]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 | 2 | /* |
c54fce6e | 3 | * kernel/workqueue.c - generic async execution with shared worker pool |
1da177e4 | 4 | * |
c54fce6e | 5 | * Copyright (C) 2002 Ingo Molnar |
1da177e4 | 6 | * |
c54fce6e TH |
7 | * Derived from the taskqueue/keventd code by: |
8 | * David Woodhouse <dwmw2@infradead.org> | |
9 | * Andrew Morton | |
10 | * Kai Petzke <wpp@marie.physik.tu-berlin.de> | |
11 | * Theodore Ts'o <tytso@mit.edu> | |
1da177e4 | 12 | * |
c54fce6e | 13 | * Made to use alloc_percpu by Christoph Lameter. |
1da177e4 | 14 | * |
c54fce6e TH |
15 | * Copyright (C) 2010 SUSE Linux Products GmbH |
16 | * Copyright (C) 2010 Tejun Heo <tj@kernel.org> | |
89ada679 | 17 | * |
c54fce6e TH |
18 | * This is the generic async execution mechanism. Work items as are |
19 | * executed in process context. The worker pool is shared and | |
b11895c4 L |
20 | * automatically managed. There are two worker pools for each CPU (one for |
21 | * normal work items and the other for high priority ones) and some extra | |
22 | * pools for workqueues which are not bound to any specific CPU - the | |
23 | * number of these backing pools is dynamic. | |
c54fce6e | 24 | * |
9a261491 | 25 | * Please read Documentation/core-api/workqueue.rst for details. |
1da177e4 LT |
26 | */ |
27 | ||
9984de1a | 28 | #include <linux/export.h> |
1da177e4 LT |
29 | #include <linux/kernel.h> |
30 | #include <linux/sched.h> | |
31 | #include <linux/init.h> | |
32 | #include <linux/signal.h> | |
33 | #include <linux/completion.h> | |
34 | #include <linux/workqueue.h> | |
35 | #include <linux/slab.h> | |
36 | #include <linux/cpu.h> | |
37 | #include <linux/notifier.h> | |
38 | #include <linux/kthread.h> | |
1fa44eca | 39 | #include <linux/hardirq.h> |
46934023 | 40 | #include <linux/mempolicy.h> |
341a5958 | 41 | #include <linux/freezer.h> |
d5abe669 | 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> |
c98a9805 | 51 | #include <linux/sched/isolation.h> |
62635ea8 | 52 | #include <linux/nmi.h> |
e22bee78 | 53 | |
ea138446 | 54 | #include "workqueue_internal.h" |
1da177e4 | 55 | |
c8e55f36 | 56 | enum { |
24647570 TH |
57 | /* |
58 | * worker_pool flags | |
bc2ae0f5 | 59 | * |
24647570 | 60 | * A bound pool is either associated or disassociated with its CPU. |
bc2ae0f5 TH |
61 | * While associated (!DISASSOCIATED), all workers are bound to the |
62 | * CPU and none has %WORKER_UNBOUND set and concurrency management | |
63 | * is in effect. | |
64 | * | |
65 | * While DISASSOCIATED, the cpu may be offline and all workers have | |
66 | * %WORKER_UNBOUND set and concurrency management disabled, and may | |
24647570 | 67 | * be executing on any CPU. The pool behaves as an unbound one. |
bc2ae0f5 | 68 | * |
bc3a1afc | 69 | * Note that DISASSOCIATED should be flipped only while holding |
1258fae7 | 70 | * wq_pool_attach_mutex to avoid changing binding state while |
4736cbf7 | 71 | * worker_attach_to_pool() is in progress. |
bc2ae0f5 | 72 | */ |
692b4825 | 73 | POOL_MANAGER_ACTIVE = 1 << 0, /* being managed */ |
24647570 | 74 | POOL_DISASSOCIATED = 1 << 2, /* cpu can't serve workers */ |
db7bccf4 | 75 | |
c8e55f36 | 76 | /* worker flags */ |
c8e55f36 TH |
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 | |
8698a745 | 103 | * all cpus. Give MIN_NICE. |
e22bee78 | 104 | */ |
8698a745 DY |
105 | RESCUER_NICE_LEVEL = MIN_NICE, |
106 | HIGHPRI_NICE_LEVEL = MIN_NICE, | |
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 | * |
1258fae7 | 127 | * A: wq_pool_attach_mutex protected. |
822d8405 | 128 | * |
68e13a67 | 129 | * PL: wq_pool_mutex protected. |
5bcab335 | 130 | * |
24acfb71 | 131 | * PR: wq_pool_mutex protected for writes. RCU protected for reads. |
76af4d93 | 132 | * |
5b95e1af LJ |
133 | * PW: wq_pool_mutex and wq->mutex protected for writes. Either for reads. |
134 | * | |
135 | * PWR: wq_pool_mutex and wq->mutex protected for writes. Either or | |
24acfb71 | 136 | * RCU for reads. |
5b95e1af | 137 | * |
3c25a55d LJ |
138 | * WQ: wq->mutex protected. |
139 | * | |
24acfb71 | 140 | * WR: wq->mutex protected for writes. RCU protected for reads. |
2e109a28 TH |
141 | * |
142 | * MD: wq_mayday_lock protected. | |
1da177e4 | 143 | */ |
1da177e4 | 144 | |
2eaebdb3 | 145 | /* struct worker is defined in workqueue_internal.h */ |
c34056a3 | 146 | |
bd7bdd43 | 147 | struct worker_pool { |
d565ed63 | 148 | spinlock_t lock; /* the pool lock */ |
d84ff051 | 149 | int cpu; /* I: the associated cpu */ |
f3f90ad4 | 150 | int node; /* I: the associated node ID */ |
9daf9e67 | 151 | int id; /* I: pool ID */ |
11ebea50 | 152 | unsigned int flags; /* X: flags */ |
bd7bdd43 | 153 | |
82607adc TH |
154 | unsigned long watchdog_ts; /* L: watchdog timestamp */ |
155 | ||
bd7bdd43 | 156 | struct list_head worklist; /* L: list of pending works */ |
ea1abd61 | 157 | |
5826cc8f LJ |
158 | int nr_workers; /* L: total number of workers */ |
159 | int nr_idle; /* L: currently idle workers */ | |
bd7bdd43 TH |
160 | |
161 | struct list_head idle_list; /* X: list of idle workers */ | |
162 | struct timer_list idle_timer; /* L: worker idle timeout */ | |
163 | struct timer_list mayday_timer; /* L: SOS timer for workers */ | |
164 | ||
c5aa87bb | 165 | /* a workers is either on busy_hash or idle_list, or the manager */ |
c9e7cf27 TH |
166 | DECLARE_HASHTABLE(busy_hash, BUSY_WORKER_HASH_ORDER); |
167 | /* L: hash of busy workers */ | |
168 | ||
2607d7a6 | 169 | struct worker *manager; /* L: purely informational */ |
92f9c5c4 | 170 | struct list_head workers; /* A: attached workers */ |
60f5a4bc | 171 | struct completion *detach_completion; /* all workers detached */ |
e19e397a | 172 | |
7cda9aae | 173 | struct ida worker_ida; /* worker IDs for task name */ |
e19e397a | 174 | |
7a4e344c | 175 | struct workqueue_attrs *attrs; /* I: worker attributes */ |
68e13a67 LJ |
176 | struct hlist_node hash_node; /* PL: unbound_pool_hash node */ |
177 | int refcnt; /* PL: refcnt for unbound pools */ | |
7a4e344c | 178 | |
e19e397a TH |
179 | /* |
180 | * The current concurrency level. As it's likely to be accessed | |
181 | * from other CPUs during try_to_wake_up(), put it in a separate | |
182 | * cacheline. | |
183 | */ | |
184 | atomic_t nr_running ____cacheline_aligned_in_smp; | |
29c91e99 TH |
185 | |
186 | /* | |
24acfb71 | 187 | * Destruction of pool is RCU protected to allow dereferences |
29c91e99 TH |
188 | * from get_work_pool(). |
189 | */ | |
190 | struct rcu_head rcu; | |
8b03ae3c TH |
191 | } ____cacheline_aligned_in_smp; |
192 | ||
1da177e4 | 193 | /* |
112202d9 TH |
194 | * The per-pool workqueue. While queued, the lower WORK_STRUCT_FLAG_BITS |
195 | * of work_struct->data are used for flags and the remaining high bits | |
196 | * point to the pwq; thus, pwqs need to be aligned at two's power of the | |
197 | * number of flag bits. | |
1da177e4 | 198 | */ |
112202d9 | 199 | struct pool_workqueue { |
bd7bdd43 | 200 | struct worker_pool *pool; /* I: the associated pool */ |
4690c4ab | 201 | struct workqueue_struct *wq; /* I: the owning workqueue */ |
73f53c4a TH |
202 | int work_color; /* L: current color */ |
203 | int flush_color; /* L: flushing color */ | |
8864b4e5 | 204 | int refcnt; /* L: reference count */ |
73f53c4a TH |
205 | int nr_in_flight[WORK_NR_COLORS]; |
206 | /* L: nr of in_flight works */ | |
1e19ffc6 | 207 | int nr_active; /* L: nr of active works */ |
a0a1a5fd | 208 | int max_active; /* L: max active works */ |
1e19ffc6 | 209 | struct list_head delayed_works; /* L: delayed works */ |
3c25a55d | 210 | struct list_head pwqs_node; /* WR: node on wq->pwqs */ |
2e109a28 | 211 | struct list_head mayday_node; /* MD: node on wq->maydays */ |
8864b4e5 TH |
212 | |
213 | /* | |
214 | * Release of unbound pwq is punted to system_wq. See put_pwq() | |
215 | * and pwq_unbound_release_workfn() for details. pool_workqueue | |
24acfb71 | 216 | * itself is also RCU protected so that the first pwq can be |
b09f4fd3 | 217 | * determined without grabbing wq->mutex. |
8864b4e5 TH |
218 | */ |
219 | struct work_struct unbound_release_work; | |
220 | struct rcu_head rcu; | |
e904e6c2 | 221 | } __aligned(1 << WORK_STRUCT_FLAG_BITS); |
1da177e4 | 222 | |
73f53c4a TH |
223 | /* |
224 | * Structure used to wait for workqueue flush. | |
225 | */ | |
226 | struct wq_flusher { | |
3c25a55d LJ |
227 | struct list_head list; /* WQ: list of flushers */ |
228 | int flush_color; /* WQ: flush color waiting for */ | |
73f53c4a TH |
229 | struct completion done; /* flush completion */ |
230 | }; | |
231 | ||
226223ab TH |
232 | struct wq_device; |
233 | ||
1da177e4 | 234 | /* |
c5aa87bb TH |
235 | * The externally visible workqueue. It relays the issued work items to |
236 | * the appropriate worker_pool through its pool_workqueues. | |
1da177e4 LT |
237 | */ |
238 | struct workqueue_struct { | |
3c25a55d | 239 | struct list_head pwqs; /* WR: all pwqs of this wq */ |
e2dca7ad | 240 | struct list_head list; /* PR: list of all workqueues */ |
73f53c4a | 241 | |
3c25a55d LJ |
242 | struct mutex mutex; /* protects this wq */ |
243 | int work_color; /* WQ: current work color */ | |
244 | int flush_color; /* WQ: current flush color */ | |
112202d9 | 245 | atomic_t nr_pwqs_to_flush; /* flush in progress */ |
3c25a55d LJ |
246 | struct wq_flusher *first_flusher; /* WQ: first flusher */ |
247 | struct list_head flusher_queue; /* WQ: flush waiters */ | |
248 | struct list_head flusher_overflow; /* WQ: flush overflow list */ | |
73f53c4a | 249 | |
2e109a28 | 250 | struct list_head maydays; /* MD: pwqs requesting rescue */ |
30ae2fc0 | 251 | struct worker *rescuer; /* MD: rescue worker */ |
e22bee78 | 252 | |
87fc741e | 253 | int nr_drainers; /* WQ: drain in progress */ |
a357fc03 | 254 | int saved_max_active; /* WQ: saved pwq max_active */ |
226223ab | 255 | |
5b95e1af LJ |
256 | struct workqueue_attrs *unbound_attrs; /* PW: only for unbound wqs */ |
257 | struct pool_workqueue *dfl_pwq; /* PW: only for unbound wqs */ | |
6029a918 | 258 | |
226223ab TH |
259 | #ifdef CONFIG_SYSFS |
260 | struct wq_device *wq_dev; /* I: for sysfs interface */ | |
261 | #endif | |
4e6045f1 | 262 | #ifdef CONFIG_LOCKDEP |
669de8bd BVA |
263 | char *lock_name; |
264 | struct lock_class_key key; | |
4690c4ab | 265 | struct lockdep_map lockdep_map; |
4e6045f1 | 266 | #endif |
ecf6881f | 267 | char name[WQ_NAME_LEN]; /* I: workqueue name */ |
2728fd2f | 268 | |
e2dca7ad | 269 | /* |
24acfb71 TG |
270 | * Destruction of workqueue_struct is RCU protected to allow walking |
271 | * the workqueues list without grabbing wq_pool_mutex. | |
e2dca7ad TH |
272 | * This is used to dump all workqueues from sysrq. |
273 | */ | |
274 | struct rcu_head rcu; | |
275 | ||
2728fd2f TH |
276 | /* hot fields used during command issue, aligned to cacheline */ |
277 | unsigned int flags ____cacheline_aligned; /* WQ: WQ_* flags */ | |
278 | struct pool_workqueue __percpu *cpu_pwqs; /* I: per-cpu pwqs */ | |
5b95e1af | 279 | struct pool_workqueue __rcu *numa_pwq_tbl[]; /* PWR: unbound pwqs indexed by node */ |
1da177e4 LT |
280 | }; |
281 | ||
e904e6c2 TH |
282 | static struct kmem_cache *pwq_cache; |
283 | ||
bce90380 TH |
284 | static cpumask_var_t *wq_numa_possible_cpumask; |
285 | /* possible CPUs of each node */ | |
286 | ||
d55262c4 TH |
287 | static bool wq_disable_numa; |
288 | module_param_named(disable_numa, wq_disable_numa, bool, 0444); | |
289 | ||
cee22a15 | 290 | /* see the comment above the definition of WQ_POWER_EFFICIENT */ |
552f530c | 291 | static bool wq_power_efficient = IS_ENABLED(CONFIG_WQ_POWER_EFFICIENT_DEFAULT); |
cee22a15 VK |
292 | module_param_named(power_efficient, wq_power_efficient, bool, 0444); |
293 | ||
863b710b | 294 | static bool wq_online; /* can kworkers be created yet? */ |
3347fa09 | 295 | |
bce90380 TH |
296 | static bool wq_numa_enabled; /* unbound NUMA affinity enabled */ |
297 | ||
4c16bd32 TH |
298 | /* buf for wq_update_unbound_numa_attrs(), protected by CPU hotplug exclusion */ |
299 | static struct workqueue_attrs *wq_update_unbound_numa_attrs_buf; | |
300 | ||
68e13a67 | 301 | static DEFINE_MUTEX(wq_pool_mutex); /* protects pools and workqueues list */ |
1258fae7 | 302 | static DEFINE_MUTEX(wq_pool_attach_mutex); /* protects worker attach/detach */ |
2e109a28 | 303 | static DEFINE_SPINLOCK(wq_mayday_lock); /* protects wq->maydays list */ |
692b4825 | 304 | static DECLARE_WAIT_QUEUE_HEAD(wq_manager_wait); /* wait for manager to go away */ |
5bcab335 | 305 | |
e2dca7ad | 306 | static LIST_HEAD(workqueues); /* PR: list of all workqueues */ |
68e13a67 | 307 | static bool workqueue_freezing; /* PL: have wqs started freezing? */ |
7d19c5ce | 308 | |
ef557180 MG |
309 | /* PL: allowable cpus for unbound wqs and work items */ |
310 | static cpumask_var_t wq_unbound_cpumask; | |
311 | ||
312 | /* CPU where unbound work was last round robin scheduled from this CPU */ | |
313 | static DEFINE_PER_CPU(int, wq_rr_cpu_last); | |
b05a7928 | 314 | |
f303fccb TH |
315 | /* |
316 | * Local execution of unbound work items is no longer guaranteed. The | |
317 | * following always forces round-robin CPU selection on unbound work items | |
318 | * to uncover usages which depend on it. | |
319 | */ | |
320 | #ifdef CONFIG_DEBUG_WQ_FORCE_RR_CPU | |
321 | static bool wq_debug_force_rr_cpu = true; | |
322 | #else | |
323 | static bool wq_debug_force_rr_cpu = false; | |
324 | #endif | |
325 | module_param_named(debug_force_rr_cpu, wq_debug_force_rr_cpu, bool, 0644); | |
326 | ||
7d19c5ce | 327 | /* the per-cpu worker pools */ |
25528213 | 328 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct worker_pool [NR_STD_WORKER_POOLS], cpu_worker_pools); |
7d19c5ce | 329 | |
68e13a67 | 330 | static DEFINE_IDR(worker_pool_idr); /* PR: idr of all pools */ |
7d19c5ce | 331 | |
68e13a67 | 332 | /* PL: hash of all unbound pools keyed by pool->attrs */ |
29c91e99 TH |
333 | static DEFINE_HASHTABLE(unbound_pool_hash, UNBOUND_POOL_HASH_ORDER); |
334 | ||
c5aa87bb | 335 | /* I: attributes used when instantiating standard unbound pools on demand */ |
29c91e99 TH |
336 | static struct workqueue_attrs *unbound_std_wq_attrs[NR_STD_WORKER_POOLS]; |
337 | ||
8a2b7538 TH |
338 | /* I: attributes used when instantiating ordered pools on demand */ |
339 | static struct workqueue_attrs *ordered_wq_attrs[NR_STD_WORKER_POOLS]; | |
340 | ||
d320c038 | 341 | struct workqueue_struct *system_wq __read_mostly; |
ad7b1f84 | 342 | EXPORT_SYMBOL(system_wq); |
044c782c | 343 | struct workqueue_struct *system_highpri_wq __read_mostly; |
1aabe902 | 344 | EXPORT_SYMBOL_GPL(system_highpri_wq); |
044c782c | 345 | struct workqueue_struct *system_long_wq __read_mostly; |
d320c038 | 346 | EXPORT_SYMBOL_GPL(system_long_wq); |
044c782c | 347 | struct workqueue_struct *system_unbound_wq __read_mostly; |
f3421797 | 348 | EXPORT_SYMBOL_GPL(system_unbound_wq); |
044c782c | 349 | struct workqueue_struct *system_freezable_wq __read_mostly; |
24d51add | 350 | EXPORT_SYMBOL_GPL(system_freezable_wq); |
0668106c VK |
351 | struct workqueue_struct *system_power_efficient_wq __read_mostly; |
352 | EXPORT_SYMBOL_GPL(system_power_efficient_wq); | |
353 | struct workqueue_struct *system_freezable_power_efficient_wq __read_mostly; | |
354 | EXPORT_SYMBOL_GPL(system_freezable_power_efficient_wq); | |
d320c038 | 355 | |
7d19c5ce | 356 | static int worker_thread(void *__worker); |
6ba94429 | 357 | static void workqueue_sysfs_unregister(struct workqueue_struct *wq); |
c29eb853 | 358 | static void show_pwq(struct pool_workqueue *pwq); |
7d19c5ce | 359 | |
97bd2347 TH |
360 | #define CREATE_TRACE_POINTS |
361 | #include <trace/events/workqueue.h> | |
362 | ||
68e13a67 | 363 | #define assert_rcu_or_pool_mutex() \ |
24acfb71 | 364 | RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \ |
f78f5b90 | 365 | !lockdep_is_held(&wq_pool_mutex), \ |
24acfb71 | 366 | "RCU or wq_pool_mutex should be held") |
5bcab335 | 367 | |
5b95e1af | 368 | #define assert_rcu_or_wq_mutex_or_pool_mutex(wq) \ |
24acfb71 | 369 | RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \ |
f78f5b90 PM |
370 | !lockdep_is_held(&wq->mutex) && \ |
371 | !lockdep_is_held(&wq_pool_mutex), \ | |
24acfb71 | 372 | "RCU, wq->mutex or wq_pool_mutex should be held") |
5b95e1af | 373 | |
f02ae73a TH |
374 | #define for_each_cpu_worker_pool(pool, cpu) \ |
375 | for ((pool) = &per_cpu(cpu_worker_pools, cpu)[0]; \ | |
376 | (pool) < &per_cpu(cpu_worker_pools, cpu)[NR_STD_WORKER_POOLS]; \ | |
7a62c2c8 | 377 | (pool)++) |
4ce62e9e | 378 | |
17116969 TH |
379 | /** |
380 | * for_each_pool - iterate through all worker_pools in the system | |
381 | * @pool: iteration cursor | |
611c92a0 | 382 | * @pi: integer used for iteration |
fa1b54e6 | 383 | * |
24acfb71 | 384 | * This must be called either with wq_pool_mutex held or RCU read |
68e13a67 LJ |
385 | * locked. If the pool needs to be used beyond the locking in effect, the |
386 | * caller is responsible for guaranteeing that the pool stays online. | |
fa1b54e6 TH |
387 | * |
388 | * The if/else clause exists only for the lockdep assertion and can be | |
389 | * ignored. | |
17116969 | 390 | */ |
611c92a0 TH |
391 | #define for_each_pool(pool, pi) \ |
392 | idr_for_each_entry(&worker_pool_idr, pool, pi) \ | |
68e13a67 | 393 | if (({ assert_rcu_or_pool_mutex(); false; })) { } \ |
fa1b54e6 | 394 | else |
17116969 | 395 | |
822d8405 TH |
396 | /** |
397 | * for_each_pool_worker - iterate through all workers of a worker_pool | |
398 | * @worker: iteration cursor | |
822d8405 TH |
399 | * @pool: worker_pool to iterate workers of |
400 | * | |
1258fae7 | 401 | * This must be called with wq_pool_attach_mutex. |
822d8405 TH |
402 | * |
403 | * The if/else clause exists only for the lockdep assertion and can be | |
404 | * ignored. | |
405 | */ | |
da028469 LJ |
406 | #define for_each_pool_worker(worker, pool) \ |
407 | list_for_each_entry((worker), &(pool)->workers, node) \ | |
1258fae7 | 408 | if (({ lockdep_assert_held(&wq_pool_attach_mutex); false; })) { } \ |
822d8405 TH |
409 | else |
410 | ||
49e3cf44 TH |
411 | /** |
412 | * for_each_pwq - iterate through all pool_workqueues of the specified workqueue | |
413 | * @pwq: iteration cursor | |
414 | * @wq: the target workqueue | |
76af4d93 | 415 | * |
24acfb71 | 416 | * This must be called either with wq->mutex held or RCU read locked. |
794b18bc TH |
417 | * If the pwq needs to be used beyond the locking in effect, the caller is |
418 | * responsible for guaranteeing that the pwq stays online. | |
76af4d93 TH |
419 | * |
420 | * The if/else clause exists only for the lockdep assertion and can be | |
421 | * ignored. | |
49e3cf44 TH |
422 | */ |
423 | #define for_each_pwq(pwq, wq) \ | |
49e9d1a9 | 424 | list_for_each_entry_rcu((pwq), &(wq)->pwqs, pwqs_node, \ |
5a644662 | 425 | lockdep_is_held(&(wq->mutex))) |
f3421797 | 426 | |
dc186ad7 TG |
427 | #ifdef CONFIG_DEBUG_OBJECTS_WORK |
428 | ||
429 | static struct debug_obj_descr work_debug_descr; | |
430 | ||
99777288 SG |
431 | static void *work_debug_hint(void *addr) |
432 | { | |
433 | return ((struct work_struct *) addr)->func; | |
434 | } | |
435 | ||
b9fdac7f DC |
436 | static bool work_is_static_object(void *addr) |
437 | { | |
438 | struct work_struct *work = addr; | |
439 | ||
440 | return test_bit(WORK_STRUCT_STATIC_BIT, work_data_bits(work)); | |
441 | } | |
442 | ||
dc186ad7 TG |
443 | /* |
444 | * fixup_init is called when: | |
445 | * - an active object is initialized | |
446 | */ | |
02a982a6 | 447 | static bool work_fixup_init(void *addr, enum debug_obj_state state) |
dc186ad7 TG |
448 | { |
449 | struct work_struct *work = addr; | |
450 | ||
451 | switch (state) { | |
452 | case ODEBUG_STATE_ACTIVE: | |
453 | cancel_work_sync(work); | |
454 | debug_object_init(work, &work_debug_descr); | |
02a982a6 | 455 | return true; |
dc186ad7 | 456 | default: |
02a982a6 | 457 | return false; |
dc186ad7 TG |
458 | } |
459 | } | |
460 | ||
dc186ad7 TG |
461 | /* |
462 | * fixup_free is called when: | |
463 | * - an active object is freed | |
464 | */ | |
02a982a6 | 465 | static bool work_fixup_free(void *addr, enum debug_obj_state state) |
dc186ad7 TG |
466 | { |
467 | struct work_struct *work = addr; | |
468 | ||
469 | switch (state) { | |
470 | case ODEBUG_STATE_ACTIVE: | |
471 | cancel_work_sync(work); | |
472 | debug_object_free(work, &work_debug_descr); | |
02a982a6 | 473 | return true; |
dc186ad7 | 474 | default: |
02a982a6 | 475 | return false; |
dc186ad7 TG |
476 | } |
477 | } | |
478 | ||
479 | static struct debug_obj_descr work_debug_descr = { | |
480 | .name = "work_struct", | |
99777288 | 481 | .debug_hint = work_debug_hint, |
b9fdac7f | 482 | .is_static_object = work_is_static_object, |
dc186ad7 | 483 | .fixup_init = work_fixup_init, |
dc186ad7 TG |
484 | .fixup_free = work_fixup_free, |
485 | }; | |
486 | ||
487 | static inline void debug_work_activate(struct work_struct *work) | |
488 | { | |
489 | debug_object_activate(work, &work_debug_descr); | |
490 | } | |
491 | ||
492 | static inline void debug_work_deactivate(struct work_struct *work) | |
493 | { | |
494 | debug_object_deactivate(work, &work_debug_descr); | |
495 | } | |
496 | ||
497 | void __init_work(struct work_struct *work, int onstack) | |
498 | { | |
499 | if (onstack) | |
500 | debug_object_init_on_stack(work, &work_debug_descr); | |
501 | else | |
502 | debug_object_init(work, &work_debug_descr); | |
503 | } | |
504 | EXPORT_SYMBOL_GPL(__init_work); | |
505 | ||
506 | void destroy_work_on_stack(struct work_struct *work) | |
507 | { | |
508 | debug_object_free(work, &work_debug_descr); | |
509 | } | |
510 | EXPORT_SYMBOL_GPL(destroy_work_on_stack); | |
511 | ||
ea2e64f2 TG |
512 | void destroy_delayed_work_on_stack(struct delayed_work *work) |
513 | { | |
514 | destroy_timer_on_stack(&work->timer); | |
515 | debug_object_free(&work->work, &work_debug_descr); | |
516 | } | |
517 | EXPORT_SYMBOL_GPL(destroy_delayed_work_on_stack); | |
518 | ||
dc186ad7 TG |
519 | #else |
520 | static inline void debug_work_activate(struct work_struct *work) { } | |
521 | static inline void debug_work_deactivate(struct work_struct *work) { } | |
522 | #endif | |
523 | ||
4e8b22bd LB |
524 | /** |
525 | * worker_pool_assign_id - allocate ID and assing it to @pool | |
526 | * @pool: the pool pointer of interest | |
527 | * | |
528 | * Returns 0 if ID in [0, WORK_OFFQ_POOL_NONE) is allocated and assigned | |
529 | * successfully, -errno on failure. | |
530 | */ | |
9daf9e67 TH |
531 | static int worker_pool_assign_id(struct worker_pool *pool) |
532 | { | |
533 | int ret; | |
534 | ||
68e13a67 | 535 | lockdep_assert_held(&wq_pool_mutex); |
5bcab335 | 536 | |
4e8b22bd LB |
537 | ret = idr_alloc(&worker_pool_idr, pool, 0, WORK_OFFQ_POOL_NONE, |
538 | GFP_KERNEL); | |
229641a6 | 539 | if (ret >= 0) { |
e68035fb | 540 | pool->id = ret; |
229641a6 TH |
541 | return 0; |
542 | } | |
fa1b54e6 | 543 | return ret; |
7c3eed5c TH |
544 | } |
545 | ||
df2d5ae4 TH |
546 | /** |
547 | * unbound_pwq_by_node - return the unbound pool_workqueue for the given node | |
548 | * @wq: the target workqueue | |
549 | * @node: the node ID | |
550 | * | |
24acfb71 | 551 | * This must be called with any of wq_pool_mutex, wq->mutex or RCU |
5b95e1af | 552 | * read locked. |
df2d5ae4 TH |
553 | * If the pwq needs to be used beyond the locking in effect, the caller is |
554 | * responsible for guaranteeing that the pwq stays online. | |
d185af30 YB |
555 | * |
556 | * Return: The unbound pool_workqueue for @node. | |
df2d5ae4 TH |
557 | */ |
558 | static struct pool_workqueue *unbound_pwq_by_node(struct workqueue_struct *wq, | |
559 | int node) | |
560 | { | |
5b95e1af | 561 | assert_rcu_or_wq_mutex_or_pool_mutex(wq); |
d6e022f1 TH |
562 | |
563 | /* | |
564 | * XXX: @node can be NUMA_NO_NODE if CPU goes offline while a | |
565 | * delayed item is pending. The plan is to keep CPU -> NODE | |
566 | * mapping valid and stable across CPU on/offlines. Once that | |
567 | * happens, this workaround can be removed. | |
568 | */ | |
569 | if (unlikely(node == NUMA_NO_NODE)) | |
570 | return wq->dfl_pwq; | |
571 | ||
df2d5ae4 TH |
572 | return rcu_dereference_raw(wq->numa_pwq_tbl[node]); |
573 | } | |
574 | ||
73f53c4a TH |
575 | static unsigned int work_color_to_flags(int color) |
576 | { | |
577 | return color << WORK_STRUCT_COLOR_SHIFT; | |
578 | } | |
579 | ||
580 | static int get_work_color(struct work_struct *work) | |
581 | { | |
582 | return (*work_data_bits(work) >> WORK_STRUCT_COLOR_SHIFT) & | |
583 | ((1 << WORK_STRUCT_COLOR_BITS) - 1); | |
584 | } | |
585 | ||
586 | static int work_next_color(int color) | |
587 | { | |
588 | return (color + 1) % WORK_NR_COLORS; | |
589 | } | |
1da177e4 | 590 | |
14441960 | 591 | /* |
112202d9 TH |
592 | * While queued, %WORK_STRUCT_PWQ is set and non flag bits of a work's data |
593 | * contain the pointer to the queued pwq. Once execution starts, the flag | |
7c3eed5c | 594 | * is cleared and the high bits contain OFFQ flags and pool ID. |
7a22ad75 | 595 | * |
112202d9 TH |
596 | * set_work_pwq(), set_work_pool_and_clear_pending(), mark_work_canceling() |
597 | * and clear_work_data() can be used to set the pwq, pool or clear | |
bbb68dfa TH |
598 | * work->data. These functions should only be called while the work is |
599 | * owned - ie. while the PENDING bit is set. | |
7a22ad75 | 600 | * |
112202d9 | 601 | * get_work_pool() and get_work_pwq() can be used to obtain the pool or pwq |
7c3eed5c | 602 | * corresponding to a work. Pool is available once the work has been |
112202d9 | 603 | * queued anywhere after initialization until it is sync canceled. pwq is |
7c3eed5c | 604 | * available only while the work item is queued. |
7a22ad75 | 605 | * |
bbb68dfa TH |
606 | * %WORK_OFFQ_CANCELING is used to mark a work item which is being |
607 | * canceled. While being canceled, a work item may have its PENDING set | |
608 | * but stay off timer and worklist for arbitrarily long and nobody should | |
609 | * try to steal the PENDING bit. | |
14441960 | 610 | */ |
7a22ad75 TH |
611 | static inline void set_work_data(struct work_struct *work, unsigned long data, |
612 | unsigned long flags) | |
365970a1 | 613 | { |
6183c009 | 614 | WARN_ON_ONCE(!work_pending(work)); |
7a22ad75 TH |
615 | atomic_long_set(&work->data, data | flags | work_static(work)); |
616 | } | |
365970a1 | 617 | |
112202d9 | 618 | static void set_work_pwq(struct work_struct *work, struct pool_workqueue *pwq, |
7a22ad75 TH |
619 | unsigned long extra_flags) |
620 | { | |
112202d9 TH |
621 | set_work_data(work, (unsigned long)pwq, |
622 | WORK_STRUCT_PENDING | WORK_STRUCT_PWQ | extra_flags); | |
365970a1 DH |
623 | } |
624 | ||
4468a00f LJ |
625 | static void set_work_pool_and_keep_pending(struct work_struct *work, |
626 | int pool_id) | |
627 | { | |
628 | set_work_data(work, (unsigned long)pool_id << WORK_OFFQ_POOL_SHIFT, | |
629 | WORK_STRUCT_PENDING); | |
630 | } | |
631 | ||
7c3eed5c TH |
632 | static void set_work_pool_and_clear_pending(struct work_struct *work, |
633 | int pool_id) | |
7a22ad75 | 634 | { |
23657bb1 TH |
635 | /* |
636 | * The following wmb is paired with the implied mb in | |
637 | * test_and_set_bit(PENDING) and ensures all updates to @work made | |
638 | * here are visible to and precede any updates by the next PENDING | |
639 | * owner. | |
640 | */ | |
641 | smp_wmb(); | |
7c3eed5c | 642 | set_work_data(work, (unsigned long)pool_id << WORK_OFFQ_POOL_SHIFT, 0); |
346c09f8 RP |
643 | /* |
644 | * The following mb guarantees that previous clear of a PENDING bit | |
645 | * will not be reordered with any speculative LOADS or STORES from | |
646 | * work->current_func, which is executed afterwards. This possible | |
8bdc6201 | 647 | * reordering can lead to a missed execution on attempt to queue |
346c09f8 RP |
648 | * the same @work. E.g. consider this case: |
649 | * | |
650 | * CPU#0 CPU#1 | |
651 | * ---------------------------- -------------------------------- | |
652 | * | |
653 | * 1 STORE event_indicated | |
654 | * 2 queue_work_on() { | |
655 | * 3 test_and_set_bit(PENDING) | |
656 | * 4 } set_..._and_clear_pending() { | |
657 | * 5 set_work_data() # clear bit | |
658 | * 6 smp_mb() | |
659 | * 7 work->current_func() { | |
660 | * 8 LOAD event_indicated | |
661 | * } | |
662 | * | |
663 | * Without an explicit full barrier speculative LOAD on line 8 can | |
664 | * be executed before CPU#0 does STORE on line 1. If that happens, | |
665 | * CPU#0 observes the PENDING bit is still set and new execution of | |
666 | * a @work is not queued in a hope, that CPU#1 will eventually | |
667 | * finish the queued @work. Meanwhile CPU#1 does not see | |
668 | * event_indicated is set, because speculative LOAD was executed | |
669 | * before actual STORE. | |
670 | */ | |
671 | smp_mb(); | |
7a22ad75 | 672 | } |
f756d5e2 | 673 | |
7a22ad75 | 674 | static void clear_work_data(struct work_struct *work) |
1da177e4 | 675 | { |
7c3eed5c TH |
676 | smp_wmb(); /* see set_work_pool_and_clear_pending() */ |
677 | set_work_data(work, WORK_STRUCT_NO_POOL, 0); | |
1da177e4 LT |
678 | } |
679 | ||
112202d9 | 680 | static struct pool_workqueue *get_work_pwq(struct work_struct *work) |
b1f4ec17 | 681 | { |
e120153d | 682 | unsigned long data = atomic_long_read(&work->data); |
7a22ad75 | 683 | |
112202d9 | 684 | if (data & WORK_STRUCT_PWQ) |
e120153d TH |
685 | return (void *)(data & WORK_STRUCT_WQ_DATA_MASK); |
686 | else | |
687 | return NULL; | |
4d707b9f ON |
688 | } |
689 | ||
7c3eed5c TH |
690 | /** |
691 | * get_work_pool - return the worker_pool a given work was associated with | |
692 | * @work: the work item of interest | |
693 | * | |
68e13a67 | 694 | * Pools are created and destroyed under wq_pool_mutex, and allows read |
24acfb71 TG |
695 | * access under RCU read lock. As such, this function should be |
696 | * called under wq_pool_mutex or inside of a rcu_read_lock() region. | |
fa1b54e6 TH |
697 | * |
698 | * All fields of the returned pool are accessible as long as the above | |
699 | * mentioned locking is in effect. If the returned pool needs to be used | |
700 | * beyond the critical section, the caller is responsible for ensuring the | |
701 | * returned pool is and stays online. | |
d185af30 YB |
702 | * |
703 | * Return: The worker_pool @work was last associated with. %NULL if none. | |
7c3eed5c TH |
704 | */ |
705 | static struct worker_pool *get_work_pool(struct work_struct *work) | |
365970a1 | 706 | { |
e120153d | 707 | unsigned long data = atomic_long_read(&work->data); |
7c3eed5c | 708 | int pool_id; |
7a22ad75 | 709 | |
68e13a67 | 710 | assert_rcu_or_pool_mutex(); |
fa1b54e6 | 711 | |
112202d9 TH |
712 | if (data & WORK_STRUCT_PWQ) |
713 | return ((struct pool_workqueue *) | |
7c3eed5c | 714 | (data & WORK_STRUCT_WQ_DATA_MASK))->pool; |
7a22ad75 | 715 | |
7c3eed5c TH |
716 | pool_id = data >> WORK_OFFQ_POOL_SHIFT; |
717 | if (pool_id == WORK_OFFQ_POOL_NONE) | |
7a22ad75 TH |
718 | return NULL; |
719 | ||
fa1b54e6 | 720 | return idr_find(&worker_pool_idr, pool_id); |
7c3eed5c TH |
721 | } |
722 | ||
723 | /** | |
724 | * get_work_pool_id - return the worker pool ID a given work is associated with | |
