create_and_start_worker(struct worker_pool *pool) | |
1796 | { | |
1797 | struct worker *worker; | |
1798 | ||
cd549687 TH |
1799 | mutex_lock(&pool->manager_mutex); |
1800 | ||
ebf44d16 TH |
1801 | worker = create_worker(pool); |
1802 | if (worker) { | |
1803 | spin_lock_irq(&pool->lock); | |
1804 | start_worker(worker); | |
1805 | spin_unlock_irq(&pool->lock); | |
1806 | } | |
1807 | ||
cd549687 TH |
1808 | mutex_unlock(&pool->manager_mutex); |
1809 | ||
ebf44d16 TH |
1810 | return worker ? 0 : -ENOMEM; |
1811 | } | |
1812 | ||
c34056a3 TH |
1813 | /** |
1814 | * destroy_worker - destroy a workqueue worker | |
1815 | * @worker: worker to be destroyed | |
1816 | * | |
706026c2 | 1817 | * Destroy @worker and adjust @pool stats accordingly. |
c8e55f36 TH |
1818 | * |
1819 | * CONTEXT: | |
d565ed63 | 1820 | * spin_lock_irq(pool->lock) which is released and regrabbed. |
c34056a3 TH |
1821 | */ |
1822 | static void destroy_worker(struct worker *worker) | |
1823 | { | |
bd7bdd43 | 1824 | struct worker_pool *pool = worker->pool; |
c34056a3 | 1825 | |
cd549687 TH |
1826 | lockdep_assert_held(&pool->manager_mutex); |
1827 | lockdep_assert_held(&pool->lock); | |
1828 | ||
c34056a3 | 1829 | /* sanity check frenzy */ |
6183c009 TH |
1830 | if (WARN_ON(worker->current_work) || |
1831 | WARN_ON(!list_empty(&worker->scheduled))) | |
1832 | return; | |
c34056a3 | 1833 | |
c8e55f36 | 1834 | if (worker->flags & WORKER_STARTED) |
bd7bdd43 | 1835 | pool->nr_workers--; |
c8e55f36 | 1836 | if (worker->flags & WORKER_IDLE) |
bd7bdd43 | 1837 | pool->nr_idle--; |
c8e55f36 | 1838 | |
5bdfff96 LJ |
1839 | /* |
1840 | * Once WORKER_DIE is set, the kworker may destroy itself at any | |
1841 | * point. Pin to ensure the task stays until we're done with it. | |
1842 | */ | |
1843 | get_task_struct(worker->task); | |
1844 | ||
c8e55f36 | 1845 | list_del_init(&worker->entry); |
cb444766 | 1846 | worker->flags |= WORKER_DIE; |
c8e55f36 | 1847 | |
822d8405 TH |
1848 | idr_remove(&pool->worker_idr, worker->id); |
1849 | ||
d565ed63 | 1850 | spin_unlock_irq(&pool->lock); |
c8e55f36 | 1851 | |
c34056a3 | 1852 | kthread_stop(worker->task); |
5bdfff96 | 1853 | put_task_struct(worker->task); |
c34056a3 TH |
1854 | kfree(worker); |
1855 | ||
d565ed63 | 1856 | spin_lock_irq(&pool->lock); |
c34056a3 TH |
1857 | } |
1858 | ||
63d95a91 | 1859 | static void idle_worker_timeout(unsigned long __pool) |
e22bee78 | 1860 | { |
63d95a91 | 1861 | struct worker_pool *pool = (void *)__pool; |
e22bee78 | 1862 | |
d565ed63 | 1863 | spin_lock_irq(&pool->lock); |
e22bee78 | 1864 | |
63d95a91 | 1865 | if (too_many_workers(pool)) { |
e22bee78 TH |
1866 | struct worker *worker; |
1867 | unsigned long expires; | |
1868 | ||
1869 | /* idle_list is kept in LIFO order, check the last one */ | |
63d95a91 | 1870 | worker = list_entry(pool->idle_list.prev, struct worker, entry); |
e22bee78 TH |
1871 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; |
1872 | ||
1873 | if (time_before(jiffies, expires)) | |
63d95a91 | 1874 | mod_timer(&pool->idle_timer, expires); |
e22bee78 TH |
1875 | else { |
1876 | /* it's been idle for too long, wake up manager */ | |
11ebea50 | 1877 | pool->flags |= POOL_MANAGE_WORKERS; |
63d95a91 | 1878 | wake_up_worker(pool); |
d5abe669 | 1879 | } |
e22bee78 TH |
1880 | } |
1881 | ||
d565ed63 | 1882 | spin_unlock_irq(&pool->lock); |
e22bee78 | 1883 | } |
d5abe669 | 1884 | |
493a1724 | 1885 | static void send_mayday(struct work_struct *work) |
e22bee78 | 1886 | { |
112202d9 TH |
1887 | struct pool_workqueue *pwq = get_work_pwq(work); |
1888 | struct workqueue_struct *wq = pwq->wq; | |
493a1724 | 1889 | |
2e109a28 | 1890 | lockdep_assert_held(&wq_mayday_lock); |
e22bee78 | 1891 | |
493008a8 | 1892 | if (!wq->rescuer) |
493a1724 | 1893 | return; |
e22bee78 TH |
1894 | |
1895 | /* mayday mayday mayday */ | |
493a1724 | 1896 | if (list_empty(&pwq->mayday_node)) { |
77668c8b LJ |
1897 | /* |
1898 | * If @pwq is for an unbound wq, its base ref may be put at | |
1899 | * any time due to an attribute change. Pin @pwq until the | |
1900 | * rescuer is done with it. | |
1901 | */ | |
1902 | get_pwq(pwq); | |
493a1724 | 1903 | list_add_tail(&pwq->mayday_node, &wq->maydays); |
e22bee78 | 1904 | wake_up_process(wq->rescuer->task); |
493a1724 | 1905 | } |
e22bee78 TH |
1906 | } |
1907 | ||
706026c2 | 1908 | static void pool_mayday_timeout(unsigned long __pool) |
e22bee78 | 1909 | { |
63d95a91 | 1910 | struct worker_pool *pool = (void *)__pool; |
e22bee78 TH |
1911 | struct work_struct *work; |
1912 | ||
2e109a28 | 1913 | spin_lock_irq(&wq_mayday_lock); /* for wq->maydays */ |
493a1724 | 1914 | spin_lock(&pool->lock); |
e22bee78 | 1915 | |
63d95a91 | 1916 | if (need_to_create_worker(pool)) { |
e22bee78 TH |
1917 | /* |
1918 | * We've been trying to create a new worker but | |
1919 | * haven't been successful. We might be hitting an | |
1920 | * allocation deadlock. Send distress signals to | |
1921 | * rescuers. | |
1922 | */ | |
63d95a91 | 1923 | list_for_each_entry(work, &pool->worklist, entry) |
e22bee78 | 1924 | send_mayday(work); |
1da177e4 | 1925 | } |
e22bee78 | 1926 | |
493a1724 | 1927 | spin_unlock(&pool->lock); |
2e109a28 | 1928 | spin_unlock_irq(&wq_mayday_lock); |
e22bee78 | 1929 | |
63d95a91 | 1930 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INTERVAL); |
1da177e4 LT |
1931 | } |
1932 | ||
e22bee78 TH |
1933 | /** |
1934 | * maybe_create_worker - create a new worker if necessary | |
63d95a91 | 1935 | * @pool: pool to create a new worker for |
e22bee78 | 1936 | * |
63d95a91 | 1937 | * Create a new worker for @pool if necessary. @pool is guaranteed to |
e22bee78 TH |
1938 | * have at least one idle worker on return from this function. If |
1939 | * creating a new worker takes longer than MAYDAY_INTERVAL, mayday is | |
63d95a91 | 1940 | * sent to all rescuers with works scheduled on @pool to resolve |
e22bee78 TH |
1941 | * possible allocation deadlock. |
1942 | * | |
c5aa87bb TH |
1943 | * On return, need_to_create_worker() is guaranteed to be %false and |
1944 | * may_start_working() %true. | |
e22bee78 TH |
1945 | * |
1946 | * LOCKING: | |
d565ed63 | 1947 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
e22bee78 TH |
1948 | * multiple times. Does GFP_KERNEL allocations. Called only from |
1949 | * manager. | |
1950 | * | |
d185af30 | 1951 | * Return: |
c5aa87bb | 1952 | * %false if no action was taken and pool->lock stayed locked, %true |
e22bee78 TH |
1953 | * otherwise. |
1954 | */ | |
63d95a91 | 1955 | static bool maybe_create_worker(struct worker_pool *pool) |
d565ed63 TH |
1956 | __releases(&pool->lock) |
1957 | __acquires(&pool->lock) | |
1da177e4 | 1958 | { |
63d95a91 | 1959 | if (!need_to_create_worker(pool)) |
e22bee78 TH |
1960 | return false; |
1961 | restart: | |
d565ed63 | 1962 | spin_unlock_irq(&pool->lock); |
9f9c2364 | 1963 | |
e22bee78 | 1964 | /* if we don't make progress in MAYDAY_INITIAL_TIMEOUT, call for help */ |
63d95a91 | 1965 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INITIAL_TIMEOUT); |
e22bee78 TH |
1966 | |
1967 | while (true) { | |
1968 | struct worker *worker; | |
1969 | ||
bc2ae0f5 | 1970 | worker = create_worker(pool); |
e22bee78 | 1971 | if (worker) { |
63d95a91 | 1972 | del_timer_sync(&pool->mayday_timer); |
d565ed63 | 1973 | spin_lock_irq(&pool->lock); |
e22bee78 | 1974 | start_worker(worker); |
6183c009 TH |
1975 | if (WARN_ON_ONCE(need_to_create_worker(pool))) |
1976 | goto restart; | |
e22bee78 TH |
1977 | return true; |
1978 | } | |
1979 | ||
63d95a91 | 1980 | if (!need_to_create_worker(pool)) |
e22bee78 | 1981 | break; |
1da177e4 | 1982 | |
e22bee78 TH |
1983 | __set_current_state(TASK_INTERRUPTIBLE); |
1984 | schedule_timeout(CREATE_COOLDOWN); | |
9f9c2364 | 1985 | |
63d95a91 | 1986 | if (!need_to_create_worker(pool)) |
e22bee78 TH |
1987 | break; |
1988 | } | |
1989 | ||
63d95a91 | 1990 | del_timer_sync(&pool->mayday_timer); |
d565ed63 | 1991 | spin_lock_irq(&pool->lock); |
63d95a91 | 1992 | if (need_to_create_worker(pool)) |
e22bee78 TH |
1993 | goto restart; |
1994 | return true; | |
1995 | } | |
1996 | ||
1997 | /** | |
1998 | * maybe_destroy_worker - destroy workers which have been idle for a while | |
63d95a91 | 1999 | * @pool: pool to destroy workers for |
e22bee78 | 2000 | * |
63d95a91 | 2001 | * Destroy @pool workers which have been idle for longer than |
e22bee78 TH |
2002 | * IDLE_WORKER_TIMEOUT. |
2003 | * | |
2004 | * LOCKING: | |
d565ed63 | 2005 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
e22bee78 TH |
2006 | * multiple times. Called only from manager. |
2007 | * | |
d185af30 | 2008 | * Return: |
c5aa87bb | 2009 | * %false if no action was taken and pool->lock stayed locked, %true |
e22bee78 TH |
2010 | * otherwise. |
2011 | */ | |
63d95a91 | 2012 | static bool maybe_destroy_workers(struct worker_pool *pool) |
e22bee78 TH |
2013 | { |
2014 | bool ret = false; | |
1da177e4 | 2015 | |
63d95a91 | 2016 | while (too_many_workers(pool)) { |
e22bee78 TH |
2017 | struct worker *worker; |
2018 | unsigned long expires; | |
3af24433 | 2019 | |
63d95a91 | 2020 | worker = list_entry(pool->idle_list.prev, struct worker, entry); |
e22bee78 | 2021 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; |
85f4186a | 2022 | |
e22bee78 | 2023 | if (time_before(jiffies, expires)) { |
63d95a91 | 2024 | mod_timer(&pool->idle_timer, expires); |
3af24433 | 2025 | break; |
e22bee78 | 2026 | } |
1da177e4 | 2027 | |
e22bee78 TH |
2028 | destroy_worker(worker); |
2029 | ret = true; | |
1da177e4 | 2030 | } |
1e19ffc6 | 2031 | |
e22bee78 | 2032 | return ret; |
1e19ffc6 TH |
2033 | } |
2034 | ||
73f53c4a | 2035 | /** |
e22bee78 TH |
2036 | * manage_workers - manage worker pool |
2037 | * @worker: self | |
73f53c4a | 2038 | * |
706026c2 | 2039 | * Assume the manager role and manage the worker pool @worker belongs |
e22bee78 | 2040 | * to. At any given time, there can be only zero or one manager per |
706026c2 | 2041 | * pool. The exclusion is handled automatically by this function. |
e22bee78 TH |
2042 | * |
2043 | * The caller can safely start processing works on false return. On | |
2044 | * true return, it's guaranteed that need_to_create_worker() is false | |
2045 | * and may_start_working() is true. | |
73f53c4a TH |
2046 | * |
2047 | * CONTEXT: | |
d565ed63 | 2048 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
e22bee78 TH |
2049 | * multiple times. Does GFP_KERNEL allocations. |
2050 | * | |
d185af30 | 2051 | * Return: |
2d498db9 L |
2052 | * %false if the pool don't need management and the caller can safely start |
2053 | * processing works, %true indicates that the function released pool->lock | |
2054 | * and reacquired it to perform some management function and that the | |
2055 | * conditions that the caller verified while holding the lock before | |
2056 | * calling the function might no longer be true. | |
73f53c4a | 2057 | */ |
e22bee78 | 2058 | static bool manage_workers(struct worker *worker) |
73f53c4a | 2059 | { |
63d95a91 | 2060 | struct worker_pool *pool = worker->pool; |
e22bee78 | 2061 | bool ret = false; |
73f53c4a | 2062 | |
bc3a1afc TH |
2063 | /* |
2064 | * Managership is governed by two mutexes - manager_arb and | |
2065 | * manager_mutex. manager_arb handles arbitration of manager role. | |
2066 | * Anyone who successfully grabs manager_arb wins the arbitration | |
2067 | * and becomes the manager. mutex_trylock() on pool->manager_arb | |
2068 | * failure while holding pool->lock reliably indicates that someone | |
2069 | * else is managing the pool and the worker which failed trylock | |
2070 | * can proceed to executing work items. This means that anyone | |
2071 | * grabbing manager_arb is responsible for actually performing | |
2072 | * manager duties. If manager_arb is grabbed and released without | |
2073 | * actual management, the pool may stall indefinitely. | |
2074 | * | |
2075 | * manager_mutex is used for exclusion of actual management | |
2076 | * operations. The holder of manager_mutex can be sure that none | |
2077 | * of management operations, including creation and destruction of | |
2078 | * workers, won't take place until the mutex is released. Because | |
2079 | * manager_mutex doesn't interfere with manager role arbitration, | |
2080 | * it is guaranteed that the pool's management, while may be | |
2081 | * delayed, won't be disturbed by someone else grabbing | |
2082 | * manager_mutex. | |
2083 | */ | |
34a06bd6 | 2084 | if (!mutex_trylock(&pool->manager_arb)) |
e22bee78 | 2085 | return ret; |
1e19ffc6 | 2086 | |
ee378aa4 | 2087 | /* |
bc3a1afc TH |
2088 | * With manager arbitration won, manager_mutex would be free in |
2089 | * most cases. trylock first without dropping @pool->lock. | |
ee378aa4 | 2090 | */ |
bc3a1afc | 2091 | if (unlikely(!mutex_trylock(&pool->manager_mutex))) { |
d565ed63 | 2092 | spin_unlock_irq(&pool->lock); |
bc3a1afc | 2093 | mutex_lock(&pool->manager_mutex); |
8f174b11 | 2094 | spin_lock_irq(&pool->lock); |
ee378aa4 LJ |
2095 | ret = true; |
2096 | } | |
73f53c4a | 2097 | |
11ebea50 | 2098 | pool->flags &= ~POOL_MANAGE_WORKERS; |
73f53c4a TH |
2099 | |
2100 | /* | |
e22bee78 TH |
2101 | * Destroy and then create so that may_start_working() is true |
2102 | * on return. | |
73f53c4a | 2103 | */ |
63d95a91 TH |
2104 | ret |= maybe_destroy_workers(pool); |
2105 | ret |= maybe_create_worker(pool); | |
e22bee78 | 2106 | |
bc3a1afc | 2107 | mutex_unlock(&pool->manager_mutex); |
34a06bd6 | 2108 | mutex_unlock(&pool->manager_arb); |
e22bee78 | 2109 | return ret; |
73f53c4a TH |
2110 | } |
2111 | ||
a62428c0 TH |
2112 | /** |
2113 | * process_one_work - process single work | |
c34056a3 | 2114 | * @worker: self |
a62428c0 TH |
2115 | * @work: work to process |
2116 | * | |
2117 | * Process @work. This function contains all the logics necessary to | |
2118 | * process a single work including synchronization against and | |
2119 | * interaction with other workers on the same cpu, queueing and | |
2120 | * flushing. As long as context requirement is met, any worker can | |
2121 | * call this function to process a work. | |
2122 | * | |
2123 | * CONTEXT: | |
d565ed63 | 2124 | * spin_lock_irq(pool->lock) which is released and regrabbed. |
a62428c0 | 2125 | */ |
c34056a3 | 2126 | static void process_one_work(struct worker *worker, struct work_struct *work) |
d565ed63 TH |
2127 | __releases(&pool->lock) |
2128 | __acquires(&pool->lock) | |
a62428c0 | 2129 | { |
112202d9 | 2130 | struct pool_workqueue *pwq = get_work_pwq(work); |
bd7bdd43 | 2131 | struct worker_pool *pool = worker->pool; |
112202d9 | 2132 | bool cpu_intensive = pwq->wq->flags & WQ_CPU_INTENSIVE; |
73f53c4a | 2133 | int work_color; |
7e11629d | 2134 | struct worker *collision; |
a62428c0 TH |
2135 | #ifdef CONFIG_LOCKDEP |
2136 | /* | |
2137 | * It is permissible to free the struct work_struct from | |
2138 | * inside the function that is called from it, this we need to | |
2139 | * take into account for lockdep too. To avoid bogus "held | |
2140 | * lock freed" warnings as well as problems when looking into | |
2141 | * work->lockdep_map, make a copy and use that here. | |
2142 | */ | |
4d82a1de PZ |
2143 | struct lockdep_map lockdep_map; |
2144 | ||
2145 | lockdep_copy_map(&lockdep_map, &work->lockdep_map); | |
a62428c0 | 2146 | #endif |
6fec10a1 TH |
2147 | /* |
2148 | * Ensure we're on the correct CPU. DISASSOCIATED test is | |
2149 | * necessary to avoid spurious warnings from rescuers servicing the | |
24647570 | 2150 | * unbound or a disassociated pool. |
6fec10a1 | 2151 | */ |
5f7dabfd | 2152 | WARN_ON_ONCE(!(worker->flags & WORKER_UNBOUND) && |
24647570 | 2153 | !(pool->flags & POOL_DISASSOCIATED) && |
ec22ca5e | 2154 | raw_smp_processor_id() != pool->cpu); |
25511a47 | 2155 | |
7e11629d TH |
2156 | /* |
2157 | * A single work shouldn't be executed concurrently by | |
2158 | * multiple workers on a single cpu. Check whether anyone is | |
2159 | * already processing the work. If so, defer the work to the | |
2160 | * currently executing one. | |
2161 | */ | |
c9e7cf27 | 2162 | collision = find_worker_executing_work(pool, work); |
7e11629d TH |
2163 | if (unlikely(collision)) { |
2164 | move_linked_works(work, &collision->scheduled, NULL); | |
2165 | return; | |
2166 | } | |
2167 | ||
8930caba | 2168 | /* claim and dequeue */ |
a62428c0 | 2169 | debug_work_deactivate(work); |
c9e7cf27 | 2170 | hash_add(pool->busy_hash, &worker->hentry, (unsigned long)work); |
c34056a3 | 2171 | worker->current_work = work; |
a2c1c57b | 2172 | worker->current_func = work->func; |
112202d9 | 2173 | worker->current_pwq = pwq; |
73f53c4a | 2174 | work_color = get_work_color(work); |
7a22ad75 | 2175 | |
a62428c0 TH |
2176 | list_del_init(&work->entry); |
2177 | ||
fb0e7beb TH |
2178 | /* |
2179 | * CPU intensive works don't participate in concurrency | |
2180 | * management. They're the scheduler's responsibility. | |
2181 | */ | |
2182 | if (unlikely(cpu_intensive)) | |
2183 | worker_set_flags(worker, WORKER_CPU_INTENSIVE, true); | |
2184 | ||
974271c4 | 2185 | /* |
d565ed63 | 2186 | * Unbound pool isn't concurrency managed and work items should be |