725 | * @work: the work item of interest | |
726 | * | |
d185af30 | 727 | * Return: The worker_pool ID @work was last associated with. |
7c3eed5c TH |
728 | * %WORK_OFFQ_POOL_NONE if none. |
729 | */ | |
730 | static int get_work_pool_id(struct work_struct *work) | |
731 | { | |
54d5b7d0 LJ |
732 | unsigned long data = atomic_long_read(&work->data); |
733 | ||
112202d9 TH |
734 | if (data & WORK_STRUCT_PWQ) |
735 | return ((struct pool_workqueue *) | |
54d5b7d0 | 736 | (data & WORK_STRUCT_WQ_DATA_MASK))->pool->id; |
7c3eed5c | 737 | |
54d5b7d0 | 738 | return data >> WORK_OFFQ_POOL_SHIFT; |
7c3eed5c TH |
739 | } |
740 | ||
bbb68dfa TH |
741 | static void mark_work_canceling(struct work_struct *work) |
742 | { | |
7c3eed5c | 743 | unsigned long pool_id = get_work_pool_id(work); |
bbb68dfa | 744 | |
7c3eed5c TH |
745 | pool_id <<= WORK_OFFQ_POOL_SHIFT; |
746 | set_work_data(work, pool_id | WORK_OFFQ_CANCELING, WORK_STRUCT_PENDING); | |
bbb68dfa TH |
747 | } |
748 | ||
749 | static bool work_is_canceling(struct work_struct *work) | |
750 | { | |
751 | unsigned long data = atomic_long_read(&work->data); | |
752 | ||
112202d9 | 753 | return !(data & WORK_STRUCT_PWQ) && (data & WORK_OFFQ_CANCELING); |
bbb68dfa TH |
754 | } |
755 | ||
e22bee78 | 756 | /* |
3270476a TH |
757 | * Policy functions. These define the policies on how the global worker |
758 | * pools are managed. Unless noted otherwise, these functions assume that | |
d565ed63 | 759 | * they're being called with pool->lock held. |
e22bee78 TH |
760 | */ |
761 | ||
63d95a91 | 762 | static bool __need_more_worker(struct worker_pool *pool) |
a848e3b6 | 763 | { |
e19e397a | 764 | return !atomic_read(&pool->nr_running); |
a848e3b6 ON |
765 | } |
766 | ||
4594bf15 | 767 | /* |
e22bee78 TH |
768 | * Need to wake up a worker? Called from anything but currently |
769 | * running workers. | |
974271c4 TH |
770 | * |
771 | * Note that, because unbound workers never contribute to nr_running, this | |
706026c2 | 772 | * function will always return %true for unbound pools as long as the |
974271c4 | 773 | * worklist isn't empty. |
4594bf15 | 774 | */ |
63d95a91 | 775 | static bool need_more_worker(struct worker_pool *pool) |
365970a1 | 776 | { |
63d95a91 | 777 | return !list_empty(&pool->worklist) && __need_more_worker(pool); |
e22bee78 | 778 | } |
4594bf15 | 779 | |
e22bee78 | 780 | /* Can I start working? Called from busy but !running workers. */ |
63d95a91 | 781 | static bool may_start_working(struct worker_pool *pool) |
e22bee78 | 782 | { |
63d95a91 | 783 | return pool->nr_idle; |
e22bee78 TH |
784 | } |
785 | ||
786 | /* Do I need to keep working? Called from currently running workers. */ | |
63d95a91 | 787 | static bool keep_working(struct worker_pool *pool) |
e22bee78 | 788 | { |
e19e397a TH |
789 | return !list_empty(&pool->worklist) && |
790 | atomic_read(&pool->nr_running) <= 1; | |
e22bee78 TH |
791 | } |
792 | ||
793 | /* Do we need a new worker? Called from manager. */ | |
63d95a91 | 794 | static bool need_to_create_worker(struct worker_pool *pool) |
e22bee78 | 795 | { |
63d95a91 | 796 | return need_more_worker(pool) && !may_start_working(pool); |
e22bee78 | 797 | } |
365970a1 | 798 | |
e22bee78 | 799 | /* Do we have too many workers and should some go away? */ |
63d95a91 | 800 | static bool too_many_workers(struct worker_pool *pool) |
e22bee78 | 801 | { |
692b4825 | 802 | bool managing = pool->flags & POOL_MANAGER_ACTIVE; |
63d95a91 TH |
803 | int nr_idle = pool->nr_idle + managing; /* manager is considered idle */ |
804 | int nr_busy = pool->nr_workers - nr_idle; | |
e22bee78 TH |
805 | |
806 | return nr_idle > 2 && (nr_idle - 2) * MAX_IDLE_WORKERS_RATIO >= nr_busy; | |
365970a1 DH |
807 | } |
808 | ||
4d707b9f | 809 | /* |
e22bee78 TH |
810 | * Wake up functions. |
811 | */ | |
812 | ||
1037de36 LJ |
813 | /* Return the first idle worker. Safe with preemption disabled */ |
814 | static struct worker *first_idle_worker(struct worker_pool *pool) | |
7e11629d | 815 | { |
63d95a91 | 816 | if (unlikely(list_empty(&pool->idle_list))) |
7e11629d TH |
817 | return NULL; |
818 | ||
63d95a91 | 819 | return list_first_entry(&pool->idle_list, struct worker, entry); |
7e11629d TH |
820 | } |
821 | ||
822 | /** | |
823 | * wake_up_worker - wake up an idle worker | |
63d95a91 | 824 | * @pool: worker pool to wake worker from |
7e11629d | 825 | * |
63d95a91 | 826 | * Wake up the first idle worker of @pool. |
7e11629d TH |
827 | * |
828 | * CONTEXT: | |
d565ed63 | 829 | * spin_lock_irq(pool->lock). |
7e11629d | 830 | */ |
63d95a91 | 831 | static void wake_up_worker(struct worker_pool *pool) |
7e11629d | 832 | { |
1037de36 | 833 | struct worker *worker = first_idle_worker(pool); |
7e11629d TH |
834 | |
835 | if (likely(worker)) | |
836 | wake_up_process(worker->task); | |
837 | } | |
838 | ||
d302f017 | 839 | /** |
6d25be57 | 840 | * wq_worker_running - a worker is running again |
e22bee78 | 841 | * @task: task waking up |
e22bee78 | 842 | * |
6d25be57 | 843 | * This function is called when a worker returns from schedule() |
e22bee78 | 844 | */ |
6d25be57 | 845 | void wq_worker_running(struct task_struct *task) |
e22bee78 TH |
846 | { |
847 | struct worker *worker = kthread_data(task); | |
848 | ||
6d25be57 TG |
849 | if (!worker->sleeping) |
850 | return; | |
851 | if (!(worker->flags & WORKER_NOT_RUNNING)) | |
e19e397a | 852 | atomic_inc(&worker->pool->nr_running); |
6d25be57 | 853 | worker->sleeping = 0; |
e22bee78 TH |
854 | } |
855 | ||
856 | /** | |
857 | * wq_worker_sleeping - a worker is going to sleep | |
858 | * @task: task going to sleep | |
e22bee78 | 859 | * |
6d25be57 | 860 | * This function is called from schedule() when a busy worker is |
62849a96 SAS |
861 | * going to sleep. Preemption needs to be disabled to protect ->sleeping |
862 | * assignment. | |
e22bee78 | 863 | */ |
6d25be57 | 864 | void wq_worker_sleeping(struct task_struct *task) |
e22bee78 | 865 | { |
6d25be57 | 866 | struct worker *next, *worker = kthread_data(task); |
111c225a | 867 | struct worker_pool *pool; |
e22bee78 | 868 | |
111c225a TH |
869 | /* |
870 | * Rescuers, which may not have all the fields set up like normal | |
871 | * workers, also reach here, let's not access anything before | |
872 | * checking NOT_RUNNING. | |
873 | */ | |
2d64672e | 874 | if (worker->flags & WORKER_NOT_RUNNING) |
6d25be57 | 875 | return; |
e22bee78 | 876 | |
111c225a | 877 | pool = worker->pool; |
111c225a | 878 | |
62849a96 SAS |
879 | /* Return if preempted before wq_worker_running() was reached */ |
880 | if (worker->sleeping) | |
6d25be57 TG |
881 | return; |
882 | ||
883 | worker->sleeping = 1; | |
884 | spin_lock_irq(&pool->lock); | |
e22bee78 TH |
885 | |
886 | /* | |
887 | * The counterpart of the following dec_and_test, implied mb, | |
888 | * worklist not empty test sequence is in insert_work(). | |
889 | * Please read comment there. | |
890 | * | |
628c78e7 TH |
891 | * NOT_RUNNING is clear. This means that we're bound to and |
892 | * running on the local cpu w/ rq lock held and preemption | |
893 | * disabled, which in turn means that none else could be | |
d565ed63 | 894 | * manipulating idle_list, so dereferencing idle_list without pool |
628c78e7 | 895 | * lock is safe. |
e22bee78 | 896 | */ |
e19e397a | 897 | if (atomic_dec_and_test(&pool->nr_running) && |
6d25be57 TG |
898 | !list_empty(&pool->worklist)) { |
899 | next = first_idle_worker(pool); | |
900 | if (next) | |
901 | wake_up_process(next->task); | |
902 | } | |
903 | spin_unlock_irq(&pool->lock); | |
e22bee78 TH |
904 | } |
905 | ||
1b69ac6b JW |
906 | /** |
907 | * wq_worker_last_func - retrieve worker's last work function | |
8194fe94 | 908 | * @task: Task to retrieve last work function of. |
1b69ac6b JW |
909 | * |
910 | * Determine the last function a worker executed. This is called from | |
911 | * the scheduler to get a worker's last known identity. | |
912 | * | |
913 | * CONTEXT: | |
914 | * spin_lock_irq(rq->lock) | |
915 | * | |
4b047002 JW |
916 | * This function is called during schedule() when a kworker is going |
917 | * to sleep. It's used by psi to identify aggregation workers during | |
918 | * dequeuing, to allow periodic aggregation to shut-off when that | |
919 | * worker is the last task in the system or cgroup to go to sleep. | |
920 | * | |
921 | * As this function doesn't involve any workqueue-related locking, it | |
922 | * only returns stable values when called from inside the scheduler's | |
923 | * queuing and dequeuing paths, when @task, which must be a kworker, | |
924 | * is guaranteed to not be processing any works. | |
925 | * | |
1b69ac6b JW |
926 | * Return: |
927 | * The last work function %current executed as a worker, NULL if it | |
928 | * hasn't executed any work yet. | |
929 | */ | |
930 | work_func_t wq_worker_last_func(struct task_struct *task) | |
931 | { | |
932 | struct worker *worker = kthread_data(task); | |
933 | ||
934 | return worker->last_func; | |
935 | } | |
936 | ||
e22bee78 TH |
937 | /** |
938 | * worker_set_flags - set worker flags and adjust nr_running accordingly | |
cb444766 | 939 | * @worker: self |
d302f017 | 940 | * @flags: flags to set |
d302f017 | 941 | * |
228f1d00 | 942 | * Set @flags in @worker->flags and adjust nr_running accordingly. |
d302f017 | 943 | * |
cb444766 | 944 | * CONTEXT: |
d565ed63 | 945 | * spin_lock_irq(pool->lock) |
d302f017 | 946 | */ |
228f1d00 | 947 | static inline void worker_set_flags(struct worker *worker, unsigned int flags) |
d302f017 | 948 | { |
bd7bdd43 | 949 | struct worker_pool *pool = worker->pool; |
e22bee78 | 950 | |
cb444766 TH |
951 | WARN_ON_ONCE(worker->task != current); |
952 | ||
228f1d00 | 953 | /* If transitioning into NOT_RUNNING, adjust nr_running. */ |
e22bee78 TH |
954 | if ((flags & WORKER_NOT_RUNNING) && |
955 | !(worker->flags & WORKER_NOT_RUNNING)) { | |
228f1d00 | 956 | atomic_dec(&pool->nr_running); |
e22bee78 TH |
957 | } |
958 | ||
d302f017 TH |
959 | worker->flags |= flags; |
960 | } | |
961 | ||
962 | /** | |
e22bee78 | 963 | * worker_clr_flags - clear worker flags and adjust nr_running accordingly |
cb444766 | 964 | * @worker: self |
d302f017 TH |
965 | * @flags: flags to clear |
966 | * | |
e22bee78 | 967 | * Clear @flags in @worker->flags and adjust nr_running accordingly. |
d302f017 | 968 | * |
cb444766 | 969 | * CONTEXT: |
d565ed63 | 970 | * spin_lock_irq(pool->lock) |
d302f017 TH |
971 | */ |
972 | static inline void worker_clr_flags(struct worker *worker, unsigned int flags) | |
973 | { | |
63d95a91 | 974 | struct worker_pool *pool = worker->pool; |
e22bee78 TH |
975 | unsigned int oflags = worker->flags; |
976 | ||
cb444766 TH |
977 | WARN_ON_ONCE(worker->task != current); |
978 | ||
d302f017 | 979 | worker->flags &= ~flags; |
e22bee78 | 980 | |
42c025f3 TH |
981 | /* |
982 | * If transitioning out of NOT_RUNNING, increment nr_running. Note | |
983 | * that the nested NOT_RUNNING is not a noop. NOT_RUNNING is mask | |
984 | * of multiple flags, not a single flag. | |
985 | */ | |
e22bee78 TH |
986 | if ((flags & WORKER_NOT_RUNNING) && (oflags & WORKER_NOT_RUNNING)) |
987 | if (!(worker->flags & WORKER_NOT_RUNNING)) | |
e19e397a | 988 | atomic_inc(&pool->nr_running); |
d302f017 TH |
989 | } |
990 | ||
8cca0eea TH |
991 | /** |
992 | * find_worker_executing_work - find worker which is executing a work | |
c9e7cf27 | 993 | * @pool: pool of interest |
8cca0eea TH |
994 | * @work: work to find worker for |
995 | * | |
c9e7cf27 TH |
996 | * Find a worker which is executing @work on @pool by searching |
997 | * @pool->busy_hash which is keyed by the address of @work. For a worker | |
a2c1c57b TH |
998 | * to match, its current execution should match the address of @work and |
999 | * its work function. This is to avoid unwanted dependency between | |
1000 | * unrelated work executions through a work item being recycled while still | |
1001 | * being executed. | |
1002 | * | |
1003 | * This is a bit tricky. A work item may be freed once its execution | |
1004 | * starts and nothing prevents the freed area from being recycled for | |
1005 | * another work item. If the same work item address ends up being reused | |
1006 | * before the original execution finishes, workqueue will identify the | |
1007 | * recycled work item as currently executing and make it wait until the | |
1008 | * current execution finishes, introducing an unwanted dependency. | |
1009 | * | |
c5aa87bb TH |
1010 | * This function checks the work item address and work function to avoid |
1011 | * false positives. Note that this isn't complete as one may construct a | |
1012 | * work function which can introduce dependency onto itself through a | |
1013 | * recycled work item. Well, if somebody wants to shoot oneself in the | |
1014 | * foot that badly, there's only so much we can do, and if such deadlock | |
1015 | * actually occurs, it should be easy to locate the culprit work function. | |
8cca0eea TH |
1016 | * |
1017 | * CONTEXT: | |
d565ed63 | 1018 | * spin_lock_irq(pool->lock). |
8cca0eea | 1019 | * |
d185af30 YB |
1020 | * Return: |
1021 | * Pointer to worker which is executing @work if found, %NULL | |
8cca0eea | 1022 | * otherwise. |
4d707b9f | 1023 | */ |
c9e7cf27 | 1024 | static struct worker *find_worker_executing_work(struct worker_pool *pool, |
8cca0eea | 1025 | struct work_struct *work) |
4d707b9f | 1026 | { |
42f8570f | 1027 | struct worker *worker; |
42f8570f | 1028 | |
b67bfe0d | 1029 | hash_for_each_possible(pool->busy_hash, worker, hentry, |
a2c1c57b TH |
1030 | (unsigned long)work) |
1031 | if (worker->current_work == work && | |
1032 | worker->current_func == work->func) | |
42f8570f SL |
1033 | return worker; |
1034 | ||
1035 | return NULL; | |
4d707b9f ON |
1036 | } |
1037 | ||
bf4ede01 TH |
1038 | /** |
1039 | * move_linked_works - move linked works to a list | |
1040 | * @work: start of series of works to be scheduled | |
1041 | * @head: target list to append @work to | |
402dd89d | 1042 | * @nextp: out parameter for nested worklist walking |
bf4ede01 TH |
1043 | * |
1044 | * Schedule linked works starting from @work to @head. Work series to | |
1045 | * be scheduled starts at @work and includes any consecutive work with | |
1046 | * WORK_STRUCT_LINKED set in its predecessor. | |
1047 | * | |
1048 | * If @nextp is not NULL, it's updated to point to the next work of | |
1049 | * the last scheduled work. This allows move_linked_works() to be | |
1050 | * nested inside outer list_for_each_entry_safe(). | |
1051 | * | |
1052 | * CONTEXT: | |
d565ed63 | 1053 | * spin_lock_irq(pool->lock). |
bf4ede01 TH |
1054 | */ |
1055 | static void move_linked_works(struct work_struct *work, struct list_head *head, | |
1056 | struct work_struct **nextp) | |
1057 | { | |
1058 | struct work_struct *n; | |
1059 | ||
1060 | /* | |
1061 | * Linked worklist will always end before the end of the list, | |
1062 | * use NULL for list head. | |
1063 | */ | |
1064 | list_for_each_entry_safe_from(work, n, NULL, entry) { | |
1065 | list_move_tail(&work->entry, head); | |
1066 | if (!(*work_data_bits(work) & WORK_STRUCT_LINKED)) | |
1067 | break; | |
1068 | } | |
1069 | ||
1070 | /* | |
1071 | * If we're already inside safe list traversal and have moved | |
1072 | * multiple works to the scheduled queue, the next position | |
1073 | * needs to be updated. | |
1074 | */ | |
1075 | if (nextp) | |
1076 | *nextp = n; | |
1077 | } | |
1078 | ||
8864b4e5 TH |
1079 | /** |
1080 | * get_pwq - get an extra reference on the specified pool_workqueue | |
1081 | * @pwq: pool_workqueue to get | |
1082 | * | |
1083 | * Obtain an extra reference on @pwq. The caller should guarantee that | |
1084 | * @pwq has positive refcnt and be holding the matching pool->lock. | |
1085 | */ | |
1086 | static void get_pwq(struct pool_workqueue *pwq) | |
1087 | { | |
1088 | lockdep_assert_held(&pwq->pool->lock); | |
1089 | WARN_ON_ONCE(pwq->refcnt <= 0); | |
1090 | pwq->refcnt++; | |
1091 | } | |
1092 | ||
1093 | /** | |
1094 | * put_pwq - put a pool_workqueue reference | |
1095 | * @pwq: pool_workqueue to put | |
1096 | * | |
1097 | * Drop a reference of @pwq. If its refcnt reaches zero, schedule its | |
1098 | * destruction. The caller should be holding the matching pool->lock. | |
1099 | */ | |
1100 | static void put_pwq(struct pool_workqueue *pwq) | |
1101 | { | |
1102 | lockdep_assert_held(&pwq->pool->lock); | |
1103 | if (likely(--pwq->refcnt)) | |
1104 | return; | |
1105 | if (WARN_ON_ONCE(!(pwq->wq->flags & WQ_UNBOUND))) | |
1106 | return; | |
1107 | /* | |
1108 | * @pwq can't be released under pool->lock, bounce to | |
1109 | * pwq_unbound_release_workfn(). This never recurses on the same | |
1110 | * pool->lock as this path is taken only for unbound workqueues and | |
1111 | * the release work item is scheduled on a per-cpu workqueue. To | |
1112 | * avoid lockdep warning, unbound pool->locks are given lockdep | |
1113 | * subclass of 1 in get_unbound_pool(). | |
1114 | */ | |
1115 | schedule_work(&pwq->unbound_release_work); | |
1116 | } | |
1117 | ||
dce90d47 TH |
1118 | /** |
1119 | * put_pwq_unlocked - put_pwq() with surrounding pool lock/unlock | |
1120 | * @pwq: pool_workqueue to put (can be %NULL) | |
1121 | * | |
1122 | * put_pwq() with locking. This function also allows %NULL @pwq. | |
1123 | */ | |
1124 | static void put_pwq_unlocked(struct pool_workqueue *pwq) | |
1125 | { | |
1126 | if (pwq) { | |
1127 | /* | |
24acfb71 | 1128 | * As both pwqs and pools are RCU protected, the |
dce90d47 TH |
1129 | * following lock operations are safe. |
1130 | */ | |
1131 | spin_lock_irq(&pwq->pool->lock); | |
1132 | put_pwq(pwq); | |
1133 | spin_unlock_irq(&pwq->pool->lock); | |
1134 | } | |
1135 | } | |
1136 | ||
112202d9 | 1137 | static void pwq_activate_delayed_work(struct work_struct *work) |
bf4ede01 | 1138 | { |
112202d9 | 1139 | struct pool_workqueue *pwq = get_work_pwq(work); |
bf4ede01 TH |
1140 | |
1141 | trace_workqueue_activate_work(work); | |
82607adc TH |
1142 | if (list_empty(&pwq->pool->worklist)) |
1143 | pwq->pool->watchdog_ts = jiffies; | |
112202d9 | 1144 | move_linked_works(work, &pwq->pool->worklist, NULL); |
bf4ede01 | 1145 | __clear_bit(WORK_STRUCT_DELAYED_BIT, work_data_bits(work)); |
112202d9 | 1146 | pwq->nr_active++; |
bf4ede01 TH |
1147 | } |
1148 | ||
112202d9 | 1149 | static void pwq_activate_first_delayed(struct pool_workqueue *pwq) |
3aa62497 | 1150 | { |
112202d9 | 1151 | struct work_struct *work = list_first_entry(&pwq->delayed_works, |
3aa62497 LJ |
1152 | struct work_struct, entry); |
1153 | ||
112202d9 | 1154 | pwq_activate_delayed_work(work); |
3aa62497 LJ |
1155 | } |
1156 | ||
bf4ede01 | 1157 | /** |
112202d9 TH |
1158 | * pwq_dec_nr_in_flight - decrement pwq's nr_in_flight |
1159 | * @pwq: pwq of interest | |
bf4ede01 | 1160 | * @color: color of work which left the queue |
bf4ede01 TH |
1161 | * |
1162 | * A work either has completed or is removed from pending queue, | |
112202d9 | 1163 | * decrement nr_in_flight of its pwq and handle workqueue flushing. |
bf4ede01 TH |
1164 | * |
1165 | * CONTEXT: | |
d565ed63 | 1166 | * spin_lock_irq(pool->lock). |
bf4ede01 | 1167 | */ |
112202d9 | 1168 | static void pwq_dec_nr_in_flight(struct pool_workqueue *pwq, int color) |
bf4ede01 | 1169 | { |
8864b4e5 | 1170 | /* uncolored work items don't participate in flushing or nr_active */ |
bf4ede01 | 1171 | if (color == WORK_NO_COLOR) |
8864b4e5 | 1172 | goto out_put; |
bf4ede01 | 1173 | |
112202d9 | 1174 | pwq->nr_in_flight[color]--; |
bf4ede01 | 1175 | |
112202d9 TH |
1176 | pwq->nr_active--; |
1177 | if (!list_empty(&pwq->delayed_works)) { | |
b3f9f405 | 1178 | /* one down, submit a delayed one */ |
112202d9 TH |
1179 | if (pwq->nr_active < pwq->max_active) |
1180 | pwq_activate_first_delayed(pwq); | |
bf4ede01 TH |
1181 | } |
1182 | ||
1183 | /* is flush in progress and are we at the flushing tip? */ | |
112202d9 | 1184 | if (likely(pwq->flush_color != color)) |
8864b4e5 | 1185 | goto out_put; |
bf4ede01 TH |
1186 | |
1187 | /* are there still in-flight works? */ | |
112202d9 | 1188 | if (pwq->nr_in_flight[color]) |
8864b4e5 | 1189 | goto out_put; |
bf4ede01 | 1190 | |
112202d9 TH |
1191 | /* this pwq is done, clear flush_color */ |
1192 | pwq->flush_color = -1; | |
bf4ede01 TH |
1193 | |
1194 | /* | |
112202d9 | 1195 | * If this was the last pwq, wake up the first flusher. It |
bf4ede01 TH |
1196 | * will handle the rest. |
1197 | */ | |
112202d9 TH |
1198 | if (atomic_dec_and_test(&pwq->wq->nr_pwqs_to_flush)) |
1199 | complete(&pwq->wq->first_flusher->done); | |
8864b4e5 TH |
1200 | out_put: |
1201 | put_pwq(pwq); | |
bf4ede01 TH |
1202 | } |
1203 | ||
36e227d2 | 1204 | /** |
bbb68dfa | 1205 | * try_to_grab_pending - steal work item from worklist and disable irq |
36e227d2 TH |
1206 | * @work: work item to steal |
1207 | * @is_dwork: @work is a delayed_work | |
bbb68dfa | 1208 | * @flags: place to store irq state |
36e227d2 TH |
1209 | * |
1210 | * Try to grab PENDING bit of @work. This function can handle @work in any | |
d185af30 | 1211 | * stable state - idle, on timer or on worklist. |
36e227d2 | 1212 | * |
d185af30 | 1213 | * Return: |
36e227d2 TH |
1214 | * 1 if @work was pending and we successfully stole PENDING |
1215 | * 0 if @work was idle and we claimed PENDING | |
1216 | * -EAGAIN if PENDING couldn't be grabbed at the moment, safe to busy-retry | |
bbb68dfa TH |
1217 | * -ENOENT if someone else is canceling @work, this state may persist |
1218 | * for arbitrarily long | |
36e227d2 | 1219 | * |
d185af30 | 1220 | * Note: |
bbb68dfa | 1221 | * On >= 0 return, the caller owns @work's PENDING bit. To avoid getting |
e0aecdd8 TH |
1222 | * interrupted while holding PENDING and @work off queue, irq must be |
1223 | * disabled on entry. This, combined with delayed_work->timer being | |
1224 | * irqsafe, ensures that we return -EAGAIN for finite short period of time. | |
bbb68dfa TH |
1225 | * |
1226 | * On successful return, >= 0, irq is disabled and the caller is | |
1227 | * responsible for releasing it using local_irq_restore(*@flags). | |
1228 | * | |
e0aecdd8 | 1229 | * This function is safe to call from any context including IRQ handler. |
bf4ede01 | 1230 | */ |
bbb68dfa TH |
1231 | static int try_to_grab_pending(struct work_struct *work, bool is_dwork, |
1232 | unsigned long *flags) | |
bf4ede01 | 1233 | { |
d565ed63 | 1234 | struct worker_pool *pool; |
112202d9 | 1235 | struct pool_workqueue *pwq; |
bf4ede01 | 1236 | |
bbb68dfa TH |
1237 | local_irq_save(*flags); |
1238 | ||
36e227d2 TH |
1239 | /* try to steal the timer if it exists */ |
1240 | if (is_dwork) { | |
1241 | struct delayed_work *dwork = to_delayed_work(work); | |
1242 | ||
e0aecdd8 TH |
1243 | /* |
1244 | * dwork->timer is irqsafe. If del_timer() fails, it's | |
1245 | * guaranteed that the timer is not queued anywhere and not | |
1246 | * running on the local CPU. | |
1247 | */ | |
36e227d2 TH |
1248 | if (likely(del_timer(&dwork->timer))) |
1249 | return 1; | |
1250 | } | |
1251 | ||
1252 | /* try to claim PENDING the normal way */ | |
bf4ede01 TH |
1253 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) |
1254 | return 0; | |
1255 | ||
24acfb71 | 1256 | rcu_read_lock(); |
bf4ede01 TH |
1257 | /* |
1258 | * The queueing is in progress, or it is already queued. Try to | |
1259 | * steal it from ->worklist without clearing WORK_STRUCT_PENDING. | |
1260 | */ | |
d565ed63 TH |
1261 | pool = get_work_pool(work); |
1262 | if (!pool) | |
bbb68dfa | 1263 | goto fail; |
bf4ede01 | 1264 | |
d565ed63 | 1265 | spin_lock(&pool->lock); |
0b3dae68 | 1266 | /* |
112202d9 TH |
1267 | * work->data is guaranteed to point to pwq only while the work |
1268 | * item is queued on pwq->wq, and both updating work->data to point | |
1269 | * to pwq on queueing and to pool on dequeueing are done under | |
1270 | * pwq->pool->lock. This in turn guarantees that, if work->data | |
1271 | * points to pwq which is associated with a locked pool, the work | |
0b3dae68 LJ |
1272 | * item is currently queued on that pool. |
1273 | */ | |
112202d9 TH |
1274 | pwq = get_work_pwq(work); |
1275 | if (pwq && pwq->pool == pool) { | |
16062836 TH |
1276 | debug_work_deactivate(work); |
1277 | ||
1278 | /* | |
1279 | * A delayed work item cannot be grabbed directly because | |
1280 | * it might have linked NO_COLOR work items which, if left | |
112202d9 | 1281 | * on the delayed_list, will confuse pwq->nr_active |
16062836 TH |
1282 | * management later on and cause stall. Make sure the work |
1283 | * item is activated before grabbing. | |
1284 | */ | |
1285 | if (*work_data_bits(work) & WORK_STRUCT_DELAYED) | |
112202d9 | 1286 | pwq_activate_delayed_work(work); |
16062836 TH |
1287 | |
1288 | list_del_init(&work->entry); | |
9c34a704 | 1289 | pwq_dec_nr_in_flight(pwq, get_work_color(work)); |
16062836 | 1290 | |
112202d9 | 1291 | /* work->data points to pwq iff queued, point to pool */ |
16062836 TH |
1292 | set_work_pool_and_keep_pending(work, pool->id); |
1293 | ||
1294 | spin_unlock(&pool->lock); | |
24acfb71 | 1295 | rcu_read_unlock(); |
16062836 | 1296 | return 1; |
bf4ede01 | 1297 | } |
d565ed63 | 1298 | spin_unlock(&pool->lock); |
bbb68dfa | 1299 | fail: |
24acfb71 | 1300 | rcu_read_unlock(); |
bbb68dfa TH |
1301 | local_irq_restore(*flags); |
1302 | if (work_is_canceling(work)) | |
1303 | return -ENOENT; | |
1304 | cpu_relax(); | |
36e227d2 | 1305 | return -EAGAIN; |
bf4ede01 TH |
1306 | } |
1307 | ||
4690c4ab | 1308 | /** |
706026c2 | 1309 | * insert_work - insert a work into a pool |
112202d9 | 1310 | * @pwq: pwq @work belongs to |
4690c4ab TH |
1311 | * @work: work to insert |
1312 | * @head: insertion point | |
1313 | * @extra_flags: extra WORK_STRUCT_* flags to set | |
1314 | * | |
112202d9 | 1315 | * Insert @work which belongs to @pwq after @head. @extra_flags is or'd to |
706026c2 | 1316 | * work_struct flags. |
4690c4ab TH |
1317 | * |
1318 | * CONTEXT: | |
d565ed63 | 1319 | * spin_lock_irq(pool->lock). |
4690c4ab | 1320 | */ |
112202d9 TH |
1321 | static void insert_work(struct pool_workqueue *pwq, struct work_struct *work, |
1322 | struct list_head *head, unsigned int extra_flags) | |
b89deed3 | 1323 | { |
112202d9 | 1324 | struct worker_pool *pool = pwq->pool; |
e22bee78 | 1325 | |
4690c4ab | 1326 | /* we own @work, set data and link */ |
112202d9 | 1327 | set_work_pwq(work, pwq, extra_flags); |
1a4d9b0a | 1328 | list_add_tail(&work->entry, head); |
8864b4e5 | 1329 | get_pwq(pwq); |
e22bee78 TH |
1330 | |
1331 | /* | |
c5aa87bb TH |
1332 | * Ensure either wq_worker_sleeping() sees the above |
1333 | * list_add_tail() or we see zero nr_running to avoid workers lying | |
1334 | * around lazily while there are works to be processed. | |
e22bee78 TH |
1335 | */ |
1336 | smp_mb(); | |
1337 | ||
63d95a91 TH |
1338 | if (__need_more_worker(pool)) |
1339 | wake_up_worker(pool); | |
b89deed3 ON |
1340 | } |
1341 | ||
c8efcc25 TH |
1342 | /* |
1343 | * Test whether @work is being queued from another work executing on the | |
8d03ecfe | 1344 | * same workqueue. |
c8efcc25 TH |
1345 | */ |
1346 | static bool is_chained_work(struct workqueue_struct *wq) | |
1347 | { | |
8d03ecfe TH |
1348 | struct worker *worker; |
1349 | ||
1350 | worker = current_wq_worker(); | |
1351 | /* | |
bf393fd4 | 1352 | * Return %true iff I'm a worker executing a work item on @wq. If |
8d03ecfe TH |
1353 | * I'm @worker, it's safe to dereference it without locking. |
1354 | */ | |
112202d9 | 1355 | return worker && worker->current_pwq->wq == wq; |
c8efcc25 TH |
1356 | } |
1357 | ||
ef557180 MG |
1358 | /* |
1359 | * When queueing an unbound work item to a wq, prefer local CPU if allowed | |
1360 | * by wq_unbound_cpumask. Otherwise, round robin among the allowed ones to | |
1361 | * avoid perturbing sensitive tasks. | |
1362 | */ | |
1363 | static int wq_select_unbound_cpu(int cpu) | |
1364 | { | |
f303fccb | 1365 | static bool printed_dbg_warning; |
ef557180 MG |
1366 | int new_cpu; |
1367 | ||
f303fccb TH |
1368 | if (likely(!wq_debug_force_rr_cpu)) { |
1369 | if (cpumask_test_cpu(cpu, wq_unbound_cpumask)) | |
1370 | return cpu; | |
1371 | } else if (!printed_dbg_warning) { | |
1372 | pr_warn("workqueue: round-robin CPU selection forced, expect performance impact\n"); | |
1373 | printed_dbg_warning = true; | |
1374 | } | |
1375 | ||
ef557180 MG |
1376 | if (cpumask_empty(wq_unbound_cpumask)) |
1377 | return cpu; | |
1378 | ||
1379 | new_cpu = __this_cpu_read(wq_rr_cpu_last); | |
1380 | new_cpu = cpumask_next_and(new_cpu, wq_unbound_cpumask, cpu_online_mask); | |
1381 | if (unlikely(new_cpu >= nr_cpu_ids)) { | |
1382 | new_cpu = cpumask_first_and(wq_unbound_cpumask, cpu_online_mask); | |
1383 | if (unlikely(new_cpu >= nr_cpu_ids)) | |
1384 | return cpu; | |
1385 | } | |
1386 | __this_cpu_write(wq_rr_cpu_last, new_cpu); | |
1387 | ||
1388 | return new_cpu; | |
1389 | } | |
1390 | ||
d84ff051 | 1391 | static void __queue_work(int cpu, struct workqueue_struct *wq, |
1da177e4 LT |
1392 | struct work_struct *work) |
1393 | { | |
112202d9 | 1394 | struct pool_workqueue *pwq; |
c9178087 | 1395 | struct worker_pool *last_pool; |
1e19ffc6 | 1396 | struct list_head *worklist; |
8a2e8e5d | 1397 | unsigned int work_flags; |
b75cac93 | 1398 | unsigned int req_cpu = cpu; |
8930caba TH |
1399 | |
1400 | /* | |
1401 | * While a work item is PENDING && off queue, a task trying to | |
1402 | * steal the PENDING will busy-loop waiting for it to either get | |
1403 | * queued or lose PENDING. Grabbing PENDING and queueing should | |
1404 | * happen with IRQ disabled. | |
1405 | */ | |
8e8eb730 | 1406 | lockdep_assert_irqs_disabled(); |
1da177e4 | 1407 | |
dc186ad7 | 1408 | debug_work_activate(work); |
1e19ffc6 | 1409 | |
9ef28a73 | 1410 | /* if draining, only works from the same workqueue are allowed */ |
618b01eb | 1411 | if (unlikely(wq->flags & __WQ_DRAINING) && |
c8efcc25 | 1412 | WARN_ON_ONCE(!is_chained_work(wq))) |
e41e704b | 1413 | return; |
24acfb71 | 1414 | rcu_read_lock(); |
9e8cd2f5 | 1415 | retry: |
c9178087 | 1416 | /* pwq which will be used unless @work is executing elsewhere */ |
aa202f1f HD |
1417 | if (wq->flags & WQ_UNBOUND) { |
1418 | if (req_cpu == WORK_CPU_UNBOUND) | |
1419 | cpu = wq_select_unbound_cpu(raw_smp_processor_id()); | |
df2d5ae4 | 1420 | pwq = unbound_pwq_by_node(wq, cpu_to_node(cpu)); |
aa202f1f HD |
1421 | } else { |
1422 | if (req_cpu == WORK_CPU_UNBOUND) | |
1423 | cpu = raw_smp_processor_id(); | |
1424 | pwq = per_cpu_ptr(wq->cpu_pwqs, cpu); | |
1425 | } | |
dbf2576e | 1426 | |
c9178087 TH |
1427 | /* |
1428 | * If @work was previously on a different pool, it might still be | |
1429 | * running there, in which case the work needs to be queued on that | |
1430 | * pool to guarantee non-reentrancy. | |
1431 | */ | |
1432 | last_pool = get_work_pool(work); | |
1433 | if (last_pool && last_pool != pwq->pool) { | |
1434 | struct worker *worker; | |
18aa9eff | 1435 | |
c9178087 | 1436 | spin_lock(&last_pool->lock); |
18aa9eff | 1437 | |
c9178087 | 1438 | worker = find_worker_executing_work(last_pool, work); |
18aa9eff | 1439 | |
c9178087 TH |
1440 | if (worker && worker->current_pwq->wq == wq) { |
1441 | pwq = worker->current_pwq; | |
8930caba | 1442 | } else { |
c9178087 TH |
1443 | /* meh... not running there, queue here */ |
1444 | spin_unlock(&last_pool->lock); | |
112202d9 | 1445 | spin_lock(&pwq->pool->lock); |
8930caba | 1446 | } |
f3421797 | 1447 | } else { |
112202d9 | 1448 | spin_lock(&pwq->pool->lock); |
502ca9d8 TH |
1449 | } |
1450 | ||
9e8cd2f5 TH |
1451 | /* |
1452 | * pwq is determined and locked. For unbound pools, we could have | |
1453 | * raced with pwq release and it could already be dead. If its | |
1454 | * refcnt is zero, repeat pwq selection. Note that pwqs never die | |
df2d5ae4 TH |
1455 | * without another pwq replacing it in the numa_pwq_tbl or while |
1456 | * work items are executing on it, so the retrying is guaranteed to | |
9e8cd2f5 TH |
1457 | * make forward-progress. |
1458 | */ | |
1459 | if (unlikely(!pwq->refcnt)) { | |
1460 | if (wq->flags & WQ_UNBOUND) { | |
1461 | spin_unlock(&pwq->pool->lock); | |
1462 | cpu_relax(); | |
1463 | goto retry; | |
1464 | } | |
1465 | /* oops */ | |
1466 | WARN_ONCE(true, "workqueue: per-cpu pwq for %s on cpu%d has 0 refcnt", | |
1467 | wq->name, cpu); | |
1468 | } | |
1469 | ||
112202d9 TH |
1470 | /* pwq determined, queue */ |
1471 | trace_workqueue_queue_work(req_cpu, pwq, work); | |
502ca9d8 | 1472 | |
24acfb71 TG |
1473 | if (WARN_ON(!list_empty(&work->entry))) |
1474 | goto out; | |
1e19ffc6 | 1475 | |
112202d9 TH |
1476 | pwq->nr_in_flight[pwq->work_color]++; |
1477 | work_flags = work_color_to_flags(pwq->work_color); | |
1e19ffc6 | 1478 | |
112202d9 | 1479 | if (likely(pwq->nr_active < pwq->max_active)) { |
cdadf009 | 1480 | trace_workqueue_activate_work(work); |
112202d9 TH |
1481 | pwq->nr_active++; |
1482 | worklist = &pwq->pool->worklist; | |
82607adc TH |
1483 | if (list_empty(worklist)) |
1484 | pwq->pool->watchdog_ts = jiffies; | |
8a2e8e5d TH |
1485 | } else { |
1486 | work_flags |= WORK_STRUCT_DELAYED; | |
112202d9 | 1487 | worklist = &pwq->delayed_works; |
8a2e8e5d | 1488 | } |
1e19ffc6 | 1489 | |
112202d9 | 1490 | insert_work(pwq, work, worklist, work_flags); |
1e19ffc6 | 1491 | |
24acfb71 | 1492 | out: |
112202d9 | 1493 | spin_unlock(&pwq->pool->lock); |
24acfb71 | 1494 | rcu_read_unlock(); |
1da177e4 LT |
1495 | } |
1496 | ||
0fcb78c2 | 1497 | /** |
c1a220e7 ZR |
1498 | * queue_work_on - queue work on specific cpu |
1499 | * @cpu: CPU number to execute work on | |
0fcb78c2 REB |
1500 | * @wq: workqueue to use |
1501 | * @work: work to queue | |
1502 | * | |
c1a220e7 ZR |
1503 | * We queue the work to a specific CPU, the caller must ensure it |
1504 | * can't go away. | |
d185af30 YB |
1505 | * |
1506 | * Return: %false if @work was already on a queue, %true otherwise. | |
1da177e4 | 1507 | */ |
d4283e93 TH |
1508 | bool queue_work_on(int cpu, struct workqueue_struct *wq, |
1509 | struct work_struct *work) | |
1da177e4 | 1510 | { |
d4283e93 | 1511 | bool ret = false; |
8930caba | 1512 | unsigned long flags; |
ef1ca236 | 1513 | |
8930caba | 1514 | local_irq_save(flags); |
c1a220e7 | 1515 | |
22df02bb | 1516 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
4690c4ab | 1517 | __queue_work(cpu, wq, work); |
d4283e93 | 1518 | ret = true; |
c1a220e7 | 1519 | } |
ef1ca236 | 1520 | |
8930caba | 1521 | local_irq_restore(flags); |
1da177e4 LT |
1522 | return ret; |
1523 | } | |
ad7b1f84 | 1524 | EXPORT_SYMBOL(queue_work_on); |
1da177e4 | 1525 | |
8204e0c1 AD |
1526 | /** |
1527 | * workqueue_select_cpu_near - Select a CPU based on NUMA node | |
1528 | * @node: NUMA node ID that we want to select a CPU from | |
1529 | * | |
1530 | * This function will attempt to find a "random" cpu available on a given | |
1531 | * node. If there are no CPUs available on the given node it will return | |
1532 | * WORK_CPU_UNBOUND indicating that we should just schedule to any | |
1533 | * available CPU if we need to schedule this work. | |
1534 | */ | |
1535 | static int workqueue_select_cpu_near(int node) | |
1536 | { | |
1537 | int cpu; | |
1538 | ||
1539 | /* No point in doing this if NUMA isn't enabled for workqueues */ | |
1540 | if (!wq_numa_enabled) | |
1541 | return WORK_CPU_UNBOUND; | |
1542 | ||
1543 | /* Delay binding to CPU if node is not valid or online */ | |
1544 | if (node < 0 || node >= MAX_NUMNODES || !node_online(node)) | |
1545 | return WORK_CPU_UNBOUND; | |
1546 | ||
1547 | /* Use local node/cpu if we are already there */ | |
1548 | cpu = raw_smp_processor_id(); | |
1549 | if (node == cpu_to_node(cpu)) | |
1550 | return cpu; | |
1551 | ||
1552 | /* Use "random" otherwise know as "first" online CPU of node */ | |
1553 | cpu = cpumask_any_and(cpumask_of_node(node), cpu_online_mask); | |
1554 | ||
1555 | /* If CPU is valid return that, otherwise just defer */ | |
1556 | return cpu < nr_cpu_ids ? cpu : WORK_CPU_UNBOUND; | |
1557 | } | |
1558 | ||
1559 | /** | |
1560 | * queue_work_node - queue work on a "random" cpu for a given NUMA node | |
1561 | * @node: NUMA node that we are targeting the work for | |
1562 | * @wq: workqueue to use | |
1563 | * @work: work to queue | |
1564 | * | |
1565 | * We queue the work to a "random" CPU within a given NUMA node. The basic | |
1566 | * idea here is to provide a way to somehow associate work with a given | |
1567 | * NUMA node. | |
1568 | * | |
1569 | * This function will only make a best effort attempt at getting this onto | |
1570 | * the right NUMA node. If no node is requested or the requested node is | |
1571 | * offline then we just fall back to standard queue_work behavior. | |
1572 | * | |
1573 | * Currently the "random" CPU ends up being the first available CPU in the | |
1574 | * intersection of cpu_online_mask and the cpumask of the node, unless we | |
1575 | * are running on the node. In that case we just use the current CPU. | |
1576 | * | |
1577 | * Return: %false if @work was already on a queue, %true otherwise. | |
1578 | */ | |
1579 | bool queue_work_node(int node, struct workqueue_struct *wq, | |
1580 | struct work_struct *work) | |
1581 | { | |
1582 | unsigned long flags; | |
1583 | bool ret = false; | |
1584 | ||
1585 | /* | |
1586 | * This current implementation is specific to unbound workqueues. | |
1587 | * Specifically we only return the first available CPU for a given | |
1588 | * node instead of cycling through individual CPUs within the node. | |
1589 | * | |
1590 | * If this is used with a per-cpu workqueue then the logic in | |
1591 | * workqueue_select_cpu_near would need to be updated to allow for | |
1592 | * some round robin type logic. | |
1593 | */ | |
1594 | WARN_ON_ONCE(!(wq->flags & WQ_UNBOUND)); | |
1595 | ||
1596 | local_irq_save(flags); | |
1597 | ||
1598 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { | |
1599 | int cpu = workqueue_select_cpu_near(node); | |
1600 | ||
1601 | __queue_work(cpu, wq, work); | |
1602 | ret = true; | |
1603 | } | |
1604 | ||
1605 | local_irq_restore(flags); | |
1606 | return ret; | |
1607 | } | |
1608 | EXPORT_SYMBOL_GPL(queue_work_node); | |
1609 | ||
8c20feb6 | 1610 | void delayed_work_timer_fn(struct timer_list *t) |
1da177e4 | 1611 | { |
8c20feb6 | 1612 | struct delayed_work *dwork = from_timer(dwork, t, timer); |
1da177e4 | 1613 | |
e0aecdd8 | 1614 | /* should have been called from irqsafe timer with irq already off */ |
60c057bc | 1615 | __queue_work(dwork->cpu, dwork->wq, &dwork->work); |
1da177e4 | 1616 | } |
1438ade5 | 1617 | EXPORT_SYMBOL(delayed_work_timer_fn); |
1da177e4 | 1618 | |
7beb2edf TH |
1619 | static void __queue_delayed_work(int cpu, struct workqueue_struct *wq, |
1620 | struct delayed_work *dwork, unsigned long delay) | |
1da177e4 | 1621 | { |
7beb2edf TH |
1622 | struct timer_list *timer = &dwork->timer; |
1623 | struct work_struct *work = &dwork->work; | |
7beb2edf | 1624 | |
637fdbae | 1625 | WARN_ON_ONCE(!wq); |
841b86f3 | 1626 | WARN_ON_ONCE(timer->function != delayed_work_timer_fn); |
fc4b514f TH |
1627 | WARN_ON_ONCE(timer_pending(timer)); |
1628 | WARN_ON_ONCE(!list_empty(&work->entry)); | |
7beb2edf | 1629 | |
8852aac2 TH |
1630 | /* |
1631 | * If @delay is 0, queue @dwork->work immediately. This is for | |
1632 | * both optimization and correctness. The earliest @timer can | |
1633 | * expire is on the closest next tick and delayed_work users depend | |
1634 | * on that there's no such delay when @delay is 0. | |
1635 | */ | |
1636 | if (!delay) { | |
1637 | __queue_work(cpu, wq, &dwork->work); | |
1638 | return; | |
1639 | } | |
1640 | ||
60c057bc | 1641 | dwork->wq = wq; |
1265057f | 1642 | dwork->cpu = cpu; |
7beb2edf TH |
1643 | timer->expires = jiffies + delay; |
1644 | ||
041bd12e TH |
1645 | if (unlikely(cpu != WORK_CPU_UNBOUND)) |
1646 | add_timer_on(timer, cpu); | |
1647 | else | |
1648 | add_timer(timer); | |
1da177e4 LT |
1649 | } |
1650 | ||
0fcb78c2 REB |
1651 | /** |
1652 | * queue_delayed_work_on - queue work on specific CPU after delay | |
1653 | * @cpu: CPU number to execute work on | |
1654 | * @wq: workqueue to use | |
af9997e4 | 1655 | * @dwork: work to queue |
0fcb78c2 REB |
1656 | * @delay: number of jiffies to wait before queueing |
1657 | * | |
d185af30 | 1658 | * Return: %false if @work was already on a queue, %true otherwise. If |
715f1300 TH |
1659 | * @delay is zero and @dwork is idle, it will be scheduled for immediate |
1660 | * execution. | |
0fcb78c2 | 1661 | */ |
d4283e93 TH |
1662 | bool queue_delayed_work_on(int cpu, struct workqueue_struct *wq, |
1663 | struct delayed_work *dwork, unsigned long delay) | |
7a6bc1cd | 1664 | { |
52bad64d | 1665 | struct work_struct *work = &dwork->work; |
d4283e93 | 1666 | bool ret = false; |
8930caba | 1667 | unsigned long flags; |
7a6bc1cd | 1668 | |
8930caba TH |
1669 | /* read the comment in __queue_work() */ |
1670 | local_irq_save(flags); | |
7a6bc1cd | 1671 | |
22df02bb | 1672 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
7beb2edf | 1673 | __queue_delayed_work(cpu, wq, dwork, delay); |
d4283e93 | 1674 | ret = true; |
7a6bc1cd | 1675 | } |
8a3e77cc | 1676 | |
8930caba | 1677 | local_irq_restore(flags); |
7a6bc1cd VP |
1678 | return ret; |
1679 | } | |
ad7b1f84 | 1680 | EXPORT_SYMBOL(queue_delayed_work_on); |
c7fc77f7 | 1681 | |
8376fe22 TH |
1682 | /** |
1683 | * mod_delayed_work_on - modify delay of or queue a delayed work on specific CPU | |
1684 | * @cpu: CPU number to execute work on | |
1685 | * @wq: workqueue to use | |
1686 | * @dwork: work to queue | |
1687 | * @delay: number of jiffies to wait before queueing | |
1688 | * | |
1689 | * If @dwork is idle, equivalent to queue_delayed_work_on(); otherwise, | |
1690 | * modify @dwork's timer so that it expires after @delay. If @delay is | |
1691 | * zero, @work is guaranteed to be scheduled immediately regardless of its | |
1692 | * current state. | |
1693 | * | |
d185af30 | 1694 | * Return: %false if @dwork was idle and queued, %true if @dwork was |
8376fe22 TH |
1695 | * pending and its timer was modified. |
1696 | * | |
e0aecdd8 | 1697 | * This function is safe to call from any context including IRQ handler. |
8376fe22 TH |
1698 | * See try_to_grab_pending() for details. |
1699 | */ | |
1700 | bool mod_delayed_work_on(int cpu, struct workqueue_struct *wq, | |
1701 | struct delayed_work *dwork, unsigned long delay) | |
1702 | { | |
1703 | unsigned long flags; | |
1704 | int ret; | |
c7fc77f7 | 1705 | |
8376fe22 TH |
1706 | do { |
1707 | ret = try_to_grab_pending(&dwork->work, true, &flags); | |
1708 | } while (unlikely(ret == -EAGAIN)); | |
63bc0362 | 1709 | |
8376fe22 TH |
1710 | if (likely(ret >= 0)) { |
1711 | __queue_delayed_work(cpu, wq, dwork, delay); | |
1712 | local_irq_restore(flags); | |
7a6bc1cd | 1713 | } |
8376fe22 TH |
1714 | |
1715 | /* -ENOENT from try_to_grab_pending() becomes %true */ | |
7a6bc1cd VP |
1716 | return ret; |
1717 | } | |
8376fe22 TH |
1718 | EXPORT_SYMBOL_GPL(mod_delayed_work_on); |
1719 | ||
05f0fe6b TH |
1720 | static void rcu_work_rcufn(struct rcu_head *rcu) |
1721 | { | |
1722 | struct rcu_work *rwork = container_of(rcu, struct rcu_work, rcu); | |
1723 | ||
1724 | /* read the comment in __queue_work() */ | |
1725 | local_irq_disable(); | |
1726 | __queue_work(WORK_CPU_UNBOUND, rwork->wq, &rwork->work); | |
1727 | local_irq_enable(); | |
1728 | } | |
1729 | ||
1730 | /** | |
1731 | * queue_rcu_work - queue work after a RCU grace period | |
1732 | * @wq: workqueue to use | |
1733 | * @rwork: work to queue | |
1734 | * | |
1735 | * Return: %false if @rwork was already pending, %true otherwise. Note | |
1736 | * that a full RCU grace period is guaranteed only after a %true return. | |
bf393fd4 | 1737 | * While @rwork is guaranteed to be executed after a %false return, the |
05f0fe6b TH |
1738 | * execution may happen before a full RCU grace period has passed. |
1739 | */ | |
1740 | bool queue_rcu_work(struct workqueue_struct *wq, struct rcu_work *rwork) | |
1741 | { | |
1742 | struct work_struct *work = &rwork->work; | |
1743 | ||
1744 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { | |
1745 | rwork->wq = wq; | |
1746 | call_rcu(&rwork->rcu, rcu_work_rcufn); | |
1747 | return true; | |
1748 | } | |
1749 | ||
1750 | return false; | |
1751 | } | |
1752 | EXPORT_SYMBOL(queue_rcu_work); | |
1753 | ||
c8e55f36 TH |
1754 | /** |
1755 | * worker_enter_idle - enter idle state | |
1756 | * @worker: worker which is entering idle state | |
1757 | * | |
1758 | * @worker is entering idle state. Update stats and idle timer if | |
1759 | * necessary. | |
1760 | * | |
1761 | * LOCKING: | |
d565ed63 | 1762 | * spin_lock_irq(pool->lock). |
c8e55f36 TH |
1763 | */ |
1764 | static void worker_enter_idle(struct worker *worker) | |
1da177e4 | 1765 | { |
bd7bdd43 | 1766 | struct worker_pool *pool = worker->pool; |
c8e55f36 | 1767 | |
6183c009 TH |
1768 | if (WARN_ON_ONCE(worker->flags & WORKER_IDLE) || |
1769 | WARN_ON_ONCE(!list_empty(&worker->entry) && | |
1770 | (worker->hentry.next || worker->hentry.pprev))) | |
1771 | return; | |
c8e55f36 | 1772 | |
051e1850 | 1773 | /* can't use worker_set_flags(), also called from create_worker() */ |
cb444766 | 1774 | worker->flags |= WORKER_IDLE; |
bd7bdd43 | 1775 | pool->nr_idle++; |
e22bee78 | 1776 | worker->last_active = jiffies; |
c8e55f36 TH |
1777 | |
1778 | /* idle_list is LIFO */ | |
bd7bdd43 | 1779 | list_add(&worker->entry, &pool->idle_list); |
db7bccf4 | 1780 | |
628c78e7 TH |
1781 | if (too_many_workers(pool) && !timer_pending(&pool->idle_timer)) |
1782 | mod_timer(&pool->idle_timer, jiffies + IDLE_WORKER_TIMEOUT); | |
cb444766 | 1783 | |
544ecf31 | 1784 | /* |
e8b3f8db | 1785 | * Sanity check nr_running. Because unbind_workers() releases |
d565ed63 | 1786 | * pool->lock between setting %WORKER_UNBOUND and zapping |
628c78e7 TH |
1787 | * nr_running, the warning may trigger spuriously. Check iff |
1788 | * unbind is not in progress. | |
544ecf31 | 1789 | */ |
24647570 | 1790 | WARN_ON_ONCE(!(pool->flags & POOL_DISASSOCIATED) && |
bd7bdd43 | 1791 | pool->nr_workers == pool->nr_idle && |
e19e397a | 1792 | atomic_read(&pool->nr_running)); |
c8e55f36 TH |
1793 | } |
1794 | ||
1795 | /** | |
1796 | * worker_leave_idle - leave idle state | |
1797 | * @worker: worker which is leaving idle state | |
1798 | * | |
1799 | * @worker is leaving idle state. Update stats. | |
1800 | * | |
1801 | * LOCKING: | |
d565ed63 | 1802 | * spin_lock_irq(pool->lock). |
c8e55f36 TH |
1803 | */ |
1804 | static void worker_leave_idle(struct worker *worker) | |
1805 | { | |
bd7bdd43 | 1806 | struct worker_pool *pool = worker->pool; |
c8e55f36 | 1807 | |
6183c009 TH |
1808 | if (WARN_ON_ONCE(!(worker->flags & WORKER_IDLE))) |
1809 | return; | |
d302f017 | 1810 | worker_clr_flags(worker, WORKER_IDLE); |
bd7bdd43 | 1811 | pool->nr_idle--; |
c8e55f36 TH |
1812 | list_del_init(&worker->entry); |
1813 | } | |
1814 | ||
f7537df5 | 1815 | static struct worker *alloc_worker(int node) |
c34056a3 TH |
1816 | { |
1817 | struct worker *worker; | |
1818 | ||
f7537df5 | 1819 | worker = kzalloc_node(sizeof(*worker), GFP_KERNEL, node); |
c8e55f36 TH |
1820 | if (worker) { |
1821 | INIT_LIST_HEAD(&worker->entry); | |
affee4b2 | 1822 | INIT_LIST_HEAD(&worker->scheduled); |
da028469 | 1823 | INIT_LIST_HEAD(&worker->node); |
e22bee78 TH |
1824 | /* on creation a worker is in !idle && prep state */ |
1825 | worker->flags = WORKER_PREP; | |
c8e55f36 | 1826 | } |
c34056a3 TH |
1827 | return worker; |
1828 | } | |
1829 | ||
4736cbf7 LJ |
1830 | /** |
1831 | * worker_attach_to_pool() - attach a worker to a pool | |
1832 | * @worker: worker to be attached | |
1833 | * @pool: the target pool | |
1834 | * | |
1835 | * Attach @worker to @pool. Once attached, the %WORKER_UNBOUND flag and | |
1836 | * cpu-binding of @worker are kept coordinated with the pool across | |
1837 | * cpu-[un]hotplugs. | |
1838 | */ | |
1839 | static void worker_attach_to_pool(struct worker *worker, | |
1840 | struct worker_pool *pool) | |
1841 | { | |
1258fae7 | 1842 | mutex_lock(&wq_pool_attach_mutex); |
4736cbf7 LJ |
1843 | |
1844 | /* | |
1845 | * set_cpus_allowed_ptr() will fail if the cpumask doesn't have any | |
1846 | * online CPUs. It'll be re-applied when any of the CPUs come up. | |
1847 | */ | |
1848 | set_cpus_allowed_ptr(worker->task, pool->attrs->cpumask); | |
1849 | ||
1850 | /* | |
1258fae7 TH |
1851 | * The wq_pool_attach_mutex ensures %POOL_DISASSOCIATED remains |
1852 | * stable across this function. See the comments above the flag | |
1853 | * definition for details. | |
4736cbf7 LJ |
1854 | */ |
1855 | if (pool->flags & POOL_DISASSOCIATED) | |
1856 | worker->flags |= WORKER_UNBOUND; | |
1857 | ||
1858 | list_add_tail(&worker->node, &pool->workers); | |
a2d812a2 | 1859 | worker->pool = pool; |
4736cbf7 | 1860 | |
1258fae7 | 1861 | mutex_unlock(&wq_pool_attach_mutex); |
4736cbf7 LJ |
1862 | } |
1863 | ||
60f5a4bc LJ |
1864 | /** |
1865 | * worker_detach_from_pool() - detach a worker from its pool | |
1866 | * @worker: worker which is attached to its pool | |
60f5a4bc | 1867 | * |
4736cbf7 LJ |
1868 | * Undo the attaching which had been done in worker_attach_to_pool(). The |
1869 | * caller worker shouldn't access to the pool after detached except it has | |
1870 | * other reference to the pool. | |
60f5a4bc | 1871 | */ |
a2d812a2 | 1872 | static void worker_detach_from_pool(struct worker *worker) |
60f5a4bc | 1873 | { |
a2d812a2 | 1874 | struct worker_pool *pool = worker->pool; |
60f5a4bc LJ |
1875 | struct completion *detach_completion = NULL; |
1876 | ||
1258fae7 | 1877 | mutex_lock(&wq_pool_attach_mutex); |
a2d812a2 | 1878 | |
da028469 | 1879 | list_del(&worker->node); |
a2d812a2 TH |
1880 | worker->pool = NULL; |
1881 | ||
da028469 | 1882 | if (list_empty(&pool->workers)) |
60f5a4bc | 1883 | detach_completion = pool->detach_completion; |
1258fae7 | 1884 | mutex_unlock(&wq_pool_attach_mutex); |
60f5a4bc | 1885 | |
b62c0751 LJ |
1886 | /* clear leftover flags without pool->lock after it is detached */ |
1887 | worker->flags &= ~(WORKER_UNBOUND | WORKER_REBOUND); | |
1888 | ||
60f5a4bc LJ |
1889 | if (detach_completion) |
1890 | complete(detach_completion); | |
1891 | } | |
1892 | ||
c34056a3 TH |
1893 | /** |
1894 | * create_worker - create a new workqueue worker | |
63d95a91 | 1895 | * @pool: pool the new worker will belong to |
c34056a3 | 1896 | * |
051e1850 | 1897 | * Create and start a new worker which is attached to @pool. |
c34056a3 TH |
1898 | * |
1899 | * CONTEXT: | |
1900 | * Might sleep. Does GFP_KERNEL allocations. | |
1901 | * | |
d185af30 | 1902 | * Return: |
c34056a3 TH |
1903 | * Pointer to the newly created worker. |
1904 | */ | |
bc2ae0f5 | 1905 | static struct worker *create_worker(struct worker_pool *pool) |
c34056a3 | 1906 | { |
c34056a3 | 1907 | struct worker *worker = NULL; |
f3421797 | 1908 | int id = -1; |
e3c916a4 | 1909 | char id_buf[16]; |
c34056a3 | 1910 | |
7cda9aae LJ |
1911 | /* ID is needed to determine kthread name */ |
1912 | id = ida_simple_get(&pool->worker_ida, 0, 0, GFP_KERNEL); | |
822d8405 TH |
1913 | if (id < 0) |
1914 | goto fail; | |
c34056a3 | 1915 | |
f7537df5 | 1916 | worker = alloc_worker(pool->node); |
c34056a3 TH |
1917 | if (!worker) |
1918 | goto fail; | |
1919 | ||
c34056a3 TH |
1920 | worker->id = id; |
1921 | ||
29c91e99 | 1922 | if (pool->cpu >= 0) |
e3c916a4 TH |
1923 | snprintf(id_buf, sizeof(id_buf), "%d:%d%s", pool->cpu, id, |
1924 | pool->attrs->nice < 0 ? "H" : ""); | |
f3421797 | 1925 | else |
e3c916a4 TH |
1926 | snprintf(id_buf, sizeof(id_buf), "u%d:%d", pool->id, id); |
1927 | ||
f3f90ad4 | 1928 | worker->task = kthread_create_on_node(worker_thread, worker, pool->node, |
e3c916a4 | 1929 | "kworker/%s", id_buf); |
c34056a3 TH |
1930 | if (IS_ERR(worker->task)) |
1931 | goto fail; | |
1932 | ||
91151228 | 1933 | set_user_nice(worker->task, pool->attrs->nice); |
25834c73 | 1934 | kthread_bind_mask(worker->task, pool->attrs->cpumask); |
91151228 | 1935 | |
da028469 | 1936 | /* successful, attach the worker to the pool */ |
4736cbf7 | 1937 | worker_attach_to_pool(worker, pool); |
822d8405 | 1938 | |
051e1850 LJ |
1939 | /* start the newly created worker */ |
1940 | spin_lock_irq(&pool->lock); | |
1941 | worker->pool->nr_workers++; | |
1942 | worker_enter_idle(worker); | |
1943 | wake_up_process(worker->task); | |
1944 | spin_unlock_irq(&pool->lock); | |
1945 | ||
c34056a3 | 1946 | return worker; |
822d8405 | 1947 | |
c34056a3 | 1948 | fail: |
9625ab17 | 1949 | if (id >= 0) |
7cda9aae | 1950 | ida_simple_remove(&pool->worker_ida, id); |
c34056a3 TH |
1951 | kfree(worker); |
1952 | return NULL; | |
1953 | } | |
1954 | ||
c34056a3 TH |
1955 | /** |
1956 | * destroy_worker - destroy a workqueue worker | |
1957 | * @worker: worker to be destroyed | |
1958 | * | |
73eb7fe7 LJ |
1959 | * Destroy @worker and adjust @pool stats accordingly. The worker should |
1960 | * be idle. | |
c8e55f36 TH |
1961 | * |
1962 | * CONTEXT: | |
60f5a4bc | 1963 | * spin_lock_irq(pool->lock). |
c34056a3 TH |
1964 | */ |
1965 | static void destroy_worker(struct worker *worker) | |
1966 | { | |
bd7bdd43 | 1967 | struct worker_pool *pool = worker->pool; |
c34056a3 | 1968 | |
cd549687 TH |
1969 | lockdep_assert_held(&pool->lock); |
1970 | ||
c34056a3 | 1971 | /* sanity check frenzy */ |
6183c009 | 1972 | if (WARN_ON(worker->current_work) || |
73eb7fe7 LJ |
1973 | WARN_ON(!list_empty(&worker->scheduled)) || |
1974 | WARN_ON(!(worker->flags & WORKER_IDLE))) | |
6183c009 | 1975 | return; |
c34056a3 | 1976 | |
73eb7fe7 LJ |
1977 | pool->nr_workers--; |
1978 | pool->nr_idle--; | |
5bdfff96 | 1979 | |
c8e55f36 | 1980 | list_del_init(&worker->entry); |
cb444766 | 1981 | worker->flags |= WORKER_DIE; |
60f5a4bc | 1982 | wake_up_process(worker->task); |
c34056a3 TH |
1983 | } |
1984 | ||
32a6c723 | 1985 | static void idle_worker_timeout(struct timer_list *t) |
e22bee78 | 1986 | { |
32a6c723 | 1987 | struct worker_pool *pool = from_timer(pool, t, idle_timer); |
e22bee78 | 1988 | |
d565ed63 | 1989 | spin_lock_irq(&pool->lock); |
e22bee78 | 1990 | |
3347fc9f | 1991 | while (too_many_workers(pool)) { |
e22bee78 TH |
1992 | struct worker *worker; |
1993 | unsigned long expires; | |
1994 | ||
1995 | /* idle_list is kept in LIFO order, check the last one */ | |
63d95a91 | 1996 | worker = list_entry(pool->idle_list.prev, struct worker, entry); |
e22bee78 TH |
1997 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; |
1998 | ||
3347fc9f | 1999 | if (time_before(jiffies, expires)) { |
63d95a91 | 2000 | mod_timer(&pool->idle_timer, expires); |
3347fc9f | 2001 | break; |
d5abe669 | 2002 | } |
3347fc9f LJ |
2003 | |
2004 | destroy_worker(worker); | |
e22bee78 TH |
2005 | } |
2006 | ||
d565ed63 | 2007 | spin_unlock_irq(&pool->lock); |
e22bee78 | 2008 | } |
d5abe669 | 2009 | |
493a1724 | 2010 | static void send_mayday(struct work_struct *work) |
e22bee78 | 2011 | { |
112202d9 TH |
2012 | struct pool_workqueue *pwq = get_work_pwq(work); |
2013 | struct workqueue_struct *wq = pwq->wq; | |
493a1724 | 2014 | |
2e109a28 | 2015 | lockdep_assert_held(&wq_mayday_lock); |
e22bee78 | 2016 | |
493008a8 | 2017 | if (!wq->rescuer) |
493a1724 | 2018 | return; |
e22bee78 TH |
2019 | |
2020 | /* mayday mayday mayday */ | |
493a1724 | 2021 | if (list_empty(&pwq->mayday_node)) { |
77668c8b LJ |
2022 | /* |
2023 | * If @pwq is for an unbound wq, its base ref may be put at | |
2024 | * any time due to an attribute change. Pin @pwq until the | |
2025 | * rescuer is done with it. | |
2026 | */ | |
2027 | get_pwq(pwq); | |
493a1724 | 2028 | list_add_tail(&pwq->mayday_node, &wq->maydays); |
e22bee78 | 2029 | wake_up_process(wq->rescuer->task); |
493a1724 | 2030 | } |
e22bee78 TH |
2031 | } |
2032 | ||
32a6c723 | 2033 | static void pool_mayday_timeout(struct timer_list *t) |
e22bee78 | 2034 | { |
32a6c723 | 2035 | struct worker_pool *pool = from_timer(pool, t, mayday_timer); |
e22bee78 TH |
2036 | struct work_struct *work; |
2037 | ||
b2d82909 TH |
2038 | spin_lock_irq(&pool->lock); |
2039 | spin_lock(&wq_mayday_lock); /* for wq->maydays */ | |
e22bee78 | 2040 | |
63d95a91 | 2041 | if (need_to_create_worker(pool)) { |
e22bee78 TH |
2042 | /* |
2043 | * We've been trying to create a new worker but | |
2044 | * haven't been successful. We might be hitting an | |
2045 | * allocation deadlock. Send distress signals to | |
2046 | * rescuers. | |
2047 | */ | |
63d95a91 | 2048 | list_for_each_entry(work, &pool->worklist, entry) |
e22bee78 | 2049 | send_mayday(work); |
1da177e4 | 2050 | } |
e22bee78 | 2051 | |
b2d82909 TH |
2052 | spin_unlock(&wq_mayday_lock); |
2053 | spin_unlock_irq(&pool->lock); | |
e22bee78 | 2054 | |
63d95a91 | 2055 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INTERVAL); |
1da177e4 LT |
2056 | } |
2057 | ||
e22bee78 TH |
2058 | /** |
2059 | * maybe_create_worker - create a new worker if necessary | |
63d95a91 | 2060 | * @pool: pool to create a new worker for |
e22bee78 | 2061 | * |
63d95a91 | 2062 | * Create a new worker for @pool if necessary. @pool is guaranteed to |
e22bee78 TH |
2063 | * have at least one idle worker on return from this function. If |
2064 | * creating a new worker takes longer than MAYDAY_INTERVAL, mayday is | |
63d95a91 | 2065 | * sent to all rescuers with works scheduled on @pool to resolve |
e22bee78 TH |
2066 | * possible allocation deadlock. |
2067 | * | |
c5aa87bb TH |
2068 | * On return, need_to_create_worker() is guaranteed to be %false and |
2069 | * may_start_working() %true. | |
e22bee78 TH |
2070 | * |
2071 | * LOCKING: | |
d565ed63 | 2072 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
e22bee78 TH |
2073 | * multiple times. Does GFP_KERNEL allocations. Called only from |
2074 | * manager. | |
e22bee78 | 2075 | */ |
29187a9e | 2076 | static void maybe_create_worker(struct worker_pool *pool) |
d565ed63 TH |
2077 | __releases(&pool->lock) |
2078 | __acquires(&pool->lock) | |
1da177e4 | 2079 | { |
e22bee78 | 2080 | restart: |
d565ed63 | 2081 | spin_unlock_irq(&pool->lock); |
9f9c2364 | 2082 | |
e22bee78 | 2083 | /* if we don't make progress in MAYDAY_INITIAL_TIMEOUT, call for help */ |
63d95a91 | 2084 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INITIAL_TIMEOUT); |
e22bee78 TH |
2085 | |
2086 | while (true) { | |
051e1850 | 2087 | if (create_worker(pool) || !need_to_create_worker(pool)) |
e22bee78 | 2088 | break; |
1da177e4 | 2089 | |
e212f361 | 2090 | schedule_timeout_interruptible(CREATE_COOLDOWN); |
9f9c2364 | 2091 | |
63d95a91 | 2092 | if (!need_to_create_worker(pool)) |
e22bee78 TH |
2093 | break; |
2094 | } | |
2095 | ||
63d95a91 | 2096 | del_timer_sync(&pool->mayday_timer); |
d565ed63 | 2097 | spin_lock_irq(&pool->lock); |
051e1850 LJ |
2098 | /* |
2099 | * This is necessary even after a new worker was just successfully | |
2100 | * created as @pool->lock was dropped and the new worker might have | |
2101 | * already become busy. | |
2102 | */ | |
63d95a91 | 2103 | if (need_to_create_worker(pool)) |
e22bee78 | 2104 | goto restart; |
e22bee78 TH |
2105 | } |
2106 | ||
73f53c4a | 2107 | /** |
e22bee78 TH |
2108 | * manage_workers - manage worker pool |
2109 | * @worker: self | |
73f53c4a | 2110 | * |
706026c2 | 2111 | * Assume the manager role and manage the worker pool @worker belongs |
e22bee78 | 2112 | * to. At any given time, there can be only zero or one manager per |
706026c2 | 2113 | * pool. The exclusion is handled automatically by this function. |
e22bee78 TH |
2114 | * |
2115 | * The caller can safely start processing works on false return. On | |
2116 | * true return, it's guaranteed that need_to_create_worker() is false | |
2117 | * and may_start_working() is true. | |
73f53c4a TH |
2118 | * |
2119 | * CONTEXT: | |
d565ed63 | 2120 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
e22bee78 TH |
2121 | * multiple times. Does GFP_KERNEL allocations. |
2122 | * | |
d185af30 | 2123 | * Return: |
29187a9e TH |
2124 | * %false if the pool doesn't need management and the caller can safely |
2125 | * start processing works, %true if management function was performed and | |
2126 | * the conditions that the caller verified before calling the function may | |
2127 | * no longer be true. | |
73f53c4a | 2128 | */ |
e22bee78 | 2129 | static bool manage_workers(struct worker *worker) |
73f53c4a | 2130 | { |
63d95a91 | 2131 | struct worker_pool *pool = worker->pool; |
73f53c4a | 2132 | |
692b4825 | 2133 | if (pool->flags & POOL_MANAGER_ACTIVE) |
29187a9e | 2134 | return false; |
692b4825 TH |
2135 | |
2136 | pool->flags |= POOL_MANAGER_ACTIVE; | |
2607d7a6 | 2137 | pool->manager = worker; |
1e19ffc6 | 2138 | |
29187a9e | 2139 | maybe_create_worker(pool); |
e22bee78 | 2140 | |
2607d7a6 | 2141 | pool->manager = NULL; |
692b4825 TH |
2142 | pool->flags &= ~POOL_MANAGER_ACTIVE; |
2143 | wake_up(&wq_manager_wait); | |
29187a9e | 2144 | return true; |
73f53c4a TH |
2145 | } |
2146 | ||
a62428c0 TH |
2147 | /** |
2148 | * process_one_work - process single work | |
c34056a3 | 2149 | * @worker: self |
a62428c0 TH |
2150 | * @work: work to process |
2151 | * | |
2152 | * Process @work. This function contains all the logics necessary to | |
2153 | * process a single work including synchronization against and | |
2154 | * interaction with other workers on the same cpu, queueing and | |
2155 | * flushing. As long as context requirement is met, any worker can | |
2156 | * call this function to process a work. | |
2157 | * | |
2158 | * CONTEXT: | |
d565ed63 | 2159 | * spin_lock_irq(pool->lock) which is released and regrabbed. |
a62428c0 | 2160 | */ |
c34056a3 | 2161 | static void process_one_work(struct worker *worker, struct work_struct *work) |
d565ed63 TH |
2162 | __releases(&pool->lock) |
2163 | __acquires(&pool->lock) | |
a62428c0 | 2164 | { |
112202d9 | 2165 | struct pool_workqueue *pwq = get_work_pwq(work); |
bd7bdd43 | 2166 | struct worker_pool *pool = worker->pool; |
112202d9 | 2167 | bool cpu_intensive = pwq->wq->flags & WQ_CPU_INTENSIVE; |
73f53c4a | 2168 | int work_color; |
7e11629d | 2169 | struct worker *collision; |
a62428c0 TH |
2170 | #ifdef CONFIG_LOCKDEP |
2171 | /* | |
2172 | * It is permissible to free the struct work_struct from | |
2173 | * inside the function that is called from it, this we need to | |