974271c4 TH |
2187 | * executed ASAP. Wake up another worker if necessary. |
2188 | */ | |
63d95a91 TH |
2189 | if ((worker->flags & WORKER_UNBOUND) && need_more_worker(pool)) |
2190 | wake_up_worker(pool); | |
974271c4 | 2191 | |
8930caba | 2192 | /* |
7c3eed5c | 2193 | * Record the last pool and clear PENDING which should be the last |
d565ed63 | 2194 | * update to @work. Also, do this inside @pool->lock so that |
23657bb1 TH |
2195 | * PENDING and queued state changes happen together while IRQ is |
2196 | * disabled. | |
8930caba | 2197 | */ |
7c3eed5c | 2198 | set_work_pool_and_clear_pending(work, pool->id); |
a62428c0 | 2199 | |
d565ed63 | 2200 | spin_unlock_irq(&pool->lock); |
a62428c0 | 2201 | |
112202d9 | 2202 | lock_map_acquire_read(&pwq->wq->lockdep_map); |
a62428c0 | 2203 | lock_map_acquire(&lockdep_map); |
e36c886a | 2204 | trace_workqueue_execute_start(work); |
a2c1c57b | 2205 | worker->current_func(work); |
e36c886a AV |
2206 | /* |
2207 | * While we must be careful to not use "work" after this, the trace | |
2208 | * point will only record its address. | |
2209 | */ | |
2210 | trace_workqueue_execute_end(work); | |
a62428c0 | 2211 | lock_map_release(&lockdep_map); |
112202d9 | 2212 | lock_map_release(&pwq->wq->lockdep_map); |
a62428c0 TH |
2213 | |
2214 | if (unlikely(in_atomic() || lockdep_depth(current) > 0)) { | |
044c782c VI |
2215 | pr_err("BUG: workqueue leaked lock or atomic: %s/0x%08x/%d\n" |
2216 | " last function: %pf\n", | |
a2c1c57b TH |
2217 | current->comm, preempt_count(), task_pid_nr(current), |
2218 | worker->current_func); | |
a62428c0 TH |
2219 | debug_show_held_locks(current); |
2220 | dump_stack(); | |
2221 | } | |
2222 | ||
b22ce278 TH |
2223 | /* |
2224 | * The following prevents a kworker from hogging CPU on !PREEMPT | |
2225 | * kernels, where a requeueing work item waiting for something to | |
2226 | * happen could deadlock with stop_machine as such work item could | |
2227 | * indefinitely requeue itself while all other CPUs are trapped in | |
2228 | * stop_machine. | |
2229 | */ | |
2230 | cond_resched(); | |
2231 | ||
d565ed63 | 2232 | spin_lock_irq(&pool->lock); |
a62428c0 | 2233 | |
fb0e7beb TH |
2234 | /* clear cpu intensive status */ |
2235 | if (unlikely(cpu_intensive)) | |
2236 | worker_clr_flags(worker, WORKER_CPU_INTENSIVE); | |
2237 | ||
a62428c0 | 2238 | /* we're done with it, release */ |
42f8570f | 2239 | hash_del(&worker->hentry); |
c34056a3 | 2240 | worker->current_work = NULL; |
a2c1c57b | 2241 | worker->current_func = NULL; |
112202d9 | 2242 | worker->current_pwq = NULL; |
3d1cb205 | 2243 | worker->desc_valid = false; |
112202d9 | 2244 | pwq_dec_nr_in_flight(pwq, work_color); |
a62428c0 TH |
2245 | } |
2246 | ||
affee4b2 TH |
2247 | /** |
2248 | * process_scheduled_works - process scheduled works | |
2249 | * @worker: self | |
2250 | * | |
2251 | * Process all scheduled works. Please note that the scheduled list | |
2252 | * may change while processing a work, so this function repeatedly | |
2253 | * fetches a work from the top and executes it. | |
2254 | * | |
2255 | * CONTEXT: | |
d565ed63 | 2256 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
affee4b2 TH |
2257 | * multiple times. |
2258 | */ | |
2259 | static void process_scheduled_works(struct worker *worker) | |
1da177e4 | 2260 | { |
affee4b2 TH |
2261 | while (!list_empty(&worker->scheduled)) { |
2262 | struct work_struct *work = list_first_entry(&worker->scheduled, | |
1da177e4 | 2263 | struct work_struct, entry); |
c34056a3 | 2264 | process_one_work(worker, work); |
1da177e4 | 2265 | } |
1da177e4 LT |
2266 | } |
2267 | ||
4690c4ab TH |
2268 | /** |
2269 | * worker_thread - the worker thread function | |
c34056a3 | 2270 | * @__worker: self |
4690c4ab | 2271 | * |
c5aa87bb TH |
2272 | * The worker thread function. All workers belong to a worker_pool - |
2273 | * either a per-cpu one or dynamic unbound one. These workers process all | |
2274 | * work items regardless of their specific target workqueue. The only | |
2275 | * exception is work items which belong to workqueues with a rescuer which | |
2276 | * will be explained in rescuer_thread(). | |
d185af30 YB |
2277 | * |
2278 | * Return: 0 | |
4690c4ab | 2279 | */ |
c34056a3 | 2280 | static int worker_thread(void *__worker) |
1da177e4 | 2281 | { |
c34056a3 | 2282 | struct worker *worker = __worker; |
bd7bdd43 | 2283 | struct worker_pool *pool = worker->pool; |
1da177e4 | 2284 | |
e22bee78 TH |
2285 | /* tell the scheduler that this is a workqueue worker */ |
2286 | worker->task->flags |= PF_WQ_WORKER; | |
c8e55f36 | 2287 | woke_up: |
d565ed63 | 2288 | spin_lock_irq(&pool->lock); |
1da177e4 | 2289 | |
a9ab775b TH |
2290 | /* am I supposed to die? */ |
2291 | if (unlikely(worker->flags & WORKER_DIE)) { | |
d565ed63 | 2292 | spin_unlock_irq(&pool->lock); |
a9ab775b TH |
2293 | WARN_ON_ONCE(!list_empty(&worker->entry)); |
2294 | worker->task->flags &= ~PF_WQ_WORKER; | |
2295 | return 0; | |
c8e55f36 | 2296 | } |
affee4b2 | 2297 | |
c8e55f36 | 2298 | worker_leave_idle(worker); |
db7bccf4 | 2299 | recheck: |
e22bee78 | 2300 | /* no more worker necessary? */ |
63d95a91 | 2301 | if (!need_more_worker(pool)) |
e22bee78 TH |
2302 | goto sleep; |
2303 | ||
2304 | /* do we need to manage? */ | |
63d95a91 | 2305 | if (unlikely(!may_start_working(pool)) && manage_workers(worker)) |
e22bee78 TH |
2306 | goto recheck; |
2307 | ||
c8e55f36 TH |
2308 | /* |
2309 | * ->scheduled list can only be filled while a worker is | |
2310 | * preparing to process a work or actually processing it. | |
2311 | * Make sure nobody diddled with it while I was sleeping. | |
2312 | */ | |
6183c009 | 2313 | WARN_ON_ONCE(!list_empty(&worker->scheduled)); |
c8e55f36 | 2314 | |
e22bee78 | 2315 | /* |
a9ab775b TH |
2316 | * Finish PREP stage. We're guaranteed to have at least one idle |
2317 | * worker or that someone else has already assumed the manager | |
2318 | * role. This is where @worker starts participating in concurrency | |
2319 | * management if applicable and concurrency management is restored | |
2320 | * after being rebound. See rebind_workers() for details. | |
e22bee78 | 2321 | */ |
a9ab775b | 2322 | worker_clr_flags(worker, WORKER_PREP | WORKER_REBOUND); |
e22bee78 TH |
2323 | |
2324 | do { | |
c8e55f36 | 2325 | struct work_struct *work = |
bd7bdd43 | 2326 | list_first_entry(&pool->worklist, |
c8e55f36 TH |
2327 | struct work_struct, entry); |
2328 | ||
2329 | if (likely(!(*work_data_bits(work) & WORK_STRUCT_LINKED))) { | |
2330 | /* optimization path, not strictly necessary */ | |
2331 | process_one_work(worker, work); | |
2332 | if (unlikely(!list_empty(&worker->scheduled))) | |
affee4b2 | 2333 | process_scheduled_works(worker); |
c8e55f36 TH |
2334 | } else { |
2335 | move_linked_works(work, &worker->scheduled, NULL); | |
2336 | process_scheduled_works(worker); | |
affee4b2 | 2337 | } |
63d95a91 | 2338 | } while (keep_working(pool)); |
e22bee78 TH |
2339 | |
2340 | worker_set_flags(worker, WORKER_PREP, false); | |
d313dd85 | 2341 | sleep: |
63d95a91 | 2342 | if (unlikely(need_to_manage_workers(pool)) && manage_workers(worker)) |
e22bee78 | 2343 | goto recheck; |
d313dd85 | 2344 | |
c8e55f36 | 2345 | /* |
d565ed63 TH |
2346 | * pool->lock is held and there's no work to process and no need to |
2347 | * manage, sleep. Workers are woken up only while holding | |
2348 | * pool->lock or from local cpu, so setting the current state | |
2349 | * before releasing pool->lock is enough to prevent losing any | |
2350 | * event. | |
c8e55f36 TH |
2351 | */ |
2352 | worker_enter_idle(worker); | |
2353 | __set_current_state(TASK_INTERRUPTIBLE); | |
d565ed63 | 2354 | spin_unlock_irq(&pool->lock); |
c8e55f36 TH |
2355 | schedule(); |
2356 | goto woke_up; | |
1da177e4 LT |
2357 | } |
2358 | ||
e22bee78 TH |
2359 | /** |
2360 | * rescuer_thread - the rescuer thread function | |
111c225a | 2361 | * @__rescuer: self |
e22bee78 TH |
2362 | * |
2363 | * Workqueue rescuer thread function. There's one rescuer for each | |
493008a8 | 2364 | * workqueue which has WQ_MEM_RECLAIM set. |
e22bee78 | 2365 | * |
706026c2 | 2366 | * Regular work processing on a pool may block trying to create a new |
e22bee78 TH |
2367 | * worker which uses GFP_KERNEL allocation which has slight chance of |
2368 | * developing into deadlock if some works currently on the same queue | |
2369 | * need to be processed to satisfy the GFP_KERNEL allocation. This is | |
2370 | * the problem rescuer solves. | |
2371 | * | |
706026c2 TH |
2372 | * When such condition is possible, the pool summons rescuers of all |
2373 | * workqueues which have works queued on the pool and let them process | |
e22bee78 TH |
2374 | * those works so that forward progress can be guaranteed. |
2375 | * | |
2376 | * This should happen rarely. | |
d185af30 YB |
2377 | * |
2378 | * Return: 0 | |
e22bee78 | 2379 | */ |
111c225a | 2380 | static int rescuer_thread(void *__rescuer) |
e22bee78 | 2381 | { |
111c225a TH |
2382 | struct worker *rescuer = __rescuer; |
2383 | struct workqueue_struct *wq = rescuer->rescue_wq; | |
e22bee78 | 2384 | struct list_head *scheduled = &rescuer->scheduled; |
4d595b86 | 2385 | bool should_stop; |
e22bee78 TH |
2386 | |
2387 | set_user_nice(current, RESCUER_NICE_LEVEL); | |
111c225a TH |
2388 | |
2389 | /* | |
2390 | * Mark rescuer as worker too. As WORKER_PREP is never cleared, it | |
2391 | * doesn't participate in concurrency management. | |
2392 | */ | |
2393 | rescuer->task->flags |= PF_WQ_WORKER; | |
e22bee78 TH |
2394 | repeat: |
2395 | set_current_state(TASK_INTERRUPTIBLE); | |
2396 | ||
4d595b86 LJ |
2397 | /* |
2398 | * By the time the rescuer is requested to stop, the workqueue | |
2399 | * shouldn't have any work pending, but @wq->maydays may still have | |
2400 | * pwq(s) queued. This can happen by non-rescuer workers consuming | |
2401 | * all the work items before the rescuer got to them. Go through | |
2402 | * @wq->maydays processing before acting on should_stop so that the | |
2403 | * list is always empty on exit. | |
2404 | */ | |
2405 | should_stop = kthread_should_stop(); | |
e22bee78 | 2406 | |
493a1724 | 2407 | /* see whether any pwq is asking for help */ |
2e109a28 | 2408 | spin_lock_irq(&wq_mayday_lock); |
493a1724 TH |
2409 | |
2410 | while (!list_empty(&wq->maydays)) { | |
2411 | struct pool_workqueue *pwq = list_first_entry(&wq->maydays, | |
2412 | struct pool_workqueue, mayday_node); | |
112202d9 | 2413 | struct worker_pool *pool = pwq->pool; |
e22bee78 TH |
2414 | struct work_struct *work, *n; |
2415 | ||
2416 | __set_current_state(TASK_RUNNING); | |
493a1724 TH |
2417 | list_del_init(&pwq->mayday_node); |
2418 | ||
2e109a28 | 2419 | spin_unlock_irq(&wq_mayday_lock); |
e22bee78 TH |
2420 | |
2421 | /* migrate to the target cpu if possible */ | |
f36dc67b | 2422 | worker_maybe_bind_and_lock(pool); |
b3104104 | 2423 | rescuer->pool = pool; |
e22bee78 TH |
2424 | |
2425 | /* | |
2426 | * Slurp in all works issued via this workqueue and | |
2427 | * process'em. | |
2428 | */ | |
6183c009 | 2429 | WARN_ON_ONCE(!list_empty(&rescuer->scheduled)); |
bd7bdd43 | 2430 | list_for_each_entry_safe(work, n, &pool->worklist, entry) |
112202d9 | 2431 | if (get_work_pwq(work) == pwq) |
e22bee78 TH |
2432 | move_linked_works(work, scheduled, &n); |
2433 | ||
2434 | process_scheduled_works(rescuer); | |
7576958a | 2435 | |
77668c8b LJ |
2436 | /* |
2437 | * Put the reference grabbed by send_mayday(). @pool won't | |
2438 | * go away while we're holding its lock. | |
2439 | */ | |
2440 | put_pwq(pwq); | |
2441 | ||
7576958a | 2442 | /* |
d565ed63 | 2443 | * Leave this pool. If keep_working() is %true, notify a |
7576958a TH |
2444 | * regular worker; otherwise, we end up with 0 concurrency |
2445 | * and stalling the execution. | |
2446 | */ | |
63d95a91 TH |
2447 | if (keep_working(pool)) |
2448 | wake_up_worker(pool); | |
7576958a | 2449 | |
b3104104 | 2450 | rescuer->pool = NULL; |
493a1724 | 2451 | spin_unlock(&pool->lock); |
2e109a28 | 2452 | spin_lock(&wq_mayday_lock); |
e22bee78 TH |
2453 | } |
2454 | ||
2e109a28 | 2455 | spin_unlock_irq(&wq_mayday_lock); |
493a1724 | 2456 | |
4d595b86 LJ |
2457 | if (should_stop) { |
2458 | __set_current_state(TASK_RUNNING); | |
2459 | rescuer->task->flags &= ~PF_WQ_WORKER; | |
2460 | return 0; | |
2461 | } | |
2462 | ||
111c225a TH |
2463 | /* rescuers should never participate in concurrency management */ |
2464 | WARN_ON_ONCE(!(rescuer->flags & WORKER_NOT_RUNNING)); | |
e22bee78 TH |
2465 | schedule(); |
2466 | goto repeat; | |
1da177e4 LT |
2467 | } |
2468 | ||
fc2e4d70 ON |
2469 | struct wq_barrier { |
2470 | struct work_struct work; | |
2471 | struct completion done; | |
2472 | }; | |
2473 | ||
2474 | static void wq_barrier_func(struct work_struct *work) | |
2475 | { | |
2476 | struct wq_barrier *barr = container_of(work, struct wq_barrier, work); | |
2477 | complete(&barr->done); | |
2478 | } | |
2479 | ||
4690c4ab TH |
2480 | /** |
2481 | * insert_wq_barrier - insert a barrier work | |
112202d9 | 2482 | * @pwq: pwq to insert barrier into |
4690c4ab | 2483 | * @barr: wq_barrier to insert |
affee4b2 TH |
2484 | * @target: target work to attach @barr to |
2485 | * @worker: worker currently executing @target, NULL if @target is not executing | |
4690c4ab | 2486 | * |
affee4b2 TH |
2487 | * @barr is linked to @target such that @barr is completed only after |
2488 | * @target finishes execution. Please note that the ordering | |
2489 | * guarantee is observed only with respect to @target and on the local | |
2490 | * cpu. | |
2491 | * | |
2492 | * Currently, a queued barrier can't be canceled. This is because | |
2493 | * try_to_grab_pending() can't determine whether the work to be | |
2494 | * grabbed is at the head of the queue and thus can't clear LINKED | |
2495 | * flag of the previous work while there must be a valid next work | |
2496 | * after a work with LINKED flag set. | |
2497 | * | |
2498 | * Note that when @worker is non-NULL, @target may be modified | |
112202d9 | 2499 | * underneath us, so we can't reliably determine pwq from @target. |
4690c4ab TH |
2500 | * |
2501 | * CONTEXT: | |
d565ed63 | 2502 | * spin_lock_irq(pool->lock). |
4690c4ab | 2503 | */ |
112202d9 | 2504 | static void insert_wq_barrier(struct pool_workqueue *pwq, |
affee4b2 TH |
2505 | struct wq_barrier *barr, |
2506 | struct work_struct *target, struct worker *worker) | |
fc2e4d70 | 2507 | { |
affee4b2 TH |
2508 | struct list_head *head; |
2509 | unsigned int linked = 0; | |
2510 | ||
dc186ad7 | 2511 | /* |
d565ed63 | 2512 | * debugobject calls are safe here even with pool->lock locked |
dc186ad7 TG |
2513 | * as we know for sure that this will not trigger any of the |
2514 | * checks and call back into the fixup functions where we | |
2515 | * might deadlock. | |
2516 | */ | |
ca1cab37 | 2517 | INIT_WORK_ONSTACK(&barr->work, wq_barrier_func); |
22df02bb | 2518 | __set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(&barr->work)); |
fc2e4d70 | 2519 | init_completion(&barr->done); |
83c22520 | 2520 | |
affee4b2 TH |
2521 | /* |
2522 | * If @target is currently being executed, schedule the | |
2523 | * barrier to the worker; otherwise, put it after @target. | |
2524 | */ | |
2525 | if (worker) | |
2526 | head = worker->scheduled.next; | |
2527 | else { | |
2528 | unsigned long *bits = work_data_bits(target); | |
2529 | ||
2530 | head = target->entry.next; | |
2531 | /* there can already be other linked works, inherit and set */ | |
2532 | linked = *bits & WORK_STRUCT_LINKED; | |
2533 | __set_bit(WORK_STRUCT_LINKED_BIT, bits); | |
2534 | } | |
2535 | ||
dc186ad7 | 2536 | debug_work_activate(&barr->work); |
112202d9 | 2537 | insert_work(pwq, &barr->work, head, |
affee4b2 | 2538 | work_color_to_flags(WORK_NO_COLOR) | linked); |
fc2e4d70 ON |
2539 | } |
2540 | ||
73f53c4a | 2541 | /** |
112202d9 | 2542 | * flush_workqueue_prep_pwqs - prepare pwqs for workqueue flushing |
73f53c4a TH |
2543 | * @wq: workqueue being flushed |
2544 | * @flush_color: new flush color, < 0 for no-op | |
2545 | * @work_color: new work color, < 0 for no-op | |
2546 | * | |
112202d9 | 2547 | * Prepare pwqs for workqueue flushing. |
73f53c4a | 2548 | * |
112202d9 TH |
2549 | * If @flush_color is non-negative, flush_color on all pwqs should be |
2550 | * -1. If no pwq has in-flight commands at the specified color, all | |
2551 | * pwq->flush_color's stay at -1 and %false is returned. If any pwq | |
2552 | * has in flight commands, its pwq->flush_color is set to | |
2553 | * @flush_color, @wq->nr_pwqs_to_flush is updated accordingly, pwq | |
73f53c4a TH |
2554 | * wakeup logic is armed and %true is returned. |
2555 | * | |
2556 | * The caller should have initialized @wq->first_flusher prior to | |
2557 | * calling this function with non-negative @flush_color. If | |
2558 | * @flush_color is negative, no flush color update is done and %false | |
2559 | * is returned. | |
2560 | * | |
112202d9 | 2561 | * If @work_color is non-negative, all pwqs should have the same |
73f53c4a TH |