2174 | * take into account for lockdep too. To avoid bogus "held | |
2175 | * lock freed" warnings as well as problems when looking into | |
2176 | * work->lockdep_map, make a copy and use that here. | |
2177 | */ | |
4d82a1de PZ |
2178 | struct lockdep_map lockdep_map; |
2179 | ||
2180 | lockdep_copy_map(&lockdep_map, &work->lockdep_map); | |
a62428c0 | 2181 | #endif |
807407c0 | 2182 | /* ensure we're on the correct CPU */ |
85327af6 | 2183 | WARN_ON_ONCE(!(pool->flags & POOL_DISASSOCIATED) && |
ec22ca5e | 2184 | raw_smp_processor_id() != pool->cpu); |
25511a47 | 2185 | |
7e11629d TH |
2186 | /* |
2187 | * A single work shouldn't be executed concurrently by | |
2188 | * multiple workers on a single cpu. Check whether anyone is | |
2189 | * already processing the work. If so, defer the work to the | |
2190 | * currently executing one. | |
2191 | */ | |
c9e7cf27 | 2192 | collision = find_worker_executing_work(pool, work); |
7e11629d TH |
2193 | if (unlikely(collision)) { |
2194 | move_linked_works(work, &collision->scheduled, NULL); | |
2195 | return; | |
2196 | } | |
2197 | ||
8930caba | 2198 | /* claim and dequeue */ |
a62428c0 | 2199 | debug_work_deactivate(work); |
c9e7cf27 | 2200 | hash_add(pool->busy_hash, &worker->hentry, (unsigned long)work); |
c34056a3 | 2201 | worker->current_work = work; |
a2c1c57b | 2202 | worker->current_func = work->func; |
112202d9 | 2203 | worker->current_pwq = pwq; |
73f53c4a | 2204 | work_color = get_work_color(work); |
7a22ad75 | 2205 | |
8bf89593 TH |
2206 | /* |
2207 | * Record wq name for cmdline and debug reporting, may get | |
2208 | * overridden through set_worker_desc(). | |
2209 | */ | |
2210 | strscpy(worker->desc, pwq->wq->name, WORKER_DESC_LEN); | |
2211 | ||
a62428c0 TH |
2212 | list_del_init(&work->entry); |
2213 | ||
fb0e7beb | 2214 | /* |
228f1d00 LJ |
2215 | * CPU intensive works don't participate in concurrency management. |
2216 | * They're the scheduler's responsibility. This takes @worker out | |
2217 | * of concurrency management and the next code block will chain | |
2218 | * execution of the pending work items. | |
fb0e7beb TH |
2219 | */ |
2220 | if (unlikely(cpu_intensive)) | |
228f1d00 | 2221 | worker_set_flags(worker, WORKER_CPU_INTENSIVE); |
fb0e7beb | 2222 | |
974271c4 | 2223 | /* |
a489a03e LJ |
2224 | * Wake up another worker if necessary. The condition is always |
2225 | * false for normal per-cpu workers since nr_running would always | |
2226 | * be >= 1 at this point. This is used to chain execution of the | |
2227 | * pending work items for WORKER_NOT_RUNNING workers such as the | |
228f1d00 | 2228 | * UNBOUND and CPU_INTENSIVE ones. |
974271c4 | 2229 | */ |
a489a03e | 2230 | if (need_more_worker(pool)) |
63d95a91 | 2231 | wake_up_worker(pool); |
974271c4 | 2232 | |
8930caba | 2233 | /* |
7c3eed5c | 2234 | * Record the last pool and clear PENDING which should be the last |
d565ed63 | 2235 | * update to @work. Also, do this inside @pool->lock so that |
23657bb1 TH |
2236 | * PENDING and queued state changes happen together while IRQ is |
2237 | * disabled. | |
8930caba | 2238 | */ |
7c3eed5c | 2239 | set_work_pool_and_clear_pending(work, pool->id); |
a62428c0 | 2240 | |
d565ed63 | 2241 | spin_unlock_irq(&pool->lock); |
a62428c0 | 2242 | |
a1d14934 | 2243 | lock_map_acquire(&pwq->wq->lockdep_map); |
a62428c0 | 2244 | lock_map_acquire(&lockdep_map); |
e6f3faa7 | 2245 | /* |
f52be570 PZ |
2246 | * Strictly speaking we should mark the invariant state without holding |
2247 | * any locks, that is, before these two lock_map_acquire()'s. | |
e6f3faa7 PZ |
2248 | * |
2249 | * However, that would result in: | |
2250 | * | |
2251 | * A(W1) | |
2252 | * WFC(C) | |
2253 | * A(W1) | |
2254 | * C(C) | |
2255 | * | |
2256 | * Which would create W1->C->W1 dependencies, even though there is no | |
2257 | * actual deadlock possible. There are two solutions, using a | |
2258 | * read-recursive acquire on the work(queue) 'locks', but this will then | |
f52be570 | 2259 | * hit the lockdep limitation on recursive locks, or simply discard |
e6f3faa7 PZ |
2260 | * these locks. |
2261 | * | |
2262 | * AFAICT there is no possible deadlock scenario between the | |
2263 | * flush_work() and complete() primitives (except for single-threaded | |
2264 | * workqueues), so hiding them isn't a problem. | |
2265 | */ | |
f52be570 | 2266 | lockdep_invariant_state(true); |
e36c886a | 2267 | trace_workqueue_execute_start(work); |
a2c1c57b | 2268 | worker->current_func(work); |
e36c886a AV |
2269 | /* |
2270 | * While we must be careful to not use "work" after this, the trace | |
2271 | * point will only record its address. | |
2272 | */ | |
1c5da0ec | 2273 | trace_workqueue_execute_end(work, worker->current_func); |
a62428c0 | 2274 | lock_map_release(&lockdep_map); |
112202d9 | 2275 | lock_map_release(&pwq->wq->lockdep_map); |
a62428c0 TH |
2276 | |
2277 | if (unlikely(in_atomic() || lockdep_depth(current) > 0)) { | |
044c782c | 2278 | pr_err("BUG: workqueue leaked lock or atomic: %s/0x%08x/%d\n" |
d75f773c | 2279 | " last function: %ps\n", |
a2c1c57b TH |
2280 | current->comm, preempt_count(), task_pid_nr(current), |
2281 | worker->current_func); | |
a62428c0 TH |
2282 | debug_show_held_locks(current); |
2283 | dump_stack(); | |
2284 | } | |
2285 | ||
b22ce278 | 2286 | /* |
025f50f3 | 2287 | * The following prevents a kworker from hogging CPU on !PREEMPTION |
b22ce278 TH |
2288 | * kernels, where a requeueing work item waiting for something to |
2289 | * happen could deadlock with stop_machine as such work item could | |
2290 | * indefinitely requeue itself while all other CPUs are trapped in | |
789cbbec JL |
2291 | * stop_machine. At the same time, report a quiescent RCU state so |
2292 | * the same condition doesn't freeze RCU. | |
b22ce278 | 2293 | */ |
a7e6425e | 2294 | cond_resched(); |
b22ce278 | 2295 | |
d565ed63 | 2296 | spin_lock_irq(&pool->lock); |
a62428c0 | 2297 | |
fb0e7beb TH |
2298 | /* clear cpu intensive status */ |
2299 | if (unlikely(cpu_intensive)) | |
2300 | worker_clr_flags(worker, WORKER_CPU_INTENSIVE); | |
2301 | ||
1b69ac6b JW |
2302 | /* tag the worker for identification in schedule() */ |
2303 | worker->last_func = worker->current_func; | |
2304 | ||
a62428c0 | 2305 | /* we're done with it, release */ |
42f8570f | 2306 | hash_del(&worker->hentry); |
c34056a3 | 2307 | worker->current_work = NULL; |
a2c1c57b | 2308 | worker->current_func = NULL; |
112202d9 TH |
2309 | worker->current_pwq = NULL; |
2310 | pwq_dec_nr_in_flight(pwq, work_color); | |
a62428c0 TH |
2311 | } |
2312 | ||
affee4b2 TH |
2313 | /** |
2314 | * process_scheduled_works - process scheduled works | |
2315 | * @worker: self | |
2316 | * | |
2317 | * Process all scheduled works. Please note that the scheduled list | |
2318 | * may change while processing a work, so this function repeatedly | |
2319 | * fetches a work from the top and executes it. | |
2320 | * | |
2321 | * CONTEXT: | |
d565ed63 | 2322 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
affee4b2 TH |
2323 | * multiple times. |
2324 | */ | |
2325 | static void process_scheduled_works(struct worker *worker) | |
1da177e4 | 2326 | { |
affee4b2 TH |
2327 | while (!list_empty(&worker->scheduled)) { |
2328 | struct work_struct *work = list_first_entry(&worker->scheduled, | |
1da177e4 | 2329 | struct work_struct, entry); |
c34056a3 | 2330 | process_one_work(worker, work); |
1da177e4 | 2331 | } |
1da177e4 LT |
2332 | } |
2333 | ||
197f6acc TH |
2334 | static void set_pf_worker(bool val) |
2335 | { | |
2336 | mutex_lock(&wq_pool_attach_mutex); | |
2337 | if (val) | |
2338 | current->flags |= PF_WQ_WORKER; | |
2339 | else | |
2340 | current->flags &= ~PF_WQ_WORKER; | |
2341 | mutex_unlock(&wq_pool_attach_mutex); | |
2342 | } | |
2343 | ||
4690c4ab TH |
2344 | /** |
2345 | * worker_thread - the worker thread function | |
c34056a3 | 2346 | * @__worker: self |
4690c4ab | 2347 | * |
c5aa87bb TH |
2348 | * The worker thread function. All workers belong to a worker_pool - |
2349 | * either a per-cpu one or dynamic unbound one. These workers process all | |
2350 | * work items regardless of their specific target workqueue. The only | |
2351 | * exception is work items which belong to workqueues with a rescuer which | |
2352 | * will be explained in rescuer_thread(). | |
d185af30 YB |
2353 | * |
2354 | * Return: 0 | |
4690c4ab | 2355 | */ |
c34056a3 | 2356 | static int worker_thread(void *__worker) |
1da177e4 | 2357 | { |
c34056a3 | 2358 | struct worker *worker = __worker; |
bd7bdd43 | 2359 | struct worker_pool *pool = worker->pool; |
1da177e4 | 2360 | |
e22bee78 | 2361 | /* tell the scheduler that this is a workqueue worker */ |
197f6acc | 2362 | set_pf_worker(true); |
c8e55f36 | 2363 | woke_up: |
d565ed63 | 2364 | spin_lock_irq(&pool->lock); |
1da177e4 | 2365 | |
a9ab775b TH |
2366 | /* am I supposed to die? */ |
2367 | if (unlikely(worker->flags & WORKER_DIE)) { | |
d565ed63 | 2368 | spin_unlock_irq(&pool->lock); |
a9ab775b | 2369 | WARN_ON_ONCE(!list_empty(&worker->entry)); |
197f6acc | 2370 | set_pf_worker(false); |
60f5a4bc LJ |
2371 | |
2372 | set_task_comm(worker->task, "kworker/dying"); | |
7cda9aae | 2373 | ida_simple_remove(&pool->worker_ida, worker->id); |
a2d812a2 | 2374 | worker_detach_from_pool(worker); |
60f5a4bc | 2375 | kfree(worker); |
a9ab775b | 2376 | return 0; |
c8e55f36 | 2377 | } |
affee4b2 | 2378 | |
c8e55f36 | 2379 | worker_leave_idle(worker); |
db7bccf4 | 2380 | recheck: |
e22bee78 | 2381 | /* no more worker necessary? */ |
63d95a91 | 2382 | if (!need_more_worker(pool)) |
e22bee78 TH |
2383 | goto sleep; |
2384 | ||
2385 | /* do we need to manage? */ | |
63d95a91 | 2386 | if (unlikely(!may_start_working(pool)) && manage_workers(worker)) |
e22bee78 TH |
2387 | goto recheck; |
2388 | ||
c8e55f36 TH |
2389 | /* |
2390 | * ->scheduled list can only be filled while a worker is | |
2391 | * preparing to process a work or actually processing it. | |
2392 | * Make sure nobody diddled with it while I was sleeping. | |
2393 | */ | |
6183c009 | 2394 | WARN_ON_ONCE(!list_empty(&worker->scheduled)); |
c8e55f36 | 2395 | |
e22bee78 | 2396 | /* |
a9ab775b TH |
2397 | * Finish PREP stage. We're guaranteed to have at least one idle |
2398 | * worker or that someone else has already assumed the manager | |
2399 | * role. This is where @worker starts participating in concurrency | |
2400 | * management if applicable and concurrency management is restored | |
2401 | * after being rebound. See rebind_workers() for details. | |
e22bee78 | 2402 | */ |
a9ab775b | 2403 | worker_clr_flags(worker, WORKER_PREP | WORKER_REBOUND); |
e22bee78 TH |
2404 | |
2405 | do { | |
c8e55f36 | 2406 | struct work_struct *work = |
bd7bdd43 | 2407 | list_first_entry(&pool->worklist, |
c8e55f36 TH |
2408 | struct work_struct, entry); |
2409 | ||
82607adc TH |
2410 | pool->watchdog_ts = jiffies; |
2411 | ||
c8e55f36 TH |
2412 | if (likely(!(*work_data_bits(work) & WORK_STRUCT_LINKED))) { |
2413 | /* optimization path, not strictly necessary */ | |
2414 | process_one_work(worker, work); | |
2415 | if (unlikely(!list_empty(&worker->scheduled))) | |
affee4b2 | 2416 | process_scheduled_works(worker); |
c8e55f36 TH |
2417 | } else { |
2418 | move_linked_works(work, &worker->scheduled, NULL); | |
2419 | process_scheduled_works(worker); | |
affee4b2 | 2420 | } |
63d95a91 | 2421 | } while (keep_working(pool)); |
e22bee78 | 2422 | |
228f1d00 | 2423 | worker_set_flags(worker, WORKER_PREP); |
d313dd85 | 2424 | sleep: |
c8e55f36 | 2425 | /* |
d565ed63 TH |
2426 | * pool->lock is held and there's no work to process and no need to |
2427 | * manage, sleep. Workers are woken up only while holding | |
2428 | * pool->lock or from local cpu, so setting the current state | |
2429 | * before releasing pool->lock is enough to prevent losing any | |
2430 | * event. | |
c8e55f36 TH |
2431 | */ |
2432 | worker_enter_idle(worker); | |
c5a94a61 | 2433 | __set_current_state(TASK_IDLE); |
d565ed63 | 2434 | spin_unlock_irq(&pool->lock); |
c8e55f36 TH |
2435 | schedule(); |
2436 | goto woke_up; | |
1da177e4 LT |
2437 | } |
2438 | ||
e22bee78 TH |
2439 | /** |
2440 | * rescuer_thread - the rescuer thread function | |
111c225a | 2441 | * @__rescuer: self |
e22bee78 TH |
2442 | * |
2443 | * Workqueue rescuer thread function. There's one rescuer for each | |
493008a8 | 2444 | * workqueue which has WQ_MEM_RECLAIM set. |
e22bee78 | 2445 | * |
706026c2 | 2446 | * Regular work processing on a pool may block trying to create a new |
e22bee78 TH |
2447 | * worker which uses GFP_KERNEL allocation which has slight chance of |
2448 | * developing into deadlock if some works currently on the same queue | |
2449 | * need to be processed to satisfy the GFP_KERNEL allocation. This is | |
2450 | * the problem rescuer solves. | |
2451 | * | |
706026c2 TH |
2452 | * When such condition is possible, the pool summons rescuers of all |
2453 | * workqueues which have works queued on the pool and let them process | |
e22bee78 TH |
2454 | * those works so that forward progress can be guaranteed. |
2455 | * | |
2456 | * This should happen rarely. | |
d185af30 YB |
2457 | * |
2458 | * Return: 0 | |
e22bee78 | 2459 | */ |
111c225a | 2460 | static int rescuer_thread(void *__rescuer) |
e22bee78 | 2461 | { |
111c225a TH |
2462 | struct worker *rescuer = __rescuer; |
2463 | struct workqueue_struct *wq = rescuer->rescue_wq; | |
e22bee78 | 2464 | struct list_head *scheduled = &rescuer->scheduled; |
4d595b86 | 2465 | bool should_stop; |
e22bee78 TH |
2466 | |
2467 | set_user_nice(current, RESCUER_NICE_LEVEL); | |
111c225a TH |
2468 | |
2469 | /* | |
2470 | * Mark rescuer as worker too. As WORKER_PREP is never cleared, it | |
2471 | * doesn't participate in concurrency management. | |
2472 | */ | |
197f6acc | 2473 | set_pf_worker(true); |
e22bee78 | 2474 | repeat: |
c5a94a61 | 2475 | set_current_state(TASK_IDLE); |
e22bee78 | 2476 | |
4d595b86 LJ |
2477 | /* |
2478 | * By the time the rescuer is requested to stop, the workqueue | |
2479 | * shouldn't have any work pending, but @wq->maydays may still have | |
2480 | * pwq(s) queued. This can happen by non-rescuer workers consuming | |
2481 | * all the work items before the rescuer got to them. Go through | |
2482 | * @wq->maydays processing before acting on should_stop so that the | |
2483 | * list is always empty on exit. | |
2484 | */ | |
2485 | should_stop = kthread_should_stop(); | |
e22bee78 | 2486 | |
493a1724 | 2487 | /* see whether any pwq is asking for help */ |
2e109a28 | 2488 | spin_lock_irq(&wq_mayday_lock); |
493a1724 TH |
2489 | |
2490 | while (!list_empty(&wq->maydays)) { | |
2491 | struct pool_workqueue *pwq = list_first_entry(&wq->maydays, | |
2492 | struct pool_workqueue, mayday_node); | |
112202d9 | 2493 | struct worker_pool *pool = pwq->pool; |
e22bee78 | 2494 | struct work_struct *work, *n; |
82607adc | 2495 | bool first = true; |
e22bee78 TH |
2496 | |
2497 | __set_current_state(TASK_RUNNING); | |
493a1724 TH |
2498 | list_del_init(&pwq->mayday_node); |
2499 | ||
2e109a28 | 2500 | spin_unlock_irq(&wq_mayday_lock); |
e22bee78 | 2501 | |
51697d39 LJ |
2502 | worker_attach_to_pool(rescuer, pool); |
2503 | ||
2504 | spin_lock_irq(&pool->lock); | |
e22bee78 TH |
2505 | |
2506 | /* | |
2507 | * Slurp in all works issued via this workqueue and | |
2508 | * process'em. | |
2509 | */ | |
0479c8c5 | 2510 | WARN_ON_ONCE(!list_empty(scheduled)); |
82607adc TH |
2511 | list_for_each_entry_safe(work, n, &pool->worklist, entry) { |
2512 | if (get_work_pwq(work) == pwq) { | |
2513 | if (first) | |
2514 | pool->watchdog_ts = jiffies; | |
e22bee78 | 2515 | move_linked_works(work, scheduled, &n); |
82607adc TH |
2516 | } |
2517 | first = false; | |
2518 | } | |
e22bee78 | 2519 | |
008847f6 N |
2520 | if (!list_empty(scheduled)) { |
2521 | process_scheduled_works(rescuer); | |
2522 | ||
2523 | /* | |
2524 | * The above execution of rescued work items could | |
2525 | * have created more to rescue through | |
2526 | * pwq_activate_first_delayed() or chained | |
2527 | * queueing. Let's put @pwq back on mayday list so | |
2528 | * that such back-to-back work items, which may be | |
2529 | * being used to relieve memory pressure, don't | |
2530 | * incur MAYDAY_INTERVAL delay inbetween. | |
2531 | */ | |
2532 | if (need_to_create_worker(pool)) { | |
2533 | spin_lock(&wq_mayday_lock); | |
e66b39af TH |
2534 | /* |
2535 | * Queue iff we aren't racing destruction | |
2536 | * and somebody else hasn't queued it already. | |
2537 | */ | |
2538 | if (wq->rescuer && list_empty(&pwq->mayday_node)) { | |
2539 | get_pwq(pwq); | |
2540 | list_add_tail(&pwq->mayday_node, &wq->maydays); | |
2541 | } | |
008847f6 N |
2542 | spin_unlock(&wq_mayday_lock); |
2543 | } | |
2544 | } | |
7576958a | 2545 | |
77668c8b LJ |
2546 | /* |
2547 | * Put the reference grabbed by send_mayday(). @pool won't | |
13b1d625 | 2548 | * go away while we're still attached to it. |
77668c8b LJ |
2549 | */ |
2550 | put_pwq(pwq); | |
2551 | ||
7576958a | 2552 | /* |
d8ca83e6 | 2553 | * Leave this pool. If need_more_worker() is %true, notify a |
7576958a TH |
2554 | * regular worker; otherwise, we end up with 0 concurrency |
2555 | * and stalling the execution. | |
2556 | */ | |
d8ca83e6 | 2557 | if (need_more_worker(pool)) |
63d95a91 | 2558 | wake_up_worker(pool); |
7576958a | 2559 | |
13b1d625 LJ |
2560 | spin_unlock_irq(&pool->lock); |
2561 | ||
a2d812a2 | 2562 | worker_detach_from_pool(rescuer); |
13b1d625 LJ |
2563 | |
2564 | spin_lock_irq(&wq_mayday_lock); | |
e22bee78 TH |
2565 | } |
2566 | ||
2e109a28 | 2567 | spin_unlock_irq(&wq_mayday_lock); |
493a1724 | 2568 | |
4d595b86 LJ |
2569 | if (should_stop) { |
2570 | __set_current_state(TASK_RUNNING); | |
197f6acc | 2571 | set_pf_worker(false); |
4d595b86 LJ |
2572 | return 0; |
2573 | } | |
2574 | ||
111c225a TH |
2575 | /* rescuers should never participate in concurrency management */ |
2576 | WARN_ON_ONCE(!(rescuer->flags & WORKER_NOT_RUNNING)); | |
e22bee78 TH |
2577 | schedule(); |
2578 | goto repeat; | |
1da177e4 LT |
2579 | } |
2580 | ||
fca839c0 TH |
2581 | /** |
2582 | * check_flush_dependency - check for flush dependency sanity | |
2583 | * @target_wq: workqueue being flushed | |
2584 | * @target_work: work item being flushed (NULL for workqueue flushes) | |
2585 | * | |
2586 | * %current is trying to flush the whole @target_wq or @target_work on it. | |
2587 | * If @target_wq doesn't have %WQ_MEM_RECLAIM, verify that %current is not | |
2588 | * reclaiming memory or running on a workqueue which doesn't have | |
2589 | * %WQ_MEM_RECLAIM as that can break forward-progress guarantee leading to | |
2590 | * a deadlock. | |
2591 | */ | |
2592 | static void check_flush_dependency(struct workqueue_struct *target_wq, | |
2593 | struct work_struct *target_work) | |
2594 | { | |
2595 | work_func_t target_func = target_work ? target_work->func : NULL; | |
2596 | struct worker *worker; | |
2597 | ||
2598 | if (target_wq->flags & WQ_MEM_RECLAIM) | |
2599 | return; | |
2600 | ||
2601 | worker = current_wq_worker(); | |
2602 | ||
2603 | WARN_ONCE(current->flags & PF_MEMALLOC, | |
d75f773c | 2604 | "workqueue: PF_MEMALLOC task %d(%s) is flushing !WQ_MEM_RECLAIM %s:%ps", |
fca839c0 | 2605 | current->pid, current->comm, target_wq->name, target_func); |
23d11a58 TH |
2606 | WARN_ONCE(worker && ((worker->current_pwq->wq->flags & |
2607 | (WQ_MEM_RECLAIM | __WQ_LEGACY)) == WQ_MEM_RECLAIM), | |
d75f773c | 2608 | "workqueue: WQ_MEM_RECLAIM %s:%ps is flushing !WQ_MEM_RECLAIM %s:%ps", |
fca839c0 TH |
2609 | worker->current_pwq->wq->name, worker->current_func, |
2610 | target_wq->name, target_func); | |
2611 | } | |
2612 | ||
fc2e4d70 ON |
2613 | struct wq_barrier { |
2614 | struct work_struct work; | |
2615 | struct completion done; | |
2607d7a6 | 2616 | struct task_struct *task; /* purely informational */ |
fc2e4d70 ON |
2617 | }; |
2618 | ||
2619 | static void wq_barrier_func(struct work_struct *work) | |
2620 | { | |
2621 | struct wq_barrier *barr = container_of(work, struct wq_barrier, work); | |
2622 | complete(&barr->done); | |
2623 | } | |
2624 | ||
4690c4ab TH |
2625 | /** |
2626 | * insert_wq_barrier - insert a barrier work | |
112202d9 | 2627 | * @pwq: pwq to insert barrier into |
4690c4ab | 2628 | * @barr: wq_barrier to insert |
affee4b2 TH |
2629 | * @target: target work to attach @barr to |
2630 | * @worker: worker currently executing @target, NULL if @target is not executing | |
4690c4ab | 2631 | * |
affee4b2 TH |
2632 | * @barr is linked to @target such that @barr is completed only after |
2633 | * @target finishes execution. Please note that the ordering | |
2634 | * guarantee is observed only with respect to @target and on the local | |
2635 | * cpu. | |
2636 | * | |
2637 | * Currently, a queued barrier can't be canceled. This is because | |
2638 | * try_to_grab_pending() can't determine whether the work to be | |
2639 | * grabbed is at the head of the queue and thus can't clear LINKED | |
2640 | * flag of the previous work while there must be a valid next work | |
2641 | * after a work with LINKED flag set. | |
2642 | * | |
2643 | * Note that when @worker is non-NULL, @target may be modified | |
112202d9 | 2644 | * underneath us, so we can't reliably determine pwq from @target. |
4690c4ab TH |
2645 | * |
2646 | * CONTEXT: | |
d565ed63 | 2647 | * spin_lock_irq(pool->lock). |
4690c4ab | 2648 | */ |
112202d9 | 2649 | static void insert_wq_barrier(struct pool_workqueue *pwq, |
affee4b2 TH |
2650 | struct wq_barrier *barr, |
2651 | struct work_struct *target, struct worker *worker) | |
fc2e4d70 | 2652 | { |
affee4b2 TH |
2653 | struct list_head *head; |
2654 | unsigned int linked = 0; | |
2655 | ||
dc186ad7 | 2656 | /* |
d565ed63 | 2657 | * debugobject calls are safe here even with pool->lock locked |
dc186ad7 TG |
2658 | * as we know for sure that this will not trigger any of the |
2659 | * checks and call back into the fixup functions where we | |
2660 | * might deadlock. | |
2661 | */ | |
ca1cab37 | 2662 | INIT_WORK_ONSTACK(&barr->work, wq_barrier_func); |
22df02bb | 2663 | __set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(&barr->work)); |
52fa5bc5 | 2664 | |
fd1a5b04 BP |
2665 | init_completion_map(&barr->done, &target->lockdep_map); |
2666 | ||
2607d7a6 | 2667 | barr->task = current; |
83c22520 | 2668 | |
affee4b2 TH |
2669 | /* |
2670 | * If @target is currently being executed, schedule the | |
2671 | * barrier to the worker; otherwise, put it after @target. | |
2672 | */ | |
2673 | if (worker) | |
2674 | head = worker->scheduled.next; | |
2675 | else { | |
2676 | unsigned long *bits = work_data_bits(target); | |
2677 | ||
2678 | head = target->entry.next; | |
2679 | /* there can already be other linked works, inherit and set */ | |
2680 | linked = *bits & WORK_STRUCT_LINKED; | |
2681 | __set_bit(WORK_STRUCT_LINKED_BIT, bits); | |
2682 | } | |
2683 | ||
dc186ad7 | 2684 | debug_work_activate(&barr->work); |
112202d9 | 2685 | insert_work(pwq, &barr->work, head, |
affee4b2 | 2686 | work_color_to_flags(WORK_NO_COLOR) | linked); |
fc2e4d70 ON |
2687 | } |
2688 | ||
73f53c4a | 2689 | /** |
112202d9 | 2690 | * flush_workqueue_prep_pwqs - prepare pwqs for workqueue flushing |
73f53c4a TH |
2691 | * @wq: workqueue being flushed |
2692 | * @flush_color: new flush color, < 0 for no-op | |
2693 | * @work_color: new work color, < 0 for no-op | |
2694 | * | |
112202d9 | 2695 | * Prepare pwqs for workqueue flushing. |
73f53c4a | 2696 | * |
112202d9 TH |
2697 | * If @flush_color is non-negative, flush_color on all pwqs should be |
2698 | * -1. If no pwq has in-flight commands at the specified color, all | |
2699 | * pwq->flush_color's stay at -1 and %false is returned. If any pwq | |
2700 | * has in flight commands, its pwq->flush_color is set to | |
2701 | * @flush_color, @wq->nr_pwqs_to_flush is updated accordingly, pwq | |
73f53c4a TH |
2702 | * wakeup logic is armed and %true is returned. |
2703 | * | |
2704 | * The caller should have initialized @wq->first_flusher prior to | |
2705 | * calling this function with non-negative @flush_color. If | |
2706 | * @flush_color is negative, no flush color update is done and %false | |
2707 | * is returned. | |
2708 | * | |
112202d9 | 2709 | * If @work_color is non-negative, all pwqs should have the same |
73f53c4a TH |
2710 | * work_color which is previous to @work_color and all will be |
2711 | * advanced to @work_color. | |
2712 | * | |
2713 | * CONTEXT: | |
3c25a55d | 2714 | * mutex_lock(wq->mutex). |
73f53c4a | 2715 | * |
d185af30 | 2716 | * Return: |
73f53c4a TH |
2717 | * %true if @flush_color >= 0 and there's something to flush. %false |
2718 | * otherwise. | |
2719 | */ | |
112202d9 | 2720 | static bool flush_workqueue_prep_pwqs(struct workqueue_struct *wq, |
73f53c4a | 2721 | int flush_color, int work_color) |
1da177e4 | 2722 | { |
73f53c4a | 2723 | bool wait = false; |
49e3cf44 | 2724 | struct pool_workqueue *pwq; |
1da177e4 | 2725 | |
73f53c4a | 2726 | if (flush_color >= 0) { |
6183c009 | 2727 | WARN_ON_ONCE(atomic_read(&wq->nr_pwqs_to_flush)); |
112202d9 | 2728 | atomic_set(&wq->nr_pwqs_to_flush, 1); |
1da177e4 | 2729 | } |
2355b70f | 2730 | |
49e3cf44 | 2731 | for_each_pwq(pwq, wq) { |
112202d9 | 2732 | struct worker_pool *pool = pwq->pool; |
fc2e4d70 | 2733 | |
b09f4fd3 | 2734 | spin_lock_irq(&pool->lock); |
83c22520 | 2735 | |
73f53c4a | 2736 | if (flush_color >= 0) { |
6183c009 | 2737 | WARN_ON_ONCE(pwq->flush_color != -1); |
fc2e4d70 | 2738 | |
112202d9 TH |
2739 | if (pwq->nr_in_flight[flush_color]) { |
2740 | pwq->flush_color = flush_color; | |
2741 | atomic_inc(&wq->nr_pwqs_to_flush); | |
73f53c4a TH |
2742 | wait = true; |
2743 | } | |
2744 | } | |
1da177e4 | 2745 | |
73f53c4a | 2746 | if (work_color >= 0) { |
6183c009 | 2747 | WARN_ON_ONCE(work_color != work_next_color(pwq->work_color)); |
112202d9 | 2748 | pwq->work_color = work_color; |
73f53c4a | 2749 | } |
1da177e4 | 2750 | |
b09f4fd3 | 2751 | spin_unlock_irq(&pool->lock); |
1da177e4 | 2752 | } |
2355b70f | 2753 | |
112202d9 | 2754 | if (flush_color >= 0 && atomic_dec_and_test(&wq->nr_pwqs_to_flush)) |
73f53c4a | 2755 | complete(&wq->first_flusher->done); |
14441960 | 2756 | |
73f53c4a | 2757 | return wait; |
1da177e4 LT |
2758 | } |
2759 | ||
0fcb78c2 | 2760 | /** |
1da177e4 | 2761 | * flush_workqueue - ensure that any scheduled work has run to completion. |
0fcb78c2 | 2762 | * @wq: workqueue to flush |
1da177e4 | 2763 | * |
c5aa87bb TH |
2764 | * This function sleeps until all work items which were queued on entry |
2765 | * have finished execution, but it is not livelocked by new incoming ones. | |
1da177e4 | 2766 | */ |
7ad5b3a5 | 2767 | void flush_workqueue(struct workqueue_struct *wq) |
1da177e4 | 2768 | { |
73f53c4a TH |
2769 | struct wq_flusher this_flusher = { |
2770 | .list = LIST_HEAD_INIT(this_flusher.list), | |
2771 | .flush_color = -1, | |
fd1a5b04 | 2772 | .done = COMPLETION_INITIALIZER_ONSTACK_MAP(this_flusher.done, wq->lockdep_map), |
73f53c4a TH |
2773 | }; |
2774 | int next_color; | |
1da177e4 | 2775 | |
3347fa09 TH |
2776 | if (WARN_ON(!wq_online)) |
2777 | return; | |
2778 | ||
87915adc JB |
2779 | lock_map_acquire(&wq->lockdep_map); |
2780 | lock_map_release(&wq->lockdep_map); | |
2781 | ||
3c25a55d | 2782 | mutex_lock(&wq->mutex); |
73f53c4a TH |
2783 | |
2784 | /* | |
2785 | * Start-to-wait phase | |
2786 | */ | |
2787 | next_color = work_next_color(wq->work_color); | |
2788 | ||
2789 | if (next_color != wq->flush_color) { | |
2790 | /* | |
2791 | * Color space is not full. The current work_color | |
2792 | * becomes our flush_color and work_color is advanced | |
2793 | * by one. | |
2794 | */ | |
6183c009 | 2795 | WARN_ON_ONCE(!list_empty(&wq->flusher_overflow)); |
73f53c4a TH |
2796 | this_flusher.flush_color = wq->work_color; |
2797 | wq->work_color = next_color; | |
2798 | ||
2799 | if (!wq->first_flusher) { | |
2800 | /* no flush in progress, become the first flusher */ | |
6183c009 | 2801 | WARN_ON_ONCE(wq->flush_color != this_flusher.flush_color); |
73f53c4a TH |
2802 | |
2803 | wq->first_flusher = &this_flusher; | |
2804 | ||
112202d9 | 2805 | if (!flush_workqueue_prep_pwqs(wq, wq->flush_color, |
73f53c4a TH |
2806 | wq->work_color)) { |
2807 | /* nothing to flush, done */ | |
2808 | wq->flush_color = next_color; | |
2809 | wq->first_flusher = NULL; | |
2810 | goto out_unlock; | |
2811 | } | |
2812 | } else { | |
2813 | /* wait in queue */ | |
6183c009 | 2814 | WARN_ON_ONCE(wq->flush_color == this_flusher.flush_color); |
73f53c4a | 2815 | list_add_tail(&this_flusher.list, &wq->flusher_queue); |
112202d9 | 2816 | flush_workqueue_prep_pwqs(wq, -1, wq->work_color); |
73f53c4a TH |
2817 | } |
2818 | } else { | |
2819 | /* | |
2820 | * Oops, color space is full, wait on overflow queue. | |
2821 | * The next flush completion will assign us | |
2822 | * flush_color and transfer to flusher_queue. | |
2823 | */ | |
2824 | list_add_tail(&this_flusher.list, &wq->flusher_overflow); | |
2825 | } | |
2826 | ||
fca839c0 TH |
2827 | check_flush_dependency(wq, NULL); |
2828 | ||
3c25a55d | 2829 | mutex_unlock(&wq->mutex); |
73f53c4a TH |
2830 | |
2831 | wait_for_completion(&this_flusher.done); | |
2832 | ||
2833 | /* | |
2834 | * Wake-up-and-cascade phase | |
2835 | * | |
2836 | * First flushers are responsible for cascading flushes and | |
2837 | * handling overflow. Non-first flushers can simply return. | |
2838 | */ | |
00d5d15b | 2839 | if (READ_ONCE(wq->first_flusher) != &this_flusher) |
73f53c4a TH |
2840 | return; |
2841 | ||
3c25a55d | 2842 | mutex_lock(&wq->mutex); |
73f53c4a | 2843 | |
4ce48b37 TH |
2844 | /* we might have raced, check again with mutex held */ |
2845 | if (wq->first_flusher != &this_flusher) | |
2846 | goto out_unlock; | |
2847 | ||
00d5d15b | 2848 | WRITE_ONCE(wq->first_flusher, NULL); |
73f53c4a | 2849 | |
6183c009 TH |
2850 | WARN_ON_ONCE(!list_empty(&this_flusher.list)); |
2851 | WARN_ON_ONCE(wq->flush_color != this_flusher.flush_color); | |
73f53c4a TH |
2852 | |
2853 | while (true) { | |
2854 | struct wq_flusher *next, *tmp; | |
2855 | ||
2856 | /* complete all the flushers sharing the current flush color */ | |
2857 | list_for_each_entry_safe(next, tmp, &wq->flusher_queue, list) { | |
2858 | if (next->flush_color != wq->flush_color) | |
2859 | break; | |
2860 | list_del_init(&next->list); | |
2861 | complete(&next->done); | |
2862 | } | |
2863 | ||
6183c009 TH |
2864 | WARN_ON_ONCE(!list_empty(&wq->flusher_overflow) && |
2865 | wq->flush_color != work_next_color(wq->work_color)); | |
73f53c4a TH |
2866 | |
2867 | /* this flush_color is finished, advance by one */ | |
2868 | wq->flush_color = work_next_color(wq->flush_color); | |
2869 | ||
2870 | /* one color has been freed, handle overflow queue */ | |
2871 | if (!list_empty(&wq->flusher_overflow)) { | |
2872 | /* | |
2873 | * Assign the same color to all overflowed | |
2874 | * flushers, advance work_color and append to | |
2875 | * flusher_queue. This is the start-to-wait | |
2876 | * phase for these overflowed flushers. | |
2877 | */ | |
2878 | list_for_each_entry(tmp, &wq->flusher_overflow, list) | |
2879 | tmp->flush_color = wq->work_color; | |
2880 | ||
2881 | wq->work_color = work_next_color(wq->work_color); | |
2882 | ||
2883 | list_splice_tail_init(&wq->flusher_overflow, | |
2884 | &wq->flusher_queue); | |
112202d9 | 2885 | flush_workqueue_prep_pwqs(wq, -1, wq->work_color); |
73f53c4a TH |
2886 | } |
2887 | ||
2888 | if (list_empty(&wq->flusher_queue)) { | |
6183c009 | 2889 | WARN_ON_ONCE(wq->flush_color != wq->work_color); |
73f53c4a TH |
2890 | break; |
2891 | } | |
2892 | ||
2893 | /* | |
2894 | * Need to flush more colors. Make the next flusher | |
112202d9 | 2895 | * the new first flusher and arm pwqs. |
73f53c4a | 2896 | */ |
6183c009 TH |
2897 | WARN_ON_ONCE(wq->flush_color == wq->work_color); |
2898 | WARN_ON_ONCE(wq->flush_color != next->flush_color); | |
73f53c4a TH |
2899 | |
2900 | list_del_init(&next->list); | |
2901 | wq->first_flusher = next; | |
2902 | ||
112202d9 | 2903 | if (flush_workqueue_prep_pwqs(wq, wq->flush_color, -1)) |
73f53c4a TH |
2904 | break; |
2905 | ||
2906 | /* | |
2907 | * Meh... this color is already done, clear first | |
2908 | * flusher and repeat cascading. | |
2909 | */ | |
2910 | wq->first_flusher = NULL; | |
2911 | } | |
2912 | ||
2913 | out_unlock: | |
3c25a55d | 2914 | mutex_unlock(&wq->mutex); |
1da177e4 | 2915 | } |
1dadafa8 | 2916 | EXPORT_SYMBOL(flush_workqueue); |
1da177e4 | 2917 | |
9c5a2ba7 TH |
2918 | /** |
2919 | * drain_workqueue - drain a workqueue | |
2920 | * @wq: workqueue to drain | |
2921 | * | |
2922 | * Wait until the workqueue becomes empty. While draining is in progress, | |
2923 | * only chain queueing is allowed. IOW, only currently pending or running | |
2924 | * work items on @wq can queue further work items on it. @wq is flushed | |
b749b1b6 | 2925 | * repeatedly until it becomes empty. The number of flushing is determined |
9c5a2ba7 TH |
2926 | * by the depth of chaining and should be relatively short. Whine if it |
2927 | * takes too long. | |
2928 | */ | |
2929 | void drain_workqueue(struct workqueue_struct *wq) | |
2930 | { | |
2931 | unsigned int flush_cnt = 0; | |
49e3cf44 | 2932 | struct pool_workqueue *pwq; |
9c5a2ba7 TH |
2933 | |
2934 | /* | |
2935 | * __queue_work() needs to test whether there are drainers, is much | |
2936 | * hotter than drain_workqueue() and already looks at @wq->flags. | |
618b01eb | 2937 | * Use __WQ_DRAINING so that queue doesn't have to check nr_drainers. |
9c5a2ba7 | 2938 | */ |
87fc741e | 2939 | mutex_lock(&wq->mutex); |
9c5a2ba7 | 2940 | if (!wq->nr_drainers++) |
618b01eb | 2941 | wq->flags |= __WQ_DRAINING; |
87fc741e | 2942 | mutex_unlock(&wq->mutex); |
9c5a2ba7 TH |
2943 | reflush: |
2944 | flush_workqueue(wq); | |
2945 | ||
b09f4fd3 | 2946 | mutex_lock(&wq->mutex); |
76af4d93 | 2947 | |
49e3cf44 | 2948 | for_each_pwq(pwq, wq) { |
fa2563e4 | 2949 | bool drained; |
9c5a2ba7 | 2950 | |
b09f4fd3 | 2951 | spin_lock_irq(&pwq->pool->lock); |
112202d9 | 2952 | drained = !pwq->nr_active && list_empty(&pwq->delayed_works); |
b09f4fd3 | 2953 | spin_unlock_irq(&pwq->pool->lock); |
fa2563e4 TT |
2954 | |
2955 | if (drained) | |
9c5a2ba7 TH |
2956 | continue; |
2957 | ||
2958 | if (++flush_cnt == 10 || | |
2959 | (flush_cnt % 100 == 0 && flush_cnt <= 1000)) | |
c5aa87bb | 2960 | pr_warn("workqueue %s: drain_workqueue() isn't complete after %u tries\n", |
044c782c | 2961 | wq->name, flush_cnt); |
76af4d93 | 2962 | |
b09f4fd3 | 2963 | mutex_unlock(&wq->mutex); |
9c5a2ba7 TH |
2964 | goto reflush; |
2965 | } | |
2966 | ||
9c5a2ba7 | 2967 | if (!--wq->nr_drainers) |
618b01eb | 2968 | wq->flags &= ~__WQ_DRAINING; |
87fc741e | 2969 | mutex_unlock(&wq->mutex); |
9c5a2ba7 TH |
2970 | } |
2971 | EXPORT_SYMBOL_GPL(drain_workqueue); | |
2972 | ||
d6e89786 JB |
2973 | static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr, |
2974 | bool from_cancel) | |
db700897 | 2975 | { |
affee4b2 | 2976 | struct worker *worker = NULL; |
c9e7cf27 | 2977 | struct worker_pool *pool; |
112202d9 | 2978 | struct pool_workqueue *pwq; |
db700897 ON |
2979 | |
2980 | might_sleep(); | |
fa1b54e6 | 2981 | |
24acfb71 | 2982 | rcu_read_lock(); |
c9e7cf27 | 2983 | pool = get_work_pool(work); |
fa1b54e6 | 2984 | if (!pool) { |
24acfb71 | 2985 | rcu_read_unlock(); |
baf59022 | 2986 | return false; |
fa1b54e6 | 2987 | } |
db700897 | 2988 | |
24acfb71 | 2989 | spin_lock_irq(&pool->lock); |
0b3dae68 | 2990 | /* see the comment in try_to_grab_pending() with the same code */ |
112202d9 TH |
2991 | pwq = get_work_pwq(work); |
2992 | if (pwq) { | |
2993 | if (unlikely(pwq->pool != pool)) | |
4690c4ab | 2994 | goto already_gone; |
606a5020 | 2995 | } else { |
c9e7cf27 | 2996 | worker = find_worker_executing_work(pool, work); |
affee4b2 | 2997 | if (!worker) |
4690c4ab | 2998 | goto already_gone; |
112202d9 | 2999 | pwq = worker->current_pwq; |
606a5020 | 3000 | } |
db700897 | 3001 | |
fca839c0 TH |
3002 | check_flush_dependency(pwq->wq, work); |
3003 | ||
112202d9 | 3004 | insert_wq_barrier(pwq, barr, work, worker); |
d565ed63 | 3005 | spin_unlock_irq(&pool->lock); |
7a22ad75 | 3006 | |
e159489b | 3007 | /* |
a1d14934 PZ |
3008 | * Force a lock recursion deadlock when using flush_work() inside a |
3009 | * single-threaded or rescuer equipped workqueue. | |
3010 | * | |
3011 | * For single threaded workqueues the deadlock happens when the work | |
3012 | * is after the work issuing the flush_work(). For rescuer equipped | |
3013 | * workqueues the deadlock happens when the rescuer stalls, blocking | |
3014 | * forward progress. | |
e159489b | 3015 | */ |
d6e89786 JB |
3016 | if (!from_cancel && |
3017 | (pwq->wq->saved_max_active == 1 || pwq->wq->rescuer)) { | |
112202d9 | 3018 | lock_map_acquire(&pwq->wq->lockdep_map); |
a1d14934 PZ |
3019 | lock_map_release(&pwq->wq->lockdep_map); |
3020 | } | |
24acfb71 | 3021 | rcu_read_unlock(); |
401a8d04 | 3022 | return true; |
4690c4ab | 3023 | already_gone: |
d565ed63 | 3024 | spin_unlock_irq(&pool->lock); |
24acfb71 | 3025 | rcu_read_unlock(); |
401a8d04 | 3026 | return false; |
db700897 | 3027 | } |
baf59022 | 3028 | |
d6e89786 JB |
3029 | static bool __flush_work(struct work_struct *work, bool from_cancel) |
3030 | { | |
3031 | struct wq_barrier barr; | |
3032 | ||
3033 | if (WARN_ON(!wq_online)) | |
3034 | return false; | |
3035 | ||
4d43d395 TH |
3036 | if (WARN_ON(!work->func)) |
3037 | return false; | |
3038 | ||
87915adc JB |
3039 | if (!from_cancel) { |
3040 | lock_map_acquire(&work->lockdep_map); | |
3041 | lock_map_release(&work->lockdep_map); | |
3042 | } | |
3043 | ||
d6e89786 JB |
3044 | if (start_flush_work(work, &barr, from_cancel)) { |
3045 | wait_for_completion(&barr.done); | |
3046 | destroy_work_on_stack(&barr.work); | |
3047 | return true; | |
3048 | } else { | |
3049 | return false; | |
3050 | } | |
3051 | } | |
3052 | ||
baf59022 TH |
3053 | /** |
3054 | * flush_work - wait for a work to finish executing the last queueing instance | |
3055 | * @work: the work to flush | |
3056 | * | |
606a5020 TH |
3057 | * Wait until @work has finished execution. @work is guaranteed to be idle |
3058 | * on return if it hasn't been requeued since flush started. | |
baf59022 | 3059 | * |
d185af30 | 3060 | * Return: |
baf59022 TH |
3061 | * %true if flush_work() waited for the work to finish execution, |
3062 | * %false if it was already idle. | |
3063 | */ | |
3064 | bool flush_work(struct work_struct *work) | |
3065 | { | |
d6e89786 | 3066 | return __flush_work(work, false); |
6e84d644 | 3067 | } |
606a5020 | 3068 | EXPORT_SYMBOL_GPL(flush_work); |
6e84d644 | 3069 | |
8603e1b3 | 3070 | struct cwt_wait { |
ac6424b9 | 3071 | wait_queue_entry_t wait; |
8603e1b3 TH |
3072 | struct work_struct *work; |
3073 | }; | |
3074 | ||
ac6424b9 | 3075 | static int cwt_wakefn(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
8603e1b3 TH |
3076 | { |
3077 | struct cwt_wait *cwait = container_of(wait, struct cwt_wait, wait); | |
3078 | ||
3079 | if (cwait->work != key) | |
3080 | return 0; | |
3081 | return autoremove_wake_function(wait, mode, sync, key); | |
3082 | } | |
3083 | ||
36e227d2 | 3084 | static bool __cancel_work_timer(struct work_struct *work, bool is_dwork) |
1f1f642e | 3085 | { |
8603e1b3 | 3086 | static DECLARE_WAIT_QUEUE_HEAD(cancel_waitq); |
bbb68dfa | 3087 | unsigned long flags; |
1f1f642e ON |
3088 | int ret; |
3089 | ||
3090 | do { | |
bbb68dfa TH |
3091 | ret = try_to_grab_pending(work, is_dwork, &flags); |
3092 | /* | |
8603e1b3 TH |
3093 | * If someone else is already canceling, wait for it to |
3094 | * finish. flush_work() doesn't work for PREEMPT_NONE | |
3095 | * because we may get scheduled between @work's completion | |
3096 | * and the other canceling task resuming and clearing | |
3097 | * CANCELING - flush_work() will return false immediately | |
3098 | * as @work is no longer busy, try_to_grab_pending() will | |
3099 | * return -ENOENT as @work is still being canceled and the | |
3100 | * other canceling task won't be able to clear CANCELING as | |
3101 | * we're hogging the CPU. | |
3102 | * | |
3103 | * Let's wait for completion using a waitqueue. As this | |
3104 | * may lead to the thundering herd problem, use a custom | |
3105 | * wake function which matches @work along with exclusive | |
3106 | * wait and wakeup. | |
bbb68dfa | 3107 | */ |
8603e1b3 TH |
3108 | if (unlikely(ret == -ENOENT)) { |
3109 | struct cwt_wait cwait; | |
3110 | ||
3111 | init_wait(&cwait.wait); | |
3112 | cwait.wait.func = cwt_wakefn; | |
3113 | cwait.work = work; | |
3114 | ||
3115 | prepare_to_wait_exclusive(&cancel_waitq, &cwait.wait, | |
3116 | TASK_UNINTERRUPTIBLE); | |
3117 | if (work_is_canceling(work)) | |
3118 | schedule(); | |
3119 | finish_wait(&cancel_waitq, &cwait.wait); | |
3120 | } | |
1f1f642e ON |
3121 | } while (unlikely(ret < 0)); |
3122 | ||
bbb68dfa TH |
3123 | /* tell other tasks trying to grab @work to back off */ |
3124 | mark_work_canceling(work); | |
3125 | local_irq_restore(flags); | |
3126 | ||
3347fa09 TH |
3127 | /* |
3128 | * This allows canceling during early boot. We know that @work | |
3129 | * isn't executing. | |
3130 | */ | |
3131 | if (wq_online) | |
d6e89786 | 3132 | __flush_work(work, true); |
3347fa09 | 3133 | |
7a22ad75 | 3134 | clear_work_data(work); |
8603e1b3 TH |
3135 | |
3136 | /* | |
3137 | * Paired with prepare_to_wait() above so that either | |
3138 | * waitqueue_active() is visible here or !work_is_canceling() is | |
3139 | * visible there. | |
3140 | */ | |
3141 | smp_mb(); | |
3142 | if (waitqueue_active(&cancel_waitq)) | |
3143 | __wake_up(&cancel_waitq, TASK_NORMAL, 1, work); | |
3144 | ||
1f1f642e ON |
3145 | return ret; |
3146 | } | |
3147 | ||
6e84d644 | 3148 | /** |
401a8d04 TH |
3149 | * cancel_work_sync - cancel a work and wait for it to finish |
3150 | * @work: the work to cancel | |
6e84d644 | 3151 | * |
401a8d04 TH |
3152 | * Cancel @work and wait for its execution to finish. This function |
3153 | * can be used even if the work re-queues itself or migrates to | |
3154 | * another workqueue. On return from this function, @work is | |
3155 | * guaranteed to be not pending or executing on any CPU. | |
1f1f642e | 3156 | * |
401a8d04 TH |
3157 | * cancel_work_sync(&delayed_work->work) must not be used for |
3158 | * delayed_work's. Use cancel_delayed_work_sync() instead. | |
6e84d644 | 3159 | * |
401a8d04 | 3160 | * The caller must ensure that the workqueue on which @work was last |
6e84d644 | 3161 | * queued can't be destroyed before this function returns. |
401a8d04 | 3162 | * |
d185af30 | 3163 | * Return: |
401a8d04 | 3164 | * %true if @work was pending, %false otherwise. |
6e84d644 | 3165 | */ |
401a8d04 | 3166 | bool cancel_work_sync(struct work_struct *work) |
6e84d644 | 3167 | { |
36e227d2 | 3168 | return __cancel_work_timer(work, false); |
b89deed3 | 3169 | } |
28e53bdd | 3170 | EXPORT_SYMBOL_GPL(cancel_work_sync); |
b89deed3 | 3171 | |
6e84d644 | 3172 | /** |
401a8d04 TH |
3173 | * flush_delayed_work - wait for a dwork to finish executing the last queueing |
3174 | * @dwork: the delayed work to flush | |
6e84d644 | 3175 | * |
401a8d04 TH |
3176 | * Delayed timer is cancelled and the pending work is queued for |
3177 | * immediate execution. Like flush_work(), this function only | |
3178 | * considers the last queueing instance of @dwork. | |
1f1f642e | 3179 | * |
d185af30 | 3180 | * Return: |
401a8d04 TH |
3181 | * %true if flush_work() waited for the work to finish execution, |
3182 | * %false if it was already idle. | |
6e84d644 | 3183 | */ |
401a8d04 TH |
3184 | bool flush_delayed_work(struct delayed_work *dwork) |
3185 | { | |
8930caba | 3186 | local_irq_disable(); |
401a8d04 | 3187 | if (del_timer_sync(&dwork->timer)) |
60c057bc | 3188 | __queue_work(dwork->cpu, dwork->wq, &dwork->work); |
8930caba | 3189 | local_irq_enable(); |
401a8d04 TH |
3190 | return flush_work(&dwork->work); |
3191 | } | |
3192 | EXPORT_SYMBOL(flush_delayed_work); | |
3193 | ||
05f0fe6b TH |
3194 | /** |
3195 | * flush_rcu_work - wait for a rwork to finish executing the last queueing | |
3196 | * @rwork: the rcu work to flush | |
3197 | * | |
3198 | * Return: | |
3199 | * %true if flush_rcu_work() waited for the work to finish execution, | |
3200 | * %false if it was already idle. | |
3201 | */ | |
3202 | bool flush_rcu_work(struct rcu_work *rwork) | |
3203 | { | |
3204 | if (test_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(&rwork->work))) { | |
3205 | rcu_barrier(); | |
3206 | flush_work(&rwork->work); | |
3207 | return true; | |
3208 | } else { | |
3209 | return flush_work(&rwork->work); | |
3210 | } | |
3211 | } | |
3212 | EXPORT_SYMBOL(flush_rcu_work); | |
3213 | ||
f72b8792 JA |
3214 | static bool __cancel_work(struct work_struct *work, bool is_dwork) |
3215 | { | |
3216 | unsigned long flags; | |
3217 | int ret; | |
3218 | ||
3219 | do { | |
3220 | ret = try_to_grab_pending(work, is_dwork, &flags); | |
3221 | } while (unlikely(ret == -EAGAIN)); | |
3222 | ||
3223 | if (unlikely(ret < 0)) | |
3224 | return false; | |
3225 | ||
3226 | set_work_pool_and_clear_pending(work, get_work_pool_id(work)); | |
3227 | local_irq_restore(flags); | |
3228 | return ret; | |
3229 | } | |
3230 | ||
09383498 | 3231 | /** |
57b30ae7 TH |
3232 | * cancel_delayed_work - cancel a delayed work |
3233 | * @dwork: delayed_work to cancel | |
09383498 | 3234 | * |
d185af30 YB |
3235 | * Kill off a pending delayed_work. |
3236 | * | |
3237 | * Return: %true if @dwork was pending and canceled; %false if it wasn't | |
3238 | * pending. | |
3239 | * | |
3240 | * Note: | |
3241 | * The work callback function may still be running on return, unless | |
3242 | * it returns %true and the work doesn't re-arm itself. Explicitly flush or | |
3243 | * use cancel_delayed_work_sync() to wait on it. | |
09383498 | 3244 | * |
57b30ae7 | 3245 | * This function is safe to call from any context including IRQ handler. |
09383498 | 3246 | */ |
57b30ae7 | 3247 | bool cancel_delayed_work(struct delayed_work *dwork) |
09383498 | 3248 | { |
f72b8792 | 3249 | return __cancel_work(&dwork->work, true); |
09383498 | 3250 | } |
57b30ae7 | 3251 | EXPORT_SYMBOL(cancel_delayed_work); |
09383498 | 3252 | |
401a8d04 TH |
3253 | /** |
3254 | * cancel_delayed_work_sync - cancel a delayed work and wait for it to finish | |
3255 | * @dwork: the delayed work cancel | |
3256 | * | |
3257 | * This is cancel_work_sync() for delayed works. | |
3258 | * | |
d185af30 | 3259 | * Return: |
401a8d04 TH |
3260 | * %true if @dwork was pending, %false otherwise. |
3261 | */ | |
3262 | bool cancel_delayed_work_sync(struct delayed_work *dwork) | |
6e84d644 | 3263 | { |
36e227d2 | 3264 | return __cancel_work_timer(&dwork->work, true); |
6e84d644 | 3265 | } |
f5a421a4 | 3266 | EXPORT_SYMBOL(cancel_delayed_work_sync); |
1da177e4 | 3267 | |
b6136773 | 3268 | /** |
31ddd871 | 3269 | * schedule_on_each_cpu - execute a function synchronously on each online CPU |
b6136773 | 3270 | * @func: the function to call |
b6136773 | 3271 | * |
31ddd871 TH |
3272 | * schedule_on_each_cpu() executes @func on each online CPU using the |
3273 | * system workqueue and blocks until all CPUs have completed. | |
b6136773 | 3274 | * schedule_on_each_cpu() is very slow. |
31ddd871 | 3275 | * |
d185af30 | 3276 | * Return: |
31ddd871 | 3277 | * 0 on success, -errno on failure. |
b6136773 | 3278 | */ |
65f27f38 | 3279 | int schedule_on_each_cpu(work_func_t func) |
15316ba8 CL |
3280 | { |
3281 | int cpu; | |
38f51568 | 3282 | struct work_struct __percpu *works; |
15316ba8 | 3283 | |
b6136773 AM |
3284 | works = alloc_percpu(struct work_struct); |
3285 | if (!works) | |
15316ba8 | 3286 | return -ENOMEM; |
b6136773 | 3287 | |
93981800 TH |
3288 | get_online_cpus(); |
3289 | ||
15316ba8 | 3290 | for_each_online_cpu(cpu) { |
9bfb1839 IM |
3291 | struct work_struct *work = per_cpu_ptr(works, cpu); |
3292 | ||
3293 | INIT_WORK(work, func); | |
b71ab8c2 | 3294 | schedule_work_on(cpu, work); |
65a64464 | 3295 | } |
93981800 TH |
3296 | |
3297 | for_each_online_cpu(cpu) | |
3298 | flush_work(per_cpu_ptr(works, cpu)); | |
3299 | ||
95402b38 | 3300 | put_online_cpus(); |
b6136773 | 3301 | free_percpu(works); |
15316ba8 CL |
3302 | return 0; |
3303 | } | |
3304 | ||
1fa44eca JB |
3305 | /** |
3306 | * execute_in_process_context - reliably execute the routine with user context | |
3307 | * @fn: the function to execute | |
1fa44eca JB |
3308 | * @ew: guaranteed storage for the execute work structure (must |
3309 | * be available when the work executes) | |
3310 | * | |
3311 | * Executes the function immediately if process context is available, | |
3312 | * otherwise schedules the function for delayed execution. | |
3313 | * | |
d185af30 | 3314 | * Return: 0 - function was executed |
1fa44eca JB |
3315 | * 1 - function was scheduled for execution |
3316 | */ | |
65f27f38 | 3317 | int execute_in_process_context(work_func_t fn, struct execute_work *ew) |
1fa44eca JB |
3318 | { |
3319 | if (!in_interrupt()) { | |
65f27f38 | 3320 | fn(&ew->work); |
1fa44eca JB |
3321 | return 0; |
3322 | } | |
3323 | ||
65f27f38 | 3324 | INIT_WORK(&ew->work, fn); |
1fa44eca JB |
3325 | schedule_work(&ew->work); |
3326 | ||
3327 | return 1; | |
3328 | } | |
3329 | EXPORT_SYMBOL_GPL(execute_in_process_context); | |
3330 | ||
6ba94429 FW |
3331 | /** |
3332 | * free_workqueue_attrs - free a workqueue_attrs | |
3333 | * @attrs: workqueue_attrs to free | |
226223ab | 3334 | * |
6ba94429 | 3335 | * Undo alloc_workqueue_attrs(). |
226223ab | 3336 | */ |
513c98d0 | 3337 | void free_workqueue_attrs(struct workqueue_attrs *attrs) |
226223ab | 3338 | { |
6ba94429 FW |
3339 | if (attrs) { |
3340 | free_cpumask_var(attrs->cpumask); | |
3341 | kfree(attrs); | |
3342 | } | |
226223ab TH |
3343 | } |
3344 | ||
6ba94429 FW |
3345 | /** |
3346 | * alloc_workqueue_attrs - allocate a workqueue_attrs | |
6ba94429 FW |
3347 | * |
3348 | * Allocate a new workqueue_attrs, initialize with default settings and | |
3349 | * return it. | |
3350 | * | |
3351 | * Return: The allocated new workqueue_attr on success. %NULL on failure. | |
3352 | */ | |
513c98d0 | 3353 | struct workqueue_attrs *alloc_workqueue_attrs(void) |
226223ab | 3354 | { |
6ba94429 | 3355 | struct workqueue_attrs *attrs; |
226223ab | 3356 | |
be69d00d | 3357 | attrs = kzalloc(sizeof(*attrs), GFP_KERNEL); |
6ba94429 FW |
3358 | if (!attrs) |
3359 | goto fail; | |
be69d00d | 3360 | if (!alloc_cpumask_var(&attrs->cpumask, GFP_KERNEL)) |
6ba94429 FW |
3361 | goto fail; |
3362 | ||
3363 | cpumask_copy(attrs->cpumask, cpu_possible_mask); | |
3364 | return attrs; | |
3365 | fail: | |
3366 | free_workqueue_attrs(attrs); | |
3367 | return NULL; | |
226223ab TH |
3368 | } |
3369 | ||
6ba94429 FW |
3370 | static void copy_workqueue_attrs(struct workqueue_attrs *to, |
3371 | const struct workqueue_attrs *from) | |
226223ab | 3372 | { |
6ba94429 FW |
3373 | to->nice = from->nice; |
3374 | cpumask_copy(to->cpumask, from->cpumask); | |
3375 | /* | |
3376 | * Unlike hash and equality test, this function doesn't ignore | |
3377 | * ->no_numa as it is used for both pool and wq attrs. Instead, | |
3378 | * get_unbound_pool() explicitly clears ->no_numa after copying. | |
3379 | */ | |
3380 | to->no_numa = from->no_numa; | |
226223ab TH |
3381 | } |
3382 | ||
6ba94429 FW |
3383 | /* hash value of the content of @attr */ |
3384 | static u32 wqattrs_hash(const struct workqueue_attrs *attrs) | |
226223ab | 3385 | { |
6ba94429 | 3386 | u32 hash = 0; |
226223ab | 3387 | |
6ba94429 FW |
3388 | hash = jhash_1word(attrs->nice, hash); |
3389 | hash = jhash(cpumask_bits(attrs->cpumask), | |
3390 | BITS_TO_LONGS(nr_cpumask_bits) * sizeof(long), hash); | |
3391 | return hash; | |
226223ab | 3392 | } |
226223ab | 3393 | |
6ba94429 FW |
3394 | /* content equality test */ |
3395 | static bool wqattrs_equal(const struct workqueue_attrs *a, | |
3396 | const struct workqueue_attrs *b) | |
226223ab | 3397 | { |
6ba94429 FW |
3398 | if (a->nice != b->nice) |
3399 | return false; | |
3400 | if (!cpumask_equal(a->cpumask, b->cpumask)) | |
3401 | return false; | |
3402 | return true; | |
226223ab TH |
3403 | } |
3404 | ||
6ba94429 FW |
3405 | /** |
3406 | * init_worker_pool - initialize a newly zalloc'd worker_pool | |
3407 | * @pool: worker_pool to initialize | |
3408 | * | |
402dd89d | 3409 | * Initialize a newly zalloc'd @pool. It also allocates @pool->attrs. |
6ba94429 FW |
3410 | * |
3411 | * Return: 0 on success, -errno on failure. Even on failure, all fields | |
3412 | * inside @pool proper are initialized and put_unbound_pool() can be called | |
3413 | * on @pool safely to release it. | |
3414 | */ | |
3415 | static int init_worker_pool(struct worker_pool *pool) | |
226223ab | 3416 | { |
6ba94429 FW |
3417 | spin_lock_init(&pool->lock); |
3418 | pool->id = -1; | |
3419 | pool->cpu = -1; | |
3420 | pool->node = NUMA_NO_NODE; | |
3421 | pool->flags |= POOL_DISASSOCIATED; | |
82607adc | 3422 | pool->watchdog_ts = jiffies; |
6ba94429 FW |
3423 | INIT_LIST_HEAD(&pool->worklist); |
3424 | INIT_LIST_HEAD(&pool->idle_list); | |
3425 | hash_init(pool->busy_hash); | |
226223ab | 3426 | |
32a6c723 | 3427 | timer_setup(&pool->idle_timer, idle_worker_timeout, TIMER_DEFERRABLE); |
226223ab | 3428 | |
32a6c723 | 3429 | timer_setup(&pool->mayday_timer, pool_mayday_timeout, 0); |
226223ab | 3430 | |
6ba94429 | 3431 | INIT_LIST_HEAD(&pool->workers); |
226223ab | 3432 | |
6ba94429 FW |
3433 | ida_init(&pool->worker_ida); |
3434 | INIT_HLIST_NODE(&pool->hash_node); | |
3435 | pool->refcnt = 1; | |
226223ab | 3436 | |
6ba94429 | 3437 | /* shouldn't fail above this point */ |
be69d00d | 3438 | pool->attrs = alloc_workqueue_attrs(); |
6ba94429 FW |
3439 | if (!pool->attrs) |
3440 | return -ENOMEM; | |
3441 | return 0; | |
226223ab TH |
3442 | } |
3443 | ||
669de8bd BVA |
3444 | #ifdef CONFIG_LOCKDEP |
3445 | static void wq_init_lockdep(struct workqueue_struct *wq) | |
3446 | { | |
3447 | char *lock_name; | |
3448 | ||
3449 | lockdep_register_key(&wq->key); | |
3450 | lock_name = kasprintf(GFP_KERNEL, "%s%s", "(wq_completion)", wq->name); | |
3451 | if (!lock_name) | |
3452 | lock_name = wq->name; | |
69a106c0 QC |
3453 | |
3454 | wq->lock_name = lock_name; | |
669de8bd BVA |
3455 | lockdep_init_map(&wq->lockdep_map, lock_name, &wq->key, 0); |
3456 | } | |
3457 | ||
3458 | static void wq_unregister_lockdep(struct workqueue_struct *wq) | |
3459 | { | |
3460 | lockdep_unregister_key(&wq->key); | |
3461 | } | |
3462 | ||
3463 | static void wq_free_lockdep(struct workqueue_struct *wq) | |
3464 | { | |
3465 | if (wq->lock_name != wq->name) | |
3466 | kfree(wq->lock_name); | |
3467 | } | |
3468 | #else | |
3469 | static void wq_init_lockdep(struct workqueue_struct *wq) | |
3470 | { | |
3471 | } | |
3472 | ||
3473 | static void wq_unregister_lockdep(struct workqueue_struct *wq) | |
3474 | { | |
3475 | } | |
3476 | ||
3477 | static void wq_free_lockdep(struct workqueue_struct *wq) | |
3478 | { | |
3479 | } | |
3480 | #endif | |
3481 | ||
6ba94429 | 3482 | static void rcu_free_wq(struct rcu_head *rcu) |
226223ab | 3483 | { |
6ba94429 FW |
3484 | struct workqueue_struct *wq = |
3485 | container_of(rcu, struct workqueue_struct, rcu); | |
226223ab | 3486 | |
669de8bd BVA |
3487 | wq_free_lockdep(wq); |
3488 | ||
6ba94429 FW |
3489 | if (!(wq->flags & WQ_UNBOUND)) |
3490 | free_percpu(wq->cpu_pwqs); | |
226223ab | 3491 | else |
6ba94429 | 3492 | free_workqueue_attrs(wq->unbound_attrs); |
226223ab | 3493 | |
6ba94429 FW |
3494 | kfree(wq->rescuer); |
3495 | kfree(wq); | |
226223ab TH |
3496 | } |
3497 | ||
6ba94429 | 3498 | static void rcu_free_pool(struct rcu_head *rcu) |
226223ab | 3499 | { |
6ba94429 | 3500 | struct worker_pool *pool = container_of(rcu, struct worker_pool, rcu); |
226223ab | 3501 | |
6ba94429 FW |
3502 | ida_destroy(&pool->worker_ida); |
3503 | free_workqueue_attrs(pool->attrs); | |
3504 | kfree(pool); | |
226223ab TH |
3505 | } |
3506 | ||
6ba94429 FW |
3507 | /** |
3508 | * put_unbound_pool - put a worker_pool | |
3509 | * @pool: worker_pool to put | |
3510 | * | |
24acfb71 | 3511 | * Put @pool. If its refcnt reaches zero, it gets destroyed in RCU |
6ba94429 FW |
3512 | * safe manner. get_unbound_pool() calls this function on its failure path |
3513 | * and this function should be able to release pools which went through, | |
3514 | * successfully or not, init_worker_pool(). | |
3515 | * | |
3516 | * Should be called with wq_pool_mutex held. | |
3517 | */ | |
3518 | static void put_unbound_pool(struct worker_pool *pool) | |
226223ab | 3519 | { |
6ba94429 FW |
3520 | DECLARE_COMPLETION_ONSTACK(detach_completion); |
3521 | struct worker *worker; | |
226223ab | 3522 | |
6ba94429 | 3523 | lockdep_assert_held(&wq_pool_mutex); |
226223ab | 3524 | |
6ba94429 FW |
3525 | if (--pool->refcnt) |
3526 | return; | |
226223ab | 3527 | |
6ba94429 FW |
3528 | /* sanity checks */ |
3529 | if (WARN_ON(!(pool->cpu < 0)) || | |
3530 | WARN_ON(!list_empty(&pool->worklist))) | |
3531 | return; | |
226223ab | 3532 | |
6ba94429 FW |
3533 | /* release id and unhash */ |
3534 | if (pool->id >= 0) | |
3535 | idr_remove(&worker_pool_idr, pool->id); | |
3536 | hash_del(&pool->hash_node); | |
d55262c4 | 3537 | |
6ba94429 | 3538 | /* |
692b4825 TH |
3539 | * Become the manager and destroy all workers. This prevents |
3540 | * @pool's workers from blocking on attach_mutex. We're the last | |
3541 | * manager and @pool gets freed with the flag set. | |
6ba94429 | 3542 | */ |
6ba94429 | 3543 | spin_lock_irq(&pool->lock); |
692b4825 TH |
3544 | wait_event_lock_irq(wq_manager_wait, |
3545 | !(pool->flags & POOL_MANAGER_ACTIVE), pool->lock); | |
3546 | pool->flags |= POOL_MANAGER_ACTIVE; | |
3547 | ||
6ba94429 FW |
3548 | while ((worker = first_idle_worker(pool))) |
3549 | destroy_worker(worker); | |
3550 | WARN_ON(pool->nr_workers || pool->nr_idle); | |
3551 | spin_unlock_irq(&pool->lock); | |
d55262c4 | 3552 | |
1258fae7 | 3553 | mutex_lock(&wq_pool_attach_mutex); |
6ba94429 FW |
3554 | if (!list_empty(&pool->workers)) |
3555 | pool->detach_completion = &detach_completion; | |
1258fae7 | 3556 | mutex_unlock(&wq_pool_attach_mutex); |
226223ab | 3557 | |
6ba94429 FW |
3558 | if (pool->detach_completion) |
3559 | wait_for_completion(pool->detach_completion); | |
226223ab | 3560 | |
6ba94429 FW |
3561 | /* shut down the timers */ |
3562 | del_timer_sync(&pool->idle_timer); | |
3563 | del_timer_sync(&pool->mayday_timer); | |
226223ab | 3564 | |
24acfb71 | 3565 | /* RCU protected to allow dereferences from get_work_pool() */ |
25b00775 | 3566 | call_rcu(&pool->rcu, rcu_free_pool); |
226223ab TH |
3567 | } |
3568 | ||
3569 | /** | |
6ba94429 FW |
3570 | * get_unbound_pool - get a worker_pool with the specified attributes |
3571 | * @attrs: the attributes of the worker_pool to get | |
226223ab | 3572 | * |
6ba94429 FW |
3573 | * Obtain a worker_pool which has the same attributes as @attrs, bump the |
3574 | * reference count and return it. If there already is a matching | |
3575 | * worker_pool, it will be used; otherwise, this function attempts to | |
3576 | * create a new one. | |
226223ab | 3577 | * |
6ba94429 | 3578 | * Should be called with wq_pool_mutex held. |
226223ab | 3579 | * |
6ba94429 FW |
3580 | * Return: On success, a worker_pool with the same attributes as @attrs. |
3581 | * On failure, %NULL. | |
226223ab | 3582 | */ |
6ba94429 | 3583 | static struct worker_pool *get_unbound_pool(const struct workqueue_attrs *attrs) |
226223ab | 3584 | { |
6ba94429 FW |
3585 | u32 hash = wqattrs_hash(attrs); |
3586 | struct worker_pool *pool; | |
3587 | int node; | |
e2273584 | 3588 | int target_node = NUMA_NO_NODE; |
226223ab | 3589 | |
6ba94429 | 3590 | lockdep_assert_held(&wq_pool_mutex); |
226223ab | 3591 | |
6ba94429 FW |
3592 | /* do we already have a matching pool? */ |
3593 | hash_for_each_possible(unbound_pool_hash, pool, hash_node, hash) { | |
3594 | if (wqattrs_equal(pool->attrs, attrs)) { | |
3595 | pool->refcnt++; | |
3596 | return pool; | |
3597 | } | |
3598 | } | |
226223ab | 3599 | |
e2273584 XP |
3600 | /* if cpumask is contained inside a NUMA node, we belong to that node */ |
3601 | if (wq_numa_enabled) { | |
3602 | for_each_node(node) { | |
3603 | if (cpumask_subset(attrs->cpumask, | |
3604 | wq_numa_possible_cpumask[node])) { | |
3605 | target_node = node; | |
3606 | break; | |
3607 | } | |
3608 | } | |
3609 | } | |
3610 | ||
6ba94429 | 3611 | /* nope, create a new one */ |
e2273584 | 3612 | pool = kzalloc_node(sizeof(*pool), GFP_KERNEL, target_node); |
6ba94429 FW |
3613 | if (!pool || init_worker_pool(pool) < 0) |
3614 | goto fail; | |
3615 | ||
3616 | lockdep_set_subclass(&pool->lock, 1); /* see put_pwq() */ | |
3617 | copy_workqueue_attrs(pool->attrs, attrs); | |
e2273584 | 3618 | pool->node = target_node; |
226223ab TH |
3619 | |
3620 | /* | |
6ba94429 FW |
3621 | * no_numa isn't a worker_pool attribute, always clear it. See |
3622 | * 'struct workqueue_attrs' comments for detail. | |
226223ab | 3623 | */ |
6ba94429 | 3624 | pool->attrs->no_numa = false; |
226223ab | 3625 | |
6ba94429 FW |
3626 | if (worker_pool_assign_id(pool) < 0) |
3627 | goto fail; | |
226223ab | 3628 | |
6ba94429 | 3629 | /* create and start the initial worker */ |
3347fa09 | 3630 | if (wq_online && !create_worker(pool)) |
6ba94429 | 3631 | goto fail; |
226223ab | 3632 | |
6ba94429 FW |
3633 | /* install */ |
3634 | hash_add(unbound_pool_hash, &pool->hash_node, hash); | |
226223ab | 3635 | |
6ba94429 FW |
3636 | return pool; |
3637 | fail: | |
3638 | if (pool) | |
3639 | put_unbound_pool(pool); | |
3640 | return NULL; | |
226223ab | 3641 | } |
226223ab | 3642 | |
6ba94429 | 3643 | static void rcu_free_pwq(struct rcu_head *rcu) |
7a4e344c | 3644 | { |
6ba94429 FW |
3645 | kmem_cache_free(pwq_cache, |
3646 | container_of(rcu, struct pool_workqueue, rcu)); | |
7a4e344c TH |
3647 | } |
3648 | ||
6ba94429 FW |
3649 | /* |
3650 | * Scheduled on system_wq by put_pwq() when an unbound pwq hits zero refcnt | |
3651 | * and needs to be destroyed. | |
7a4e344c | 3652 | */ |
6ba94429 | 3653 | static void pwq_unbound_release_workfn(struct work_struct *work) |
7a4e344c | 3654 | { |
6ba94429 FW |