2562 | * work_color which is previous to @work_color and all will be |
2563 | * advanced to @work_color. | |
2564 | * | |
2565 | * CONTEXT: | |
3c25a55d | 2566 | * mutex_lock(wq->mutex). |
73f53c4a | 2567 | * |
d185af30 | 2568 | * Return: |
73f53c4a TH |
2569 | * %true if @flush_color >= 0 and there's something to flush. %false |
2570 | * otherwise. | |
2571 | */ | |
112202d9 | 2572 | static bool flush_workqueue_prep_pwqs(struct workqueue_struct *wq, |
73f53c4a | 2573 | int flush_color, int work_color) |
1da177e4 | 2574 | { |
73f53c4a | 2575 | bool wait = false; |
49e3cf44 | 2576 | struct pool_workqueue *pwq; |
1da177e4 | 2577 | |
73f53c4a | 2578 | if (flush_color >= 0) { |
6183c009 | 2579 | WARN_ON_ONCE(atomic_read(&wq->nr_pwqs_to_flush)); |
112202d9 | 2580 | atomic_set(&wq->nr_pwqs_to_flush, 1); |
1da177e4 | 2581 | } |
2355b70f | 2582 | |
49e3cf44 | 2583 | for_each_pwq(pwq, wq) { |
112202d9 | 2584 | struct worker_pool *pool = pwq->pool; |
fc2e4d70 | 2585 | |
b09f4fd3 | 2586 | spin_lock_irq(&pool->lock); |
83c22520 | 2587 | |
73f53c4a | 2588 | if (flush_color >= 0) { |
6183c009 | 2589 | WARN_ON_ONCE(pwq->flush_color != -1); |
fc2e4d70 | 2590 | |
112202d9 TH |
2591 | if (pwq->nr_in_flight[flush_color]) { |
2592 | pwq->flush_color = flush_color; | |
2593 | atomic_inc(&wq->nr_pwqs_to_flush); | |
73f53c4a TH |
2594 | wait = true; |
2595 | } | |
2596 | } | |
1da177e4 | 2597 | |
73f53c4a | 2598 | if (work_color >= 0) { |
6183c009 | 2599 | WARN_ON_ONCE(work_color != work_next_color(pwq->work_color)); |
112202d9 | 2600 | pwq->work_color = work_color; |
73f53c4a | 2601 | } |
1da177e4 | 2602 | |
b09f4fd3 | 2603 | spin_unlock_irq(&pool->lock); |
1da177e4 | 2604 | } |
2355b70f | 2605 | |
112202d9 | 2606 | if (flush_color >= 0 && atomic_dec_and_test(&wq->nr_pwqs_to_flush)) |
73f53c4a | 2607 | complete(&wq->first_flusher->done); |
14441960 | 2608 | |
73f53c4a | 2609 | return wait; |
1da177e4 LT |
2610 | } |
2611 | ||
0fcb78c2 | 2612 | /** |
1da177e4 | 2613 | * flush_workqueue - ensure that any scheduled work has run to completion. |
0fcb78c2 | 2614 | * @wq: workqueue to flush |
1da177e4 | 2615 | * |
c5aa87bb TH |
2616 | * This function sleeps until all work items which were queued on entry |
2617 | * have finished execution, but it is not livelocked by new incoming ones. | |
1da177e4 | 2618 | */ |
7ad5b3a5 | 2619 | void flush_workqueue(struct workqueue_struct *wq) |
1da177e4 | 2620 | { |
73f53c4a TH |
2621 | struct wq_flusher this_flusher = { |
2622 | .list = LIST_HEAD_INIT(this_flusher.list), | |
2623 | .flush_color = -1, | |
2624 | .done = COMPLETION_INITIALIZER_ONSTACK(this_flusher.done), | |
2625 | }; | |
2626 | int next_color; | |
1da177e4 | 2627 | |
3295f0ef IM |
2628 | lock_map_acquire(&wq->lockdep_map); |
2629 | lock_map_release(&wq->lockdep_map); | |
73f53c4a | 2630 | |
3c25a55d | 2631 | mutex_lock(&wq->mutex); |
73f53c4a TH |
2632 | |
2633 | /* | |
2634 | * Start-to-wait phase | |
2635 | */ | |
2636 | next_color = work_next_color(wq->work_color); | |
2637 | ||
2638 | if (next_color != wq->flush_color) { | |
2639 | /* | |
2640 | * Color space is not full. The current work_color | |
2641 | * becomes our flush_color and work_color is advanced | |
2642 | * by one. | |
2643 | */ | |
6183c009 | 2644 | WARN_ON_ONCE(!list_empty(&wq->flusher_overflow)); |
73f53c4a TH |
2645 | this_flusher.flush_color = wq->work_color; |
2646 | wq->work_color = next_color; | |
2647 | ||
2648 | if (!wq->first_flusher) { | |
2649 | /* no flush in progress, become the first flusher */ | |
6183c009 | 2650 | WARN_ON_ONCE(wq->flush_color != this_flusher.flush_color); |
73f53c4a TH |
2651 | |
2652 | wq->first_flusher = &this_flusher; | |
2653 | ||
112202d9 | 2654 | if (!flush_workqueue_prep_pwqs(wq, wq->flush_color, |
73f53c4a TH |
2655 | wq->work_color)) { |
2656 | /* nothing to flush, done */ | |
2657 | wq->flush_color = next_color; | |
2658 | wq->first_flusher = NULL; | |
2659 | goto out_unlock; | |
2660 | } | |
2661 | } else { | |
2662 | /* wait in queue */ | |
6183c009 | 2663 | WARN_ON_ONCE(wq->flush_color == this_flusher.flush_color); |
73f53c4a | 2664 | list_add_tail(&this_flusher.list, &wq->flusher_queue); |
112202d9 | 2665 | flush_workqueue_prep_pwqs(wq, -1, wq->work_color); |
73f53c4a TH |
2666 | } |
2667 | } else { | |
2668 | /* | |
2669 | * Oops, color space is full, wait on overflow queue. | |
2670 | * The next flush completion will assign us | |
2671 | * flush_color and transfer to flusher_queue. | |
2672 | */ | |
2673 | list_add_tail(&this_flusher.list, &wq->flusher_overflow); | |
2674 | } | |
2675 | ||
3c25a55d | 2676 | mutex_unlock(&wq->mutex); |
73f53c4a TH |
2677 | |
2678 | wait_for_completion(&this_flusher.done); | |
2679 | ||
2680 | /* | |
2681 | * Wake-up-and-cascade phase | |
2682 | * | |
2683 | * First flushers are responsible for cascading flushes and | |
2684 | * handling overflow. Non-first flushers can simply return. | |
2685 | */ | |
2686 | if (wq->first_flusher != &this_flusher) | |
2687 | return; | |
2688 | ||
3c25a55d | 2689 | mutex_lock(&wq->mutex); |
73f53c4a | 2690 | |
4ce48b37 TH |
2691 | /* we might have raced, check again with mutex held */ |
2692 | if (wq->first_flusher != &this_flusher) | |
2693 | goto out_unlock; | |
2694 | ||
73f53c4a TH |
2695 | wq->first_flusher = NULL; |
2696 | ||
6183c009 TH |
2697 | WARN_ON_ONCE(!list_empty(&this_flusher.list)); |
2698 | WARN_ON_ONCE(wq->flush_color != this_flusher.flush_color); | |
73f53c4a TH |
2699 | |
2700 | while (true) { | |
2701 | struct wq_flusher *next, *tmp; | |
2702 | ||
2703 | /* complete all the flushers sharing the current flush color */ | |
2704 | list_for_each_entry_safe(next, tmp, &wq->flusher_queue, list) { | |
2705 | if (next->flush_color != wq->flush_color) | |
2706 | break; | |
2707 | list_del_init(&next->list); | |
2708 | complete(&next->done); | |
2709 | } | |
2710 | ||
6183c009 TH |
2711 | WARN_ON_ONCE(!list_empty(&wq->flusher_overflow) && |
2712 | wq->flush_color != work_next_color(wq->work_color)); | |
73f53c4a TH |
2713 | |
2714 | /* this flush_color is finished, advance by one */ | |
2715 | wq->flush_color = work_next_color(wq->flush_color); | |
2716 | ||
2717 | /* one color has been freed, handle overflow queue */ | |
2718 | if (!list_empty(&wq->flusher_overflow)) { | |
2719 | /* | |
2720 | * Assign the same color to all overflowed | |
2721 | * flushers, advance work_color and append to | |
2722 | * flusher_queue. This is the start-to-wait | |
2723 | * phase for these overflowed flushers. | |
2724 | */ | |
2725 | list_for_each_entry(tmp, &wq->flusher_overflow, list) | |
2726 | tmp->flush_color = wq->work_color; | |
2727 | ||
2728 | wq->work_color = work_next_color(wq->work_color); | |
2729 | ||
2730 | list_splice_tail_init(&wq->flusher_overflow, | |
2731 | &wq->flusher_queue); | |
112202d9 | 2732 | flush_workqueue_prep_pwqs(wq, -1, wq->work_color); |
73f53c4a TH |
2733 | } |
2734 | ||
2735 | if (list_empty(&wq->flusher_queue)) { | |
6183c009 | 2736 | WARN_ON_ONCE(wq->flush_color != wq->work_color); |
73f53c4a TH |
2737 | break; |
2738 | } | |
2739 | ||
2740 | /* | |
2741 | * Need to flush more colors. Make the next flusher | |
112202d9 | 2742 | * the new first flusher and arm pwqs. |
73f53c4a | 2743 | */ |
6183c009 TH |
2744 | WARN_ON_ONCE(wq->flush_color == wq->work_color); |
2745 | WARN_ON_ONCE(wq->flush_color != next->flush_color); | |
73f53c4a TH |
2746 | |
2747 | list_del_init(&next->list); | |
2748 | wq->first_flusher = next; | |
2749 | ||
112202d9 | 2750 | if (flush_workqueue_prep_pwqs(wq, wq->flush_color, -1)) |
73f53c4a TH |
2751 | break; |
2752 | ||
2753 | /* | |
2754 | * Meh... this color is already done, clear first | |
2755 | * flusher and repeat cascading. | |
2756 | */ | |
2757 | wq->first_flusher = NULL; | |
2758 | } | |
2759 | ||
2760 | out_unlock: | |
3c25a55d | 2761 | mutex_unlock(&wq->mutex); |
1da177e4 | 2762 | } |
ae90dd5d | 2763 | EXPORT_SYMBOL_GPL(flush_workqueue); |
1da177e4 | 2764 | |
9c5a2ba7 TH |
2765 | /** |
2766 | * drain_workqueue - drain a workqueue | |
2767 | * @wq: workqueue to drain | |
2768 | * | |
2769 | * Wait until the workqueue becomes empty. While draining is in progress, | |
2770 | * only chain queueing is allowed. IOW, only currently pending or running | |
2771 | * work items on @wq can queue further work items on it. @wq is flushed | |
2772 | * repeatedly until it becomes empty. The number of flushing is detemined | |
2773 | * by the depth of chaining and should be relatively short. Whine if it | |
2774 | * takes too long. | |
2775 | */ | |
2776 | void drain_workqueue(struct workqueue_struct *wq) | |
2777 | { | |
2778 | unsigned int flush_cnt = 0; | |
49e3cf44 | 2779 | struct pool_workqueue *pwq; |
9c5a2ba7 TH |
2780 | |
2781 | /* | |
2782 | * __queue_work() needs to test whether there are drainers, is much | |
2783 | * hotter than drain_workqueue() and already looks at @wq->flags. | |
618b01eb | 2784 | * Use __WQ_DRAINING so that queue doesn't have to check nr_drainers. |
9c5a2ba7 | 2785 | */ |
87fc741e | 2786 | mutex_lock(&wq->mutex); |
9c5a2ba7 | 2787 | if (!wq->nr_drainers++) |
618b01eb | 2788 | wq->flags |= __WQ_DRAINING; |
87fc741e | 2789 | mutex_unlock(&wq->mutex); |
9c5a2ba7 TH |
2790 | reflush: |
2791 | flush_workqueue(wq); | |
2792 | ||
b09f4fd3 | 2793 | mutex_lock(&wq->mutex); |
76af4d93 | 2794 | |
49e3cf44 | 2795 | for_each_pwq(pwq, wq) { |
fa2563e4 | 2796 | bool drained; |
9c5a2ba7 | 2797 | |
b09f4fd3 | 2798 | spin_lock_irq(&pwq->pool->lock); |
112202d9 | 2799 | drained = !pwq->nr_active && list_empty(&pwq->delayed_works); |
b09f4fd3 | 2800 | spin_unlock_irq(&pwq->pool->lock); |
fa2563e4 TT |
2801 | |
2802 | if (drained) | |
9c5a2ba7 TH |
2803 | continue; |
2804 | ||
2805 | if (++flush_cnt == 10 || | |
2806 | (flush_cnt % 100 == 0 && flush_cnt <= 1000)) | |
c5aa87bb | 2807 | pr_warn("workqueue %s: drain_workqueue() isn't complete after %u tries\n", |
044c782c | 2808 | wq->name, flush_cnt); |
76af4d93 | 2809 | |
b09f4fd3 | 2810 | mutex_unlock(&wq->mutex); |
9c5a2ba7 TH |
2811 | goto reflush; |
2812 | } | |
2813 | ||
9c5a2ba7 | 2814 | if (!--wq->nr_drainers) |
618b01eb | 2815 | wq->flags &= ~__WQ_DRAINING; |
87fc741e | 2816 | mutex_unlock(&wq->mutex); |
9c5a2ba7 TH |
2817 | } |
2818 | EXPORT_SYMBOL_GPL(drain_workqueue); | |
2819 | ||
606a5020 | 2820 | static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr) |
db700897 | 2821 | { |
affee4b2 | 2822 | struct worker *worker = NULL; |
c9e7cf27 | 2823 | struct worker_pool *pool; |
112202d9 | 2824 | struct pool_workqueue *pwq; |
db700897 ON |
2825 | |
2826 | might_sleep(); | |
fa1b54e6 TH |
2827 | |
2828 | local_irq_disable(); | |
c9e7cf27 | 2829 | pool = get_work_pool(work); |
fa1b54e6 TH |
2830 | if (!pool) { |
2831 | local_irq_enable(); | |
baf59022 | 2832 | return false; |
fa1b54e6 | 2833 | } |
db700897 | 2834 | |
fa1b54e6 | 2835 | spin_lock(&pool->lock); |
0b3dae68 | 2836 | /* see the comment in try_to_grab_pending() with the same code */ |
112202d9 TH |
2837 | pwq = get_work_pwq(work); |
2838 | if (pwq) { | |
2839 | if (unlikely(pwq->pool != pool)) | |
4690c4ab | 2840 | goto already_gone; |
606a5020 | 2841 | } else { |
c9e7cf27 | 2842 | worker = find_worker_executing_work(pool, work); |
affee4b2 | 2843 | if (!worker) |
4690c4ab | 2844 | goto already_gone; |
112202d9 | 2845 | pwq = worker->current_pwq; |
606a5020 | 2846 | } |
db700897 | 2847 | |
112202d9 | 2848 | insert_wq_barrier(pwq, barr, work, worker); |
d565ed63 | 2849 | spin_unlock_irq(&pool->lock); |
7a22ad75 | 2850 | |
e159489b TH |
2851 | /* |
2852 | * If @max_active is 1 or rescuer is in use, flushing another work | |
2853 | * item on the same workqueue may lead to deadlock. Make sure the | |
2854 | * flusher is not running on the same workqueue by verifying write | |
2855 | * access. | |
2856 | */ | |
493008a8 | 2857 | if (pwq->wq->saved_max_active == 1 || pwq->wq->rescuer) |
112202d9 | 2858 | lock_map_acquire(&pwq->wq->lockdep_map); |
e159489b | 2859 | else |
112202d9 TH |
2860 | lock_map_acquire_read(&pwq->wq->lockdep_map); |
2861 | lock_map_release(&pwq->wq->lockdep_map); | |
e159489b | 2862 | |
401a8d04 | 2863 | return true; |
4690c4ab | 2864 | already_gone: |
d565ed63 | 2865 | spin_unlock_irq(&pool->lock); |
401a8d04 | 2866 | return false; |
db700897 | 2867 | } |
baf59022 TH |
2868 | |
2869 | /** | |
2870 | * flush_work - wait for a work to finish executing the last queueing instance | |
2871 | * @work: the work to flush | |
2872 | * | |
606a5020 TH |
2873 | * Wait until @work has finished execution. @work is guaranteed to be idle |
2874 | * on return if it hasn't been requeued since flush started. | |
baf59022 | 2875 | * |
d185af30 | 2876 | * Return: |
baf59022 TH |
2877 | * %true if flush_work() waited for the work to finish execution, |
2878 | * %false if it was already idle. | |
2879 | */ | |
2880 | bool flush_work(struct work_struct *work) | |
2881 | { | |
12997d1a BH |
2882 | struct wq_barrier barr; |
2883 | ||
0976dfc1 SB |
2884 | lock_map_acquire(&work->lockdep_map); |
2885 | lock_map_release(&work->lockdep_map); | |
2886 | ||
12997d1a BH |
2887 | if (start_flush_work(work, &barr)) { |
2888 | wait_for_completion(&barr.done); | |
2889 | destroy_work_on_stack(&barr.work); | |
2890 | return true; | |
2891 | } else { | |
2892 | return false; | |
2893 | } | |
6e84d644 | 2894 | } |
606a5020 | 2895 | EXPORT_SYMBOL_GPL(flush_work); |
6e84d644 | 2896 | |
36e227d2 | 2897 | static bool __cancel_work_timer(struct work_struct *work, bool is_dwork) |
1f1f642e | 2898 | { |
bbb68dfa | 2899 | unsigned long flags; |
1f1f642e ON |
2900 | int ret; |
2901 | ||
2902 | do { | |
bbb68dfa TH |
2903 | ret = try_to_grab_pending(work, is_dwork, &flags); |
2904 | /* | |
2905 | * If someone else is canceling, wait for the same event it | |
2906 | * would be waiting for before retrying. | |
2907 | */ | |
2908 | if (unlikely(ret == -ENOENT)) | |
606a5020 | 2909 | flush_work(work); |
1f1f642e ON |
2910 | } while (unlikely(ret < 0)); |
2911 | ||
bbb68dfa TH |
2912 | /* tell other tasks trying to grab @work to back off */ |
2913 | mark_work_canceling(work); | |
2914 | local_irq_restore(flags); | |
2915 | ||
606a5020 | 2916 | flush_work(work); |
7a22ad75 | 2917 | clear_work_data(work); |
1f1f642e ON |
2918 | return ret; |
2919 | } | |
2920 | ||
6e84d644 | 2921 | /** |
401a8d04 TH |
2922 | * cancel_work_sync - cancel a work and wait for it to finish |
2923 | * @work: the work to cancel | |
6e84d644 | 2924 | * |
401a8d04 TH |
2925 | * Cancel @work and wait for its execution to finish. This function |
2926 | * can be used even if the work re-queues itself or migrates to | |
2927 | * another workqueue. On return from this function, @work is | |
2928 | * guaranteed to be not pending or executing on any CPU. | |
1f1f642e | 2929 | * |
401a8d04 TH |
2930 | * cancel_work_sync(&delayed_work->work) must not be used for |
2931 | * delayed_work's. Use cancel_delayed_work_sync() instead. | |
6e84d644 | 2932 | * |
401a8d04 | 2933 | * The caller must ensure that the workqueue on which @work was last |
6e84d644 | 2934 | * queued can't be destroyed before this function returns. |
401a8d04 | 2935 | * |
d185af30 | 2936 | * Return: |
401a8d04 | 2937 | * %true if @work was pending, %false otherwise. |
6e84d644 | 2938 | */ |
401a8d04 | 2939 | bool cancel_work_sync(struct work_struct *work) |
6e84d644 | 2940 | { |
36e227d2 | 2941 | return __cancel_work_timer(work, false); |
b89deed3 | 2942 | } |
28e53bdd | 2943 | EXPORT_SYMBOL_GPL(cancel_work_sync); |
b89deed3 | 2944 | |
6e84d644 | 2945 | /** |
401a8d04 TH |
2946 | * flush_delayed_work - wait for a dwork to finish executing the last queueing |
2947 | * @dwork: the delayed work to flush | |
6e84d644 | 2948 | * |
401a8d04 TH |
2949 | * Delayed timer is cancelled and the pending work is queued for |
2950 | * immediate execution. Like flush_work(), this function only | |
2951 | * considers the last queueing instance of @dwork. | |
1f1f642e | 2952 | * |
d185af30 | 2953 | * Return: |
401a8d04 TH |
2954 | * %true if flush_work() waited for the work to finish execution, |
2955 | * %false if it was already idle. | |
6e84d644 | 2956 | */ |
401a8d04 TH |
2957 | bool flush_delayed_work(struct delayed_work *dwork) |
2958 | { | |
8930caba | 2959 | local_irq_disable(); |
401a8d04 | 2960 | if (del_timer_sync(&dwork->timer)) |
60c057bc | 2961 | __queue_work(dwork->cpu, dwork->wq, &dwork->work); |
8930caba | 2962 | local_irq_enable(); |
401a8d04 TH |
2963 | return flush_work(&dwork->work); |
2964 | } | |
2965 | EXPORT_SYMBOL(flush_delayed_work); | |
2966 | ||
09383498 | 2967 | /** |
57b30ae7 TH |
2968 | * cancel_delayed_work - cancel a delayed work |
2969 | * @dwork: delayed_work to cancel | |
09383498 | 2970 | * |
d185af30 YB |
2971 | * Kill off a pending delayed_work. |
2972 | * | |
2973 | * Return: %true if @dwork was pending and canceled; %false if it wasn't | |
2974 | * pending. | |
2975 | * | |
2976 | * Note: | |
2977 | * The work callback function may still be running on return, unless | |
2978 | * it returns %true and the work doesn't re-arm itself. Explicitly flush or | |
2979 | * use cancel_delayed_work_sync() to wait on it. | |
09383498 | 2980 | * |