3655 | struct pool_workqueue *pwq = container_of(work, struct pool_workqueue, |
3656 | unbound_release_work); | |
3657 | struct workqueue_struct *wq = pwq->wq; | |
3658 | struct worker_pool *pool = pwq->pool; | |
3659 | bool is_last; | |
7a4e344c | 3660 | |
6ba94429 FW |
3661 | if (WARN_ON_ONCE(!(wq->flags & WQ_UNBOUND))) |
3662 | return; | |
7a4e344c | 3663 | |
6ba94429 FW |
3664 | mutex_lock(&wq->mutex); |
3665 | list_del_rcu(&pwq->pwqs_node); | |
3666 | is_last = list_empty(&wq->pwqs); | |
3667 | mutex_unlock(&wq->mutex); | |
3668 | ||
3669 | mutex_lock(&wq_pool_mutex); | |
3670 | put_unbound_pool(pool); | |
3671 | mutex_unlock(&wq_pool_mutex); | |
3672 | ||
25b00775 | 3673 | call_rcu(&pwq->rcu, rcu_free_pwq); |
7a4e344c | 3674 | |
2865a8fb | 3675 | /* |
6ba94429 FW |
3676 | * If we're the last pwq going away, @wq is already dead and no one |
3677 | * is gonna access it anymore. Schedule RCU free. | |
2865a8fb | 3678 | */ |
669de8bd BVA |
3679 | if (is_last) { |
3680 | wq_unregister_lockdep(wq); | |
25b00775 | 3681 | call_rcu(&wq->rcu, rcu_free_wq); |
669de8bd | 3682 | } |
29c91e99 TH |
3683 | } |
3684 | ||
7a4e344c | 3685 | /** |
6ba94429 FW |
3686 | * pwq_adjust_max_active - update a pwq's max_active to the current setting |
3687 | * @pwq: target pool_workqueue | |
d185af30 | 3688 | * |
6ba94429 FW |
3689 | * If @pwq isn't freezing, set @pwq->max_active to the associated |
3690 | * workqueue's saved_max_active and activate delayed work items | |
3691 | * accordingly. If @pwq is freezing, clear @pwq->max_active to zero. | |
7a4e344c | 3692 | */ |
6ba94429 | 3693 | static void pwq_adjust_max_active(struct pool_workqueue *pwq) |
4e1a1f9a | 3694 | { |
6ba94429 FW |
3695 | struct workqueue_struct *wq = pwq->wq; |
3696 | bool freezable = wq->flags & WQ_FREEZABLE; | |
3347fa09 | 3697 | unsigned long flags; |
4e1a1f9a | 3698 | |
6ba94429 FW |
3699 | /* for @wq->saved_max_active */ |
3700 | lockdep_assert_held(&wq->mutex); | |
4e1a1f9a | 3701 | |
6ba94429 FW |
3702 | /* fast exit for non-freezable wqs */ |
3703 | if (!freezable && pwq->max_active == wq->saved_max_active) | |
3704 | return; | |
7a4e344c | 3705 | |
3347fa09 TH |
3706 | /* this function can be called during early boot w/ irq disabled */ |
3707 | spin_lock_irqsave(&pwq->pool->lock, flags); | |
29c91e99 | 3708 | |
6ba94429 FW |
3709 | /* |
3710 | * During [un]freezing, the caller is responsible for ensuring that | |
3711 | * this function is called at least once after @workqueue_freezing | |
3712 | * is updated and visible. | |
3713 | */ | |
3714 | if (!freezable || !workqueue_freezing) { | |
3715 | pwq->max_active = wq->saved_max_active; | |
4e1a1f9a | 3716 | |
6ba94429 FW |
3717 | while (!list_empty(&pwq->delayed_works) && |
3718 | pwq->nr_active < pwq->max_active) | |
3719 | pwq_activate_first_delayed(pwq); | |
e2dca7ad | 3720 | |
6ba94429 FW |
3721 | /* |
3722 | * Need to kick a worker after thawed or an unbound wq's | |
3723 | * max_active is bumped. It's a slow path. Do it always. | |
3724 | */ | |
3725 | wake_up_worker(pwq->pool); | |
3726 | } else { | |
3727 | pwq->max_active = 0; | |
3728 | } | |
e2dca7ad | 3729 | |
3347fa09 | 3730 | spin_unlock_irqrestore(&pwq->pool->lock, flags); |
e2dca7ad TH |
3731 | } |
3732 | ||
6ba94429 FW |
3733 | /* initialize newly alloced @pwq which is associated with @wq and @pool */ |
3734 | static void init_pwq(struct pool_workqueue *pwq, struct workqueue_struct *wq, | |
3735 | struct worker_pool *pool) | |
29c91e99 | 3736 | { |
6ba94429 | 3737 | BUG_ON((unsigned long)pwq & WORK_STRUCT_FLAG_MASK); |
29c91e99 | 3738 | |
6ba94429 FW |
3739 | memset(pwq, 0, sizeof(*pwq)); |
3740 | ||
3741 | pwq->pool = pool; | |
3742 | pwq->wq = wq; | |
3743 | pwq->flush_color = -1; | |
3744 | pwq->refcnt = 1; | |
3745 | INIT_LIST_HEAD(&pwq->delayed_works); | |
3746 | INIT_LIST_HEAD(&pwq->pwqs_node); | |
3747 | INIT_LIST_HEAD(&pwq->mayday_node); | |
3748 | INIT_WORK(&pwq->unbound_release_work, pwq_unbound_release_workfn); | |
29c91e99 TH |
3749 | } |
3750 | ||
6ba94429 FW |
3751 | /* sync @pwq with the current state of its associated wq and link it */ |
3752 | static void link_pwq(struct pool_workqueue *pwq) | |
29c91e99 | 3753 | { |
6ba94429 | 3754 | struct workqueue_struct *wq = pwq->wq; |
29c91e99 | 3755 | |
6ba94429 | 3756 | lockdep_assert_held(&wq->mutex); |
a892cacc | 3757 | |
6ba94429 FW |
3758 | /* may be called multiple times, ignore if already linked */ |
3759 | if (!list_empty(&pwq->pwqs_node)) | |
29c91e99 | 3760 | return; |
29c91e99 | 3761 | |
6ba94429 FW |
3762 | /* set the matching work_color */ |
3763 | pwq->work_color = wq->work_color; | |
29c91e99 | 3764 | |
6ba94429 FW |
3765 | /* sync max_active to the current setting */ |
3766 | pwq_adjust_max_active(pwq); | |
29c91e99 | 3767 | |
6ba94429 FW |
3768 | /* link in @pwq */ |
3769 | list_add_rcu(&pwq->pwqs_node, &wq->pwqs); | |
3770 | } | |
29c91e99 | 3771 | |
6ba94429 FW |
3772 | /* obtain a pool matching @attr and create a pwq associating the pool and @wq */ |
3773 | static struct pool_workqueue *alloc_unbound_pwq(struct workqueue_struct *wq, | |
3774 | const struct workqueue_attrs *attrs) | |
3775 | { | |
3776 | struct worker_pool *pool; | |
3777 | struct pool_workqueue *pwq; | |
60f5a4bc | 3778 | |
6ba94429 | 3779 | lockdep_assert_held(&wq_pool_mutex); |
60f5a4bc | 3780 | |
6ba94429 FW |
3781 | pool = get_unbound_pool(attrs); |
3782 | if (!pool) | |
3783 | return NULL; | |
60f5a4bc | 3784 | |
6ba94429 FW |
3785 | pwq = kmem_cache_alloc_node(pwq_cache, GFP_KERNEL, pool->node); |
3786 | if (!pwq) { | |
3787 | put_unbound_pool(pool); | |
3788 | return NULL; | |
3789 | } | |
29c91e99 | 3790 | |
6ba94429 FW |
3791 | init_pwq(pwq, wq, pool); |
3792 | return pwq; | |
3793 | } | |
29c91e99 | 3794 | |
29c91e99 | 3795 | /** |
30186c6f | 3796 | * wq_calc_node_cpumask - calculate a wq_attrs' cpumask for the specified node |
042f7df1 | 3797 | * @attrs: the wq_attrs of the default pwq of the target workqueue |
6ba94429 FW |
3798 | * @node: the target NUMA node |
3799 | * @cpu_going_down: if >= 0, the CPU to consider as offline | |
3800 | * @cpumask: outarg, the resulting cpumask | |
29c91e99 | 3801 | * |
6ba94429 FW |
3802 | * Calculate the cpumask a workqueue with @attrs should use on @node. If |
3803 | * @cpu_going_down is >= 0, that cpu is considered offline during | |
3804 | * calculation. The result is stored in @cpumask. | |
a892cacc | 3805 | * |
6ba94429 FW |
3806 | * If NUMA affinity is not enabled, @attrs->cpumask is always used. If |
3807 | * enabled and @node has online CPUs requested by @attrs, the returned | |
3808 | * cpumask is the intersection of the possible CPUs of @node and | |
3809 | * @attrs->cpumask. | |
d185af30 | 3810 | * |
6ba94429 FW |
3811 | * The caller is responsible for ensuring that the cpumask of @node stays |
3812 | * stable. | |
3813 | * | |
3814 | * Return: %true if the resulting @cpumask is different from @attrs->cpumask, | |
3815 | * %false if equal. | |
29c91e99 | 3816 | */ |
6ba94429 FW |
3817 | static bool wq_calc_node_cpumask(const struct workqueue_attrs *attrs, int node, |
3818 | int cpu_going_down, cpumask_t *cpumask) | |
29c91e99 | 3819 | { |
6ba94429 FW |
3820 | if (!wq_numa_enabled || attrs->no_numa) |
3821 | goto use_dfl; | |
29c91e99 | 3822 | |
6ba94429 FW |
3823 | /* does @node have any online CPUs @attrs wants? */ |
3824 | cpumask_and(cpumask, cpumask_of_node(node), attrs->cpumask); | |
3825 | if (cpu_going_down >= 0) | |
3826 | cpumask_clear_cpu(cpu_going_down, cpumask); | |
29c91e99 | 3827 | |
6ba94429 FW |
3828 | if (cpumask_empty(cpumask)) |
3829 | goto use_dfl; | |
4c16bd32 TH |
3830 | |
3831 | /* yeap, return possible CPUs in @node that @attrs wants */ | |
3832 | cpumask_and(cpumask, attrs->cpumask, wq_numa_possible_cpumask[node]); | |
1ad0f0a7 MB |
3833 | |
3834 | if (cpumask_empty(cpumask)) { | |
3835 | pr_warn_once("WARNING: workqueue cpumask: online intersect > " | |
3836 | "possible intersect\n"); | |
3837 | return false; | |
3838 | } | |
3839 | ||
4c16bd32 TH |
3840 | return !cpumask_equal(cpumask, attrs->cpumask); |
3841 | ||
3842 | use_dfl: | |
3843 | cpumask_copy(cpumask, attrs->cpumask); | |
3844 | return false; | |
3845 | } | |
3846 | ||
1befcf30 TH |
3847 | /* install @pwq into @wq's numa_pwq_tbl[] for @node and return the old pwq */ |
3848 | static struct pool_workqueue *numa_pwq_tbl_install(struct workqueue_struct *wq, | |
3849 | int node, | |
3850 | struct pool_workqueue *pwq) | |
3851 | { | |
3852 | struct pool_workqueue *old_pwq; | |
3853 | ||
5b95e1af | 3854 | lockdep_assert_held(&wq_pool_mutex); |
1befcf30 TH |
3855 | lockdep_assert_held(&wq->mutex); |
3856 | ||
3857 | /* link_pwq() can handle duplicate calls */ | |
3858 | link_pwq(pwq); | |
3859 | ||
3860 | old_pwq = rcu_access_pointer(wq->numa_pwq_tbl[node]); | |
3861 | rcu_assign_pointer(wq->numa_pwq_tbl[node], pwq); | |
3862 | return old_pwq; | |
3863 | } | |
3864 | ||
2d5f0764 LJ |
3865 | /* context to store the prepared attrs & pwqs before applying */ |
3866 | struct apply_wqattrs_ctx { | |
3867 | struct workqueue_struct *wq; /* target workqueue */ | |
3868 | struct workqueue_attrs *attrs; /* attrs to apply */ | |
042f7df1 | 3869 | struct list_head list; /* queued for batching commit */ |
2d5f0764 LJ |
3870 | struct pool_workqueue *dfl_pwq; |
3871 | struct pool_workqueue *pwq_tbl[]; | |
3872 | }; | |
3873 | ||
3874 | /* free the resources after success or abort */ | |
3875 | static void apply_wqattrs_cleanup(struct apply_wqattrs_ctx *ctx) | |
3876 | { | |
3877 | if (ctx) { | |
3878 | int node; | |
3879 | ||
3880 | for_each_node(node) | |
3881 | put_pwq_unlocked(ctx->pwq_tbl[node]); | |
3882 | put_pwq_unlocked(ctx->dfl_pwq); | |
3883 | ||
3884 | free_workqueue_attrs(ctx->attrs); | |
3885 | ||
3886 | kfree(ctx); | |
3887 | } | |
3888 | } | |
3889 | ||
3890 | /* allocate the attrs and pwqs for later installation */ | |
3891 | static struct apply_wqattrs_ctx * | |
3892 | apply_wqattrs_prepare(struct workqueue_struct *wq, | |
3893 | const struct workqueue_attrs *attrs) | |
9e8cd2f5 | 3894 | { |
2d5f0764 | 3895 | struct apply_wqattrs_ctx *ctx; |
4c16bd32 | 3896 | struct workqueue_attrs *new_attrs, *tmp_attrs; |
2d5f0764 | 3897 | int node; |
9e8cd2f5 | 3898 | |
2d5f0764 | 3899 | lockdep_assert_held(&wq_pool_mutex); |
9e8cd2f5 | 3900 | |
acafe7e3 | 3901 | ctx = kzalloc(struct_size(ctx, pwq_tbl, nr_node_ids), GFP_KERNEL); |
8719dcea | 3902 | |
be69d00d TG |
3903 | new_attrs = alloc_workqueue_attrs(); |
3904 | tmp_attrs = alloc_workqueue_attrs(); | |
2d5f0764 LJ |
3905 | if (!ctx || !new_attrs || !tmp_attrs) |
3906 | goto out_free; | |
13e2e556 | 3907 | |
042f7df1 LJ |
3908 | /* |
3909 | * Calculate the attrs of the default pwq. | |
3910 | * If the user configured cpumask doesn't overlap with the | |
3911 | * wq_unbound_cpumask, we fallback to the wq_unbound_cpumask. | |
3912 | */ | |
13e2e556 | 3913 | copy_workqueue_attrs(new_attrs, attrs); |
b05a7928 | 3914 | cpumask_and(new_attrs->cpumask, new_attrs->cpumask, wq_unbound_cpumask); |
042f7df1 LJ |
3915 | if (unlikely(cpumask_empty(new_attrs->cpumask))) |
3916 | cpumask_copy(new_attrs->cpumask, wq_unbound_cpumask); | |
13e2e556 | 3917 | |
4c16bd32 TH |
3918 | /* |
3919 | * We may create multiple pwqs with differing cpumasks. Make a | |
3920 | * copy of @new_attrs which will be modified and used to obtain | |
3921 | * pools. | |
3922 | */ | |
3923 | copy_workqueue_attrs(tmp_attrs, new_attrs); | |
3924 | ||
4c16bd32 TH |
3925 | /* |
3926 | * If something goes wrong during CPU up/down, we'll fall back to | |
3927 | * the default pwq covering whole @attrs->cpumask. Always create | |
3928 | * it even if we don't use it immediately. | |
3929 | */ | |
2d5f0764 LJ |
3930 | ctx->dfl_pwq = alloc_unbound_pwq(wq, new_attrs); |
3931 | if (!ctx->dfl_pwq) | |
3932 | goto out_free; | |
4c16bd32 TH |
3933 | |
3934 | for_each_node(node) { | |
042f7df1 | 3935 | if (wq_calc_node_cpumask(new_attrs, node, -1, tmp_attrs->cpumask)) { |
2d5f0764 LJ |
3936 | ctx->pwq_tbl[node] = alloc_unbound_pwq(wq, tmp_attrs); |
3937 | if (!ctx->pwq_tbl[node]) | |
3938 | goto out_free; | |
4c16bd32 | 3939 | } else { |
2d5f0764 LJ |
3940 | ctx->dfl_pwq->refcnt++; |
3941 | ctx->pwq_tbl[node] = ctx->dfl_pwq; | |
4c16bd32 TH |
3942 | } |
3943 | } | |
3944 | ||
042f7df1 LJ |
3945 | /* save the user configured attrs and sanitize it. */ |
3946 | copy_workqueue_attrs(new_attrs, attrs); | |
3947 | cpumask_and(new_attrs->cpumask, new_attrs->cpumask, cpu_possible_mask); | |
2d5f0764 | 3948 | ctx->attrs = new_attrs; |
042f7df1 | 3949 | |
2d5f0764 LJ |
3950 | ctx->wq = wq; |
3951 | free_workqueue_attrs(tmp_attrs); | |
3952 | return ctx; | |
3953 | ||
3954 | out_free: | |
3955 | free_workqueue_attrs(tmp_attrs); | |
3956 | free_workqueue_attrs(new_attrs); | |
3957 | apply_wqattrs_cleanup(ctx); | |
3958 | return NULL; | |
3959 | } | |
3960 | ||
3961 | /* set attrs and install prepared pwqs, @ctx points to old pwqs on return */ | |
3962 | static void apply_wqattrs_commit(struct apply_wqattrs_ctx *ctx) | |
3963 | { | |
3964 | int node; | |
9e8cd2f5 | 3965 | |
4c16bd32 | 3966 | /* all pwqs have been created successfully, let's install'em */ |
2d5f0764 | 3967 | mutex_lock(&ctx->wq->mutex); |
a892cacc | 3968 | |
2d5f0764 | 3969 | copy_workqueue_attrs(ctx->wq->unbound_attrs, ctx->attrs); |
4c16bd32 TH |
3970 | |
3971 | /* save the previous pwq and install the new one */ | |
f147f29e | 3972 | for_each_node(node) |
2d5f0764 LJ |
3973 | ctx->pwq_tbl[node] = numa_pwq_tbl_install(ctx->wq, node, |
3974 | ctx->pwq_tbl[node]); | |
4c16bd32 TH |
3975 | |
3976 | /* @dfl_pwq might not have been used, ensure it's linked */ | |
2d5f0764 LJ |
3977 | link_pwq(ctx->dfl_pwq); |
3978 | swap(ctx->wq->dfl_pwq, ctx->dfl_pwq); | |
f147f29e | 3979 | |
2d5f0764 LJ |
3980 | mutex_unlock(&ctx->wq->mutex); |
3981 | } | |
9e8cd2f5 | 3982 | |
a0111cf6 LJ |
3983 | static void apply_wqattrs_lock(void) |
3984 | { | |
3985 | /* CPUs should stay stable across pwq creations and installations */ | |
3986 | get_online_cpus(); | |
3987 | mutex_lock(&wq_pool_mutex); | |
3988 | } | |
3989 | ||
3990 | static void apply_wqattrs_unlock(void) | |
3991 | { | |
3992 | mutex_unlock(&wq_pool_mutex); | |
3993 | put_online_cpus(); | |
3994 | } | |
3995 | ||
3996 | static int apply_workqueue_attrs_locked(struct workqueue_struct *wq, | |
3997 | const struct workqueue_attrs *attrs) | |
2d5f0764 LJ |
3998 | { |
3999 | struct apply_wqattrs_ctx *ctx; | |
4c16bd32 | 4000 | |
2d5f0764 LJ |
4001 | /* only unbound workqueues can change attributes */ |
4002 | if (WARN_ON(!(wq->flags & WQ_UNBOUND))) | |
4003 | return -EINVAL; | |
13e2e556 | 4004 | |
2d5f0764 | 4005 | /* creating multiple pwqs breaks ordering guarantee */ |
0a94efb5 TH |
4006 | if (!list_empty(&wq->pwqs)) { |
4007 | if (WARN_ON(wq->flags & __WQ_ORDERED_EXPLICIT)) | |
4008 | return -EINVAL; | |
4009 | ||
4010 | wq->flags &= ~__WQ_ORDERED; | |
4011 | } | |
2d5f0764 | 4012 | |
2d5f0764 | 4013 | ctx = apply_wqattrs_prepare(wq, attrs); |
6201171e | 4014 | if (!ctx) |
4015 | return -ENOMEM; | |
2d5f0764 LJ |
4016 | |
4017 | /* the ctx has been prepared successfully, let's commit it */ | |
6201171e | 4018 | apply_wqattrs_commit(ctx); |
2d5f0764 LJ |
4019 | apply_wqattrs_cleanup(ctx); |
4020 | ||
6201171e | 4021 | return 0; |
9e8cd2f5 TH |
4022 | } |
4023 | ||
a0111cf6 LJ |
4024 | /** |
4025 | * apply_workqueue_attrs - apply new workqueue_attrs to an unbound workqueue | |
4026 | * @wq: the target workqueue | |
4027 | * @attrs: the workqueue_attrs to apply, allocated with alloc_workqueue_attrs() | |
4028 | * | |
4029 | * Apply @attrs to an unbound workqueue @wq. Unless disabled, on NUMA | |
4030 | * machines, this function maps a separate pwq to each NUMA node with | |
4031 | * possibles CPUs in @attrs->cpumask so that work items are affine to the | |
4032 | * NUMA node it was issued on. Older pwqs are released as in-flight work | |
4033 | * items finish. Note that a work item which repeatedly requeues itself | |
4034 | * back-to-back will stay on its current pwq. | |
4035 | * | |
4036 | * Performs GFP_KERNEL allocations. | |
4037 | * | |
509b3204 DJ |
4038 | * Assumes caller has CPU hotplug read exclusion, i.e. get_online_cpus(). |
4039 | * | |
a0111cf6 LJ |
4040 | * Return: 0 on success and -errno on failure. |
4041 | */ | |
513c98d0 | 4042 | int apply_workqueue_attrs(struct workqueue_struct *wq, |
a0111cf6 LJ |
4043 | const struct workqueue_attrs *attrs) |
4044 | { | |
4045 | int ret; | |
4046 | ||
509b3204 DJ |
4047 | lockdep_assert_cpus_held(); |
4048 | ||
4049 | mutex_lock(&wq_pool_mutex); | |
a0111cf6 | 4050 | ret = apply_workqueue_attrs_locked(wq, attrs); |
509b3204 | 4051 | mutex_unlock(&wq_pool_mutex); |
a0111cf6 LJ |
4052 | |
4053 | return ret; | |
4054 | } | |
4055 | ||
4c16bd32 TH |
4056 | /** |
4057 | * wq_update_unbound_numa - update NUMA affinity of a wq for CPU hot[un]plug | |
4058 | * @wq: the target workqueue | |
4059 | * @cpu: the CPU coming up or going down | |
4060 | * @online: whether @cpu is coming up or going down | |
4061 | * | |
4062 | * This function is to be called from %CPU_DOWN_PREPARE, %CPU_ONLINE and | |
4063 | * %CPU_DOWN_FAILED. @cpu is being hot[un]plugged, update NUMA affinity of | |
4064 | * @wq accordingly. | |
4065 | * | |
4066 | * If NUMA affinity can't be adjusted due to memory allocation failure, it | |
4067 | * falls back to @wq->dfl_pwq which may not be optimal but is always | |
4068 | * correct. | |
4069 | * | |
4070 | * Note that when the last allowed CPU of a NUMA node goes offline for a | |
4071 | * workqueue with a cpumask spanning multiple nodes, the workers which were | |
4072 | * already executing the work items for the workqueue will lose their CPU | |
4073 | * affinity and may execute on any CPU. This is similar to how per-cpu | |
4074 | * workqueues behave on CPU_DOWN. If a workqueue user wants strict | |
4075 | * affinity, it's the user's responsibility to flush the work item from | |
4076 | * CPU_DOWN_PREPARE. | |
4077 | */ | |
4078 | static void wq_update_unbound_numa(struct workqueue_struct *wq, int cpu, | |
4079 | bool online) | |
4080 | { | |
4081 | int node = cpu_to_node(cpu); | |
4082 | int cpu_off = online ? -1 : cpu; | |
4083 | struct pool_workqueue *old_pwq = NULL, *pwq; | |
4084 | struct workqueue_attrs *target_attrs; | |
4085 | cpumask_t *cpumask; | |
4086 | ||
4087 | lockdep_assert_held(&wq_pool_mutex); | |
4088 | ||
f7142ed4 LJ |
4089 | if (!wq_numa_enabled || !(wq->flags & WQ_UNBOUND) || |
4090 | wq->unbound_attrs->no_numa) | |
4c16bd32 TH |
4091 | return; |
4092 | ||
4093 | /* | |
4094 | * We don't wanna alloc/free wq_attrs for each wq for each CPU. | |
4095 | * Let's use a preallocated one. The following buf is protected by | |
4096 | * CPU hotplug exclusion. | |
4097 | */ | |
4098 | target_attrs = wq_update_unbound_numa_attrs_buf; | |
4099 | cpumask = target_attrs->cpumask; | |
4100 | ||
4c16bd32 TH |
4101 | copy_workqueue_attrs(target_attrs, wq->unbound_attrs); |
4102 | pwq = unbound_pwq_by_node(wq, node); | |
4103 | ||
4104 | /* | |
4105 | * Let's determine what needs to be done. If the target cpumask is | |
042f7df1 LJ |
4106 | * different from the default pwq's, we need to compare it to @pwq's |
4107 | * and create a new one if they don't match. If the target cpumask | |
4108 | * equals the default pwq's, the default pwq should be used. | |
4c16bd32 | 4109 | */ |
042f7df1 | 4110 | if (wq_calc_node_cpumask(wq->dfl_pwq->pool->attrs, node, cpu_off, cpumask)) { |
4c16bd32 | 4111 | if (cpumask_equal(cpumask, pwq->pool->attrs->cpumask)) |
f7142ed4 | 4112 | return; |
4c16bd32 | 4113 | } else { |
534a3fbb | 4114 | goto use_dfl_pwq; |
4c16bd32 TH |
4115 | } |
4116 | ||
4c16bd32 TH |
4117 | /* create a new pwq */ |
4118 | pwq = alloc_unbound_pwq(wq, target_attrs); | |
4119 | if (!pwq) { | |
2d916033 FF |
4120 | pr_warn("workqueue: allocation failed while updating NUMA affinity of \"%s\"\n", |
4121 | wq->name); | |
77f300b1 | 4122 | goto use_dfl_pwq; |
4c16bd32 TH |
4123 | } |
4124 | ||
f7142ed4 | 4125 | /* Install the new pwq. */ |
4c16bd32 TH |
4126 | mutex_lock(&wq->mutex); |
4127 | old_pwq = numa_pwq_tbl_install(wq, node, pwq); | |
4128 | goto out_unlock; | |
4129 | ||
4130 | use_dfl_pwq: | |
f7142ed4 | 4131 | mutex_lock(&wq->mutex); |
4c16bd32 TH |
4132 | spin_lock_irq(&wq->dfl_pwq->pool->lock); |
4133 | get_pwq(wq->dfl_pwq); | |
4134 | spin_unlock_irq(&wq->dfl_pwq->pool->lock); | |
4135 | old_pwq = numa_pwq_tbl_install(wq, node, wq->dfl_pwq); | |
4136 | out_unlock: | |
4137 | mutex_unlock(&wq->mutex); | |
4138 | put_pwq_unlocked(old_pwq); | |
4139 | } | |
4140 | ||
30cdf249 | 4141 | static int alloc_and_link_pwqs(struct workqueue_struct *wq) |
0f900049 | 4142 | { |
49e3cf44 | 4143 | bool highpri = wq->flags & WQ_HIGHPRI; |
8a2b7538 | 4144 | int cpu, ret; |
30cdf249 TH |
4145 | |
4146 | if (!(wq->flags & WQ_UNBOUND)) { | |
420c0ddb TH |
4147 | wq->cpu_pwqs = alloc_percpu(struct pool_workqueue); |
4148 | if (!wq->cpu_pwqs) | |
30cdf249 TH |
4149 | return -ENOMEM; |
4150 | ||
4151 | for_each_possible_cpu(cpu) { | |
7fb98ea7 TH |
4152 | struct pool_workqueue *pwq = |
4153 | per_cpu_ptr(wq->cpu_pwqs, cpu); | |
7a62c2c8 | 4154 | struct worker_pool *cpu_pools = |
f02ae73a | 4155 | per_cpu(cpu_worker_pools, cpu); |
f3421797 | 4156 | |
f147f29e TH |
4157 | init_pwq(pwq, wq, &cpu_pools[highpri]); |
4158 | ||
4159 | mutex_lock(&wq->mutex); | |
1befcf30 | 4160 | link_pwq(pwq); |
f147f29e | 4161 | mutex_unlock(&wq->mutex); |
30cdf249 | 4162 | } |
9e8cd2f5 | 4163 | return 0; |
509b3204 DJ |
4164 | } |
4165 | ||
4166 | get_online_cpus(); | |
4167 | if (wq->flags & __WQ_ORDERED) { | |
8a2b7538 TH |
4168 | ret = apply_workqueue_attrs(wq, ordered_wq_attrs[highpri]); |
4169 | /* there should only be single pwq for ordering guarantee */ | |
4170 | WARN(!ret && (wq->pwqs.next != &wq->dfl_pwq->pwqs_node || | |
4171 | wq->pwqs.prev != &wq->dfl_pwq->pwqs_node), | |
4172 | "ordering guarantee broken for workqueue %s\n", wq->name); | |
30cdf249 | 4173 | } else { |
509b3204 | 4174 | ret = apply_workqueue_attrs(wq, unbound_std_wq_attrs[highpri]); |
30cdf249 | 4175 | } |
509b3204 DJ |
4176 | put_online_cpus(); |
4177 | ||
4178 | return ret; | |
0f900049 TH |
4179 | } |
4180 | ||
f3421797 TH |
4181 | static int wq_clamp_max_active(int max_active, unsigned int flags, |
4182 | const char *name) | |
b71ab8c2 | 4183 | { |
f3421797 TH |
4184 | int lim = flags & WQ_UNBOUND ? WQ_UNBOUND_MAX_ACTIVE : WQ_MAX_ACTIVE; |
4185 | ||
4186 | if (max_active < 1 || max_active > lim) | |
044c782c VI |
4187 | pr_warn("workqueue: max_active %d requested for %s is out of range, clamping between %d and %d\n", |
4188 | max_active, name, 1, lim); | |
b71ab8c2 | 4189 | |
f3421797 | 4190 | return clamp_val(max_active, 1, lim); |
b71ab8c2 TH |
4191 | } |
4192 | ||
983c7515 TH |
4193 | /* |
4194 | * Workqueues which may be used during memory reclaim should have a rescuer | |
4195 | * to guarantee forward progress. | |
4196 | */ | |
4197 | static int init_rescuer(struct workqueue_struct *wq) | |
4198 | { | |
4199 | struct worker *rescuer; | |
4200 | int ret; | |
4201 | ||
4202 | if (!(wq->flags & WQ_MEM_RECLAIM)) | |
4203 | return 0; | |
4204 | ||
4205 | rescuer = alloc_worker(NUMA_NO_NODE); | |
4206 | if (!rescuer) | |
4207 | return -ENOMEM; | |
4208 | ||
4209 | rescuer->rescue_wq = wq; | |
4210 | rescuer->task = kthread_create(rescuer_thread, rescuer, "%s", wq->name); | |
4211 | ret = PTR_ERR_OR_ZERO(rescuer->task); | |
4212 | if (ret) { | |
4213 | kfree(rescuer); | |
4214 | return ret; | |
4215 | } | |
4216 | ||
4217 | wq->rescuer = rescuer; | |
4218 | kthread_bind_mask(rescuer->task, cpu_possible_mask); | |
4219 | wake_up_process(rescuer->task); | |
4220 | ||
4221 | return 0; | |
4222 | } | |
4223 | ||
a2775bbc | 4224 | __printf(1, 4) |
669de8bd BVA |
4225 | struct workqueue_struct *alloc_workqueue(const char *fmt, |
4226 | unsigned int flags, | |
4227 | int max_active, ...) | |
1da177e4 | 4228 | { |
df2d5ae4 | 4229 | size_t tbl_size = 0; |
ecf6881f | 4230 | va_list args; |
1da177e4 | 4231 | struct workqueue_struct *wq; |
49e3cf44 | 4232 | struct pool_workqueue *pwq; |
b196be89 | 4233 | |
5c0338c6 TH |
4234 | /* |
4235 | * Unbound && max_active == 1 used to imply ordered, which is no | |
4236 | * longer the case on NUMA machines due to per-node pools. While | |
4237 | * alloc_ordered_workqueue() is the right way to create an ordered | |
4238 | * workqueue, keep the previous behavior to avoid subtle breakages | |
4239 | * on NUMA. | |
4240 | */ | |
4241 | if ((flags & WQ_UNBOUND) && max_active == 1) | |
4242 | flags |= __WQ_ORDERED; | |
4243 | ||
cee22a15 VK |
4244 | /* see the comment above the definition of WQ_POWER_EFFICIENT */ |
4245 | if ((flags & WQ_POWER_EFFICIENT) && wq_power_efficient) | |
4246 | flags |= WQ_UNBOUND; | |
4247 | ||
ecf6881f | 4248 | /* allocate wq and format name */ |
df2d5ae4 | 4249 | if (flags & WQ_UNBOUND) |
ddcb57e2 | 4250 | tbl_size = nr_node_ids * sizeof(wq->numa_pwq_tbl[0]); |
df2d5ae4 TH |
4251 | |
4252 | wq = kzalloc(sizeof(*wq) + tbl_size, GFP_KERNEL); | |
b196be89 | 4253 | if (!wq) |
d2c1d404 | 4254 | return NULL; |
b196be89 | 4255 | |
6029a918 | 4256 | if (flags & WQ_UNBOUND) { |
be69d00d | 4257 | wq->unbound_attrs = alloc_workqueue_attrs(); |
6029a918 TH |
4258 | if (!wq->unbound_attrs) |
4259 | goto err_free_wq; | |
4260 | } | |
4261 | ||
669de8bd | 4262 | va_start(args, max_active); |
ecf6881f | 4263 | vsnprintf(wq->name, sizeof(wq->name), fmt, args); |
b196be89 | 4264 | va_end(args); |
1da177e4 | 4265 | |
d320c038 | 4266 | max_active = max_active ?: WQ_DFL_ACTIVE; |
b196be89 | 4267 | max_active = wq_clamp_max_active(max_active, flags, wq->name); |
3af24433 | 4268 | |
b196be89 | 4269 | /* init wq */ |
97e37d7b | 4270 | wq->flags = flags; |
a0a1a5fd | 4271 | wq->saved_max_active = max_active; |
3c25a55d | 4272 | mutex_init(&wq->mutex); |
112202d9 | 4273 | atomic_set(&wq->nr_pwqs_to_flush, 0); |
30cdf249 | 4274 | INIT_LIST_HEAD(&wq->pwqs); |
73f53c4a TH |
4275 | INIT_LIST_HEAD(&wq->flusher_queue); |
4276 | INIT_LIST_HEAD(&wq->flusher_overflow); | |
493a1724 | 4277 | INIT_LIST_HEAD(&wq->maydays); |
502ca9d8 | 4278 | |
669de8bd | 4279 | wq_init_lockdep(wq); |
cce1a165 | 4280 | INIT_LIST_HEAD(&wq->list); |
3af24433 | 4281 | |
30cdf249 | 4282 | if (alloc_and_link_pwqs(wq) < 0) |
82efcab3 | 4283 | goto err_unreg_lockdep; |
1537663f | 4284 | |
40c17f75 | 4285 | if (wq_online && init_rescuer(wq) < 0) |
983c7515 | 4286 | goto err_destroy; |
3af24433 | 4287 | |
226223ab TH |
4288 | if ((wq->flags & WQ_SYSFS) && workqueue_sysfs_register(wq)) |
4289 | goto err_destroy; | |
4290 | ||
a0a1a5fd | 4291 | /* |
68e13a67 LJ |
4292 | * wq_pool_mutex protects global freeze state and workqueues list. |
4293 | * Grab it, adjust max_active and add the new @wq to workqueues | |
4294 | * list. | |
a0a1a5fd | 4295 | */ |
68e13a67 | 4296 | mutex_lock(&wq_pool_mutex); |
a0a1a5fd | 4297 | |
a357fc03 | 4298 | mutex_lock(&wq->mutex); |
699ce097 TH |
4299 | for_each_pwq(pwq, wq) |
4300 | pwq_adjust_max_active(pwq); | |
a357fc03 | 4301 | mutex_unlock(&wq->mutex); |
a0a1a5fd | 4302 | |
e2dca7ad | 4303 | list_add_tail_rcu(&wq->list, &workqueues); |
a0a1a5fd | 4304 | |
68e13a67 | 4305 | mutex_unlock(&wq_pool_mutex); |
1537663f | 4306 | |
3af24433 | 4307 | return wq; |
d2c1d404 | 4308 | |
82efcab3 | 4309 | err_unreg_lockdep: |
009bb421 BVA |
4310 | wq_unregister_lockdep(wq); |
4311 | wq_free_lockdep(wq); | |
82efcab3 | 4312 | err_free_wq: |
6029a918 | 4313 | free_workqueue_attrs(wq->unbound_attrs); |
d2c1d404 TH |
4314 | kfree(wq); |
4315 | return NULL; | |
4316 | err_destroy: | |
4317 | destroy_workqueue(wq); | |
4690c4ab | 4318 | return NULL; |
3af24433 | 4319 | } |
669de8bd | 4320 | EXPORT_SYMBOL_GPL(alloc_workqueue); |
1da177e4 | 4321 | |
c29eb853 TH |
4322 | static bool pwq_busy(struct pool_workqueue *pwq) |
4323 | { | |
4324 | int i; | |
4325 | ||
4326 | for (i = 0; i < WORK_NR_COLORS; i++) | |
4327 | if (pwq->nr_in_flight[i]) | |
4328 | return true; | |
4329 | ||
4330 | if ((pwq != pwq->wq->dfl_pwq) && (pwq->refcnt > 1)) | |
4331 | return true; | |
4332 | if (pwq->nr_active || !list_empty(&pwq->delayed_works)) | |
4333 | return true; | |
4334 | ||
4335 | return false; | |
4336 | } | |
4337 | ||
3af24433 ON |
4338 | /** |
4339 | * destroy_workqueue - safely terminate a workqueue | |
4340 | * @wq: target workqueue | |
4341 | * | |
4342 | * Safely destroy a workqueue. All work currently pending will be done first. | |
4343 | */ | |
4344 | void destroy_workqueue(struct workqueue_struct *wq) | |
4345 | { | |
49e3cf44 | 4346 | struct pool_workqueue *pwq; |
4c16bd32 | 4347 | int node; |
3af24433 | 4348 | |
def98c84 TH |
4349 | /* |
4350 | * Remove it from sysfs first so that sanity check failure doesn't | |
4351 | * lead to sysfs name conflicts. | |
4352 | */ | |
4353 | workqueue_sysfs_unregister(wq); | |
4354 | ||
9c5a2ba7 TH |
4355 | /* drain it before proceeding with destruction */ |
4356 | drain_workqueue(wq); | |
c8efcc25 | 4357 | |
def98c84 TH |
4358 | /* kill rescuer, if sanity checks fail, leave it w/o rescuer */ |
4359 | if (wq->rescuer) { | |
4360 | struct worker *rescuer = wq->rescuer; | |
4361 | ||
4362 | /* this prevents new queueing */ | |
4363 | spin_lock_irq(&wq_mayday_lock); | |
4364 | wq->rescuer = NULL; | |
4365 | spin_unlock_irq(&wq_mayday_lock); | |
4366 | ||
4367 | /* rescuer will empty maydays list before exiting */ | |
4368 | kthread_stop(rescuer->task); | |
8efe1223 | 4369 | kfree(rescuer); |
def98c84 TH |
4370 | } |
4371 | ||
c29eb853 TH |
4372 | /* |
4373 | * Sanity checks - grab all the locks so that we wait for all | |
4374 | * in-flight operations which may do put_pwq(). | |
4375 | */ | |
4376 | mutex_lock(&wq_pool_mutex); | |
b09f4fd3 | 4377 | mutex_lock(&wq->mutex); |
49e3cf44 | 4378 | for_each_pwq(pwq, wq) { |
c29eb853 TH |
4379 | spin_lock_irq(&pwq->pool->lock); |
4380 | if (WARN_ON(pwq_busy(pwq))) { | |
1d9a6159 KW |
4381 | pr_warn("%s: %s has the following busy pwq\n", |
4382 | __func__, wq->name); | |
c29eb853 TH |
4383 | show_pwq(pwq); |
4384 | spin_unlock_irq(&pwq->pool->lock); | |
b09f4fd3 | 4385 | mutex_unlock(&wq->mutex); |
c29eb853 | 4386 | mutex_unlock(&wq_pool_mutex); |
fa07fb6a | 4387 | show_workqueue_state(); |
6183c009 | 4388 | return; |
76af4d93 | 4389 | } |
c29eb853 | 4390 | spin_unlock_irq(&pwq->pool->lock); |
6183c009 | 4391 | } |
b09f4fd3 | 4392 | mutex_unlock(&wq->mutex); |
c29eb853 | 4393 | mutex_unlock(&wq_pool_mutex); |
6183c009 | 4394 | |
a0a1a5fd TH |
4395 | /* |
4396 | * wq list is used to freeze wq, remove from list after | |
4397 | * flushing is complete in case freeze races us. | |
4398 | */ | |
68e13a67 | 4399 | mutex_lock(&wq_pool_mutex); |
e2dca7ad | 4400 | list_del_rcu(&wq->list); |
68e13a67 | 4401 | mutex_unlock(&wq_pool_mutex); |
3af24433 | 4402 | |
8864b4e5 | 4403 | if (!(wq->flags & WQ_UNBOUND)) { |
669de8bd | 4404 | wq_unregister_lockdep(wq); |
8864b4e5 TH |
4405 | /* |
4406 | * The base ref is never dropped on per-cpu pwqs. Directly | |
e2dca7ad | 4407 | * schedule RCU free. |
8864b4e5 | 4408 | */ |
25b00775 | 4409 | call_rcu(&wq->rcu, rcu_free_wq); |
8864b4e5 TH |
4410 | } else { |