57b30ae7 | 2981 | * This function is safe to call from any context including IRQ handler. |
09383498 | 2982 | */ |
57b30ae7 | 2983 | bool cancel_delayed_work(struct delayed_work *dwork) |
09383498 | 2984 | { |
57b30ae7 TH |
2985 | unsigned long flags; |
2986 | int ret; | |
2987 | ||
2988 | do { | |
2989 | ret = try_to_grab_pending(&dwork->work, true, &flags); | |
2990 | } while (unlikely(ret == -EAGAIN)); | |
2991 | ||
2992 | if (unlikely(ret < 0)) | |
2993 | return false; | |
2994 | ||
7c3eed5c TH |
2995 | set_work_pool_and_clear_pending(&dwork->work, |
2996 | get_work_pool_id(&dwork->work)); | |
57b30ae7 | 2997 | local_irq_restore(flags); |
c0158ca6 | 2998 | return ret; |
09383498 | 2999 | } |
57b30ae7 | 3000 | EXPORT_SYMBOL(cancel_delayed_work); |
09383498 | 3001 | |
401a8d04 TH |
3002 | /** |
3003 | * cancel_delayed_work_sync - cancel a delayed work and wait for it to finish | |
3004 | * @dwork: the delayed work cancel | |
3005 | * | |
3006 | * This is cancel_work_sync() for delayed works. | |
3007 | * | |
d185af30 | 3008 | * Return: |
401a8d04 TH |
3009 | * %true if @dwork was pending, %false otherwise. |
3010 | */ | |
3011 | bool cancel_delayed_work_sync(struct delayed_work *dwork) | |
6e84d644 | 3012 | { |
36e227d2 | 3013 | return __cancel_work_timer(&dwork->work, true); |
6e84d644 | 3014 | } |
f5a421a4 | 3015 | EXPORT_SYMBOL(cancel_delayed_work_sync); |
1da177e4 | 3016 | |
b6136773 | 3017 | /** |
31ddd871 | 3018 | * schedule_on_each_cpu - execute a function synchronously on each online CPU |
b6136773 | 3019 | * @func: the function to call |
b6136773 | 3020 | * |
31ddd871 TH |
3021 | * schedule_on_each_cpu() executes @func on each online CPU using the |
3022 | * system workqueue and blocks until all CPUs have completed. | |
b6136773 | 3023 | * schedule_on_each_cpu() is very slow. |
31ddd871 | 3024 | * |
d185af30 | 3025 | * Return: |
31ddd871 | 3026 | * 0 on success, -errno on failure. |
b6136773 | 3027 | */ |
65f27f38 | 3028 | int schedule_on_each_cpu(work_func_t func) |
15316ba8 CL |
3029 | { |
3030 | int cpu; | |
38f51568 | 3031 | struct work_struct __percpu *works; |
15316ba8 | 3032 | |
b6136773 AM |
3033 | works = alloc_percpu(struct work_struct); |
3034 | if (!works) | |
15316ba8 | 3035 | return -ENOMEM; |
b6136773 | 3036 | |
93981800 TH |
3037 | get_online_cpus(); |
3038 | ||
15316ba8 | 3039 | for_each_online_cpu(cpu) { |
9bfb1839 IM |
3040 | struct work_struct *work = per_cpu_ptr(works, cpu); |
3041 | ||
3042 | INIT_WORK(work, func); | |
b71ab8c2 | 3043 | schedule_work_on(cpu, work); |
65a64464 | 3044 | } |
93981800 TH |
3045 | |
3046 | for_each_online_cpu(cpu) | |
3047 | flush_work(per_cpu_ptr(works, cpu)); | |
3048 | ||
95402b38 | 3049 | put_online_cpus(); |
b6136773 | 3050 | free_percpu(works); |
15316ba8 CL |
3051 | return 0; |
3052 | } | |
3053 | ||
eef6a7d5 AS |
3054 | /** |
3055 | * flush_scheduled_work - ensure that any scheduled work has run to completion. | |
3056 | * | |
3057 | * Forces execution of the kernel-global workqueue and blocks until its | |
3058 | * completion. | |
3059 | * | |
3060 | * Think twice before calling this function! It's very easy to get into | |
3061 | * trouble if you don't take great care. Either of the following situations | |
3062 | * will lead to deadlock: | |
3063 | * | |
3064 | * One of the work items currently on the workqueue needs to acquire | |
3065 | * a lock held by your code or its caller. | |
3066 | * | |
3067 | * Your code is running in the context of a work routine. | |
3068 | * | |
3069 | * They will be detected by lockdep when they occur, but the first might not | |
3070 | * occur very often. It depends on what work items are on the workqueue and | |
3071 | * what locks they need, which you have no control over. | |
3072 | * | |
3073 | * In most situations flushing the entire workqueue is overkill; you merely | |
3074 | * need to know that a particular work item isn't queued and isn't running. | |
3075 | * In such cases you should use cancel_delayed_work_sync() or | |
3076 | * cancel_work_sync() instead. | |
3077 | */ | |
1da177e4 LT |
3078 | void flush_scheduled_work(void) |
3079 | { | |
d320c038 | 3080 | flush_workqueue(system_wq); |
1da177e4 | 3081 | } |
ae90dd5d | 3082 | EXPORT_SYMBOL(flush_scheduled_work); |
1da177e4 | 3083 | |
1fa44eca JB |
3084 | /** |
3085 | * execute_in_process_context - reliably execute the routine with user context | |
3086 | * @fn: the function to execute | |
1fa44eca JB |
3087 | * @ew: guaranteed storage for the execute work structure (must |
3088 | * be available when the work executes) | |
3089 | * | |
3090 | * Executes the function immediately if process context is available, | |
3091 | * otherwise schedules the function for delayed execution. | |
3092 | * | |
d185af30 | 3093 | * Return: 0 - function was executed |
1fa44eca JB |
3094 | * 1 - function was scheduled for execution |
3095 | */ | |
65f27f38 | 3096 | int execute_in_process_context(work_func_t fn, struct execute_work *ew) |
1fa44eca JB |
3097 | { |
3098 | if (!in_interrupt()) { | |
65f27f38 | 3099 | fn(&ew->work); |
1fa44eca JB |
3100 | return 0; |
3101 | } | |
3102 | ||
65f27f38 | 3103 | INIT_WORK(&ew->work, fn); |
1fa44eca JB |
3104 | schedule_work(&ew->work); |
3105 | ||
3106 | return 1; | |
3107 | } | |
3108 | EXPORT_SYMBOL_GPL(execute_in_process_context); | |
3109 | ||
226223ab TH |
3110 | #ifdef CONFIG_SYSFS |
3111 | /* | |
3112 | * Workqueues with WQ_SYSFS flag set is visible to userland via | |
3113 | * /sys/bus/workqueue/devices/WQ_NAME. All visible workqueues have the | |
3114 | * following attributes. | |
3115 | * | |
3116 | * per_cpu RO bool : whether the workqueue is per-cpu or unbound | |
3117 | * max_active RW int : maximum number of in-flight work items | |
3118 | * | |
3119 | * Unbound workqueues have the following extra attributes. | |
3120 | * | |
3121 | * id RO int : the associated pool ID | |
3122 | * nice RW int : nice value of the workers | |
3123 | * cpumask RW mask : bitmask of allowed CPUs for the workers | |
3124 | */ | |
3125 | struct wq_device { | |
3126 | struct workqueue_struct *wq; | |
3127 | struct device dev; | |
3128 | }; | |
3129 | ||
3130 | static struct workqueue_struct *dev_to_wq(struct device *dev) | |
3131 | { | |
3132 | struct wq_device *wq_dev = container_of(dev, struct wq_device, dev); | |
3133 | ||
3134 | return wq_dev->wq; | |
3135 | } | |
3136 | ||
1a6661da GKH |
3137 | static ssize_t per_cpu_show(struct device *dev, struct device_attribute *attr, |
3138 | char *buf) | |
226223ab TH |
3139 | { |
3140 | struct workqueue_struct *wq = dev_to_wq(dev); | |
3141 | ||
3142 | return scnprintf(buf, PAGE_SIZE, "%d\n", (bool)!(wq->flags & WQ_UNBOUND)); | |
3143 | } | |
1a6661da | 3144 | static DEVICE_ATTR_RO(per_cpu); |
226223ab | 3145 | |
1a6661da GKH |
3146 | static ssize_t max_active_show(struct device *dev, |
3147 | struct device_attribute *attr, char *buf) | |
226223ab TH |
3148 | { |
3149 | struct workqueue_struct *wq = dev_to_wq(dev); | |
3150 | ||
3151 | return scnprintf(buf, PAGE_SIZE, "%d\n", wq->saved_max_active); | |
3152 | } | |
3153 | ||
1a6661da GKH |
3154 | static ssize_t max_active_store(struct device *dev, |
3155 | struct device_attribute *attr, const char *buf, | |
3156 | size_t count) | |
226223ab TH |
3157 | { |
3158 | struct workqueue_struct *wq = dev_to_wq(dev); | |
3159 | int val; | |
3160 | ||
3161 | if (sscanf(buf, "%d", &val) != 1 || val <= 0) | |
3162 | return -EINVAL; | |
3163 | ||
3164 | workqueue_set_max_active(wq, val); | |
3165 | return count; | |
3166 | } | |
1a6661da | 3167 | static DEVICE_ATTR_RW(max_active); |
226223ab | 3168 | |
1a6661da GKH |
3169 | static struct attribute *wq_sysfs_attrs[] = { |
3170 | &dev_attr_per_cpu.attr, | |
3171 | &dev_attr_max_active.attr, | |
3172 | NULL, | |
226223ab | 3173 | }; |
1a6661da | 3174 | ATTRIBUTE_GROUPS(wq_sysfs); |
226223ab | 3175 | |
d55262c4 TH |
3176 | static ssize_t wq_pool_ids_show(struct device *dev, |
3177 | struct device_attribute *attr, char *buf) | |
226223ab TH |
3178 | { |
3179 | struct workqueue_struct *wq = dev_to_wq(dev); | |
d55262c4 TH |
3180 | const char *delim = ""; |
3181 | int node, written = 0; | |
226223ab TH |
3182 | |
3183 | rcu_read_lock_sched(); | |
d55262c4 TH |
3184 | for_each_node(node) { |
3185 | written += scnprintf(buf + written, PAGE_SIZE - written, | |
3186 | "%s%d:%d", delim, node, | |
3187 | unbound_pwq_by_node(wq, node)->pool->id); | |
3188 | delim = " "; | |
3189 | } | |
3190 | written += scnprintf(buf + written, PAGE_SIZE - written, "\n"); | |
226223ab TH |
3191 | rcu_read_unlock_sched(); |
3192 | ||
3193 | return written; | |
3194 | } | |
3195 | ||
3196 | static ssize_t wq_nice_show(struct device *dev, struct device_attribute *attr, | |
3197 | char *buf) | |
3198 | { | |
3199 | struct workqueue_struct *wq = dev_to_wq(dev); | |
3200 | int written; | |
3201 | ||
6029a918 TH |
3202 | mutex_lock(&wq->mutex); |
3203 | written = scnprintf(buf, PAGE_SIZE, "%d\n", wq->unbound_attrs->nice); | |
3204 | mutex_unlock(&wq->mutex); | |
226223ab TH |
3205 | |
3206 | return written; | |
3207 | } | |
3208 | ||
3209 | /* prepare workqueue_attrs for sysfs store operations */ | |
3210 | static struct workqueue_attrs *wq_sysfs_prep_attrs(struct workqueue_struct *wq) | |
3211 | { | |
3212 | struct workqueue_attrs *attrs; | |
3213 | ||
3214 | attrs = alloc_workqueue_attrs(GFP_KERNEL); | |
3215 | if (!attrs) | |
3216 | return NULL; | |
3217 | ||
6029a918 TH |
3218 | mutex_lock(&wq->mutex); |
3219 | copy_workqueue_attrs(attrs, wq->unbound_attrs); | |
3220 | mutex_unlock(&wq->mutex); | |
226223ab TH |
3221 | return attrs; |
3222 | } | |
3223 | ||
3224 | static ssize_t wq_nice_store(struct device *dev, struct device_attribute *attr, | |
3225 | const char *buf, size_t count) | |
3226 | { | |
3227 | struct workqueue_struct *wq = dev_to_wq(dev); | |
3228 | struct workqueue_attrs *attrs; | |
3229 | int ret; | |
3230 | ||
3231 | attrs = wq_sysfs_prep_attrs(wq); | |
3232 | if (!attrs) | |
3233 | return -ENOMEM; | |
3234 | ||
3235 | if (sscanf(buf, "%d", &attrs->nice) == 1 && | |
14481842 | 3236 | attrs->nice >= MIN_NICE && attrs->nice <= MAX_NICE) |
226223ab TH |
3237 | ret = apply_workqueue_attrs(wq, attrs); |
3238 | else | |
3239 | ret = -EINVAL; | |
3240 | ||
3241 | free_workqueue_attrs(attrs); | |
3242 | return ret ?: count; | |
3243 | } | |
3244 | ||
3245 | static ssize_t wq_cpumask_show(struct device *dev, | |
3246 | struct device_attribute *attr, char *buf) | |
3247 | { | |
3248 | struct workqueue_struct *wq = dev_to_wq(dev); | |
3249 | int written; | |
3250 | ||
6029a918 TH |
3251 | mutex_lock(&wq->mutex); |
3252 | written = cpumask_scnprintf(buf, PAGE_SIZE, wq->unbound_attrs->cpumask); | |
3253 | mutex_unlock(&wq->mutex); | |
226223ab TH |
3254 | |
3255 | written += scnprintf(buf + written, PAGE_SIZE - written, "\n"); | |
3256 | return written; | |
3257 | } | |
3258 | ||
3259 | static ssize_t wq_cpumask_store(struct device *dev, | |
3260 | struct device_attribute *attr, | |
3261 | const char *buf, size_t count) | |
3262 | { | |
3263 | struct workqueue_struct *wq = dev_to_wq(dev); | |
3264 | struct workqueue_attrs *attrs; | |
3265 | int ret; | |
3266 | ||
3267 | attrs = wq_sysfs_prep_attrs(wq); | |
3268 | if (!attrs) | |
3269 | return -ENOMEM; | |
3270 | ||
3271 | ret = cpumask_parse(buf, attrs->cpumask); | |
3272 | if (!ret) | |
3273 | ret = apply_workqueue_attrs(wq, attrs); | |
3274 | ||
3275 | free_workqueue_attrs(attrs); | |
3276 | return ret ?: count; | |
3277 | } | |
3278 | ||
d55262c4 TH |
3279 | static ssize_t wq_numa_show(struct device *dev, struct device_attribute *attr, |
3280 | char *buf) | |
3281 | { | |
3282 | struct workqueue_struct *wq = dev_to_wq(dev); | |
3283 | int written; | |
3284 | ||
3285 | mutex_lock(&wq->mutex); | |
3286 | written = scnprintf(buf, PAGE_SIZE, "%d\n", | |
3287 | !wq->unbound_attrs->no_numa); | |
3288 | mutex_unlock(&wq->mutex); | |
3289 | ||
3290 | return written; | |
3291 | } | |
3292 | ||
3293 | static ssize_t wq_numa_store(struct device *dev, struct device_attribute *attr, | |
3294 | const char *buf, size_t count) | |
3295 | { | |
3296 | struct workqueue_struct *wq = dev_to_wq(dev); | |
3297 | struct workqueue_attrs *attrs; | |
3298 | int v, ret; | |
3299 | ||
3300 | attrs = wq_sysfs_prep_attrs(wq); | |
3301 | if (!attrs) | |
3302 | return -ENOMEM; | |
3303 | ||
3304 | ret = -EINVAL; | |
3305 | if (sscanf(buf, "%d", &v) == 1) { | |
3306 | attrs->no_numa = !v; | |
3307 | ret = apply_workqueue_attrs(wq, attrs); | |
3308 | } | |
3309 | ||
3310 | free_workqueue_attrs(attrs); | |
3311 | return ret ?: count; | |
3312 | } | |
3313 | ||
226223ab | 3314 | static struct device_attribute wq_sysfs_unbound_attrs[] = { |
d55262c4 | 3315 | __ATTR(pool_ids, 0444, wq_pool_ids_show, NULL), |
226223ab TH |
3316 | __ATTR(nice, 0644, wq_nice_show, wq_nice_store), |
3317 | __ATTR(cpumask, 0644, wq_cpumask_show, wq_cpumask_store), | |
d55262c4 | 3318 | __ATTR(numa, 0644, wq_numa_show, wq_numa_store), |
226223ab TH |
3319 | __ATTR_NULL, |
3320 | }; | |
3321 | ||
3322 | static struct bus_type wq_subsys = { | |
3323 | .name = "workqueue", | |
1a6661da | 3324 | .dev_groups = wq_sysfs_groups, |
226223ab TH |
3325 | }; |
3326 | ||
3327 | static int __init wq_sysfs_init(void) | |
3328 | { | |
3329 | return subsys_virtual_register(&wq_subsys, NULL); | |
3330 | } | |
3331 | core_initcall(wq_sysfs_init); | |
3332 | ||
3333 | static void wq_device_release(struct device *dev) | |
3334 | { | |
3335 | struct wq_device *wq_dev = container_of(dev, struct wq_device, dev); | |
3336 | ||
3337 | kfree(wq_dev); | |
3338 | } | |
3339 | ||
3340 | /** | |
3341 | * workqueue_sysfs_register - make a workqueue visible in sysfs | |
3342 | * @wq: the workqueue to register | |
3343 | * | |
3344 | * Expose @wq in sysfs under /sys/bus/workqueue/devices. | |
3345 | * alloc_workqueue*() automatically calls this function if WQ_SYSFS is set | |
3346 | * which is the preferred method. | |
3347 | * | |
3348 | * Workqueue user should use this function directly iff it wants to apply | |
3349 | * workqueue_attrs before making the workqueue visible in sysfs; otherwise, | |
3350 | * apply_workqueue_attrs() may race against userland updating the | |
3351 | * attributes. | |
3352 | * | |
d185af30 | 3353 | * Return: 0 on success, -errno on failure. |
226223ab TH |
3354 | */ |
3355 | int workqueue_sysfs_register(struct workqueue_struct *wq) | |
3356 | { | |
3357 | struct wq_device *wq_dev; | |
3358 | int ret; | |
3359 | ||
3360 | /* | |
3361 | * Adjusting max_active or creating new pwqs by applyting | |
3362 | * attributes breaks ordering guarantee. Disallow exposing ordered | |
3363 | * workqueues. | |
3364 | */ | |
3365 | if (WARN_ON(wq->flags & __WQ_ORDERED)) | |
3366 | return -EINVAL; | |
3367 | ||
3368 | wq->wq_dev = wq_dev = kzalloc(sizeof(*wq_dev), GFP_KERNEL); | |
3369 | if (!wq_dev) | |
3370 | return -ENOMEM; | |
3371 | ||
3372 | wq_dev->wq = wq; | |
3373 | wq_dev->dev.bus = &wq_subsys; | |
3374 | wq_dev->dev.init_name = wq->name; | |
3375 | wq_dev->dev.release = wq_device_release; | |
3376 | ||
3377 | /* | |
3378 | * unbound_attrs are created separately. Suppress uevent until | |
3379 | * everything is ready. | |
3380 | */ | |
3381 | dev_set_uevent_suppress(&wq_dev->dev, true); | |
3382 | ||
3383 | ret = device_register(&wq_dev->dev); | |
3384 | if (ret) { | |
3385 | kfree(wq_dev); | |
3386 | wq->wq_dev = NULL; | |
3387 | return ret; | |
3388 | } | |
3389 | ||
3390 | if (wq->flags & WQ_UNBOUND) { | |
3391 | struct device_attribute *attr; | |
3392 | ||
3393 | for (attr = wq_sysfs_unbound_attrs; attr->attr.name; attr++) { | |
3394 | ret = device_create_file(&wq_dev->dev, attr); | |
3395 | if (ret) { | |
3396 | device_unregister(&wq_dev->dev); | |
3397 | wq->wq_dev = NULL; | |
3398 | return ret; | |
3399 | } | |
3400 | } | |
3401 | } | |
3402 | ||
3403 | kobject_uevent(&wq_dev->dev.kobj, KOBJ_ADD); | |
3404 | return 0; | |
3405 | } | |
3406 | ||
3407 | /** | |
3408 | * workqueue_sysfs_unregister - undo workqueue_sysfs_register() | |
3409 | * @wq: the workqueue to unregister | |
3410 | * | |
3411 | * If @wq is registered to sysfs by workqueue_sysfs_register(), unregister. | |
3412 | */ | |
3413 | static void workqueue_sysfs_unregister(struct workqueue_struct *wq) | |
3414 | { | |
3415 | struct wq_device *wq_dev = wq->wq_dev; | |
3416 | ||
3417 | if (!wq->wq_dev) | |
3418 | return; | |
3419 | ||
3420 | wq->wq_dev = NULL; | |
3421 | device_unregister(&wq_dev->dev); | |
3422 | } | |
3423 | #else /* CONFIG_SYSFS */ | |
3424 | static void workqueue_sysfs_unregister(struct workqueue_struct *wq) { } | |
3425 | #endif /* CONFIG_SYSFS */ | |
3426 | ||
7a4e344c TH |
3427 | /** |
3428 | * free_workqueue_attrs - free a workqueue_attrs | |
3429 | * @attrs: workqueue_attrs to free | |
3430 | * | |
3431 | * Undo alloc_workqueue_attrs(). | |
3432 | */ | |
3433 | void free_workqueue_attrs(struct workqueue_attrs *attrs) | |
3434 | { | |
3435 | if (attrs) { | |
3436 | free_cpumask_var(attrs->cpumask); | |
3437 | kfree(attrs); | |
3438 | } | |
3439 | } | |
3440 | ||
3441 | /** | |
3442 | * alloc_workqueue_attrs - allocate a workqueue_attrs | |
3443 | * @gfp_mask: allocation mask to use | |
3444 | * | |
3445 | * Allocate a new workqueue_attrs, initialize with default settings and | |
d185af30 YB |
3446 | * return it. |
3447 | * | |
3448 | * Return: The allocated new workqueue_attr on success. %NULL on failure. | |
7a4e344c TH |
3449 | */ |
3450 | struct workqueue_attrs *alloc_workqueue_attrs(gfp_t gfp_mask) | |
3451 | { | |
3452 | struct workqueue_attrs *attrs; | |
3453 | ||
3454 | attrs = kzalloc(sizeof(*attrs), gfp_mask); | |
3455 | if (!attrs) | |
3456 | goto fail; | |
3457 | if (!alloc_cpumask_var(&attrs->cpumask, gfp_mask)) | |
3458 | goto fail; | |
3459 | ||
13e2e556 | 3460 | cpumask_copy(attrs->cpumask, cpu_possible_mask); |
7a4e344c TH |
3461 | return attrs; |
3462 | fail: | |
3463 | free_workqueue_attrs(attrs); | |
3464 | return NULL; | |
3465 | } | |
3466 | ||
29c91e99 TH |
3467 | static void copy_workqueue_attrs(struct workqueue_attrs *to, |
3468 | const struct workqueue_attrs *from) | |
3469 | { | |
3470 | to->nice = from->nice; | |
3471 | cpumask_copy(to->cpumask, from->cpumask); | |
2865a8fb SL |
3472 | /* |
3473 | * Unlike hash and equality test, this function doesn't ignore | |
3474 | * ->no_numa as it is used for both pool and wq attrs. Instead, | |
3475 | * get_unbound_pool() explicitly clears ->no_numa after copying. | |
3476 | */ | |
3477 | to->no_numa = from->no_numa; | |
29c91e99 TH |
3478 | } |
3479 | ||
29c91e99 TH |
3480 | /* hash value of the content of @attr */ |
3481 | static u32 wqattrs_hash(const struct workqueue_attrs *attrs) | |
3482 | { | |
3483 | u32 hash = 0; | |
3484 | ||
3485 | hash = jhash_1word(attrs->nice, hash); | |
13e2e556 TH |
3486 | hash = jhash(cpumask_bits(attrs->cpumask), |
3487 | BITS_TO_LONGS(nr_cpumask_bits) * sizeof(long), hash); | |
29c91e99 TH |
3488 | return hash; |
3489 | } | |
3490 | ||
3491 | /* content equality test */ | |
3492 | static bool wqattrs_equal(const struct workqueue_attrs *a, | |
3493 | const struct workqueue_attrs *b) | |
3494 | { | |
3495 | if (a->nice != b->nice) | |
3496 | return false; | |
3497 | if (!cpumask_equal(a->cpumask, b->cpumask)) | |
3498 | return false; | |
3499 | return true; | |
3500 | } | |
3501 | ||
7a4e344c TH |
3502 | /** |
3503 | * init_worker_pool - initialize a newly zalloc'd worker_pool | |
3504 | * @pool: worker_pool to initialize | |
3505 | * | |
3506 | * Initiailize a newly zalloc'd @pool. It also allocates @pool->attrs. | |
d185af30 YB |
3507 | * |
3508 | * Return: 0 on success, -errno on failure. Even on failure, all fields | |
29c91e99 TH |
3509 | * inside @pool proper are initialized and put_unbound_pool() can be called |
3510 | * on @pool safely to release it. | |
7a4e344c TH |
3511 | */ |
3512 | static int init_worker_pool(struct worker_pool *pool) | |
4e1a1f9a TH |
3513 | { |
3514 | spin_lock_init(&pool->lock); | |
29c91e99 TH |
3515 | pool->id = -1; |
3516 | pool->cpu = -1; | |
f3f90ad4 | 3517 | pool->node = NUMA_NO_NODE; |
4e1a1f9a TH |
3518 | pool->flags |= POOL_DISASSOCIATED; |
3519 | INIT_LIST_HEAD(&pool->worklist); | |
3520 | INIT_LIST_HEAD(&pool->idle_list); | |
3521 | hash_init(pool->busy_hash); | |
3522 | ||
3523 | init_timer_deferrable(&pool->idle_timer); | |
3524 | pool->idle_timer.function = idle_worker_timeout; | |
3525 | pool->idle_timer.data = (unsigned long)pool; | |
3526 | ||
3527 | setup_timer(&pool->mayday_timer, pool_mayday_timeout, | |
3528 | (unsigned long)pool); | |
3529 | ||
3530 | mutex_init(&pool->manager_arb); | |
bc3a1afc | 3531 | mutex_init(&pool->manager_mutex); |
822d8405 | 3532 | idr_init(&pool->worker_idr); |
7a4e344c | 3533 | |
29c91e99 TH |
3534 | INIT_HLIST_NODE(&pool->hash_node); |
3535 | pool->refcnt = 1; | |
3536 | ||
3537 | /* shouldn't fail above this point */ | |
7a4e344c TH |
3538 | pool->attrs = alloc_workqueue_attrs(GFP_KERNEL); |
3539 | if (!pool->attrs) | |
3540 | return -ENOMEM; | |
3541 | return 0; | |
4e1a1f9a TH |
3542 | } |
3543 | ||
29c91e99 TH |
3544 | static void rcu_free_pool(struct rcu_head *rcu) |
3545 | { | |
3546 | struct worker_pool *pool = container_of(rcu, struct worker_pool, rcu); | |
3547 | ||
822d8405 | 3548 | idr_destroy(&pool->worker_idr); |
29c91e99 TH |
3549 | free_workqueue_attrs(pool->attrs); |
3550 | kfree(pool); | |
3551 | } | |
3552 | ||
3553 | /** | |
3554 | * put_unbound_pool - put a worker_pool | |
3555 | * @pool: worker_pool to put | |
3556 | * | |
3557 | * Put @pool. If its refcnt reaches zero, it gets destroyed in sched-RCU | |
c5aa87bb TH |
3558 | * safe manner. get_unbound_pool() calls this function on its failure path |
3559 | * and this function should be able to release pools which went through, | |
3560 | * successfully or not, init_worker_pool(). | |
a892cacc TH |
3561 | * |
3562 | * Should be called with wq_pool_mutex held. | |
29c91e99 TH |
3563 | */ |
3564 | static void put_unbound_pool(struct worker_pool *pool) | |
3565 | { | |
3566 | struct worker *worker; | |
3567 | ||
a892cacc TH |
3568 | lockdep_assert_held(&wq_pool_mutex); |
3569 | ||
3570 | if (--pool->refcnt) | |
29c91e99 | 3571 | return; |
29c91e99 TH |
3572 | |
3573 | /* sanity checks */ | |
3574 | if (WARN_ON(!(pool->flags & POOL_DISASSOCIATED)) || | |
a892cacc | 3575 | WARN_ON(!list_empty(&pool->worklist))) |
29c91e99 | 3576 | return; |
29c91e99 TH |
3577 | |
3578 | /* release id and unhash */ | |
3579 | if (pool->id >= 0) | |
3580 | idr_remove(&worker_pool_idr, pool->id); | |
3581 | hash_del(&pool->hash_node); | |
3582 | ||
c5aa87bb TH |
3583 | /* |
3584 | * Become the manager and destroy all workers. Grabbing | |
3585 | * manager_arb prevents @pool's workers from blocking on | |
3586 | * manager_mutex. | |
3587 | */ | |
29c91e99 | 3588 | mutex_lock(&pool->manager_arb); |
cd549687 | 3589 | mutex_lock(&pool->manager_mutex); |
29c91e99 TH |
3590 | spin_lock_irq(&pool->lock); |
3591 | ||
3592 | while ((worker = first_worker(pool))) | |
3593 | destroy_worker(worker); | |
3594 | WARN_ON(pool->nr_workers || pool->nr_idle); | |
3595 | ||
3596 | spin_unlock_irq(&pool->lock); | |
cd549687 | 3597 | mutex_unlock(&pool->manager_mutex); |
29c91e99 TH |
3598 | mutex_unlock(&pool->manager_arb); |
3599 | ||
3600 | /* shut down the timers */ | |
3601 | del_timer_sync(&pool->idle_timer); | |
3602 | del_timer_sync(&pool->mayday_timer); | |
3603 | ||
3604 | /* sched-RCU protected to allow dereferences from get_work_pool() */ | |
3605 | call_rcu_sched(&pool->rcu, rcu_free_pool); | |
3606 | } | |
3607 | ||
3608 | /** | |
3609 | * get_unbound_pool - get a worker_pool with the specified attributes | |
3610 | * @attrs: the attributes of the worker_pool to get | |
3611 | * | |
3612 | * Obtain a worker_pool which has the same attributes as @attrs, bump the | |
3613 | * reference count and return it. If there already is a matching | |
3614 | * worker_pool, it will be used; otherwise, this function attempts to | |
d185af30 | 3615 | * create a new one. |
a892cacc TH |
3616 | * |
3617 | * Should be called with wq_pool_mutex held. | |
d185af30 YB |
3618 | * |
3619 | * Return: On success, a worker_pool with the same attributes as @attrs. | |
3620 | * On failure, %NULL. | |
29c91e99 TH |
3621 | */ |
3622 | static struct worker_pool *get_unbound_pool(const struct workqueue_attrs *attrs) | |
3623 | { | |
29c91e99 TH |
3624 | u32 hash = wqattrs_hash(attrs); |
3625 | struct worker_pool *pool; | |
f3f90ad4 | 3626 | int node; |
29c91e99 | 3627 | |
a892cacc | 3628 | lockdep_assert_held(&wq_pool_mutex); |
29c91e99 TH |
3629 | |
3630 | /* do we already have a matching pool? */ | |
29c91e99 TH |
3631 | hash_for_each_possible(unbound_pool_hash, pool, hash_node, hash) { |
3632 | if (wqattrs_equal(pool->attrs, attrs)) { | |
3633 | pool->refcnt++; | |
3634 | goto out_unlock; | |
3635 | } | |
3636 | } | |
29c91e99 TH |
3637 | |
3638 | /* nope, create a new one */ | |
3639 | pool = kzalloc(sizeof(*pool), GFP_KERNEL); | |
3640 | if (!pool || init_worker_pool(pool) < 0) | |
3641 | goto fail; | |
3642 | ||
12ee4fc6 LJ |
3643 | if (workqueue_freezing) |
3644 | pool->flags |= POOL_FREEZING; | |
3645 | ||
8864b4e5 | 3646 | lockdep_set_subclass(&pool->lock, 1); /* see put_pwq() */ |
29c91e99 TH |
3647 | copy_workqueue_attrs(pool->attrs, attrs); |
3648 | ||
2865a8fb SL |
3649 | /* |
3650 | * no_numa isn't a worker_pool attribute, always clear it. See | |
3651 | * 'struct workqueue_attrs' comments for detail. | |
3652 | */ | |
3653 | pool->attrs->no_numa = false; | |
3654 | ||
f3f90ad4 TH |
3655 | /* if cpumask is contained inside a NUMA node, we belong to that node */ |
3656 | if (wq_numa_enabled) { | |
3657 | for_each_node(node) { | |
3658 | if (cpumask_subset(pool->attrs->cpumask, | |
3659 | wq_numa_possible_cpumask[node])) { | |
3660 | pool->node = node; | |
3661 | break; | |
3662 | } | |
3663 | } | |
3664 | } | |
3665 | ||
29c91e99 TH |
3666 | if (worker_pool_assign_id(pool) < 0) |
3667 | goto fail; | |
3668 | ||
3669 | /* create and start the initial worker */ | |
ebf44d16 | 3670 | if (create_and_start_worker(pool) < 0) |
29c91e99 TH |
3671 | goto fail; |
3672 | ||
29c91e99 | 3673 | /* install */ |
29c91e99 TH |
3674 | hash_add(unbound_pool_hash, &pool->hash_node, hash); |
3675 | out_unlock: | |
29c91e99 TH |
3676 | return pool; |
3677 | fail: | |
29c91e99 TH |
3678 | if (pool) |
3679 | put_unbound_pool(pool); | |
3680 | return NULL; | |
3681 | } | |
3682 | ||
8864b4e5 TH |
3683 | static void rcu_free_pwq(struct rcu_head *rcu) |
3684 | { | |
3685 | kmem_cache_free(pwq_cache, | |
3686 | container_of(rcu, struct pool_workqueue, rcu)); | |
3687 | } | |
3688 | ||
3689 | /* | |
3690 | * Scheduled on system_wq by put_pwq() when an unbound pwq hits zero refcnt | |
3691 | * and needs to be destroyed. | |
3692 | */ | |
3693 | static void pwq_unbound_release_workfn(struct work_struct *work) | |
3694 | { | |
3695 | struct pool_workqueue *pwq = container_of(work, struct pool_workqueue, | |
3696 | unbound_release_work); | |
3697 | struct workqueue_struct *wq = pwq->wq; | |
3698 | struct worker_pool *pool = pwq->pool; | |
bc0caf09 | 3699 | bool is_last; |
8864b4e5 TH |
3700 | |
3701 | if (WARN_ON_ONCE(!(wq->flags & WQ_UNBOUND))) | |
3702 | return; | |
3703 | ||
75ccf595 | 3704 | /* |
3c25a55d | 3705 | * Unlink @pwq. Synchronization against wq->mutex isn't strictly |
75ccf595 TH |
3706 | * necessary on release but do it anyway. It's easier to verify |
3707 | * and consistent with the linking path. | |
3708 | */ | |
3c25a55d | 3709 | mutex_lock(&wq->mutex); |
8864b4e5 | 3710 | list_del_rcu(&pwq->pwqs_node); |
bc0caf09 | 3711 | is_last = list_empty(&wq->pwqs); |
3c25a55d | 3712 | mutex_unlock(&wq->mutex); |
8864b4e5 | 3713 | |
a892cacc | 3714 | mutex_lock(&wq_pool_mutex); |
8864b4e5 | 3715 | put_unbound_pool(pool); |
a892cacc TH |
3716 | mutex_unlock(&wq_pool_mutex); |
3717 | ||
8864b4e5 TH |
3718 | call_rcu_sched(&pwq->rcu, rcu_free_pwq); |
3719 | ||
3720 | /* | |
3721 | * If we're the last pwq going away, @wq is already dead and no one | |
3722 | * is gonna access it anymore. Free it. | |
3723 | */ | |
6029a918 TH |
3724 | if (is_last) { |
3725 | free_workqueue_attrs(wq->unbound_attrs); | |
8864b4e5 | 3726 | kfree(wq); |
6029a918 | 3727 | } |
8864b4e5 TH |
3728 | } |
3729 | ||
0fbd95aa | 3730 | /** |
699ce097 | 3731 | * pwq_adjust_max_active - update a pwq's max_active to the current setting |
0fbd95aa | 3732 | * @pwq: target pool_workqueue |
0fbd95aa | 3733 | * |
699ce097 TH |
3734 | * If @pwq isn't freezing, set @pwq->max_active to the associated |
3735 | * workqueue's saved_max_active and activate delayed work items | |
3736 | * accordingly. If @pwq is freezing, clear @pwq->max_active to zero. | |
0fbd95aa | 3737 | */ |
699ce097 | 3738 | static void pwq_adjust_max_active(struct pool_workqueue *pwq) |
0fbd95aa | 3739 | { |
699ce097 TH |
3740 | struct workqueue_struct *wq = pwq->wq; |
3741 | bool freezable = wq->flags & WQ_FREEZABLE; | |
3742 | ||
3743 | /* for @wq->saved_max_active */ | |
a357fc03 | 3744 | lockdep_assert_held(&wq->mutex); |
699ce097 TH |
3745 | |
3746 | /* fast exit for non-freezable wqs */ | |
3747 | if (!freezable && pwq->max_active == wq->saved_max_active) | |
3748 | return; | |
3749 | ||
a357fc03 | 3750 | spin_lock_irq(&pwq->pool->lock); |
699ce097 TH |
3751 | |
3752 | if (!freezable || !(pwq->pool->flags & POOL_FREEZING)) { | |
3753 | pwq->max_active = wq->saved_max_active; | |
0fbd95aa | 3754 | |
699ce097 TH |
3755 | while (!list_empty(&pwq->delayed_works) && |
3756 | pwq->nr_active < pwq->max_active) | |
3757 | pwq_activate_first_delayed(pwq); | |
951a078a LJ |
3758 | |
3759 | /* | |
3760 | * Need to kick a worker after thawed or an unbound wq's | |
3761 | * max_active is bumped. It's a slow path. Do it always. | |
3762 | */ | |
3763 | wake_up_worker(pwq->pool); | |
699ce097 TH |
3764 | } else { |
3765 | pwq->max_active = 0; | |
3766 | } | |
3767 | ||
a357fc03 | 3768 | spin_unlock_irq(&pwq->pool->lock); |
0fbd95aa TH |
3769 | } |
3770 | ||
e50aba9a | 3771 | /* initialize newly alloced @pwq which is associated with @wq and @pool */ |
f147f29e TH |
3772 | static void init_pwq(struct pool_workqueue *pwq, struct workqueue_struct *wq, |
3773 | struct worker_pool *pool) | |
d2c1d404 TH |
3774 | { |
3775 | BUG_ON((unsigned long)pwq & WORK_STRUCT_FLAG_MASK); | |
3776 | ||
e50aba9a TH |
3777 | memset(pwq, 0, sizeof(*pwq)); |
3778 | ||
d2c1d404 TH |
3779 | pwq->pool = pool; |
3780 | pwq->wq = wq; | |
3781 | pwq->flush_color = -1; | |
8864b4e5 | 3782 | pwq->refcnt = 1; |
d2c1d404 | 3783 | INIT_LIST_HEAD(&pwq->delayed_works); |
1befcf30 | 3784 | INIT_LIST_HEAD(&pwq->pwqs_node); |
d2c1d404 | 3785 | INIT_LIST_HEAD(&pwq->mayday_node); |
8864b4e5 | 3786 | INIT_WORK(&pwq->unbound_release_work, pwq_unbound_release_workfn); |
f147f29e | 3787 | } |
d2c1d404 | 3788 | |
f147f29e | 3789 | /* sync @pwq with the current state of its associated wq and link it */ |
1befcf30 | 3790 | static void link_pwq(struct pool_workqueue *pwq) |
f147f29e TH |
3791 | { |
3792 | struct workqueue_struct *wq = pwq->wq; | |
3793 | ||
3794 | lockdep_assert_held(&wq->mutex); | |
75ccf595 | 3795 | |
1befcf30 TH |
3796 | /* may be called multiple times, ignore if already linked */ |
3797 | if (!list_empty(&pwq->pwqs_node)) | |
3798 | return; | |
3799 | ||
983ca25e TH |
3800 | /* |
3801 | * Set the matching work_color. This is synchronized with | |
3c25a55d | 3802 | * wq->mutex to avoid confusing flush_workqueue(). |
983ca25e | 3803 | */ |
75ccf595 | 3804 | pwq->work_color = wq->work_color; |
983ca25e TH |
3805 | |
3806 | /* sync max_active to the current setting */ | |
3807 | pwq_adjust_max_active(pwq); | |
3808 | ||
3809 | /* link in @pwq */ | |
9e8cd2f5 | 3810 | list_add_rcu(&pwq->pwqs_node, &wq->pwqs); |
f147f29e | 3811 | } |
a357fc03 | 3812 | |
f147f29e TH |
3813 | /* obtain a pool matching @attr and create a pwq associating the pool and @wq */ |
3814 | static struct pool_workqueue *alloc_unbound_pwq(struct workqueue_struct *wq, | |
3815 | const struct workqueue_attrs *attrs) | |
3816 | { | |
3817 | struct worker_pool *pool; | |
3818 | struct pool_workqueue *pwq; | |
3819 | ||
3820 | lockdep_assert_held(&wq_pool_mutex); | |
3821 | ||
3822 | pool = get_unbound_pool(attrs); | |
3823 | if (!pool) | |
3824 | return NULL; | |
3825 | ||
e50aba9a | 3826 | pwq = kmem_cache_alloc_node(pwq_cache, GFP_KERNEL, pool->node); |
f147f29e TH |
3827 | if (!pwq) { |
3828 | put_unbound_pool(pool); | |
3829 | return NULL; | |
df2d5ae4 | 3830 | } |
6029a918 | 3831 | |
f147f29e TH |
3832 | init_pwq(pwq, wq, pool); |
3833 | return pwq; | |
d2c1d404 TH |
3834 | } |
3835 | ||
4c16bd32 TH |
3836 | /* undo alloc_unbound_pwq(), used only in the error path */ |
3837 | static void free_unbound_pwq(struct pool_workqueue *pwq) | |
3838 | { | |
3839 | lockdep_assert_held(&wq_pool_mutex); | |
3840 | ||
3841 | if (pwq) { | |
3842 | put_unbound_pool(pwq->pool); | |
cece95df | 3843 | kmem_cache_free(pwq_cache, pwq); |
4c16bd32 TH |
3844 | } |
3845 | } | |
3846 | ||
3847 | /** | |
3848 | * wq_calc_node_mask - calculate a wq_attrs' cpumask for the specified node | |
3849 | * @attrs: the wq_attrs of interest | |
3850 | * @node: the target NUMA node | |
3851 | * @cpu_going_down: if >= 0, the CPU to consider as offline | |
3852 | * @cpumask: outarg, the resulting cpumask | |
3853 | * | |
3854 | * Calculate the cpumask a workqueue with @attrs should use on @node. If | |
3855 | * @cpu_going_down is >= 0, that cpu is considered offline during | |
d185af30 | 3856 | * calculation. The result is stored in @cpumask. |
4c16bd32 TH |
3857 | * |
3858 | * If NUMA affinity is not enabled, @attrs->cpumask is always used. If | |
3859 | * enabled and @node has online CPUs requested by @attrs, the returned | |
3860 | * cpumask is the intersection of the possible CPUs of @node and | |
3861 | * @attrs->cpumask. | |
3862 | * | |
3863 | * The caller is responsible for ensuring that the cpumask of @node stays | |
3864 | * stable. | |
d185af30 YB |
3865 | * |
3866 | * Return: %true if the resulting @cpumask is different from @attrs->cpumask, | |
3867 | * %false if equal. | |
4c16bd32 TH |
3868 | */ |