4411 | /* | |
4412 | * We're the sole accessor of @wq at this point. Directly | |
4c16bd32 TH |
4413 | * access numa_pwq_tbl[] and dfl_pwq to put the base refs. |
4414 | * @wq will be freed when the last pwq is released. | |
8864b4e5 | 4415 | */ |
4c16bd32 TH |
4416 | for_each_node(node) { |
4417 | pwq = rcu_access_pointer(wq->numa_pwq_tbl[node]); | |
4418 | RCU_INIT_POINTER(wq->numa_pwq_tbl[node], NULL); | |
4419 | put_pwq_unlocked(pwq); | |
4420 | } | |
4421 | ||
4422 | /* | |
4423 | * Put dfl_pwq. @wq may be freed any time after dfl_pwq is | |
4424 | * put. Don't access it afterwards. | |
4425 | */ | |
4426 | pwq = wq->dfl_pwq; | |
4427 | wq->dfl_pwq = NULL; | |
dce90d47 | 4428 | put_pwq_unlocked(pwq); |
29c91e99 | 4429 | } |
3af24433 ON |
4430 | } |
4431 | EXPORT_SYMBOL_GPL(destroy_workqueue); | |
4432 | ||
dcd989cb TH |
4433 | /** |
4434 | * workqueue_set_max_active - adjust max_active of a workqueue | |
4435 | * @wq: target workqueue | |
4436 | * @max_active: new max_active value. | |
4437 | * | |
4438 | * Set max_active of @wq to @max_active. | |
4439 | * | |
4440 | * CONTEXT: | |
4441 | * Don't call from IRQ context. | |
4442 | */ | |
4443 | void workqueue_set_max_active(struct workqueue_struct *wq, int max_active) | |
4444 | { | |
49e3cf44 | 4445 | struct pool_workqueue *pwq; |
dcd989cb | 4446 | |
8719dcea | 4447 | /* disallow meddling with max_active for ordered workqueues */ |
0a94efb5 | 4448 | if (WARN_ON(wq->flags & __WQ_ORDERED_EXPLICIT)) |
8719dcea TH |
4449 | return; |
4450 | ||
f3421797 | 4451 | max_active = wq_clamp_max_active(max_active, wq->flags, wq->name); |
dcd989cb | 4452 | |
a357fc03 | 4453 | mutex_lock(&wq->mutex); |
dcd989cb | 4454 | |
0a94efb5 | 4455 | wq->flags &= ~__WQ_ORDERED; |
dcd989cb TH |
4456 | wq->saved_max_active = max_active; |
4457 | ||
699ce097 TH |
4458 | for_each_pwq(pwq, wq) |
4459 | pwq_adjust_max_active(pwq); | |
93981800 | 4460 | |
a357fc03 | 4461 | mutex_unlock(&wq->mutex); |
15316ba8 | 4462 | } |
dcd989cb | 4463 | EXPORT_SYMBOL_GPL(workqueue_set_max_active); |
15316ba8 | 4464 | |
27d4ee03 LW |
4465 | /** |
4466 | * current_work - retrieve %current task's work struct | |
4467 | * | |
4468 | * Determine if %current task is a workqueue worker and what it's working on. | |
4469 | * Useful to find out the context that the %current task is running in. | |
4470 | * | |
4471 | * Return: work struct if %current task is a workqueue worker, %NULL otherwise. | |
4472 | */ | |
4473 | struct work_struct *current_work(void) | |
4474 | { | |
4475 | struct worker *worker = current_wq_worker(); | |
4476 | ||
4477 | return worker ? worker->current_work : NULL; | |
4478 | } | |
4479 | EXPORT_SYMBOL(current_work); | |
4480 | ||
e6267616 TH |
4481 | /** |
4482 | * current_is_workqueue_rescuer - is %current workqueue rescuer? | |
4483 | * | |
4484 | * Determine whether %current is a workqueue rescuer. Can be used from | |
4485 | * work functions to determine whether it's being run off the rescuer task. | |
d185af30 YB |
4486 | * |
4487 | * Return: %true if %current is a workqueue rescuer. %false otherwise. | |
e6267616 TH |
4488 | */ |
4489 | bool current_is_workqueue_rescuer(void) | |
4490 | { | |
4491 | struct worker *worker = current_wq_worker(); | |
4492 | ||
6a092dfd | 4493 | return worker && worker->rescue_wq; |
e6267616 TH |
4494 | } |
4495 | ||
eef6a7d5 | 4496 | /** |
dcd989cb TH |
4497 | * workqueue_congested - test whether a workqueue is congested |
4498 | * @cpu: CPU in question | |
4499 | * @wq: target workqueue | |
eef6a7d5 | 4500 | * |
dcd989cb TH |
4501 | * Test whether @wq's cpu workqueue for @cpu is congested. There is |
4502 | * no synchronization around this function and the test result is | |
4503 | * unreliable and only useful as advisory hints or for debugging. | |
eef6a7d5 | 4504 | * |
d3251859 TH |
4505 | * If @cpu is WORK_CPU_UNBOUND, the test is performed on the local CPU. |
4506 | * Note that both per-cpu and unbound workqueues may be associated with | |
4507 | * multiple pool_workqueues which have separate congested states. A | |
4508 | * workqueue being congested on one CPU doesn't mean the workqueue is also | |
4509 | * contested on other CPUs / NUMA nodes. | |
4510 | * | |
d185af30 | 4511 | * Return: |
dcd989cb | 4512 | * %true if congested, %false otherwise. |
eef6a7d5 | 4513 | */ |
d84ff051 | 4514 | bool workqueue_congested(int cpu, struct workqueue_struct *wq) |
1da177e4 | 4515 | { |
7fb98ea7 | 4516 | struct pool_workqueue *pwq; |
76af4d93 TH |
4517 | bool ret; |
4518 | ||
24acfb71 TG |
4519 | rcu_read_lock(); |
4520 | preempt_disable(); | |
7fb98ea7 | 4521 | |
d3251859 TH |
4522 | if (cpu == WORK_CPU_UNBOUND) |
4523 | cpu = smp_processor_id(); | |
4524 | ||
7fb98ea7 TH |
4525 | if (!(wq->flags & WQ_UNBOUND)) |
4526 | pwq = per_cpu_ptr(wq->cpu_pwqs, cpu); | |
4527 | else | |
df2d5ae4 | 4528 | pwq = unbound_pwq_by_node(wq, cpu_to_node(cpu)); |
dcd989cb | 4529 | |
76af4d93 | 4530 | ret = !list_empty(&pwq->delayed_works); |
24acfb71 TG |
4531 | preempt_enable(); |
4532 | rcu_read_unlock(); | |
76af4d93 TH |
4533 | |
4534 | return ret; | |
1da177e4 | 4535 | } |
dcd989cb | 4536 | EXPORT_SYMBOL_GPL(workqueue_congested); |
1da177e4 | 4537 | |
dcd989cb TH |
4538 | /** |
4539 | * work_busy - test whether a work is currently pending or running | |
4540 | * @work: the work to be tested | |
4541 | * | |
4542 | * Test whether @work is currently pending or running. There is no | |
4543 | * synchronization around this function and the test result is | |
4544 | * unreliable and only useful as advisory hints or for debugging. | |
dcd989cb | 4545 | * |
d185af30 | 4546 | * Return: |
dcd989cb TH |
4547 | * OR'd bitmask of WORK_BUSY_* bits. |
4548 | */ | |
4549 | unsigned int work_busy(struct work_struct *work) | |
1da177e4 | 4550 | { |
fa1b54e6 | 4551 | struct worker_pool *pool; |
dcd989cb TH |
4552 | unsigned long flags; |
4553 | unsigned int ret = 0; | |
1da177e4 | 4554 | |
dcd989cb TH |
4555 | if (work_pending(work)) |
4556 | ret |= WORK_BUSY_PENDING; | |
1da177e4 | 4557 | |
24acfb71 | 4558 | rcu_read_lock(); |
fa1b54e6 | 4559 | pool = get_work_pool(work); |
038366c5 | 4560 | if (pool) { |
24acfb71 | 4561 | spin_lock_irqsave(&pool->lock, flags); |
038366c5 LJ |
4562 | if (find_worker_executing_work(pool, work)) |
4563 | ret |= WORK_BUSY_RUNNING; | |
24acfb71 | 4564 | spin_unlock_irqrestore(&pool->lock, flags); |
038366c5 | 4565 | } |
24acfb71 | 4566 | rcu_read_unlock(); |
1da177e4 | 4567 | |
dcd989cb | 4568 | return ret; |
1da177e4 | 4569 | } |
dcd989cb | 4570 | EXPORT_SYMBOL_GPL(work_busy); |
1da177e4 | 4571 | |
3d1cb205 TH |
4572 | /** |
4573 | * set_worker_desc - set description for the current work item | |
4574 | * @fmt: printf-style format string | |
4575 | * @...: arguments for the format string | |
4576 | * | |
4577 | * This function can be called by a running work function to describe what | |
4578 | * the work item is about. If the worker task gets dumped, this | |
4579 | * information will be printed out together to help debugging. The | |
4580 | * description can be at most WORKER_DESC_LEN including the trailing '\0'. | |
4581 | */ | |
4582 | void set_worker_desc(const char *fmt, ...) | |
4583 | { | |
4584 | struct worker *worker = current_wq_worker(); | |
4585 | va_list args; | |
4586 | ||
4587 | if (worker) { | |
4588 | va_start(args, fmt); | |
4589 | vsnprintf(worker->desc, sizeof(worker->desc), fmt, args); | |
4590 | va_end(args); | |
3d1cb205 TH |
4591 | } |
4592 | } | |
5c750d58 | 4593 | EXPORT_SYMBOL_GPL(set_worker_desc); |
3d1cb205 TH |
4594 | |
4595 | /** | |
4596 | * print_worker_info - print out worker information and description | |
4597 | * @log_lvl: the log level to use when printing | |
4598 | * @task: target task | |
4599 | * | |
4600 | * If @task is a worker and currently executing a work item, print out the | |
4601 | * name of the workqueue being serviced and worker description set with | |
4602 | * set_worker_desc() by the currently executing work item. | |
4603 | * | |
4604 | * This function can be safely called on any task as long as the | |
4605 | * task_struct itself is accessible. While safe, this function isn't | |
4606 | * synchronized and may print out mixups or garbages of limited length. | |
4607 | */ | |
4608 | void print_worker_info(const char *log_lvl, struct task_struct *task) | |
4609 | { | |
4610 | work_func_t *fn = NULL; | |
4611 | char name[WQ_NAME_LEN] = { }; | |
4612 | char desc[WORKER_DESC_LEN] = { }; | |
4613 | struct pool_workqueue *pwq = NULL; | |
4614 | struct workqueue_struct *wq = NULL; | |
3d1cb205 TH |
4615 | struct worker *worker; |
4616 | ||
4617 | if (!(task->flags & PF_WQ_WORKER)) | |
4618 | return; | |
4619 | ||
4620 | /* | |
4621 | * This function is called without any synchronization and @task | |
4622 | * could be in any state. Be careful with dereferences. | |
4623 | */ | |
e700591a | 4624 | worker = kthread_probe_data(task); |
3d1cb205 TH |
4625 | |
4626 | /* | |
8bf89593 TH |
4627 | * Carefully copy the associated workqueue's workfn, name and desc. |
4628 | * Keep the original last '\0' in case the original is garbage. | |
3d1cb205 TH |
4629 | */ |
4630 | probe_kernel_read(&fn, &worker->current_func, sizeof(fn)); | |
4631 | probe_kernel_read(&pwq, &worker->current_pwq, sizeof(pwq)); | |
4632 | probe_kernel_read(&wq, &pwq->wq, sizeof(wq)); | |
4633 | probe_kernel_read(name, wq->name, sizeof(name) - 1); | |
8bf89593 | 4634 | probe_kernel_read(desc, worker->desc, sizeof(desc) - 1); |
3d1cb205 TH |
4635 | |
4636 | if (fn || name[0] || desc[0]) { | |
d75f773c | 4637 | printk("%sWorkqueue: %s %ps", log_lvl, name, fn); |
8bf89593 | 4638 | if (strcmp(name, desc)) |
3d1cb205 TH |
4639 | pr_cont(" (%s)", desc); |
4640 | pr_cont("\n"); | |
4641 | } | |
4642 | } | |
4643 | ||
3494fc30 TH |
4644 | static void pr_cont_pool_info(struct worker_pool *pool) |
4645 | { | |
4646 | pr_cont(" cpus=%*pbl", nr_cpumask_bits, pool->attrs->cpumask); | |
4647 | if (pool->node != NUMA_NO_NODE) | |
4648 | pr_cont(" node=%d", pool->node); | |
4649 | pr_cont(" flags=0x%x nice=%d", pool->flags, pool->attrs->nice); | |
4650 | } | |
4651 | ||
4652 | static void pr_cont_work(bool comma, struct work_struct *work) | |
4653 | { | |
4654 | if (work->func == wq_barrier_func) { | |
4655 | struct wq_barrier *barr; | |
4656 | ||
4657 | barr = container_of(work, struct wq_barrier, work); | |
4658 | ||
4659 | pr_cont("%s BAR(%d)", comma ? "," : "", | |
4660 | task_pid_nr(barr->task)); | |
4661 | } else { | |
d75f773c | 4662 | pr_cont("%s %ps", comma ? "," : "", work->func); |
3494fc30 TH |
4663 | } |
4664 | } | |
4665 | ||
4666 | static void show_pwq(struct pool_workqueue *pwq) | |
4667 | { | |
4668 | struct worker_pool *pool = pwq->pool; | |
4669 | struct work_struct *work; | |
4670 | struct worker *worker; | |
4671 | bool has_in_flight = false, has_pending = false; | |
4672 | int bkt; | |
4673 | ||
4674 | pr_info(" pwq %d:", pool->id); | |
4675 | pr_cont_pool_info(pool); | |
4676 | ||
e66b39af TH |
4677 | pr_cont(" active=%d/%d refcnt=%d%s\n", |
4678 | pwq->nr_active, pwq->max_active, pwq->refcnt, | |
3494fc30 TH |
4679 | !list_empty(&pwq->mayday_node) ? " MAYDAY" : ""); |
4680 | ||
4681 | hash_for_each(pool->busy_hash, bkt, worker, hentry) { | |
4682 | if (worker->current_pwq == pwq) { | |
4683 | has_in_flight = true; | |
4684 | break; | |
4685 | } | |
4686 | } | |
4687 | if (has_in_flight) { | |
4688 | bool comma = false; | |
4689 | ||
4690 | pr_info(" in-flight:"); | |
4691 | hash_for_each(pool->busy_hash, bkt, worker, hentry) { | |
4692 | if (worker->current_pwq != pwq) | |
4693 | continue; | |
4694 | ||
d75f773c | 4695 | pr_cont("%s %d%s:%ps", comma ? "," : "", |
3494fc30 | 4696 | task_pid_nr(worker->task), |
30ae2fc0 | 4697 | worker->rescue_wq ? "(RESCUER)" : "", |
3494fc30 TH |
4698 | worker->current_func); |
4699 | list_for_each_entry(work, &worker->scheduled, entry) | |
4700 | pr_cont_work(false, work); | |
4701 | comma = true; | |
4702 | } | |
4703 | pr_cont("\n"); | |
4704 | } | |
4705 | ||
4706 | list_for_each_entry(work, &pool->worklist, entry) { | |
4707 | if (get_work_pwq(work) == pwq) { | |
4708 | has_pending = true; | |
4709 | break; | |
4710 | } | |
4711 | } | |
4712 | if (has_pending) { | |
4713 | bool comma = false; | |
4714 | ||
4715 | pr_info(" pending:"); | |
4716 | list_for_each_entry(work, &pool->worklist, entry) { | |
4717 | if (get_work_pwq(work) != pwq) | |
4718 | continue; | |
4719 | ||
4720 | pr_cont_work(comma, work); | |
4721 | comma = !(*work_data_bits(work) & WORK_STRUCT_LINKED); | |
4722 | } | |
4723 | pr_cont("\n"); | |
4724 | } | |
4725 | ||
4726 | if (!list_empty(&pwq->delayed_works)) { | |
4727 | bool comma = false; | |
4728 | ||
4729 | pr_info(" delayed:"); | |
4730 | list_for_each_entry(work, &pwq->delayed_works, entry) { | |
4731 | pr_cont_work(comma, work); | |
4732 | comma = !(*work_data_bits(work) & WORK_STRUCT_LINKED); | |
4733 | } | |
4734 | pr_cont("\n"); | |
4735 | } | |
4736 | } | |
4737 | ||
4738 | /** | |
4739 | * show_workqueue_state - dump workqueue state | |
4740 | * | |
7b776af6 RL |
4741 | * Called from a sysrq handler or try_to_freeze_tasks() and prints out |
4742 | * all busy workqueues and pools. | |
3494fc30 TH |
4743 | */ |
4744 | void show_workqueue_state(void) | |
4745 | { | |
4746 | struct workqueue_struct *wq; | |
4747 | struct worker_pool *pool; | |
4748 | unsigned long flags; | |
4749 | int pi; | |
4750 | ||
24acfb71 | 4751 | rcu_read_lock(); |
3494fc30 TH |
4752 | |
4753 | pr_info("Showing busy workqueues and worker pools:\n"); | |
4754 | ||
4755 | list_for_each_entry_rcu(wq, &workqueues, list) { | |
4756 | struct pool_workqueue *pwq; | |
4757 | bool idle = true; | |
4758 | ||
4759 | for_each_pwq(pwq, wq) { | |
4760 | if (pwq->nr_active || !list_empty(&pwq->delayed_works)) { | |
4761 | idle = false; | |
4762 | break; | |
4763 | } | |
4764 | } | |
4765 | if (idle) | |
4766 | continue; | |
4767 | ||
4768 | pr_info("workqueue %s: flags=0x%x\n", wq->name, wq->flags); | |
4769 | ||
4770 | for_each_pwq(pwq, wq) { | |
4771 | spin_lock_irqsave(&pwq->pool->lock, flags); | |
4772 | if (pwq->nr_active || !list_empty(&pwq->delayed_works)) | |
4773 | show_pwq(pwq); | |
4774 | spin_unlock_irqrestore(&pwq->pool->lock, flags); | |
62635ea8 SS |
4775 | /* |
4776 | * We could be printing a lot from atomic context, e.g. | |
4777 | * sysrq-t -> show_workqueue_state(). Avoid triggering | |
4778 | * hard lockup. | |
4779 | */ | |
4780 | touch_nmi_watchdog(); | |
3494fc30 TH |
4781 | } |
4782 | } | |
4783 | ||
4784 | for_each_pool(pool, pi) { | |
4785 | struct worker *worker; | |
4786 | bool first = true; | |
4787 | ||
4788 | spin_lock_irqsave(&pool->lock, flags); | |
4789 | if (pool->nr_workers == pool->nr_idle) | |
4790 | goto next_pool; | |
4791 | ||
4792 | pr_info("pool %d:", pool->id); | |
4793 | pr_cont_pool_info(pool); | |
82607adc TH |
4794 | pr_cont(" hung=%us workers=%d", |
4795 | jiffies_to_msecs(jiffies - pool->watchdog_ts) / 1000, | |
4796 | pool->nr_workers); | |
3494fc30 TH |
4797 | if (pool->manager) |
4798 | pr_cont(" manager: %d", | |
4799 | task_pid_nr(pool->manager->task)); | |
4800 | list_for_each_entry(worker, &pool->idle_list, entry) { | |
4801 | pr_cont(" %s%d", first ? "idle: " : "", | |
4802 | task_pid_nr(worker->task)); | |
4803 | first = false; | |
4804 | } | |
4805 | pr_cont("\n"); | |
4806 | next_pool: | |
4807 | spin_unlock_irqrestore(&pool->lock, flags); | |
62635ea8 SS |
4808 | /* |
4809 | * We could be printing a lot from atomic context, e.g. | |
4810 | * sysrq-t -> show_workqueue_state(). Avoid triggering | |
4811 | * hard lockup. | |
4812 | */ | |
4813 | touch_nmi_watchdog(); | |
3494fc30 TH |
4814 | } |
4815 | ||
24acfb71 | 4816 | rcu_read_unlock(); |
3494fc30 TH |
4817 | } |
4818 | ||
6b59808b TH |
4819 | /* used to show worker information through /proc/PID/{comm,stat,status} */ |
4820 | void wq_worker_comm(char *buf, size_t size, struct task_struct *task) | |
4821 | { | |
6b59808b TH |
4822 | int off; |
4823 | ||
4824 | /* always show the actual comm */ | |
4825 | off = strscpy(buf, task->comm, size); | |
4826 | if (off < 0) | |
4827 | return; | |
4828 | ||
197f6acc | 4829 | /* stabilize PF_WQ_WORKER and worker pool association */ |
6b59808b TH |
4830 | mutex_lock(&wq_pool_attach_mutex); |
4831 | ||
197f6acc TH |
4832 | if (task->flags & PF_WQ_WORKER) { |
4833 | struct worker *worker = kthread_data(task); | |
4834 | struct worker_pool *pool = worker->pool; | |
6b59808b | 4835 | |
197f6acc TH |
4836 | if (pool) { |
4837 | spin_lock_irq(&pool->lock); | |
4838 | /* | |
4839 | * ->desc tracks information (wq name or | |
4840 | * set_worker_desc()) for the latest execution. If | |
4841 | * current, prepend '+', otherwise '-'. | |
4842 | */ | |
4843 | if (worker->desc[0] != '\0') { | |
4844 | if (worker->current_work) | |
4845 | scnprintf(buf + off, size - off, "+%s", | |
4846 | worker->desc); | |
4847 | else | |
4848 | scnprintf(buf + off, size - off, "-%s", | |
4849 | worker->desc); | |
4850 | } | |
4851 | spin_unlock_irq(&pool->lock); | |
6b59808b | 4852 | } |
6b59808b TH |
4853 | } |
4854 | ||
4855 | mutex_unlock(&wq_pool_attach_mutex); | |
4856 | } | |
4857 | ||
66448bc2 MM |
4858 | #ifdef CONFIG_SMP |
4859 | ||
db7bccf4 TH |
4860 | /* |
4861 | * CPU hotplug. | |
4862 | * | |
e22bee78 | 4863 | * There are two challenges in supporting CPU hotplug. Firstly, there |
112202d9 | 4864 | * are a lot of assumptions on strong associations among work, pwq and |
706026c2 | 4865 | * pool which make migrating pending and scheduled works very |
e22bee78 | 4866 | * difficult to implement without impacting hot paths. Secondly, |
94cf58bb | 4867 | * worker pools serve mix of short, long and very long running works making |
e22bee78 TH |
4868 | * blocked draining impractical. |
4869 | * | |
24647570 | 4870 | * This is solved by allowing the pools to be disassociated from the CPU |
628c78e7 TH |
4871 | * running as an unbound one and allowing it to be reattached later if the |
4872 | * cpu comes back online. | |
db7bccf4 | 4873 | */ |
1da177e4 | 4874 | |
e8b3f8db | 4875 | static void unbind_workers(int cpu) |
3af24433 | 4876 | { |
4ce62e9e | 4877 | struct worker_pool *pool; |
db7bccf4 | 4878 | struct worker *worker; |
3af24433 | 4879 | |
f02ae73a | 4880 | for_each_cpu_worker_pool(pool, cpu) { |
1258fae7 | 4881 | mutex_lock(&wq_pool_attach_mutex); |
94cf58bb | 4882 | spin_lock_irq(&pool->lock); |
3af24433 | 4883 | |
94cf58bb | 4884 | /* |
92f9c5c4 | 4885 | * We've blocked all attach/detach operations. Make all workers |
94cf58bb TH |
4886 | * unbound and set DISASSOCIATED. Before this, all workers |
4887 | * except for the ones which are still executing works from | |
4888 | * before the last CPU down must be on the cpu. After | |
4889 | * this, they may become diasporas. | |
4890 | */ | |
da028469 | 4891 | for_each_pool_worker(worker, pool) |
c9e7cf27 | 4892 | worker->flags |= WORKER_UNBOUND; |
06ba38a9 | 4893 | |
24647570 | 4894 | pool->flags |= POOL_DISASSOCIATED; |
f2d5a0ee | 4895 | |
94cf58bb | 4896 | spin_unlock_irq(&pool->lock); |
1258fae7 | 4897 | mutex_unlock(&wq_pool_attach_mutex); |
628c78e7 | 4898 | |
eb283428 LJ |
4899 | /* |
4900 | * Call schedule() so that we cross rq->lock and thus can | |
4901 | * guarantee sched callbacks see the %WORKER_UNBOUND flag. | |
4902 | * This is necessary as scheduler callbacks may be invoked | |
4903 | * from other cpus. | |
4904 | */ | |
4905 | schedule(); | |
06ba38a9 | 4906 | |
eb283428 LJ |
4907 | /* |
4908 | * Sched callbacks are disabled now. Zap nr_running. | |
4909 | * After this, nr_running stays zero and need_more_worker() | |
4910 | * and keep_working() are always true as long as the | |
4911 | * worklist is not empty. This pool now behaves as an | |
4912 | * unbound (in terms of concurrency management) pool which | |
4913 | * are served by workers tied to the pool. | |
4914 | */ | |
e19e397a | 4915 | atomic_set(&pool->nr_running, 0); |
eb283428 LJ |
4916 | |
4917 | /* | |
4918 | * With concurrency management just turned off, a busy | |
4919 | * worker blocking could lead to lengthy stalls. Kick off | |
4920 | * unbound chain execution of currently pending work items. | |
4921 | */ | |
4922 | spin_lock_irq(&pool->lock); | |
4923 | wake_up_worker(pool); | |
4924 | spin_unlock_irq(&pool->lock); | |
4925 | } | |
3af24433 | 4926 | } |
3af24433 | 4927 | |
bd7c089e TH |
4928 | /** |
4929 | * rebind_workers - rebind all workers of a pool to the associated CPU | |
4930 | * @pool: pool of interest | |
4931 | * | |
a9ab775b | 4932 | * @pool->cpu is coming online. Rebind all workers to the CPU. |
bd7c089e TH |
4933 | */ |
4934 | static void rebind_workers(struct worker_pool *pool) | |
4935 | { | |
a9ab775b | 4936 | struct worker *worker; |
bd7c089e | 4937 | |
1258fae7 | 4938 | lockdep_assert_held(&wq_pool_attach_mutex); |
bd7c089e | 4939 | |
a9ab775b TH |
4940 | /* |
4941 | * Restore CPU affinity of all workers. As all idle workers should | |
4942 | * be on the run-queue of the associated CPU before any local | |
402dd89d | 4943 | * wake-ups for concurrency management happen, restore CPU affinity |
a9ab775b TH |
4944 | * of all workers first and then clear UNBOUND. As we're called |
4945 | * from CPU_ONLINE, the following shouldn't fail. | |
4946 | */ | |
da028469 | 4947 | for_each_pool_worker(worker, pool) |
a9ab775b TH |
4948 | WARN_ON_ONCE(set_cpus_allowed_ptr(worker->task, |
4949 | pool->attrs->cpumask) < 0); | |
bd7c089e | 4950 | |
a9ab775b | 4951 | spin_lock_irq(&pool->lock); |
f7c17d26 | 4952 | |
3de5e884 | 4953 | pool->flags &= ~POOL_DISASSOCIATED; |
bd7c089e | 4954 | |
da028469 | 4955 | for_each_pool_worker(worker, pool) { |
a9ab775b | 4956 | unsigned int worker_flags = worker->flags; |
bd7c089e TH |
4957 | |
4958 | /* | |
a9ab775b TH |
4959 | * A bound idle worker should actually be on the runqueue |
4960 | * of the associated CPU for local wake-ups targeting it to | |
4961 | * work. Kick all idle workers so that they migrate to the | |
4962 | * associated CPU. Doing this in the same loop as | |
4963 | * replacing UNBOUND with REBOUND is safe as no worker will | |
4964 | * be bound before @pool->lock is released. | |
bd7c089e | 4965 | */ |
a9ab775b TH |
4966 | if (worker_flags & WORKER_IDLE) |
4967 | wake_up_process(worker->task); | |
bd7c089e | 4968 | |
a9ab775b TH |
4969 | /* |
4970 | * We want to clear UNBOUND but can't directly call | |
4971 | * worker_clr_flags() or adjust nr_running. Atomically | |
4972 | * replace UNBOUND with another NOT_RUNNING flag REBOUND. | |
4973 | * @worker will clear REBOUND using worker_clr_flags() when | |
4974 | * it initiates the next execution cycle thus restoring | |
4975 | * concurrency management. Note that when or whether | |
4976 | * @worker clears REBOUND doesn't affect correctness. | |
4977 | * | |
c95491ed | 4978 | * WRITE_ONCE() is necessary because @worker->flags may be |
a9ab775b | 4979 | * tested without holding any lock in |
6d25be57 | 4980 | * wq_worker_running(). Without it, NOT_RUNNING test may |
a9ab775b TH |
4981 | * fail incorrectly leading to premature concurrency |
4982 | * management operations. | |
4983 | */ | |
4984 | WARN_ON_ONCE(!(worker_flags & WORKER_UNBOUND)); | |
4985 | worker_flags |= WORKER_REBOUND; | |
4986 | worker_flags &= ~WORKER_UNBOUND; | |
c95491ed | 4987 | WRITE_ONCE(worker->flags, worker_flags); |
bd7c089e | 4988 | } |
a9ab775b TH |
4989 | |
4990 | spin_unlock_irq(&pool->lock); | |
bd7c089e TH |
4991 | } |
4992 | ||
7dbc725e TH |
4993 | /** |
4994 | * restore_unbound_workers_cpumask - restore cpumask of unbound workers | |
4995 | * @pool: unbound pool of interest | |
4996 | * @cpu: the CPU which is coming up | |
4997 | * | |
4998 | * An unbound pool may end up with a cpumask which doesn't have any online | |
4999 | * CPUs. When a worker of such pool get scheduled, the scheduler resets | |
5000 | * its cpus_allowed. If @cpu is in @pool's cpumask which didn't have any | |
5001 | * online CPU before, cpus_allowed of all its workers should be restored. | |
5002 | */ | |
5003 | static void restore_unbound_workers_cpumask(struct worker_pool *pool, int cpu) | |
5004 | { | |
5005 | static cpumask_t cpumask; | |
5006 | struct worker *worker; | |
7dbc725e | 5007 | |
1258fae7 | 5008 | lockdep_assert_held(&wq_pool_attach_mutex); |
7dbc725e TH |
5009 | |
5010 | /* is @cpu allowed for @pool? */ | |
5011 | if (!cpumask_test_cpu(cpu, pool->attrs->cpumask)) | |
5012 | return; | |
5013 | ||
7dbc725e | 5014 | cpumask_and(&cpumask, pool->attrs->cpumask, cpu_online_mask); |
7dbc725e TH |
5015 | |
5016 | /* as we're called from CPU_ONLINE, the following shouldn't fail */ | |
da028469 | 5017 | for_each_pool_worker(worker, pool) |
d945b5e9 | 5018 | WARN_ON_ONCE(set_cpus_allowed_ptr(worker->task, &cpumask) < 0); |
7dbc725e TH |
5019 | } |
5020 | ||
7ee681b2 TG |
5021 | int workqueue_prepare_cpu(unsigned int cpu) |
5022 | { | |
5023 | struct worker_pool *pool; | |
5024 | ||
5025 | for_each_cpu_worker_pool(pool, cpu) { | |
5026 | if (pool->nr_workers) | |
5027 | continue; | |
5028 | if (!create_worker(pool)) | |
5029 | return -ENOMEM; | |
5030 | } | |
5031 | return 0; | |
5032 | } | |
5033 | ||
5034 | int workqueue_online_cpu(unsigned int cpu) | |
3af24433 | 5035 | { |
4ce62e9e | 5036 | struct worker_pool *pool; |
4c16bd32 | 5037 | struct workqueue_struct *wq; |
7dbc725e | 5038 | int pi; |
3ce63377 | 5039 | |
7ee681b2 | 5040 | mutex_lock(&wq_pool_mutex); |
7dbc725e | 5041 | |
7ee681b2 | 5042 | for_each_pool(pool, pi) { |
1258fae7 | 5043 | mutex_lock(&wq_pool_attach_mutex); |
94cf58bb | 5044 | |
7ee681b2 TG |
5045 | if (pool->cpu == cpu) |
5046 | rebind_workers(pool); | |
5047 | else if (pool->cpu < 0) | |
5048 | restore_unbound_workers_cpumask(pool, cpu); | |
94cf58bb | 5049 | |
1258fae7 | 5050 | mutex_unlock(&wq_pool_attach_mutex); |
7ee681b2 | 5051 | } |
6ba94429 | 5052 | |
7ee681b2 TG |
5053 | /* update NUMA affinity of unbound workqueues */ |
5054 | list_for_each_entry(wq, &workqueues, list) | |
5055 | wq_update_unbound_numa(wq, cpu, true); | |
6ba94429 | 5056 | |
7ee681b2 TG |
5057 | mutex_unlock(&wq_pool_mutex); |
5058 | return 0; | |
6ba94429 FW |
5059 | } |
5060 | ||
7ee681b2 | 5061 | int workqueue_offline_cpu(unsigned int cpu) |
6ba94429 | 5062 | { |
6ba94429 FW |
5063 | struct workqueue_struct *wq; |
5064 | ||
7ee681b2 | 5065 | /* unbinding per-cpu workers should happen on the local CPU */ |
e8b3f8db LJ |
5066 | if (WARN_ON(cpu != smp_processor_id())) |
5067 | return -1; | |
5068 | ||
5069 | unbind_workers(cpu); | |
7ee681b2 TG |
5070 | |
5071 | /* update NUMA affinity of unbound workqueues */ | |
5072 | mutex_lock(&wq_pool_mutex); | |
5073 | list_for_each_entry(wq, &workqueues, list) | |
5074 | wq_update_unbound_numa(wq, cpu, false); | |
5075 | mutex_unlock(&wq_pool_mutex); | |
5076 | ||
7ee681b2 | 5077 | return 0; |
6ba94429 FW |
5078 | } |
5079 | ||
6ba94429 FW |
5080 | struct work_for_cpu { |
5081 | struct work_struct work; | |
5082 | long (*fn)(void *); | |
5083 | void *arg; | |
5084 | long ret; | |
5085 | }; | |
5086 | ||
5087 | static void work_for_cpu_fn(struct work_struct *work) | |
5088 | { | |
5089 | struct work_for_cpu *wfc = container_of(work, struct work_for_cpu, work); | |
5090 | ||
5091 | wfc->ret = wfc->fn(wfc->arg); | |
5092 | } | |
5093 | ||
5094 | /** | |
22aceb31 | 5095 | * work_on_cpu - run a function in thread context on a particular cpu |
6ba94429 FW |
5096 | * @cpu: the cpu to run on |
5097 | * @fn: the function to run | |
5098 | * @arg: the function arg | |
5099 | * | |
5100 | * It is up to the caller to ensure that the cpu doesn't go offline. | |
5101 | * The caller must not hold any locks which would prevent @fn from completing. | |
5102 | * | |
5103 | * Return: The value @fn returns. | |
5104 | */ | |
5105 | long work_on_cpu(int cpu, long (*fn)(void *), void *arg) | |
5106 | { | |
5107 | struct work_for_cpu wfc = { .fn = fn, .arg = arg }; | |
5108 | ||
5109 | INIT_WORK_ONSTACK(&wfc.work, work_for_cpu_fn); | |
5110 | schedule_work_on(cpu, &wfc.work); | |
5111 | flush_work(&wfc.work); | |
5112 | destroy_work_on_stack(&wfc.work); | |
5113 | return wfc.ret; | |
5114 | } | |
5115 | EXPORT_SYMBOL_GPL(work_on_cpu); | |
0e8d6a93 TG |
5116 | |
5117 | /** | |
5118 | * work_on_cpu_safe - run a function in thread context on a particular cpu | |
5119 | * @cpu: the cpu to run on | |
5120 | * @fn: the function to run | |
5121 | * @arg: the function argument | |
5122 | * | |
5123 | * Disables CPU hotplug and calls work_on_cpu(). The caller must not hold | |
5124 | * any locks which would prevent @fn from completing. | |
5125 | * | |
5126 | * Return: The value @fn returns. | |
5127 | */ | |
5128 | long work_on_cpu_safe(int cpu, long (*fn)(void *), void *arg) | |
5129 | { | |
5130 | long ret = -ENODEV; | |
5131 | ||
5132 | get_online_cpus(); | |
5133 | if (cpu_online(cpu)) | |
5134 | ret = work_on_cpu(cpu, fn, arg); | |
5135 | put_online_cpus(); | |
5136 | return ret; | |
5137 | } | |
5138 | EXPORT_SYMBOL_GPL(work_on_cpu_safe); | |
6ba94429 FW |
5139 | #endif /* CONFIG_SMP */ |
5140 | ||
5141 | #ifdef CONFIG_FREEZER | |
5142 | ||
5143 | /** | |
5144 | * freeze_workqueues_begin - begin freezing workqueues | |
5145 | * | |
5146 | * Start freezing workqueues. After this function returns, all freezable | |
5147 | * workqueues will queue new works to their delayed_works list instead of | |
5148 | * pool->worklist. | |
5149 | * | |
5150 | * CONTEXT: | |
5151 | * Grabs and releases wq_pool_mutex, wq->mutex and pool->lock's. | |
5152 | */ | |
5153 | void freeze_workqueues_begin(void) | |
5154 | { | |
5155 | struct workqueue_struct *wq; | |
5156 | struct pool_workqueue *pwq; | |
5157 | ||
5158 | mutex_lock(&wq_pool_mutex); | |
5159 | ||
5160 | WARN_ON_ONCE(workqueue_freezing); | |
5161 | workqueue_freezing = true; | |
5162 | ||
5163 | list_for_each_entry(wq, &workqueues, list) { | |
5164 | mutex_lock(&wq->mutex); | |
5165 | for_each_pwq(pwq, wq) | |
5166 | pwq_adjust_max_active(pwq); | |
5167 | mutex_unlock(&wq->mutex); | |
5168 | } | |
5169 | ||
5170 | mutex_unlock(&wq_pool_mutex); | |
5171 | } | |
5172 | ||
5173 | /** | |
5174 | * freeze_workqueues_busy - are freezable workqueues still busy? | |
5175 | * | |
5176 | * Check whether freezing is complete. This function must be called | |
5177 | * between freeze_workqueues_begin() and thaw_workqueues(). | |
5178 | * | |
5179 | * CONTEXT: | |
5180 | * Grabs and releases wq_pool_mutex. | |
5181 | * | |
5182 | * Return: | |
5183 | * %true if some freezable workqueues are still busy. %false if freezing | |
5184 | * is complete. | |
5185 | */ | |
5186 | bool freeze_workqueues_busy(void) | |
5187 | { | |
5188 | bool busy = false; | |
5189 | struct workqueue_struct *wq; | |
5190 | struct pool_workqueue *pwq; | |
5191 | ||
5192 | mutex_lock(&wq_pool_mutex); | |
5193 | ||
5194 | WARN_ON_ONCE(!workqueue_freezing); | |
5195 | ||
5196 | list_for_each_entry(wq, &workqueues, list) { | |
5197 | if (!(wq->flags & WQ_FREEZABLE)) | |
5198 | continue; | |
5199 | /* | |
5200 | * nr_active is monotonically decreasing. It's safe | |
5201 | * to peek without lock. | |
5202 | */ | |
24acfb71 | 5203 | rcu_read_lock(); |