3869 | static bool wq_calc_node_cpumask(const struct workqueue_attrs *attrs, int node, | |
3870 | int cpu_going_down, cpumask_t *cpumask) | |
3871 | { | |
d55262c4 | 3872 | if (!wq_numa_enabled || attrs->no_numa) |
4c16bd32 TH |
3873 | goto use_dfl; |
3874 | ||
3875 | /* does @node have any online CPUs @attrs wants? */ | |
3876 | cpumask_and(cpumask, cpumask_of_node(node), attrs->cpumask); | |
3877 | if (cpu_going_down >= 0) | |
3878 | cpumask_clear_cpu(cpu_going_down, cpumask); | |
3879 | ||
3880 | if (cpumask_empty(cpumask)) | |
3881 | goto use_dfl; | |
3882 | ||
3883 | /* yeap, return possible CPUs in @node that @attrs wants */ | |
3884 | cpumask_and(cpumask, attrs->cpumask, wq_numa_possible_cpumask[node]); | |
3885 | return !cpumask_equal(cpumask, attrs->cpumask); | |
3886 | ||
3887 | use_dfl: | |
3888 | cpumask_copy(cpumask, attrs->cpumask); | |
3889 | return false; | |
3890 | } | |
3891 | ||
1befcf30 TH |
3892 | /* install @pwq into @wq's numa_pwq_tbl[] for @node and return the old pwq */ |
3893 | static struct pool_workqueue *numa_pwq_tbl_install(struct workqueue_struct *wq, | |
3894 | int node, | |
3895 | struct pool_workqueue *pwq) | |
3896 | { | |
3897 | struct pool_workqueue *old_pwq; | |
3898 | ||
3899 | lockdep_assert_held(&wq->mutex); | |
3900 | ||
3901 | /* link_pwq() can handle duplicate calls */ | |
3902 | link_pwq(pwq); | |
3903 | ||
3904 | old_pwq = rcu_access_pointer(wq->numa_pwq_tbl[node]); | |
3905 | rcu_assign_pointer(wq->numa_pwq_tbl[node], pwq); | |
3906 | return old_pwq; | |
3907 | } | |
3908 | ||
9e8cd2f5 TH |
3909 | /** |
3910 | * apply_workqueue_attrs - apply new workqueue_attrs to an unbound workqueue | |
3911 | * @wq: the target workqueue | |
3912 | * @attrs: the workqueue_attrs to apply, allocated with alloc_workqueue_attrs() | |
3913 | * | |
4c16bd32 TH |
3914 | * Apply @attrs to an unbound workqueue @wq. Unless disabled, on NUMA |
3915 | * machines, this function maps a separate pwq to each NUMA node with | |
3916 | * possibles CPUs in @attrs->cpumask so that work items are affine to the | |
3917 | * NUMA node it was issued on. Older pwqs are released as in-flight work | |
3918 | * items finish. Note that a work item which repeatedly requeues itself | |
3919 | * back-to-back will stay on its current pwq. | |
9e8cd2f5 | 3920 | * |
d185af30 YB |
3921 | * Performs GFP_KERNEL allocations. |
3922 | * | |
3923 | * Return: 0 on success and -errno on failure. | |
9e8cd2f5 TH |
3924 | */ |
3925 | int apply_workqueue_attrs(struct workqueue_struct *wq, | |
3926 | const struct workqueue_attrs *attrs) | |
3927 | { | |
4c16bd32 TH |
3928 | struct workqueue_attrs *new_attrs, *tmp_attrs; |
3929 | struct pool_workqueue **pwq_tbl, *dfl_pwq; | |
f147f29e | 3930 | int node, ret; |
9e8cd2f5 | 3931 | |
8719dcea | 3932 | /* only unbound workqueues can change attributes */ |
9e8cd2f5 TH |
3933 | if (WARN_ON(!(wq->flags & WQ_UNBOUND))) |
3934 | return -EINVAL; | |
3935 | ||
8719dcea TH |
3936 | /* creating multiple pwqs breaks ordering guarantee */ |
3937 | if (WARN_ON((wq->flags & __WQ_ORDERED) && !list_empty(&wq->pwqs))) | |
3938 | return -EINVAL; | |
3939 | ||
4c16bd32 | 3940 | pwq_tbl = kzalloc(wq_numa_tbl_len * sizeof(pwq_tbl[0]), GFP_KERNEL); |
13e2e556 | 3941 | new_attrs = alloc_workqueue_attrs(GFP_KERNEL); |
4c16bd32 TH |
3942 | tmp_attrs = alloc_workqueue_attrs(GFP_KERNEL); |
3943 | if (!pwq_tbl || !new_attrs || !tmp_attrs) | |
13e2e556 TH |
3944 | goto enomem; |
3945 | ||
4c16bd32 | 3946 | /* make a copy of @attrs and sanitize it */ |
13e2e556 TH |
3947 | copy_workqueue_attrs(new_attrs, attrs); |
3948 | cpumask_and(new_attrs->cpumask, new_attrs->cpumask, cpu_possible_mask); | |
3949 | ||
4c16bd32 TH |
3950 | /* |
3951 | * We may create multiple pwqs with differing cpumasks. Make a | |
3952 | * copy of @new_attrs which will be modified and used to obtain | |
3953 | * pools. | |
3954 | */ | |
3955 | copy_workqueue_attrs(tmp_attrs, new_attrs); | |
3956 | ||
3957 | /* | |
3958 | * CPUs should stay stable across pwq creations and installations. | |
3959 | * Pin CPUs, determine the target cpumask for each node and create | |
3960 | * pwqs accordingly. | |
3961 | */ | |
3962 | get_online_cpus(); | |
3963 | ||
a892cacc | 3964 | mutex_lock(&wq_pool_mutex); |
4c16bd32 TH |
3965 | |
3966 | /* | |
3967 | * If something goes wrong during CPU up/down, we'll fall back to | |
3968 | * the default pwq covering whole @attrs->cpumask. Always create | |
3969 | * it even if we don't use it immediately. | |
3970 | */ | |
3971 | dfl_pwq = alloc_unbound_pwq(wq, new_attrs); | |
3972 | if (!dfl_pwq) | |
3973 | goto enomem_pwq; | |
3974 | ||
3975 | for_each_node(node) { | |
3976 | if (wq_calc_node_cpumask(attrs, node, -1, tmp_attrs->cpumask)) { | |
3977 | pwq_tbl[node] = alloc_unbound_pwq(wq, tmp_attrs); | |
3978 | if (!pwq_tbl[node]) | |
3979 | goto enomem_pwq; | |
3980 | } else { | |
3981 | dfl_pwq->refcnt++; | |
3982 | pwq_tbl[node] = dfl_pwq; | |
3983 | } | |
3984 | } | |
3985 | ||
f147f29e | 3986 | mutex_unlock(&wq_pool_mutex); |
9e8cd2f5 | 3987 | |
4c16bd32 | 3988 | /* all pwqs have been created successfully, let's install'em */ |
f147f29e | 3989 | mutex_lock(&wq->mutex); |
a892cacc | 3990 | |
f147f29e | 3991 | copy_workqueue_attrs(wq->unbound_attrs, new_attrs); |
4c16bd32 TH |
3992 | |
3993 | /* save the previous pwq and install the new one */ | |
f147f29e | 3994 | for_each_node(node) |
4c16bd32 TH |
3995 | pwq_tbl[node] = numa_pwq_tbl_install(wq, node, pwq_tbl[node]); |
3996 | ||
3997 | /* @dfl_pwq might not have been used, ensure it's linked */ | |
3998 | link_pwq(dfl_pwq); | |
3999 | swap(wq->dfl_pwq, dfl_pwq); | |
f147f29e TH |
4000 | |
4001 | mutex_unlock(&wq->mutex); | |
9e8cd2f5 | 4002 | |
4c16bd32 TH |
4003 | /* put the old pwqs */ |
4004 | for_each_node(node) | |
4005 | put_pwq_unlocked(pwq_tbl[node]); | |
4006 | put_pwq_unlocked(dfl_pwq); | |
4007 | ||
4008 | put_online_cpus(); | |
4862125b TH |
4009 | ret = 0; |
4010 | /* fall through */ | |
4011 | out_free: | |
4c16bd32 | 4012 | free_workqueue_attrs(tmp_attrs); |
4862125b | 4013 | free_workqueue_attrs(new_attrs); |
4c16bd32 | 4014 | kfree(pwq_tbl); |
4862125b | 4015 | return ret; |
13e2e556 | 4016 | |
4c16bd32 TH |
4017 | enomem_pwq: |
4018 | free_unbound_pwq(dfl_pwq); | |
4019 | for_each_node(node) | |
4020 | if (pwq_tbl && pwq_tbl[node] != dfl_pwq) | |
4021 | free_unbound_pwq(pwq_tbl[node]); | |
4022 | mutex_unlock(&wq_pool_mutex); | |
4023 | put_online_cpus(); | |
13e2e556 | 4024 | enomem: |
4862125b TH |
4025 | ret = -ENOMEM; |
4026 | goto out_free; | |
9e8cd2f5 TH |
4027 | } |
4028 | ||
4c16bd32 TH |
4029 | /** |
4030 | * wq_update_unbound_numa - update NUMA affinity of a wq for CPU hot[un]plug | |
4031 | * @wq: the target workqueue | |
4032 | * @cpu: the CPU coming up or going down | |
4033 | * @online: whether @cpu is coming up or going down | |
4034 | * | |
4035 | * This function is to be called from %CPU_DOWN_PREPARE, %CPU_ONLINE and | |
4036 | * %CPU_DOWN_FAILED. @cpu is being hot[un]plugged, update NUMA affinity of | |
4037 | * @wq accordingly. | |
4038 | * | |
4039 | * If NUMA affinity can't be adjusted due to memory allocation failure, it | |
4040 | * falls back to @wq->dfl_pwq which may not be optimal but is always | |
4041 | * correct. | |
4042 | * | |
4043 | * Note that when the last allowed CPU of a NUMA node goes offline for a | |
4044 | * workqueue with a cpumask spanning multiple nodes, the workers which were | |
4045 | * already executing the work items for the workqueue will lose their CPU | |
4046 | * affinity and may execute on any CPU. This is similar to how per-cpu | |
4047 | * workqueues behave on CPU_DOWN. If a workqueue user wants strict | |
4048 | * affinity, it's the user's responsibility to flush the work item from | |
4049 | * CPU_DOWN_PREPARE. | |
4050 | */ | |
4051 | static void wq_update_unbound_numa(struct workqueue_struct *wq, int cpu, | |
4052 | bool online) | |
4053 | { | |
4054 | int node = cpu_to_node(cpu); | |
4055 | int cpu_off = online ? -1 : cpu; | |
4056 | struct pool_workqueue *old_pwq = NULL, *pwq; | |
4057 | struct workqueue_attrs *target_attrs; | |
4058 | cpumask_t *cpumask; | |
4059 | ||
4060 | lockdep_assert_held(&wq_pool_mutex); | |
4061 | ||
4062 | if (!wq_numa_enabled || !(wq->flags & WQ_UNBOUND)) | |
4063 | return; | |
4064 | ||
4065 | /* | |
4066 | * We don't wanna alloc/free wq_attrs for each wq for each CPU. | |
4067 | * Let's use a preallocated one. The following buf is protected by | |
4068 | * CPU hotplug exclusion. | |
4069 | */ | |
4070 | target_attrs = wq_update_unbound_numa_attrs_buf; | |
4071 | cpumask = target_attrs->cpumask; | |
4072 | ||
4073 | mutex_lock(&wq->mutex); | |
d55262c4 TH |
4074 | if (wq->unbound_attrs->no_numa) |
4075 | goto out_unlock; | |
4c16bd32 TH |
4076 | |
4077 | copy_workqueue_attrs(target_attrs, wq->unbound_attrs); | |
4078 | pwq = unbound_pwq_by_node(wq, node); | |
4079 | ||
4080 | /* | |
4081 | * Let's determine what needs to be done. If the target cpumask is | |
4082 | * different from wq's, we need to compare it to @pwq's and create | |
4083 | * a new one if they don't match. If the target cpumask equals | |
534a3fbb | 4084 | * wq's, the default pwq should be used. |
4c16bd32 TH |
4085 | */ |
4086 | if (wq_calc_node_cpumask(wq->unbound_attrs, node, cpu_off, cpumask)) { | |
4087 | if (cpumask_equal(cpumask, pwq->pool->attrs->cpumask)) | |
4088 | goto out_unlock; | |
4089 | } else { | |
534a3fbb | 4090 | goto use_dfl_pwq; |
4c16bd32 TH |
4091 | } |
4092 | ||
4093 | mutex_unlock(&wq->mutex); | |
4094 | ||
4095 | /* create a new pwq */ | |
4096 | pwq = alloc_unbound_pwq(wq, target_attrs); | |
4097 | if (!pwq) { | |
2d916033 FF |
4098 | pr_warn("workqueue: allocation failed while updating NUMA affinity of \"%s\"\n", |
4099 | wq->name); | |
77f300b1 DY |
4100 | mutex_lock(&wq->mutex); |
4101 | goto use_dfl_pwq; | |
4c16bd32 TH |
4102 | } |
4103 | ||
4104 | /* | |
4105 | * Install the new pwq. As this function is called only from CPU | |
4106 | * hotplug callbacks and applying a new attrs is wrapped with | |
4107 | * get/put_online_cpus(), @wq->unbound_attrs couldn't have changed | |
4108 | * inbetween. | |
4109 | */ | |
4110 | mutex_lock(&wq->mutex); | |
4111 | old_pwq = numa_pwq_tbl_install(wq, node, pwq); | |
4112 | goto out_unlock; | |
4113 | ||
4114 | use_dfl_pwq: | |
4115 | spin_lock_irq(&wq->dfl_pwq->pool->lock); | |
4116 | get_pwq(wq->dfl_pwq); | |
4117 | spin_unlock_irq(&wq->dfl_pwq->pool->lock); | |
4118 | old_pwq = numa_pwq_tbl_install(wq, node, wq->dfl_pwq); | |
4119 | out_unlock: | |
4120 | mutex_unlock(&wq->mutex); | |
4121 | put_pwq_unlocked(old_pwq); | |
4122 | } | |
4123 | ||
30cdf249 | 4124 | static int alloc_and_link_pwqs(struct workqueue_struct *wq) |
0f900049 | 4125 | { |
49e3cf44 | 4126 | bool highpri = wq->flags & WQ_HIGHPRI; |
8a2b7538 | 4127 | int cpu, ret; |
30cdf249 TH |
4128 | |
4129 | if (!(wq->flags & WQ_UNBOUND)) { | |
420c0ddb TH |
4130 | wq->cpu_pwqs = alloc_percpu(struct pool_workqueue); |
4131 | if (!wq->cpu_pwqs) | |
30cdf249 TH |
4132 | return -ENOMEM; |
4133 | ||
4134 | for_each_possible_cpu(cpu) { | |
7fb98ea7 TH |
4135 | struct pool_workqueue *pwq = |
4136 | per_cpu_ptr(wq->cpu_pwqs, cpu); | |
7a62c2c8 | 4137 | struct worker_pool *cpu_pools = |
f02ae73a | 4138 | per_cpu(cpu_worker_pools, cpu); |
f3421797 | 4139 | |
f147f29e TH |
4140 | init_pwq(pwq, wq, &cpu_pools[highpri]); |
4141 | ||
4142 | mutex_lock(&wq->mutex); | |
1befcf30 | 4143 | link_pwq(pwq); |
f147f29e | 4144 | mutex_unlock(&wq->mutex); |
30cdf249 | 4145 | } |
9e8cd2f5 | 4146 | return 0; |
8a2b7538 TH |
4147 | } else if (wq->flags & __WQ_ORDERED) { |
4148 | ret = apply_workqueue_attrs(wq, ordered_wq_attrs[highpri]); | |
4149 | /* there should only be single pwq for ordering guarantee */ | |
4150 | WARN(!ret && (wq->pwqs.next != &wq->dfl_pwq->pwqs_node || | |
4151 | wq->pwqs.prev != &wq->dfl_pwq->pwqs_node), | |
4152 | "ordering guarantee broken for workqueue %s\n", wq->name); | |
4153 | return ret; | |
30cdf249 | 4154 | } else { |
9e8cd2f5 | 4155 | return apply_workqueue_attrs(wq, unbound_std_wq_attrs[highpri]); |
30cdf249 | 4156 | } |
0f900049 TH |
4157 | } |
4158 | ||
f3421797 TH |
4159 | static int wq_clamp_max_active(int max_active, unsigned int flags, |
4160 | const char *name) | |
b71ab8c2 | 4161 | { |
f3421797 TH |
4162 | int lim = flags & WQ_UNBOUND ? WQ_UNBOUND_MAX_ACTIVE : WQ_MAX_ACTIVE; |
4163 | ||
4164 | if (max_active < 1 || max_active > lim) | |
044c782c VI |
4165 | pr_warn("workqueue: max_active %d requested for %s is out of range, clamping between %d and %d\n", |
4166 | max_active, name, 1, lim); | |
b71ab8c2 | 4167 | |
f3421797 | 4168 | return clamp_val(max_active, 1, lim); |
b71ab8c2 TH |
4169 | } |
4170 | ||
b196be89 | 4171 | struct workqueue_struct *__alloc_workqueue_key(const char *fmt, |
d320c038 TH |
4172 | unsigned int flags, |
4173 | int max_active, | |
4174 | struct lock_class_key *key, | |
b196be89 | 4175 | const char *lock_name, ...) |
1da177e4 | 4176 | { |
df2d5ae4 | 4177 | size_t tbl_size = 0; |
ecf6881f | 4178 | va_list args; |
1da177e4 | 4179 | struct workqueue_struct *wq; |
49e3cf44 | 4180 | struct pool_workqueue *pwq; |
b196be89 | 4181 | |
cee22a15 VK |
4182 | /* see the comment above the definition of WQ_POWER_EFFICIENT */ |
4183 | if ((flags & WQ_POWER_EFFICIENT) && wq_power_efficient) | |
4184 | flags |= WQ_UNBOUND; | |
4185 | ||
ecf6881f | 4186 | /* allocate wq and format name */ |
df2d5ae4 TH |
4187 | if (flags & WQ_UNBOUND) |
4188 | tbl_size = wq_numa_tbl_len * sizeof(wq->numa_pwq_tbl[0]); | |
4189 | ||
4190 | wq = kzalloc(sizeof(*wq) + tbl_size, GFP_KERNEL); | |
b196be89 | 4191 | if (!wq) |
d2c1d404 | 4192 | return NULL; |
b196be89 | 4193 | |
6029a918 TH |
4194 | if (flags & WQ_UNBOUND) { |
4195 | wq->unbound_attrs = alloc_workqueue_attrs(GFP_KERNEL); | |
4196 | if (!wq->unbound_attrs) | |
4197 | goto err_free_wq; | |
4198 | } | |
4199 | ||
ecf6881f TH |
4200 | va_start(args, lock_name); |
4201 | vsnprintf(wq->name, sizeof(wq->name), fmt, args); | |
b196be89 | 4202 | va_end(args); |
1da177e4 | 4203 | |
d320c038 | 4204 | max_active = max_active ?: WQ_DFL_ACTIVE; |
b196be89 | 4205 | max_active = wq_clamp_max_active(max_active, flags, wq->name); |
3af24433 | 4206 | |
b196be89 | 4207 | /* init wq */ |
97e37d7b | 4208 | wq->flags = flags; |
a0a1a5fd | 4209 | wq->saved_max_active = max_active; |
3c25a55d | 4210 | mutex_init(&wq->mutex); |
112202d9 | 4211 | atomic_set(&wq->nr_pwqs_to_flush, 0); |
30cdf249 | 4212 | INIT_LIST_HEAD(&wq->pwqs); |
73f53c4a TH |
4213 | INIT_LIST_HEAD(&wq->flusher_queue); |
4214 | INIT_LIST_HEAD(&wq->flusher_overflow); | |
493a1724 | 4215 | INIT_LIST_HEAD(&wq->maydays); |
502ca9d8 | 4216 | |
eb13ba87 | 4217 | lockdep_init_map(&wq->lockdep_map, lock_name, key, 0); |
cce1a165 | 4218 | INIT_LIST_HEAD(&wq->list); |
3af24433 | 4219 | |
30cdf249 | 4220 | if (alloc_and_link_pwqs(wq) < 0) |
d2c1d404 | 4221 | goto err_free_wq; |
1537663f | 4222 | |
493008a8 TH |
4223 | /* |
4224 | * Workqueues which may be used during memory reclaim should | |
4225 | * have a rescuer to guarantee forward progress. | |
4226 | */ | |
4227 | if (flags & WQ_MEM_RECLAIM) { | |
e22bee78 TH |
4228 | struct worker *rescuer; |
4229 | ||
d2c1d404 | 4230 | rescuer = alloc_worker(); |
e22bee78 | 4231 | if (!rescuer) |
d2c1d404 | 4232 | goto err_destroy; |
e22bee78 | 4233 | |
111c225a TH |
4234 | rescuer->rescue_wq = wq; |
4235 | rescuer->task = kthread_create(rescuer_thread, rescuer, "%s", | |
b196be89 | 4236 | wq->name); |
d2c1d404 TH |
4237 | if (IS_ERR(rescuer->task)) { |
4238 | kfree(rescuer); | |
4239 | goto err_destroy; | |
4240 | } | |
e22bee78 | 4241 | |
d2c1d404 | 4242 | wq->rescuer = rescuer; |
14a40ffc | 4243 | rescuer->task->flags |= PF_NO_SETAFFINITY; |
e22bee78 | 4244 | wake_up_process(rescuer->task); |
3af24433 ON |
4245 | } |
4246 | ||
226223ab TH |
4247 | if ((wq->flags & WQ_SYSFS) && workqueue_sysfs_register(wq)) |
4248 | goto err_destroy; | |
4249 | ||
a0a1a5fd | 4250 | /* |
68e13a67 LJ |
4251 | * wq_pool_mutex protects global freeze state and workqueues list. |
4252 | * Grab it, adjust max_active and add the new @wq to workqueues | |
4253 | * list. | |
a0a1a5fd | 4254 | */ |
68e13a67 | 4255 | mutex_lock(&wq_pool_mutex); |
a0a1a5fd | 4256 | |
a357fc03 | 4257 | mutex_lock(&wq->mutex); |
699ce097 TH |
4258 | for_each_pwq(pwq, wq) |
4259 | pwq_adjust_max_active(pwq); | |
a357fc03 | 4260 | mutex_unlock(&wq->mutex); |
a0a1a5fd | 4261 | |
1537663f | 4262 | list_add(&wq->list, &workqueues); |
a0a1a5fd | 4263 | |
68e13a67 | 4264 | mutex_unlock(&wq_pool_mutex); |
1537663f | 4265 | |
3af24433 | 4266 | return wq; |
d2c1d404 TH |
4267 | |
4268 | err_free_wq: | |
6029a918 | 4269 | free_workqueue_attrs(wq->unbound_attrs); |
d2c1d404 TH |
4270 | kfree(wq); |
4271 | return NULL; | |
4272 | err_destroy: | |
4273 | destroy_workqueue(wq); | |
4690c4ab | 4274 | return NULL; |
3af24433 | 4275 | } |
d320c038 | 4276 | EXPORT_SYMBOL_GPL(__alloc_workqueue_key); |
1da177e4 | 4277 | |
3af24433 ON |
4278 | /** |
4279 | * destroy_workqueue - safely terminate a workqueue | |
4280 | * @wq: target workqueue | |
4281 | * | |
4282 | * Safely destroy a workqueue. All work currently pending will be done first. | |
4283 | */ | |
4284 | void destroy_workqueue(struct workqueue_struct *wq) | |
4285 | { | |
49e3cf44 | 4286 | struct pool_workqueue *pwq; |
4c16bd32 | 4287 | int node; |
3af24433 | 4288 | |
9c5a2ba7 TH |
4289 | /* drain it before proceeding with destruction */ |
4290 | drain_workqueue(wq); | |
c8efcc25 | 4291 | |
6183c009 | 4292 | /* sanity checks */ |
b09f4fd3 | 4293 | mutex_lock(&wq->mutex); |
49e3cf44 | 4294 | for_each_pwq(pwq, wq) { |
6183c009 TH |
4295 | int i; |
4296 | ||
76af4d93 TH |
4297 | for (i = 0; i < WORK_NR_COLORS; i++) { |
4298 | if (WARN_ON(pwq->nr_in_flight[i])) { | |
b09f4fd3 | 4299 | mutex_unlock(&wq->mutex); |
6183c009 | 4300 | return; |
76af4d93 TH |
4301 | } |
4302 | } | |
4303 | ||
5c529597 | 4304 | if (WARN_ON((pwq != wq->dfl_pwq) && (pwq->refcnt > 1)) || |
8864b4e5 | 4305 | WARN_ON(pwq->nr_active) || |
76af4d93 | 4306 | WARN_ON(!list_empty(&pwq->delayed_works))) { |
b09f4fd3 | 4307 | mutex_unlock(&wq->mutex); |
6183c009 | 4308 | return; |
76af4d93 | 4309 | } |
6183c009 | 4310 | } |
b09f4fd3 | 4311 | mutex_unlock(&wq->mutex); |
6183c009 | 4312 | |
a0a1a5fd TH |
4313 | /* |
4314 | * wq list is used to freeze wq, remove from list after | |
4315 | * flushing is complete in case freeze races us. | |
4316 | */ | |
68e13a67 | 4317 | mutex_lock(&wq_pool_mutex); |
d2c1d404 | 4318 | list_del_init(&wq->list); |
68e13a67 | 4319 | mutex_unlock(&wq_pool_mutex); |
3af24433 | 4320 | |
226223ab TH |
4321 | workqueue_sysfs_unregister(wq); |
4322 | ||
493008a8 | 4323 | if (wq->rescuer) { |
e22bee78 | 4324 | kthread_stop(wq->rescuer->task); |
8d9df9f0 | 4325 | kfree(wq->rescuer); |
493008a8 | 4326 | wq->rescuer = NULL; |
e22bee78 TH |
4327 | } |
4328 | ||
8864b4e5 TH |
4329 | if (!(wq->flags & WQ_UNBOUND)) { |
4330 | /* | |
4331 | * The base ref is never dropped on per-cpu pwqs. Directly | |
4332 | * free the pwqs and wq. | |
4333 | */ | |
4334 | free_percpu(wq->cpu_pwqs); | |
4335 | kfree(wq); | |
4336 | } else { | |
4337 | /* | |
4338 | * We're the sole accessor of @wq at this point. Directly | |
4c16bd32 TH |
4339 | * access numa_pwq_tbl[] and dfl_pwq to put the base refs. |
4340 | * @wq will be freed when the last pwq is released. | |
8864b4e5 | 4341 | */ |
4c16bd32 TH |
4342 | for_each_node(node) { |
4343 | pwq = rcu_access_pointer(wq->numa_pwq_tbl[node]); | |
4344 | RCU_INIT_POINTER(wq->numa_pwq_tbl[node], NULL); | |
4345 | put_pwq_unlocked(pwq); | |
4346 | } | |
4347 | ||
4348 | /* | |
4349 | * Put dfl_pwq. @wq may be freed any time after dfl_pwq is | |
4350 | * put. Don't access it afterwards. | |
4351 | */ | |
4352 | pwq = wq->dfl_pwq; | |
4353 | wq->dfl_pwq = NULL; | |
dce90d47 | 4354 | put_pwq_unlocked(pwq); |
29c91e99 | 4355 | } |
3af24433 ON |
4356 | } |
4357 | EXPORT_SYMBOL_GPL(destroy_workqueue); | |
4358 | ||
dcd989cb TH |
4359 | /** |
4360 | * workqueue_set_max_active - adjust max_active of a workqueue | |
4361 | * @wq: target workqueue | |
4362 | * @max_active: new max_active value. | |
4363 | * | |
4364 | * Set max_active of @wq to @max_active. | |
4365 | * | |
4366 | * CONTEXT: | |
4367 | * Don't call from IRQ context. | |
4368 | */ | |
4369 | void workqueue_set_max_active(struct workqueue_struct *wq, int max_active) | |
4370 | { | |
49e3cf44 | 4371 | struct pool_workqueue *pwq; |
dcd989cb | 4372 | |
8719dcea TH |
4373 | /* disallow meddling with max_active for ordered workqueues */ |
4374 | if (WARN_ON(wq->flags & __WQ_ORDERED)) | |
4375 | return; | |
4376 | ||
f3421797 | 4377 | max_active = wq_clamp_max_active(max_active, wq->flags, wq->name); |
dcd989cb | 4378 | |
a357fc03 | 4379 | mutex_lock(&wq->mutex); |
dcd989cb TH |
4380 | |
4381 | wq->saved_max_active = max_active; | |
4382 | ||
699ce097 TH |
4383 | for_each_pwq(pwq, wq) |
4384 | pwq_adjust_max_active(pwq); | |
93981800 | 4385 | |
a357fc03 | 4386 | mutex_unlock(&wq->mutex); |
15316ba8 | 4387 | } |
dcd989cb | 4388 | EXPORT_SYMBOL_GPL(workqueue_set_max_active); |
15316ba8 | 4389 | |
e6267616 TH |
4390 | /** |
4391 | * current_is_workqueue_rescuer - is %current workqueue rescuer? | |
4392 | * | |
4393 | * Determine whether %current is a workqueue rescuer. Can be used from | |
4394 | * work functions to determine whether it's being run off the rescuer task. | |
d185af30 YB |
4395 | * |
4396 | * Return: %true if %current is a workqueue rescuer. %false otherwise. | |
e6267616 TH |
4397 | */ |
4398 | bool current_is_workqueue_rescuer(void) | |
4399 | { | |
4400 | struct worker *worker = current_wq_worker(); | |
4401 | ||
6a092dfd | 4402 | return worker && worker->rescue_wq; |
e6267616 TH |
4403 | } |
4404 | ||
eef6a7d5 | 4405 | /** |
dcd989cb TH |
4406 | * workqueue_congested - test whether a workqueue is congested |
4407 | * @cpu: CPU in question | |
4408 | * @wq: target workqueue | |
eef6a7d5 | 4409 | * |
dcd989cb TH |
4410 | * Test whether @wq's cpu workqueue for @cpu is congested. There is |
4411 | * no synchronization around this function and the test result is | |
4412 | * unreliable and only useful as advisory hints or for debugging. | |
eef6a7d5 | 4413 | * |
d3251859 TH |
4414 | * If @cpu is WORK_CPU_UNBOUND, the test is performed on the local CPU. |
4415 | * Note that both per-cpu and unbound workqueues may be associated with | |
4416 | * multiple pool_workqueues which have separate congested states. A | |
4417 | * workqueue being congested on one CPU doesn't mean the workqueue is also | |
4418 | * contested on other CPUs / NUMA nodes. | |
4419 | * | |
d185af30 | 4420 | * Return: |
dcd989cb | 4421 | * %true if congested, %false otherwise. |
eef6a7d5 | 4422 | */ |
d84ff051 | 4423 | bool workqueue_congested(int cpu, struct workqueue_struct *wq) |
1da177e4 | 4424 | { |
7fb98ea7 | 4425 | struct pool_workqueue *pwq; |
76af4d93 TH |
4426 | bool ret; |
4427 | ||
88109453 | 4428 | rcu_read_lock_sched(); |
7fb98ea7 | 4429 | |
d3251859 TH |
4430 | if (cpu == WORK_CPU_UNBOUND) |
4431 | cpu = smp_processor_id(); | |
4432 | ||
7fb98ea7 TH |
4433 | if (!(wq->flags & WQ_UNBOUND)) |
4434 | pwq = per_cpu_ptr(wq->cpu_pwqs, cpu); | |
4435 | else | |
df2d5ae4 | 4436 | pwq = unbound_pwq_by_node(wq, cpu_to_node(cpu)); |
dcd989cb | 4437 | |
76af4d93 | 4438 | ret = !list_empty(&pwq->delayed_works); |
88109453 | 4439 | rcu_read_unlock_sched(); |
76af4d93 TH |
4440 | |
4441 | return ret; | |
1da177e4 | 4442 | } |
dcd989cb | 4443 | EXPORT_SYMBOL_GPL(workqueue_congested); |
1da177e4 | 4444 | |
dcd989cb TH |
4445 | /** |
4446 | * work_busy - test whether a work is currently pending or running | |
4447 | * @work: the work to be tested | |
4448 | * | |
4449 | * Test whether @work is currently pending or running. There is no | |
4450 | * synchronization around this function and the test result is | |
4451 | * unreliable and only useful as advisory hints or for debugging. | |
dcd989cb | 4452 | * |
d185af30 | 4453 | * Return: |
dcd989cb TH |
4454 | * OR'd bitmask of WORK_BUSY_* bits. |
4455 | */ | |
4456 | unsigned int work_busy(struct work_struct *work) | |
1da177e4 | 4457 | { |
fa1b54e6 | 4458 | struct worker_pool *pool; |
dcd989cb TH |
4459 | unsigned long flags; |
4460 | unsigned int ret = 0; | |
1da177e4 | 4461 | |
dcd989cb TH |
4462 | if (work_pending(work)) |
4463 | ret |= WORK_BUSY_PENDING; | |
1da177e4 | 4464 | |
fa1b54e6 TH |
4465 | local_irq_save(flags); |
4466 | pool = get_work_pool(work); | |
038366c5 | 4467 | if (pool) { |
fa1b54e6 | 4468 | spin_lock(&pool->lock); |
038366c5 LJ |
4469 | if (find_worker_executing_work(pool, work)) |
4470 | ret |= WORK_BUSY_RUNNING; | |
fa1b54e6 | 4471 | spin_unlock(&pool->lock); |
038366c5 | 4472 | } |
fa1b54e6 | 4473 | local_irq_restore(flags); |
1da177e4 | 4474 | |
dcd989cb | 4475 | return ret; |
1da177e4 | 4476 | } |
dcd989cb | 4477 | EXPORT_SYMBOL_GPL(work_busy); |
1da177e4 | 4478 | |
3d1cb205 TH |
4479 | /** |
4480 | * set_worker_desc - set description for the current work item | |
4481 | * @fmt: printf-style format string | |
4482 | * @...: arguments for the format string | |
4483 | * | |
4484 | * This function can be called by a running work function to describe what | |
4485 | * the work item is about. If the worker task gets dumped, this | |
4486 | * information will be printed out together to help debugging. The | |
4487 | * description can be at most WORKER_DESC_LEN including the trailing '\0'. | |
4488 | */ | |
4489 | void set_worker_desc(const char *fmt, ...) | |
4490 | { | |
4491 | struct worker *worker = current_wq_worker(); | |
4492 | va_list args; | |
4493 | ||
4494 | if (worker) { | |
4495 | va_start(args, fmt); | |
4496 | vsnprintf(worker->desc, sizeof(worker->desc), fmt, args); | |
4497 | va_end(args); | |
4498 | worker->desc_valid = true; | |
4499 | } | |
4500 | } | |
4501 | ||
4502 | /** | |
4503 | * print_worker_info - print out worker information and description | |
4504 | * @log_lvl: the log level to use when printing | |
4505 | * @task: target task | |
4506 | * | |
4507 | * If @task is a worker and currently executing a work item, print out the | |
4508 | * name of the workqueue being serviced and worker description set with | |
4509 | * set_worker_desc() by the currently executing work item. | |
4510 | * | |
4511 | * This function can be safely called on any task as long as the | |
4512 | * task_struct itself is accessible. While safe, this function isn't | |
4513 | * synchronized and may print out mixups or garbages of limited length. | |
4514 | */ | |
4515 | void print_worker_info(const char *log_lvl, struct task_struct *task) | |
4516 | { | |
4517 | work_func_t *fn = NULL; | |
4518 | char name[WQ_NAME_LEN] = { }; | |
4519 | char desc[WORKER_DESC_LEN] = { }; | |
4520 | struct pool_workqueue *pwq = NULL; | |
4521 | struct workqueue_struct *wq = NULL; | |
4522 | bool desc_valid = false; | |
4523 | struct worker *worker; | |
4524 | ||
4525 | if (!(task->flags & PF_WQ_WORKER)) | |
4526 | return; | |
4527 | ||
4528 | /* | |
4529 | * This function is called without any synchronization and @task | |
4530 | * could be in any state. Be careful with dereferences. | |
4531 | */ | |
4532 | worker = probe_kthread_data(task); | |
4533 | ||
4534 | /* | |
4535 | * Carefully copy the associated workqueue's workfn and name. Keep | |
4536 | * the original last '\0' in case the original contains garbage. | |
4537 | */ | |
4538 | probe_kernel_read(&fn, &worker->current_func, sizeof(fn)); | |
4539 | probe_kernel_read(&pwq, &worker->current_pwq, sizeof(pwq)); | |
4540 | probe_kernel_read(&wq, &pwq->wq, sizeof(wq)); | |
4541 | probe_kernel_read(name, wq->name, sizeof(name) - 1); | |
4542 | ||
4543 | /* copy worker description */ | |
4544 | probe_kernel_read(&desc_valid, &worker->desc_valid, sizeof(desc_valid)); | |
4545 | if (desc_valid) | |
4546 | probe_kernel_read(desc, worker->desc, sizeof(desc) - 1); | |
4547 | ||
4548 | if (fn || name[0] || desc[0]) { | |
2d916033 | 4549 | pr_info("%sWorkqueue: %s %pf", log_lvl, name, fn); |
3d1cb205 TH |
4550 | if (desc[0]) |
4551 | pr_cont(" (%s)", desc); | |
4552 | pr_cont("\n"); | |
4553 | } | |
4554 | } | |
4555 | ||
db7bccf4 TH |
4556 | /* |
4557 | * CPU hotplug. | |
4558 | * | |
e22bee78 | 4559 | * There are two challenges in supporting CPU hotplug. Firstly, there |
112202d9 | 4560 | * are a lot of assumptions on strong associations among work, pwq and |
706026c2 | 4561 | * pool which make migrating pending and scheduled works very |
e22bee78 | 4562 | * difficult to implement without impacting hot paths. Secondly, |
94cf58bb | 4563 | * worker pools serve mix of short, long and very long running works making |
e22bee78 TH |
4564 | * blocked draining impractical. |
4565 | * | |
24647570 | 4566 | * This is solved by allowing the pools to be disassociated from the CPU |
628c78e7 TH |
4567 | * running as an unbound one and allowing it to be reattached later if the |
4568 | * cpu comes back online. | |
db7bccf4 | 4569 | */ |
1da177e4 | 4570 | |
706026c2 | 4571 | static void wq_unbind_fn(struct work_struct *work) |
3af24433 | 4572 | { |
38db41d9 | 4573 | int cpu = smp_processor_id(); |
4ce62e9e | 4574 | struct worker_pool *pool; |
db7bccf4 | 4575 | struct worker *worker; |
a9ab775b | 4576 | int wi; |
3af24433 | 4577 | |
f02ae73a | 4578 | for_each_cpu_worker_pool(pool, cpu) { |
6183c009 | 4579 | WARN_ON_ONCE(cpu != smp_processor_id()); |
db7bccf4 | 4580 | |
bc3a1afc | 4581 | mutex_lock(&pool->manager_mutex); |
94cf58bb | 4582 | spin_lock_irq(&pool->lock); |
3af24433 | 4583 | |
94cf58bb | 4584 | /* |
bc3a1afc | 4585 | * We've blocked all manager operations. Make all workers |
94cf58bb TH |
4586 | * unbound and set DISASSOCIATED. Before this, all workers |
4587 | * except for the ones which are still executing works from | |
4588 | * before the last CPU down must be on the cpu. After | |
4589 | * this, they may become diasporas. | |
4590 | */ | |
a9ab775b | 4591 | for_each_pool_worker(worker, wi, pool) |
c9e7cf27 | 4592 | worker->flags |= WORKER_UNBOUND; |
06ba38a9 | 4593 | |
24647570 | 4594 | pool->flags |= POOL_DISASSOCIATED; |
f2d5a0ee | 4595 | |
94cf58bb | 4596 | spin_unlock_irq(&pool->lock); |
bc3a1afc | 4597 | mutex_unlock(&pool->manager_mutex); |
628c78e7 | 4598 | |
eb283428 LJ |
4599 | /* |
4600 | * Call schedule() so that we cross rq->lock and thus can | |
4601 | * guarantee sched callbacks see the %WORKER_UNBOUND flag. | |
4602 | * This is necessary as scheduler callbacks may be invoked | |
4603 | * from other cpus. | |
4604 | */ | |
4605 | schedule(); | |
06ba38a9 | 4606 | |
eb283428 LJ |
4607 | /* |
4608 | * Sched callbacks are disabled now. Zap nr_running. | |
4609 | * After this, nr_running stays zero and need_more_worker() | |
4610 | * and keep_working() are always true as long as the | |
4611 | * worklist is not empty. This pool now behaves as an | |
4612 | * unbound (in terms of concurrency management) pool which | |
4613 | * are served by workers tied to the pool. | |
4614 | */ | |
e19e397a | 4615 | atomic_set(&pool->nr_running, 0); |
eb283428 LJ |
4616 | |
4617 | /* | |
4618 | * With concurrency management just turned off, a busy | |
4619 | * worker blocking could lead to lengthy stalls. Kick off | |
4620 | * unbound chain execution of currently pending work items. | |
4621 | */ | |
4622 | spin_lock_irq(&pool->lock); | |
4623 | wake_up_worker(pool); | |
4624 | spin_unlock_irq(&pool->lock); | |
4625 | } | |
3af24433 | 4626 | } |
3af24433 | 4627 | |
bd7c089e TH |
4628 | /** |
4629 | * rebind_workers - rebind all workers of a pool to the associated CPU | |
4630 | * @pool: pool of interest | |
4631 | * | |
a9ab775b | 4632 | * @pool->cpu is coming online. Rebind all workers to the CPU. |
bd7c089e TH |
4633 | */ |
4634 | static void rebind_workers(struct worker_pool *pool) | |
4635 | { | |
a9ab775b TH |
4636 | struct worker *worker; |
4637 | int wi; | |
bd7c089e TH |
4638 | |
4639 | lockdep_assert_held(&pool->manager_mutex); | |
bd7c089e | 4640 | |
a9ab775b TH |
4641 | /* |
4642 | * Restore CPU affinity of all workers. As all idle workers should | |
4643 | * be on the run-queue of the associated CPU before any local | |
4644 | * wake-ups for concurrency management happen, restore CPU affinty | |
4645 | * of all workers first and then clear UNBOUND. As we're called | |
4646 | * from CPU_ONLINE, the following shouldn't fail. | |
4647 | */ | |
4648 | for_each_pool_worker(worker, wi, pool) | |
4649 | WARN_ON_ONCE(set_cpus_allowed_ptr(worker->task, | |
4650 | pool->attrs->cpumask) < 0); | |
bd7c089e | 4651 | |
a9ab775b | 4652 | spin_lock_irq(&pool->lock); |
bd7c089e | 4653 | |
a9ab775b TH |
4654 | for_each_pool_worker(worker, wi, pool) { |
4655 | unsigned int worker_flags = worker->flags; | |
bd7c089e TH |
4656 | |
4657 | /* | |
a9ab775b TH |
4658 | * A bound idle worker should actually be on the runqueue |
4659 | * of the associated CPU for local wake-ups targeting it to | |
4660 | * work. Kick all idle workers so that they migrate to the | |
4661 | * associated CPU. Doing this in the same loop as | |
4662 | * replacing UNBOUND with REBOUND is safe as no worker will | |
4663 | * be bound before @pool->lock is released. | |
bd7c089e | 4664 | */ |
a9ab775b TH |
4665 | if (worker_flags & WORKER_IDLE) |
4666 | wake_up_process(worker->task); | |
bd7c089e | 4667 | |
a9ab775b TH |
4668 | /* |
4669 | * We want to clear UNBOUND but can't directly call | |
4670 | * worker_clr_flags() or adjust nr_running. Atomically | |
4671 | * replace UNBOUND with another NOT_RUNNING flag REBOUND. | |
4672 | * @worker will clear REBOUND using worker_clr_flags() when | |
4673 | * it initiates the next execution cycle thus restoring | |
4674 | * concurrency management. Note that when or whether | |
4675 | * @worker clears REBOUND doesn't affect correctness. | |
4676 | * | |
4677 | * ACCESS_ONCE() is necessary because @worker->flags may be | |
4678 | * tested without holding any lock in | |
4679 | * wq_worker_waking_up(). Without it, NOT_RUNNING test may | |
4680 | * fail incorrectly leading to premature concurrency | |
4681 | * management operations. | |
4682 | */ | |
4683 | WARN_ON_ONCE(!(worker_flags & WORKER_UNBOUND)); | |