6ba94429 FW |
5204 | for_each_pwq(pwq, wq) { |
5205 | WARN_ON_ONCE(pwq->nr_active < 0); | |
5206 | if (pwq->nr_active) { | |
5207 | busy = true; | |
24acfb71 | 5208 | rcu_read_unlock(); |
6ba94429 FW |
5209 | goto out_unlock; |
5210 | } | |
5211 | } | |
24acfb71 | 5212 | rcu_read_unlock(); |
6ba94429 FW |
5213 | } |
5214 | out_unlock: | |
5215 | mutex_unlock(&wq_pool_mutex); | |
5216 | return busy; | |
5217 | } | |
5218 | ||
5219 | /** | |
5220 | * thaw_workqueues - thaw workqueues | |
5221 | * | |
5222 | * Thaw workqueues. Normal queueing is restored and all collected | |
5223 | * frozen works are transferred to their respective pool worklists. | |
5224 | * | |
5225 | * CONTEXT: | |
5226 | * Grabs and releases wq_pool_mutex, wq->mutex and pool->lock's. | |
5227 | */ | |
5228 | void thaw_workqueues(void) | |
5229 | { | |
5230 | struct workqueue_struct *wq; | |
5231 | struct pool_workqueue *pwq; | |
5232 | ||
5233 | mutex_lock(&wq_pool_mutex); | |
5234 | ||
5235 | if (!workqueue_freezing) | |
5236 | goto out_unlock; | |
5237 | ||
5238 | workqueue_freezing = false; | |
5239 | ||
5240 | /* restore max_active and repopulate worklist */ | |
5241 | list_for_each_entry(wq, &workqueues, list) { | |
5242 | mutex_lock(&wq->mutex); | |
5243 | for_each_pwq(pwq, wq) | |
5244 | pwq_adjust_max_active(pwq); | |
5245 | mutex_unlock(&wq->mutex); | |
5246 | } | |
5247 | ||
5248 | out_unlock: | |
5249 | mutex_unlock(&wq_pool_mutex); | |
5250 | } | |
5251 | #endif /* CONFIG_FREEZER */ | |
5252 | ||
042f7df1 LJ |
5253 | static int workqueue_apply_unbound_cpumask(void) |
5254 | { | |
5255 | LIST_HEAD(ctxs); | |
5256 | int ret = 0; | |
5257 | struct workqueue_struct *wq; | |
5258 | struct apply_wqattrs_ctx *ctx, *n; | |
5259 | ||
5260 | lockdep_assert_held(&wq_pool_mutex); | |
5261 | ||
5262 | list_for_each_entry(wq, &workqueues, list) { | |
5263 | if (!(wq->flags & WQ_UNBOUND)) | |
5264 | continue; | |
5265 | /* creating multiple pwqs breaks ordering guarantee */ | |
5266 | if (wq->flags & __WQ_ORDERED) | |
5267 | continue; | |
5268 | ||
5269 | ctx = apply_wqattrs_prepare(wq, wq->unbound_attrs); | |
5270 | if (!ctx) { | |
5271 | ret = -ENOMEM; | |
5272 | break; | |
5273 | } | |
5274 | ||
5275 | list_add_tail(&ctx->list, &ctxs); | |
5276 | } | |
5277 | ||
5278 | list_for_each_entry_safe(ctx, n, &ctxs, list) { | |
5279 | if (!ret) | |
5280 | apply_wqattrs_commit(ctx); | |
5281 | apply_wqattrs_cleanup(ctx); | |
5282 | } | |
5283 | ||
5284 | return ret; | |
5285 | } | |
5286 | ||
5287 | /** | |
5288 | * workqueue_set_unbound_cpumask - Set the low-level unbound cpumask | |
5289 | * @cpumask: the cpumask to set | |
5290 | * | |
5291 | * The low-level workqueues cpumask is a global cpumask that limits | |
5292 | * the affinity of all unbound workqueues. This function check the @cpumask | |
5293 | * and apply it to all unbound workqueues and updates all pwqs of them. | |
5294 | * | |
5295 | * Retun: 0 - Success | |
5296 | * -EINVAL - Invalid @cpumask | |
5297 | * -ENOMEM - Failed to allocate memory for attrs or pwqs. | |
5298 | */ | |
5299 | int workqueue_set_unbound_cpumask(cpumask_var_t cpumask) | |
5300 | { | |
5301 | int ret = -EINVAL; | |
5302 | cpumask_var_t saved_cpumask; | |
5303 | ||
5304 | if (!zalloc_cpumask_var(&saved_cpumask, GFP_KERNEL)) | |
5305 | return -ENOMEM; | |
5306 | ||
c98a9805 TS |
5307 | /* |
5308 | * Not excluding isolated cpus on purpose. | |
5309 | * If the user wishes to include them, we allow that. | |
5310 | */ | |
042f7df1 LJ |
5311 | cpumask_and(cpumask, cpumask, cpu_possible_mask); |
5312 | if (!cpumask_empty(cpumask)) { | |
a0111cf6 | 5313 | apply_wqattrs_lock(); |
042f7df1 LJ |
5314 | |
5315 | /* save the old wq_unbound_cpumask. */ | |
5316 | cpumask_copy(saved_cpumask, wq_unbound_cpumask); | |
5317 | ||
5318 | /* update wq_unbound_cpumask at first and apply it to wqs. */ | |
5319 | cpumask_copy(wq_unbound_cpumask, cpumask); | |
5320 | ret = workqueue_apply_unbound_cpumask(); | |
5321 | ||
5322 | /* restore the wq_unbound_cpumask when failed. */ | |
5323 | if (ret < 0) | |
5324 | cpumask_copy(wq_unbound_cpumask, saved_cpumask); | |
5325 | ||
a0111cf6 | 5326 | apply_wqattrs_unlock(); |
042f7df1 | 5327 | } |
042f7df1 LJ |
5328 | |
5329 | free_cpumask_var(saved_cpumask); | |
5330 | return ret; | |
5331 | } | |
5332 | ||
6ba94429 FW |
5333 | #ifdef CONFIG_SYSFS |
5334 | /* | |
5335 | * Workqueues with WQ_SYSFS flag set is visible to userland via | |
5336 | * /sys/bus/workqueue/devices/WQ_NAME. All visible workqueues have the | |
5337 | * following attributes. | |
5338 | * | |
5339 | * per_cpu RO bool : whether the workqueue is per-cpu or unbound | |
5340 | * max_active RW int : maximum number of in-flight work items | |
5341 | * | |
5342 | * Unbound workqueues have the following extra attributes. | |
5343 | * | |
9a19b463 | 5344 | * pool_ids RO int : the associated pool IDs for each node |
6ba94429 FW |
5345 | * nice RW int : nice value of the workers |
5346 | * cpumask RW mask : bitmask of allowed CPUs for the workers | |
9a19b463 | 5347 | * numa RW bool : whether enable NUMA affinity |
6ba94429 FW |
5348 | */ |
5349 | struct wq_device { | |
5350 | struct workqueue_struct *wq; | |
5351 | struct device dev; | |
5352 | }; | |
5353 | ||
5354 | static struct workqueue_struct *dev_to_wq(struct device *dev) | |
5355 | { | |
5356 | struct wq_device *wq_dev = container_of(dev, struct wq_device, dev); | |
5357 | ||
5358 | return wq_dev->wq; | |
5359 | } | |
5360 | ||
5361 | static ssize_t per_cpu_show(struct device *dev, struct device_attribute *attr, | |
5362 | char *buf) | |
5363 | { | |
5364 | struct workqueue_struct *wq = dev_to_wq(dev); | |
5365 | ||
5366 | return scnprintf(buf, PAGE_SIZE, "%d\n", (bool)!(wq->flags & WQ_UNBOUND)); | |
5367 | } | |
5368 | static DEVICE_ATTR_RO(per_cpu); | |
5369 | ||
5370 | static ssize_t max_active_show(struct device *dev, | |
5371 | struct device_attribute *attr, char *buf) | |
5372 | { | |
5373 | struct workqueue_struct *wq = dev_to_wq(dev); | |
5374 | ||
5375 | return scnprintf(buf, PAGE_SIZE, "%d\n", wq->saved_max_active); | |
5376 | } | |
5377 | ||
5378 | static ssize_t max_active_store(struct device *dev, | |
5379 | struct device_attribute *attr, const char *buf, | |
5380 | size_t count) | |
5381 | { | |
5382 | struct workqueue_struct *wq = dev_to_wq(dev); | |
5383 | int val; | |
5384 | ||
5385 | if (sscanf(buf, "%d", &val) != 1 || val <= 0) | |
5386 | return -EINVAL; | |
5387 | ||
5388 | workqueue_set_max_active(wq, val); | |
5389 | return count; | |
5390 | } | |
5391 | static DEVICE_ATTR_RW(max_active); | |
5392 | ||
5393 | static struct attribute *wq_sysfs_attrs[] = { | |
5394 | &dev_attr_per_cpu.attr, | |
5395 | &dev_attr_max_active.attr, | |
5396 | NULL, | |
5397 | }; | |
5398 | ATTRIBUTE_GROUPS(wq_sysfs); | |
5399 | ||
5400 | static ssize_t wq_pool_ids_show(struct device *dev, | |
5401 | struct device_attribute *attr, char *buf) | |
5402 | { | |
5403 | struct workqueue_struct *wq = dev_to_wq(dev); | |
5404 | const char *delim = ""; | |
5405 | int node, written = 0; | |
5406 | ||
24acfb71 TG |
5407 | get_online_cpus(); |
5408 | rcu_read_lock(); | |
6ba94429 FW |
5409 | for_each_node(node) { |
5410 | written += scnprintf(buf + written, PAGE_SIZE - written, | |
5411 | "%s%d:%d", delim, node, | |
5412 | unbound_pwq_by_node(wq, node)->pool->id); | |
5413 | delim = " "; | |
5414 | } | |
5415 | written += scnprintf(buf + written, PAGE_SIZE - written, "\n"); | |
24acfb71 TG |
5416 | rcu_read_unlock(); |
5417 | put_online_cpus(); | |
6ba94429 FW |
5418 | |
5419 | return written; | |
5420 | } | |
5421 | ||
5422 | static ssize_t wq_nice_show(struct device *dev, struct device_attribute *attr, | |
5423 | char *buf) | |
5424 | { | |
5425 | struct workqueue_struct *wq = dev_to_wq(dev); | |
5426 | int written; | |
5427 | ||
5428 | mutex_lock(&wq->mutex); | |
5429 | written = scnprintf(buf, PAGE_SIZE, "%d\n", wq->unbound_attrs->nice); | |
5430 | mutex_unlock(&wq->mutex); | |
5431 | ||
5432 | return written; | |
5433 | } | |
5434 | ||
5435 | /* prepare workqueue_attrs for sysfs store operations */ | |
5436 | static struct workqueue_attrs *wq_sysfs_prep_attrs(struct workqueue_struct *wq) | |
5437 | { | |
5438 | struct workqueue_attrs *attrs; | |
5439 | ||
899a94fe LJ |
5440 | lockdep_assert_held(&wq_pool_mutex); |
5441 | ||
be69d00d | 5442 | attrs = alloc_workqueue_attrs(); |
6ba94429 FW |
5443 | if (!attrs) |
5444 | return NULL; | |
5445 | ||
6ba94429 | 5446 | copy_workqueue_attrs(attrs, wq->unbound_attrs); |
6ba94429 FW |
5447 | return attrs; |
5448 | } | |
5449 | ||
5450 | static ssize_t wq_nice_store(struct device *dev, struct device_attribute *attr, | |
5451 | const char *buf, size_t count) | |
5452 | { | |
5453 | struct workqueue_struct *wq = dev_to_wq(dev); | |
5454 | struct workqueue_attrs *attrs; | |
d4d3e257 LJ |
5455 | int ret = -ENOMEM; |
5456 | ||
5457 | apply_wqattrs_lock(); | |
6ba94429 FW |
5458 | |
5459 | attrs = wq_sysfs_prep_attrs(wq); | |
5460 | if (!attrs) | |
d4d3e257 | 5461 | goto out_unlock; |
6ba94429 FW |
5462 | |
5463 | if (sscanf(buf, "%d", &attrs->nice) == 1 && | |
5464 | attrs->nice >= MIN_NICE && attrs->nice <= MAX_NICE) | |
d4d3e257 | 5465 | ret = apply_workqueue_attrs_locked(wq, attrs); |
6ba94429 FW |
5466 | else |
5467 | ret = -EINVAL; | |
5468 | ||
d4d3e257 LJ |
5469 | out_unlock: |
5470 | apply_wqattrs_unlock(); | |
6ba94429 FW |
5471 | free_workqueue_attrs(attrs); |
5472 | return ret ?: count; | |
5473 | } | |
5474 | ||
5475 | static ssize_t wq_cpumask_show(struct device *dev, | |
5476 | struct device_attribute *attr, char *buf) | |
5477 | { | |
5478 | struct workqueue_struct *wq = dev_to_wq(dev); | |
5479 | int written; | |
5480 | ||
5481 | mutex_lock(&wq->mutex); | |
5482 | written = scnprintf(buf, PAGE_SIZE, "%*pb\n", | |
5483 | cpumask_pr_args(wq->unbound_attrs->cpumask)); | |
5484 | mutex_unlock(&wq->mutex); | |
5485 | return written; | |
5486 | } | |
5487 | ||
5488 | static ssize_t wq_cpumask_store(struct device *dev, | |
5489 | struct device_attribute *attr, | |
5490 | const char *buf, size_t count) | |
5491 | { | |
5492 | struct workqueue_struct *wq = dev_to_wq(dev); | |
5493 | struct workqueue_attrs *attrs; | |
d4d3e257 LJ |
5494 | int ret = -ENOMEM; |
5495 | ||
5496 | apply_wqattrs_lock(); | |
6ba94429 FW |
5497 | |
5498 | attrs = wq_sysfs_prep_attrs(wq); | |
5499 | if (!attrs) | |
d4d3e257 | 5500 | goto out_unlock; |
6ba94429 FW |
5501 | |
5502 | ret = cpumask_parse(buf, attrs->cpumask); | |
5503 | if (!ret) | |
d4d3e257 | 5504 | ret = apply_workqueue_attrs_locked(wq, attrs); |
6ba94429 | 5505 | |
d4d3e257 LJ |
5506 | out_unlock: |
5507 | apply_wqattrs_unlock(); | |
6ba94429 FW |
5508 | free_workqueue_attrs(attrs); |
5509 | return ret ?: count; | |
5510 | } | |
5511 | ||
5512 | static ssize_t wq_numa_show(struct device *dev, struct device_attribute *attr, | |
5513 | char *buf) | |
5514 | { | |
5515 | struct workqueue_struct *wq = dev_to_wq(dev); | |
5516 | int written; | |
7dbc725e | 5517 | |
6ba94429 FW |
5518 | mutex_lock(&wq->mutex); |
5519 | written = scnprintf(buf, PAGE_SIZE, "%d\n", | |
5520 | !wq->unbound_attrs->no_numa); | |
5521 | mutex_unlock(&wq->mutex); | |
4c16bd32 | 5522 | |
6ba94429 | 5523 | return written; |
65758202 TH |
5524 | } |
5525 | ||
6ba94429 FW |
5526 | static ssize_t wq_numa_store(struct device *dev, struct device_attribute *attr, |
5527 | const char *buf, size_t count) | |
65758202 | 5528 | { |
6ba94429 FW |
5529 | struct workqueue_struct *wq = dev_to_wq(dev); |
5530 | struct workqueue_attrs *attrs; | |
d4d3e257 LJ |
5531 | int v, ret = -ENOMEM; |
5532 | ||
5533 | apply_wqattrs_lock(); | |
4c16bd32 | 5534 | |
6ba94429 FW |
5535 | attrs = wq_sysfs_prep_attrs(wq); |
5536 | if (!attrs) | |
d4d3e257 | 5537 | goto out_unlock; |
4c16bd32 | 5538 | |
6ba94429 FW |
5539 | ret = -EINVAL; |
5540 | if (sscanf(buf, "%d", &v) == 1) { | |
5541 | attrs->no_numa = !v; | |
d4d3e257 | 5542 | ret = apply_workqueue_attrs_locked(wq, attrs); |
65758202 | 5543 | } |
6ba94429 | 5544 | |
d4d3e257 LJ |
5545 | out_unlock: |
5546 | apply_wqattrs_unlock(); | |
6ba94429 FW |
5547 | free_workqueue_attrs(attrs); |
5548 | return ret ?: count; | |
65758202 TH |
5549 | } |
5550 | ||
6ba94429 FW |
5551 | static struct device_attribute wq_sysfs_unbound_attrs[] = { |
5552 | __ATTR(pool_ids, 0444, wq_pool_ids_show, NULL), | |
5553 | __ATTR(nice, 0644, wq_nice_show, wq_nice_store), | |
5554 | __ATTR(cpumask, 0644, wq_cpumask_show, wq_cpumask_store), | |
5555 | __ATTR(numa, 0644, wq_numa_show, wq_numa_store), | |
5556 | __ATTR_NULL, | |
5557 | }; | |
8ccad40d | 5558 | |
6ba94429 FW |
5559 | static struct bus_type wq_subsys = { |
5560 | .name = "workqueue", | |
5561 | .dev_groups = wq_sysfs_groups, | |
2d3854a3 RR |
5562 | }; |
5563 | ||
b05a7928 FW |
5564 | static ssize_t wq_unbound_cpumask_show(struct device *dev, |
5565 | struct device_attribute *attr, char *buf) | |
5566 | { | |
5567 | int written; | |
5568 | ||
042f7df1 | 5569 | mutex_lock(&wq_pool_mutex); |
b05a7928 FW |
5570 | written = scnprintf(buf, PAGE_SIZE, "%*pb\n", |
5571 | cpumask_pr_args(wq_unbound_cpumask)); | |
042f7df1 | 5572 | mutex_unlock(&wq_pool_mutex); |
b05a7928 FW |
5573 | |
5574 | return written; | |
5575 | } | |
5576 | ||
042f7df1 LJ |
5577 | static ssize_t wq_unbound_cpumask_store(struct device *dev, |
5578 | struct device_attribute *attr, const char *buf, size_t count) | |
5579 | { | |
5580 | cpumask_var_t cpumask; | |
5581 | int ret; | |
5582 | ||
5583 | if (!zalloc_cpumask_var(&cpumask, GFP_KERNEL)) | |
5584 | return -ENOMEM; | |
5585 | ||
5586 | ret = cpumask_parse(buf, cpumask); | |
5587 | if (!ret) | |
5588 | ret = workqueue_set_unbound_cpumask(cpumask); | |
5589 | ||
5590 | free_cpumask_var(cpumask); | |
5591 | return ret ? ret : count; | |
5592 | } | |
5593 | ||
b05a7928 | 5594 | static struct device_attribute wq_sysfs_cpumask_attr = |
042f7df1 LJ |
5595 | __ATTR(cpumask, 0644, wq_unbound_cpumask_show, |
5596 | wq_unbound_cpumask_store); | |
b05a7928 | 5597 | |
6ba94429 | 5598 | static int __init wq_sysfs_init(void) |
2d3854a3 | 5599 | { |
b05a7928 FW |
5600 | int err; |
5601 | ||
5602 | err = subsys_virtual_register(&wq_subsys, NULL); | |
5603 | if (err) | |
5604 | return err; | |
5605 | ||
5606 | return device_create_file(wq_subsys.dev_root, &wq_sysfs_cpumask_attr); | |
2d3854a3 | 5607 | } |
6ba94429 | 5608 | core_initcall(wq_sysfs_init); |
2d3854a3 | 5609 | |
6ba94429 | 5610 | static void wq_device_release(struct device *dev) |
2d3854a3 | 5611 | { |
6ba94429 | 5612 | struct wq_device *wq_dev = container_of(dev, struct wq_device, dev); |
6b44003e | 5613 | |
6ba94429 | 5614 | kfree(wq_dev); |
2d3854a3 | 5615 | } |
a0a1a5fd TH |
5616 | |
5617 | /** | |
6ba94429 FW |
5618 | * workqueue_sysfs_register - make a workqueue visible in sysfs |
5619 | * @wq: the workqueue to register | |
a0a1a5fd | 5620 | * |
6ba94429 FW |
5621 | * Expose @wq in sysfs under /sys/bus/workqueue/devices. |
5622 | * alloc_workqueue*() automatically calls this function if WQ_SYSFS is set | |
5623 | * which is the preferred method. | |
a0a1a5fd | 5624 | * |
6ba94429 FW |
5625 | * Workqueue user should use this function directly iff it wants to apply |
5626 | * workqueue_attrs before making the workqueue visible in sysfs; otherwise, | |
5627 | * apply_workqueue_attrs() may race against userland updating the | |
5628 | * attributes. | |
5629 | * | |
5630 | * Return: 0 on success, -errno on failure. | |
a0a1a5fd | 5631 | */ |
6ba94429 | 5632 | int workqueue_sysfs_register(struct workqueue_struct *wq) |
a0a1a5fd | 5633 | { |
6ba94429 FW |
5634 | struct wq_device *wq_dev; |
5635 | int ret; | |
a0a1a5fd | 5636 | |
6ba94429 | 5637 | /* |
402dd89d | 5638 | * Adjusting max_active or creating new pwqs by applying |
6ba94429 FW |
5639 | * attributes breaks ordering guarantee. Disallow exposing ordered |
5640 | * workqueues. | |
5641 | */ | |
0a94efb5 | 5642 | if (WARN_ON(wq->flags & __WQ_ORDERED_EXPLICIT)) |
6ba94429 | 5643 | return -EINVAL; |
a0a1a5fd | 5644 | |
6ba94429 FW |
5645 | wq->wq_dev = wq_dev = kzalloc(sizeof(*wq_dev), GFP_KERNEL); |
5646 | if (!wq_dev) | |
5647 | return -ENOMEM; | |
5bcab335 | 5648 | |
6ba94429 FW |
5649 | wq_dev->wq = wq; |
5650 | wq_dev->dev.bus = &wq_subsys; | |
6ba94429 | 5651 | wq_dev->dev.release = wq_device_release; |
23217b44 | 5652 | dev_set_name(&wq_dev->dev, "%s", wq->name); |
a0a1a5fd | 5653 | |
6ba94429 FW |
5654 | /* |
5655 | * unbound_attrs are created separately. Suppress uevent until | |
5656 | * everything is ready. | |
5657 | */ | |
5658 | dev_set_uevent_suppress(&wq_dev->dev, true); | |
a0a1a5fd | 5659 | |
6ba94429 FW |
5660 | ret = device_register(&wq_dev->dev); |
5661 | if (ret) { | |
537f4146 | 5662 | put_device(&wq_dev->dev); |
6ba94429 FW |
5663 | wq->wq_dev = NULL; |
5664 | return ret; | |
5665 | } | |
a0a1a5fd | 5666 | |
6ba94429 FW |
5667 | if (wq->flags & WQ_UNBOUND) { |
5668 | struct device_attribute *attr; | |
a0a1a5fd | 5669 | |
6ba94429 FW |
5670 | for (attr = wq_sysfs_unbound_attrs; attr->attr.name; attr++) { |
5671 | ret = device_create_file(&wq_dev->dev, attr); | |
5672 | if (ret) { | |
5673 | device_unregister(&wq_dev->dev); | |
5674 | wq->wq_dev = NULL; | |
5675 | return ret; | |
a0a1a5fd TH |
5676 | } |
5677 | } | |
5678 | } | |
6ba94429 FW |
5679 | |
5680 | dev_set_uevent_suppress(&wq_dev->dev, false); | |
5681 | kobject_uevent(&wq_dev->dev.kobj, KOBJ_ADD); | |
5682 | return 0; | |
a0a1a5fd TH |
5683 | } |
5684 | ||
5685 | /** | |
6ba94429 FW |
5686 | * workqueue_sysfs_unregister - undo workqueue_sysfs_register() |
5687 | * @wq: the workqueue to unregister | |
a0a1a5fd | 5688 | * |
6ba94429 | 5689 | * If @wq is registered to sysfs by workqueue_sysfs_register(), unregister. |
a0a1a5fd | 5690 | */ |
6ba94429 | 5691 | static void workqueue_sysfs_unregister(struct workqueue_struct *wq) |
a0a1a5fd | 5692 | { |
6ba94429 | 5693 | struct wq_device *wq_dev = wq->wq_dev; |
8b03ae3c | 5694 | |
6ba94429 FW |
5695 | if (!wq->wq_dev) |
5696 | return; | |
a0a1a5fd | 5697 | |
6ba94429 FW |
5698 | wq->wq_dev = NULL; |
5699 | device_unregister(&wq_dev->dev); | |
a0a1a5fd | 5700 | } |
6ba94429 FW |
5701 | #else /* CONFIG_SYSFS */ |
5702 | static void workqueue_sysfs_unregister(struct workqueue_struct *wq) { } | |
5703 | #endif /* CONFIG_SYSFS */ | |
a0a1a5fd | 5704 | |
82607adc TH |
5705 | /* |
5706 | * Workqueue watchdog. | |
5707 | * | |
5708 | * Stall may be caused by various bugs - missing WQ_MEM_RECLAIM, illegal | |
5709 | * flush dependency, a concurrency managed work item which stays RUNNING | |
5710 | * indefinitely. Workqueue stalls can be very difficult to debug as the | |
5711 | * usual warning mechanisms don't trigger and internal workqueue state is | |
5712 | * largely opaque. | |
5713 | * | |
5714 | * Workqueue watchdog monitors all worker pools periodically and dumps | |
5715 | * state if some pools failed to make forward progress for a while where | |
5716 | * forward progress is defined as the first item on ->worklist changing. | |
5717 | * | |
5718 | * This mechanism is controlled through the kernel parameter | |
5719 | * "workqueue.watchdog_thresh" which can be updated at runtime through the | |
5720 | * corresponding sysfs parameter file. | |
5721 | */ | |
5722 | #ifdef CONFIG_WQ_WATCHDOG | |
5723 | ||
82607adc | 5724 | static unsigned long wq_watchdog_thresh = 30; |
5cd79d6a | 5725 | static struct timer_list wq_watchdog_timer; |
82607adc TH |
5726 | |
5727 | static unsigned long wq_watchdog_touched = INITIAL_JIFFIES; | |
5728 | static DEFINE_PER_CPU(unsigned long, wq_watchdog_touched_cpu) = INITIAL_JIFFIES; | |
5729 | ||
5730 | static void wq_watchdog_reset_touched(void) | |
5731 | { | |
5732 | int cpu; | |
5733 | ||
5734 | wq_watchdog_touched = jiffies; | |
5735 | for_each_possible_cpu(cpu) | |
5736 | per_cpu(wq_watchdog_touched_cpu, cpu) = jiffies; | |
5737 | } | |
5738 | ||
5cd79d6a | 5739 | static void wq_watchdog_timer_fn(struct timer_list *unused) |
82607adc TH |
5740 | { |
5741 | unsigned long thresh = READ_ONCE(wq_watchdog_thresh) * HZ; | |
5742 | bool lockup_detected = false; | |
5743 | struct worker_pool *pool; | |
5744 | int pi; | |
5745 | ||
5746 | if (!thresh) | |
5747 | return; | |
5748 | ||
5749 | rcu_read_lock(); | |
5750 | ||
5751 | for_each_pool(pool, pi) { | |
5752 | unsigned long pool_ts, touched, ts; | |
5753 | ||
5754 | if (list_empty(&pool->worklist)) | |
5755 | continue; | |
5756 | ||
5757 | /* get the latest of pool and touched timestamps */ | |
5758 | pool_ts = READ_ONCE(pool->watchdog_ts); | |
5759 | touched = READ_ONCE(wq_watchdog_touched); | |
5760 | ||
5761 | if (time_after(pool_ts, touched)) | |
5762 | ts = pool_ts; | |
5763 | else | |
5764 | ts = touched; | |
5765 | ||
5766 | if (pool->cpu >= 0) { | |
5767 | unsigned long cpu_touched = | |
5768 | READ_ONCE(per_cpu(wq_watchdog_touched_cpu, | |
5769 | pool->cpu)); | |
5770 | if (time_after(cpu_touched, ts)) | |
5771 | ts = cpu_touched; | |
5772 | } | |
5773 | ||
5774 | /* did we stall? */ | |
5775 | if (time_after(jiffies, ts + thresh)) { | |
5776 | lockup_detected = true; | |
5777 | pr_emerg("BUG: workqueue lockup - pool"); | |
5778 | pr_cont_pool_info(pool); | |
5779 | pr_cont(" stuck for %us!\n", | |
5780 | jiffies_to_msecs(jiffies - pool_ts) / 1000); | |
5781 | } | |
5782 | } | |
5783 | ||
5784 | rcu_read_unlock(); | |
5785 | ||
5786 | if (lockup_detected) | |
5787 | show_workqueue_state(); | |
5788 | ||
5789 | wq_watchdog_reset_touched(); | |
5790 | mod_timer(&wq_watchdog_timer, jiffies + thresh); | |
5791 | } | |
5792 | ||
cb9d7fd5 | 5793 | notrace void wq_watchdog_touch(int cpu) |
82607adc TH |
5794 | { |
5795 | if (cpu >= 0) | |
5796 | per_cpu(wq_watchdog_touched_cpu, cpu) = jiffies; | |
5797 | else | |
5798 | wq_watchdog_touched = jiffies; | |
5799 | } | |
5800 | ||
5801 | static void wq_watchdog_set_thresh(unsigned long thresh) | |
5802 | { | |
5803 | wq_watchdog_thresh = 0; | |
5804 | del_timer_sync(&wq_watchdog_timer); | |
5805 | ||
5806 | if (thresh) { | |
5807 | wq_watchdog_thresh = thresh; | |
5808 | wq_watchdog_reset_touched(); | |
5809 | mod_timer(&wq_watchdog_timer, jiffies + thresh * HZ); | |
5810 | } | |
5811 | } | |
5812 | ||
5813 | static int wq_watchdog_param_set_thresh(const char *val, | |
5814 | const struct kernel_param *kp) | |
5815 | { | |
5816 | unsigned long thresh; | |
5817 | int ret; | |
5818 | ||
5819 | ret = kstrtoul(val, 0, &thresh); | |
5820 | if (ret) | |
5821 | return ret; | |
5822 | ||
5823 | if (system_wq) | |
5824 | wq_watchdog_set_thresh(thresh); | |
5825 | else | |
5826 | wq_watchdog_thresh = thresh; | |
5827 | ||
5828 | return 0; | |
5829 | } | |
5830 | ||
5831 | static const struct kernel_param_ops wq_watchdog_thresh_ops = { | |
5832 | .set = wq_watchdog_param_set_thresh, | |
5833 | .get = param_get_ulong, | |
5834 | }; | |
5835 | ||
5836 | module_param_cb(watchdog_thresh, &wq_watchdog_thresh_ops, &wq_watchdog_thresh, | |
5837 | 0644); | |
5838 | ||
5839 | static void wq_watchdog_init(void) | |
5840 | { | |
5cd79d6a | 5841 | timer_setup(&wq_watchdog_timer, wq_watchdog_timer_fn, TIMER_DEFERRABLE); |
82607adc TH |
5842 | wq_watchdog_set_thresh(wq_watchdog_thresh); |
5843 | } | |
5844 | ||
5845 | #else /* CONFIG_WQ_WATCHDOG */ | |
5846 | ||
5847 | static inline void wq_watchdog_init(void) { } | |
5848 | ||
5849 | #endif /* CONFIG_WQ_WATCHDOG */ | |
5850 | ||
bce90380 TH |
5851 | static void __init wq_numa_init(void) |
5852 | { | |
5853 | cpumask_var_t *tbl; | |
5854 | int node, cpu; | |
5855 | ||
bce90380 TH |
5856 | if (num_possible_nodes() <= 1) |
5857 | return; | |
5858 | ||
d55262c4 TH |
5859 | if (wq_disable_numa) { |
5860 | pr_info("workqueue: NUMA affinity support disabled\n"); | |
5861 | return; | |
5862 | } | |
5863 | ||
be69d00d | 5864 | wq_update_unbound_numa_attrs_buf = alloc_workqueue_attrs(); |
4c16bd32 TH |
5865 | BUG_ON(!wq_update_unbound_numa_attrs_buf); |
5866 | ||
bce90380 TH |
5867 | /* |
5868 | * We want masks of possible CPUs of each node which isn't readily | |
5869 | * available. Build one from cpu_to_node() which should have been | |
5870 | * fully initialized by now. | |
5871 | */ | |
6396bb22 | 5872 | tbl = kcalloc(nr_node_ids, sizeof(tbl[0]), GFP_KERNEL); |
bce90380 TH |
5873 | BUG_ON(!tbl); |
5874 | ||
5875 | for_each_node(node) | |
5a6024f1 | 5876 | BUG_ON(!zalloc_cpumask_var_node(&tbl[node], GFP_KERNEL, |
1be0c25d | 5877 | node_online(node) ? node : NUMA_NO_NODE)); |
bce90380 TH |
5878 | |
5879 | for_each_possible_cpu(cpu) { | |
5880 | node = cpu_to_node(cpu); | |
5881 | if (WARN_ON(node == NUMA_NO_NODE)) { | |
5882 | pr_warn("workqueue: NUMA node mapping not available for cpu%d, disabling NUMA support\n", cpu); | |
5883 | /* happens iff arch is bonkers, let's just proceed */ | |
5884 | return; | |
5885 | } | |
5886 | cpumask_set_cpu(cpu, tbl[node]); | |
5887 | } | |
5888 | ||
5889 | wq_numa_possible_cpumask = tbl; | |
5890 | wq_numa_enabled = true; | |
5891 | } | |
5892 | ||
3347fa09 TH |
5893 | /** |
5894 | * workqueue_init_early - early init for workqueue subsystem | |
5895 | * | |
5896 | * This is the first half of two-staged workqueue subsystem initialization | |
5897 | * and invoked as soon as the bare basics - memory allocation, cpumasks and | |
5898 | * idr are up. It sets up all the data structures and system workqueues | |
5899 | * and allows early boot code to create workqueues and queue/cancel work | |
5900 | * items. Actual work item execution starts only after kthreads can be | |
5901 | * created and scheduled right before early initcalls. | |
5902 | */ | |
2333e829 | 5903 | void __init workqueue_init_early(void) |
1da177e4 | 5904 | { |
7a4e344c | 5905 | int std_nice[NR_STD_WORKER_POOLS] = { 0, HIGHPRI_NICE_LEVEL }; |
1bda3f80 | 5906 | int hk_flags = HK_FLAG_DOMAIN | HK_FLAG_WQ; |
7a4e344c | 5907 | int i, cpu; |
c34056a3 | 5908 | |
e904e6c2 TH |
5909 | WARN_ON(__alignof__(struct pool_workqueue) < __alignof__(long long)); |
5910 | ||
b05a7928 | 5911 | BUG_ON(!alloc_cpumask_var(&wq_unbound_cpumask, GFP_KERNEL)); |
1bda3f80 | 5912 | cpumask_copy(wq_unbound_cpumask, housekeeping_cpumask(hk_flags)); |
b05a7928 | 5913 | |
e904e6c2 TH |
5914 | pwq_cache = KMEM_CACHE(pool_workqueue, SLAB_PANIC); |
5915 | ||
706026c2 | 5916 | /* initialize CPU pools */ |
29c91e99 | 5917 | for_each_possible_cpu(cpu) { |
4ce62e9e | 5918 | struct worker_pool *pool; |
8b03ae3c | 5919 | |
7a4e344c | 5920 | i = 0; |
f02ae73a | 5921 | for_each_cpu_worker_pool(pool, cpu) { |
7a4e344c | 5922 | BUG_ON(init_worker_pool(pool)); |
ec22ca5e | 5923 | pool->cpu = cpu; |
29c91e99 | 5924 | cpumask_copy(pool->attrs->cpumask, cpumask_of(cpu)); |
7a4e344c | 5925 | pool->attrs->nice = std_nice[i++]; |
f3f90ad4 | 5926 | pool->node = cpu_to_node(cpu); |
7a4e344c | 5927 | |
9daf9e67 | 5928 | /* alloc pool ID */ |
68e13a67 | 5929 | mutex_lock(&wq_pool_mutex); |
9daf9e67 | 5930 | BUG_ON(worker_pool_assign_id(pool)); |
68e13a67 | 5931 | mutex_unlock(&wq_pool_mutex); |
4ce62e9e | 5932 | } |
8b03ae3c TH |
5933 | } |
5934 | ||
8a2b7538 | 5935 | /* create default unbound and ordered wq attrs */ |
29c91e99 TH |
5936 | for (i = 0; i < NR_STD_WORKER_POOLS; i++) { |
5937 | struct workqueue_attrs *attrs; | |
5938 | ||
be69d00d | 5939 | BUG_ON(!(attrs = alloc_workqueue_attrs())); |
29c91e99 | 5940 | attrs->nice = std_nice[i]; |
29c91e99 | 5941 | unbound_std_wq_attrs[i] = attrs; |
8a2b7538 TH |
5942 | |
5943 | /* | |
5944 | * An ordered wq should have only one pwq as ordering is | |
5945 | * guaranteed by max_active which is enforced by pwqs. | |
5946 | * Turn off NUMA so that dfl_pwq is used for all nodes. | |
5947 | */ | |
be69d00d | 5948 | BUG_ON(!(attrs = alloc_workqueue_attrs())); |
8a2b7538 TH |
5949 | attrs->nice = std_nice[i]; |
5950 | attrs->no_numa = true; | |
5951 | ordered_wq_attrs[i] = attrs; | |
29c91e99 TH |
5952 | } |
5953 | ||
d320c038 | 5954 | system_wq = alloc_workqueue("events", 0, 0); |
1aabe902 | 5955 | system_highpri_wq = alloc_workqueue("events_highpri", WQ_HIGHPRI, 0); |
d320c038 | 5956 | system_long_wq = alloc_workqueue("events_long", 0, 0); |
f3421797 TH |
5957 | system_unbound_wq = alloc_workqueue("events_unbound", WQ_UNBOUND, |
5958 | WQ_UNBOUND_MAX_ACTIVE); | |
24d51add TH |
5959 | system_freezable_wq = alloc_workqueue("events_freezable", |
5960 | WQ_FREEZABLE, 0); | |
0668106c VK |
5961 | system_power_efficient_wq = alloc_workqueue("events_power_efficient", |
5962 | WQ_POWER_EFFICIENT, 0); | |
5963 | system_freezable_power_efficient_wq = alloc_workqueue("events_freezable_power_efficient", | |
5964 | WQ_FREEZABLE | WQ_POWER_EFFICIENT, | |
5965 | 0); | |
1aabe902 | 5966 | BUG_ON(!system_wq || !system_highpri_wq || !system_long_wq || |
0668106c VK |
5967 | !system_unbound_wq || !system_freezable_wq || |
5968 | !system_power_efficient_wq || | |
5969 | !system_freezable_power_efficient_wq); | |
3347fa09 TH |
5970 | } |
5971 | ||
5972 | /** | |
5973 | * workqueue_init - bring workqueue subsystem fully online | |
5974 | * | |
5975 | * This is the latter half of two-staged workqueue subsystem initialization | |
5976 | * and invoked as soon as kthreads can be created and scheduled. | |
5977 | * Workqueues have been created and work items queued on them, but there | |
5978 | * are no kworkers executing the work items yet. Populate the worker pools | |
5979 | * with the initial workers and enable future kworker creations. | |
5980 | */ | |
2333e829 | 5981 | void __init workqueue_init(void) |
3347fa09 | 5982 | { |
2186d9f9 | 5983 | struct workqueue_struct *wq; |
3347fa09 TH |
5984 | struct worker_pool *pool; |
5985 | int cpu, bkt; | |
5986 | ||
2186d9f9 TH |
5987 | /* |
5988 | * It'd be simpler to initialize NUMA in workqueue_init_early() but | |
5989 | * CPU to node mapping may not be available that early on some | |
5990 | * archs such as power and arm64. As per-cpu pools created | |
5991 | * previously could be missing node hint and unbound pools NUMA | |
5992 | * affinity, fix them up. | |
40c17f75 TH |
5993 | * |
5994 | * Also, while iterating workqueues, create rescuers if requested. | |
2186d9f9 TH |
5995 | */ |
5996 | wq_numa_init(); | |
5997 | ||
5998 | mutex_lock(&wq_pool_mutex); | |
5999 | ||
6000 | for_each_possible_cpu(cpu) { | |
6001 | for_each_cpu_worker_pool(pool, cpu) { | |
6002 | pool->node = cpu_to_node(cpu); | |
6003 | } | |
6004 | } | |
6005 | ||
40c17f75 | 6006 | list_for_each_entry(wq, &workqueues, list) { |
2186d9f9 | 6007 | wq_update_unbound_numa(wq, smp_processor_id(), true); |
40c17f75 TH |
6008 | WARN(init_rescuer(wq), |
6009 | "workqueue: failed to create early rescuer for %s", | |
6010 | wq->name); | |
6011 | } | |
2186d9f9 TH |
6012 | |
6013 | mutex_unlock(&wq_pool_mutex); | |
6014 | ||
3347fa09 TH |
6015 | /* create the initial workers */ |
6016 | for_each_online_cpu(cpu) { | |
6017 | for_each_cpu_worker_pool(pool, cpu) { | |
6018 | pool->flags &= ~POOL_DISASSOCIATED; | |
6019 | BUG_ON(!create_worker(pool)); | |
6020 | } | |
6021 | } | |
6022 | ||
6023 | hash_for_each(unbound_pool_hash, bkt, pool, hash_node) | |
6024 | BUG_ON(!create_worker(pool)); | |
6025 | ||
6026 | wq_online = true; | |
82607adc | 6027 | wq_watchdog_init(); |
1da177e4 | 6028 | } |