4684 | worker_flags |= WORKER_REBOUND; | |
4685 | worker_flags &= ~WORKER_UNBOUND; | |
4686 | ACCESS_ONCE(worker->flags) = worker_flags; | |
bd7c089e | 4687 | } |
a9ab775b TH |
4688 | |
4689 | spin_unlock_irq(&pool->lock); | |
bd7c089e TH |
4690 | } |
4691 | ||
7dbc725e TH |
4692 | /** |
4693 | * restore_unbound_workers_cpumask - restore cpumask of unbound workers | |
4694 | * @pool: unbound pool of interest | |
4695 | * @cpu: the CPU which is coming up | |
4696 | * | |
4697 | * An unbound pool may end up with a cpumask which doesn't have any online | |
4698 | * CPUs. When a worker of such pool get scheduled, the scheduler resets | |
4699 | * its cpus_allowed. If @cpu is in @pool's cpumask which didn't have any | |
4700 | * online CPU before, cpus_allowed of all its workers should be restored. | |
4701 | */ | |
4702 | static void restore_unbound_workers_cpumask(struct worker_pool *pool, int cpu) | |
4703 | { | |
4704 | static cpumask_t cpumask; | |
4705 | struct worker *worker; | |
4706 | int wi; | |
4707 | ||
4708 | lockdep_assert_held(&pool->manager_mutex); | |
4709 | ||
4710 | /* is @cpu allowed for @pool? */ | |
4711 | if (!cpumask_test_cpu(cpu, pool->attrs->cpumask)) | |
4712 | return; | |
4713 | ||
4714 | /* is @cpu the only online CPU? */ | |
4715 | cpumask_and(&cpumask, pool->attrs->cpumask, cpu_online_mask); | |
4716 | if (cpumask_weight(&cpumask) != 1) | |
4717 | return; | |
4718 | ||
4719 | /* as we're called from CPU_ONLINE, the following shouldn't fail */ | |
4720 | for_each_pool_worker(worker, wi, pool) | |
4721 | WARN_ON_ONCE(set_cpus_allowed_ptr(worker->task, | |
4722 | pool->attrs->cpumask) < 0); | |
4723 | } | |
4724 | ||
8db25e78 TH |
4725 | /* |
4726 | * Workqueues should be brought up before normal priority CPU notifiers. | |
4727 | * This will be registered high priority CPU notifier. | |
4728 | */ | |
0db0628d | 4729 | static int workqueue_cpu_up_callback(struct notifier_block *nfb, |
8db25e78 TH |
4730 | unsigned long action, |
4731 | void *hcpu) | |
3af24433 | 4732 | { |
d84ff051 | 4733 | int cpu = (unsigned long)hcpu; |
4ce62e9e | 4734 | struct worker_pool *pool; |
4c16bd32 | 4735 | struct workqueue_struct *wq; |
7dbc725e | 4736 | int pi; |
3ce63377 | 4737 | |
8db25e78 | 4738 | switch (action & ~CPU_TASKS_FROZEN) { |
3af24433 | 4739 | case CPU_UP_PREPARE: |
f02ae73a | 4740 | for_each_cpu_worker_pool(pool, cpu) { |
3ce63377 TH |
4741 | if (pool->nr_workers) |
4742 | continue; | |
ebf44d16 | 4743 | if (create_and_start_worker(pool) < 0) |
3ce63377 | 4744 | return NOTIFY_BAD; |
3af24433 | 4745 | } |
8db25e78 | 4746 | break; |
3af24433 | 4747 | |
db7bccf4 TH |
4748 | case CPU_DOWN_FAILED: |
4749 | case CPU_ONLINE: | |
68e13a67 | 4750 | mutex_lock(&wq_pool_mutex); |
7dbc725e TH |
4751 | |
4752 | for_each_pool(pool, pi) { | |
bc3a1afc | 4753 | mutex_lock(&pool->manager_mutex); |
94cf58bb | 4754 | |
7dbc725e TH |
4755 | if (pool->cpu == cpu) { |
4756 | spin_lock_irq(&pool->lock); | |
4757 | pool->flags &= ~POOL_DISASSOCIATED; | |
4758 | spin_unlock_irq(&pool->lock); | |
a9ab775b | 4759 | |
7dbc725e TH |
4760 | rebind_workers(pool); |
4761 | } else if (pool->cpu < 0) { | |
4762 | restore_unbound_workers_cpumask(pool, cpu); | |
4763 | } | |
94cf58bb | 4764 | |
bc3a1afc | 4765 | mutex_unlock(&pool->manager_mutex); |
94cf58bb | 4766 | } |
7dbc725e | 4767 | |
4c16bd32 TH |
4768 | /* update NUMA affinity of unbound workqueues */ |
4769 | list_for_each_entry(wq, &workqueues, list) | |
4770 | wq_update_unbound_numa(wq, cpu, true); | |
4771 | ||
68e13a67 | 4772 | mutex_unlock(&wq_pool_mutex); |
db7bccf4 | 4773 | break; |
00dfcaf7 | 4774 | } |
65758202 TH |
4775 | return NOTIFY_OK; |
4776 | } | |
4777 | ||
4778 | /* | |
4779 | * Workqueues should be brought down after normal priority CPU notifiers. | |
4780 | * This will be registered as low priority CPU notifier. | |
4781 | */ | |
0db0628d | 4782 | static int workqueue_cpu_down_callback(struct notifier_block *nfb, |
65758202 TH |
4783 | unsigned long action, |
4784 | void *hcpu) | |
4785 | { | |
d84ff051 | 4786 | int cpu = (unsigned long)hcpu; |
8db25e78 | 4787 | struct work_struct unbind_work; |
4c16bd32 | 4788 | struct workqueue_struct *wq; |
8db25e78 | 4789 | |
65758202 TH |
4790 | switch (action & ~CPU_TASKS_FROZEN) { |
4791 | case CPU_DOWN_PREPARE: | |
4c16bd32 | 4792 | /* unbinding per-cpu workers should happen on the local CPU */ |
706026c2 | 4793 | INIT_WORK_ONSTACK(&unbind_work, wq_unbind_fn); |
7635d2fd | 4794 | queue_work_on(cpu, system_highpri_wq, &unbind_work); |
4c16bd32 TH |
4795 | |
4796 | /* update NUMA affinity of unbound workqueues */ | |
4797 | mutex_lock(&wq_pool_mutex); | |
4798 | list_for_each_entry(wq, &workqueues, list) | |
4799 | wq_update_unbound_numa(wq, cpu, false); | |
4800 | mutex_unlock(&wq_pool_mutex); | |
4801 | ||
4802 | /* wait for per-cpu unbinding to finish */ | |
8db25e78 | 4803 | flush_work(&unbind_work); |
440a1136 | 4804 | destroy_work_on_stack(&unbind_work); |
8db25e78 | 4805 | break; |
65758202 TH |
4806 | } |
4807 | return NOTIFY_OK; | |
4808 | } | |
4809 | ||
2d3854a3 | 4810 | #ifdef CONFIG_SMP |
8ccad40d | 4811 | |
2d3854a3 | 4812 | struct work_for_cpu { |
ed48ece2 | 4813 | struct work_struct work; |
2d3854a3 RR |
4814 | long (*fn)(void *); |
4815 | void *arg; | |
4816 | long ret; | |
4817 | }; | |
4818 | ||
ed48ece2 | 4819 | static void work_for_cpu_fn(struct work_struct *work) |
2d3854a3 | 4820 | { |
ed48ece2 TH |
4821 | struct work_for_cpu *wfc = container_of(work, struct work_for_cpu, work); |
4822 | ||
2d3854a3 RR |
4823 | wfc->ret = wfc->fn(wfc->arg); |
4824 | } | |
4825 | ||
4826 | /** | |
4827 | * work_on_cpu - run a function in user context on a particular cpu | |
4828 | * @cpu: the cpu to run on | |
4829 | * @fn: the function to run | |
4830 | * @arg: the function arg | |
4831 | * | |
31ad9081 | 4832 | * It is up to the caller to ensure that the cpu doesn't go offline. |
6b44003e | 4833 | * The caller must not hold any locks which would prevent @fn from completing. |
d185af30 YB |
4834 | * |
4835 | * Return: The value @fn returns. | |
2d3854a3 | 4836 | */ |
d84ff051 | 4837 | long work_on_cpu(int cpu, long (*fn)(void *), void *arg) |
2d3854a3 | 4838 | { |
ed48ece2 | 4839 | struct work_for_cpu wfc = { .fn = fn, .arg = arg }; |
6b44003e | 4840 | |
ed48ece2 TH |
4841 | INIT_WORK_ONSTACK(&wfc.work, work_for_cpu_fn); |
4842 | schedule_work_on(cpu, &wfc.work); | |
12997d1a | 4843 | flush_work(&wfc.work); |
440a1136 | 4844 | destroy_work_on_stack(&wfc.work); |
2d3854a3 RR |
4845 | return wfc.ret; |
4846 | } | |
4847 | EXPORT_SYMBOL_GPL(work_on_cpu); | |
4848 | #endif /* CONFIG_SMP */ | |
4849 | ||
a0a1a5fd TH |
4850 | #ifdef CONFIG_FREEZER |
4851 | ||
4852 | /** | |
4853 | * freeze_workqueues_begin - begin freezing workqueues | |
4854 | * | |
58a69cb4 | 4855 | * Start freezing workqueues. After this function returns, all freezable |
c5aa87bb | 4856 | * workqueues will queue new works to their delayed_works list instead of |
706026c2 | 4857 | * pool->worklist. |
a0a1a5fd TH |
4858 | * |
4859 | * CONTEXT: | |
a357fc03 | 4860 | * Grabs and releases wq_pool_mutex, wq->mutex and pool->lock's. |
a0a1a5fd TH |
4861 | */ |
4862 | void freeze_workqueues_begin(void) | |
4863 | { | |
17116969 | 4864 | struct worker_pool *pool; |
24b8a847 TH |
4865 | struct workqueue_struct *wq; |
4866 | struct pool_workqueue *pwq; | |
611c92a0 | 4867 | int pi; |
a0a1a5fd | 4868 | |
68e13a67 | 4869 | mutex_lock(&wq_pool_mutex); |
a0a1a5fd | 4870 | |
6183c009 | 4871 | WARN_ON_ONCE(workqueue_freezing); |
a0a1a5fd TH |
4872 | workqueue_freezing = true; |
4873 | ||
24b8a847 | 4874 | /* set FREEZING */ |
611c92a0 | 4875 | for_each_pool(pool, pi) { |
5bcab335 | 4876 | spin_lock_irq(&pool->lock); |
17116969 TH |
4877 | WARN_ON_ONCE(pool->flags & POOL_FREEZING); |
4878 | pool->flags |= POOL_FREEZING; | |
5bcab335 | 4879 | spin_unlock_irq(&pool->lock); |
24b8a847 | 4880 | } |
a0a1a5fd | 4881 | |
24b8a847 | 4882 | list_for_each_entry(wq, &workqueues, list) { |
a357fc03 | 4883 | mutex_lock(&wq->mutex); |
699ce097 TH |
4884 | for_each_pwq(pwq, wq) |
4885 | pwq_adjust_max_active(pwq); | |
a357fc03 | 4886 | mutex_unlock(&wq->mutex); |
a0a1a5fd | 4887 | } |
5bcab335 | 4888 | |
68e13a67 | 4889 | mutex_unlock(&wq_pool_mutex); |
a0a1a5fd TH |
4890 | } |
4891 | ||
4892 | /** | |
58a69cb4 | 4893 | * freeze_workqueues_busy - are freezable workqueues still busy? |
a0a1a5fd TH |
4894 | * |
4895 | * Check whether freezing is complete. This function must be called | |
4896 | * between freeze_workqueues_begin() and thaw_workqueues(). | |
4897 | * | |
4898 | * CONTEXT: | |
68e13a67 | 4899 | * Grabs and releases wq_pool_mutex. |
a0a1a5fd | 4900 | * |
d185af30 | 4901 | * Return: |
58a69cb4 TH |
4902 | * %true if some freezable workqueues are still busy. %false if freezing |
4903 | * is complete. | |
a0a1a5fd TH |
4904 | */ |
4905 | bool freeze_workqueues_busy(void) | |
4906 | { | |
a0a1a5fd | 4907 | bool busy = false; |
24b8a847 TH |
4908 | struct workqueue_struct *wq; |
4909 | struct pool_workqueue *pwq; | |
a0a1a5fd | 4910 | |
68e13a67 | 4911 | mutex_lock(&wq_pool_mutex); |
a0a1a5fd | 4912 | |
6183c009 | 4913 | WARN_ON_ONCE(!workqueue_freezing); |
a0a1a5fd | 4914 | |
24b8a847 TH |
4915 | list_for_each_entry(wq, &workqueues, list) { |
4916 | if (!(wq->flags & WQ_FREEZABLE)) | |
4917 | continue; | |
a0a1a5fd TH |
4918 | /* |
4919 | * nr_active is monotonically decreasing. It's safe | |
4920 | * to peek without lock. | |
4921 | */ | |
88109453 | 4922 | rcu_read_lock_sched(); |
24b8a847 | 4923 | for_each_pwq(pwq, wq) { |
6183c009 | 4924 | WARN_ON_ONCE(pwq->nr_active < 0); |
112202d9 | 4925 | if (pwq->nr_active) { |
a0a1a5fd | 4926 | busy = true; |
88109453 | 4927 | rcu_read_unlock_sched(); |
a0a1a5fd TH |
4928 | goto out_unlock; |
4929 | } | |
4930 | } | |
88109453 | 4931 | rcu_read_unlock_sched(); |
a0a1a5fd TH |
4932 | } |
4933 | out_unlock: | |
68e13a67 | 4934 | mutex_unlock(&wq_pool_mutex); |
a0a1a5fd TH |
4935 | return busy; |
4936 | } | |
4937 | ||
4938 | /** | |
4939 | * thaw_workqueues - thaw workqueues | |
4940 | * | |
4941 | * Thaw workqueues. Normal queueing is restored and all collected | |
706026c2 | 4942 | * frozen works are transferred to their respective pool worklists. |
a0a1a5fd TH |
4943 | * |
4944 | * CONTEXT: | |
a357fc03 | 4945 | * Grabs and releases wq_pool_mutex, wq->mutex and pool->lock's. |
a0a1a5fd TH |
4946 | */ |
4947 | void thaw_workqueues(void) | |
4948 | { | |
24b8a847 TH |
4949 | struct workqueue_struct *wq; |
4950 | struct pool_workqueue *pwq; | |
4951 | struct worker_pool *pool; | |
611c92a0 | 4952 | int pi; |
a0a1a5fd | 4953 | |
68e13a67 | 4954 | mutex_lock(&wq_pool_mutex); |
a0a1a5fd TH |
4955 | |
4956 | if (!workqueue_freezing) | |
4957 | goto out_unlock; | |
4958 | ||
24b8a847 | 4959 | /* clear FREEZING */ |
611c92a0 | 4960 | for_each_pool(pool, pi) { |
5bcab335 | 4961 | spin_lock_irq(&pool->lock); |
24b8a847 TH |
4962 | WARN_ON_ONCE(!(pool->flags & POOL_FREEZING)); |
4963 | pool->flags &= ~POOL_FREEZING; | |
5bcab335 | 4964 | spin_unlock_irq(&pool->lock); |
24b8a847 | 4965 | } |
8b03ae3c | 4966 | |
24b8a847 TH |
4967 | /* restore max_active and repopulate worklist */ |
4968 | list_for_each_entry(wq, &workqueues, list) { | |
a357fc03 | 4969 | mutex_lock(&wq->mutex); |
699ce097 TH |
4970 | for_each_pwq(pwq, wq) |
4971 | pwq_adjust_max_active(pwq); | |
a357fc03 | 4972 | mutex_unlock(&wq->mutex); |
a0a1a5fd TH |
4973 | } |
4974 | ||
4975 | workqueue_freezing = false; | |
4976 | out_unlock: | |
68e13a67 | 4977 | mutex_unlock(&wq_pool_mutex); |
a0a1a5fd TH |
4978 | } |
4979 | #endif /* CONFIG_FREEZER */ | |
4980 | ||
bce90380 TH |
4981 | static void __init wq_numa_init(void) |
4982 | { | |
4983 | cpumask_var_t *tbl; | |
4984 | int node, cpu; | |
4985 | ||
4986 | /* determine NUMA pwq table len - highest node id + 1 */ | |
4987 | for_each_node(node) | |
4988 | wq_numa_tbl_len = max(wq_numa_tbl_len, node + 1); | |
4989 | ||
4990 | if (num_possible_nodes() <= 1) | |
4991 | return; | |
4992 | ||
d55262c4 TH |
4993 | if (wq_disable_numa) { |
4994 | pr_info("workqueue: NUMA affinity support disabled\n"); | |
4995 | return; | |
4996 | } | |
4997 | ||
4c16bd32 TH |
4998 | wq_update_unbound_numa_attrs_buf = alloc_workqueue_attrs(GFP_KERNEL); |
4999 | BUG_ON(!wq_update_unbound_numa_attrs_buf); | |
5000 | ||
bce90380 TH |
5001 | /* |
5002 | * We want masks of possible CPUs of each node which isn't readily | |
5003 | * available. Build one from cpu_to_node() which should have been | |
5004 | * fully initialized by now. | |
5005 | */ | |
5006 | tbl = kzalloc(wq_numa_tbl_len * sizeof(tbl[0]), GFP_KERNEL); | |
5007 | BUG_ON(!tbl); | |
5008 | ||
5009 | for_each_node(node) | |
1be0c25d TH |
5010 | BUG_ON(!alloc_cpumask_var_node(&tbl[node], GFP_KERNEL, |
5011 | node_online(node) ? node : NUMA_NO_NODE)); | |
bce90380 TH |
5012 | |
5013 | for_each_possible_cpu(cpu) { | |
5014 | node = cpu_to_node(cpu); | |
5015 | if (WARN_ON(node == NUMA_NO_NODE)) { | |
5016 | pr_warn("workqueue: NUMA node mapping not available for cpu%d, disabling NUMA support\n", cpu); | |
5017 | /* happens iff arch is bonkers, let's just proceed */ | |
5018 | return; | |
5019 | } | |
5020 | cpumask_set_cpu(cpu, tbl[node]); | |
5021 | } | |
5022 | ||
5023 | wq_numa_possible_cpumask = tbl; | |
5024 | wq_numa_enabled = true; | |
5025 | } | |
5026 | ||
6ee0578b | 5027 | static int __init init_workqueues(void) |
1da177e4 | 5028 | { |
7a4e344c TH |
5029 | int std_nice[NR_STD_WORKER_POOLS] = { 0, HIGHPRI_NICE_LEVEL }; |
5030 | int i, cpu; | |
c34056a3 | 5031 | |
e904e6c2 TH |
5032 | WARN_ON(__alignof__(struct pool_workqueue) < __alignof__(long long)); |
5033 | ||
5034 | pwq_cache = KMEM_CACHE(pool_workqueue, SLAB_PANIC); | |
5035 | ||
65758202 | 5036 | cpu_notifier(workqueue_cpu_up_callback, CPU_PRI_WORKQUEUE_UP); |
a5b4e57d | 5037 | hotcpu_notifier(workqueue_cpu_down_callback, CPU_PRI_WORKQUEUE_DOWN); |
8b03ae3c | 5038 | |
bce90380 TH |
5039 | wq_numa_init(); |
5040 | ||
706026c2 | 5041 | /* initialize CPU pools */ |
29c91e99 | 5042 | for_each_possible_cpu(cpu) { |
4ce62e9e | 5043 | struct worker_pool *pool; |
8b03ae3c | 5044 | |
7a4e344c | 5045 | i = 0; |
f02ae73a | 5046 | for_each_cpu_worker_pool(pool, cpu) { |
7a4e344c | 5047 | BUG_ON(init_worker_pool(pool)); |
ec22ca5e | 5048 | pool->cpu = cpu; |
29c91e99 | 5049 | cpumask_copy(pool->attrs->cpumask, cpumask_of(cpu)); |
7a4e344c | 5050 | pool->attrs->nice = std_nice[i++]; |
f3f90ad4 | 5051 | pool->node = cpu_to_node(cpu); |
7a4e344c | 5052 | |
9daf9e67 | 5053 | /* alloc pool ID */ |
68e13a67 | 5054 | mutex_lock(&wq_pool_mutex); |
9daf9e67 | 5055 | BUG_ON(worker_pool_assign_id(pool)); |
68e13a67 | 5056 | mutex_unlock(&wq_pool_mutex); |
4ce62e9e | 5057 | } |
8b03ae3c TH |
5058 | } |
5059 | ||
e22bee78 | 5060 | /* create the initial worker */ |
29c91e99 | 5061 | for_each_online_cpu(cpu) { |
4ce62e9e | 5062 | struct worker_pool *pool; |
e22bee78 | 5063 | |
f02ae73a | 5064 | for_each_cpu_worker_pool(pool, cpu) { |
29c91e99 | 5065 | pool->flags &= ~POOL_DISASSOCIATED; |
ebf44d16 | 5066 | BUG_ON(create_and_start_worker(pool) < 0); |
4ce62e9e | 5067 | } |
e22bee78 TH |
5068 | } |
5069 | ||
8a2b7538 | 5070 | /* create default unbound and ordered wq attrs */ |
29c91e99 TH |
5071 | for (i = 0; i < NR_STD_WORKER_POOLS; i++) { |
5072 | struct workqueue_attrs *attrs; | |
5073 | ||
5074 | BUG_ON(!(attrs = alloc_workqueue_attrs(GFP_KERNEL))); | |
29c91e99 | 5075 | attrs->nice = std_nice[i]; |
29c91e99 | 5076 | unbound_std_wq_attrs[i] = attrs; |
8a2b7538 TH |
5077 | |
5078 | /* | |
5079 | * An ordered wq should have only one pwq as ordering is | |
5080 | * guaranteed by max_active which is enforced by pwqs. | |
5081 | * Turn off NUMA so that dfl_pwq is used for all nodes. | |
5082 | */ | |
5083 | BUG_ON(!(attrs = alloc_workqueue_attrs(GFP_KERNEL))); | |
5084 | attrs->nice = std_nice[i]; | |
5085 | attrs->no_numa = true; | |
5086 | ordered_wq_attrs[i] = attrs; | |
29c91e99 TH |
5087 | } |
5088 | ||
d320c038 | 5089 | system_wq = alloc_workqueue("events", 0, 0); |
1aabe902 | 5090 | system_highpri_wq = alloc_workqueue("events_highpri", WQ_HIGHPRI, 0); |
d320c038 | 5091 | system_long_wq = alloc_workqueue("events_long", 0, 0); |
f3421797 TH |
5092 | system_unbound_wq = alloc_workqueue("events_unbound", WQ_UNBOUND, |
5093 | WQ_UNBOUND_MAX_ACTIVE); | |
24d51add TH |
5094 | system_freezable_wq = alloc_workqueue("events_freezable", |
5095 | WQ_FREEZABLE, 0); | |
0668106c VK |
5096 | system_power_efficient_wq = alloc_workqueue("events_power_efficient", |
5097 | WQ_POWER_EFFICIENT, 0); | |
5098 | system_freezable_power_efficient_wq = alloc_workqueue("events_freezable_power_efficient", | |
5099 | WQ_FREEZABLE | WQ_POWER_EFFICIENT, | |
5100 | 0); | |
1aabe902 | 5101 | BUG_ON(!system_wq || !system_highpri_wq || !system_long_wq || |
0668106c VK |
5102 | !system_unbound_wq || !system_freezable_wq || |
5103 | !system_power_efficient_wq || | |
5104 | !system_freezable_power_efficient_wq); | |
6ee0578b | 5105 | return 0; |
1da177e4 | 5106 | } |
6ee0578b | 5107 | early_initcall(init_workqueues); |