Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
Tejun Heo | c54fce6 | 2010-09-10 16:51:36 +0200 | [diff] [blame] | 2 | * kernel/workqueue.c - generic async execution with shared worker pool |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3 | * |
Tejun Heo | c54fce6 | 2010-09-10 16:51:36 +0200 | [diff] [blame] | 4 | * Copyright (C) 2002 Ingo Molnar |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5 | * |
Tejun Heo | c54fce6 | 2010-09-10 16:51:36 +0200 | [diff] [blame] | 6 | * Derived from the taskqueue/keventd code by: |
| 7 | * David Woodhouse <dwmw2@infradead.org> |
| 8 | * Andrew Morton |
| 9 | * Kai Petzke <wpp@marie.physik.tu-berlin.de> |
| 10 | * Theodore Ts'o <tytso@mit.edu> |
Christoph Lameter | 89ada67 | 2005-10-30 15:01:59 -0800 | [diff] [blame] | 11 | * |
Christoph Lameter | cde5353 | 2008-07-04 09:59:22 -0700 | [diff] [blame] | 12 | * Made to use alloc_percpu by Christoph Lameter. |
Tejun Heo | c54fce6 | 2010-09-10 16:51:36 +0200 | [diff] [blame] | 13 | * |
| 14 | * Copyright (C) 2010 SUSE Linux Products GmbH |
| 15 | * Copyright (C) 2010 Tejun Heo <tj@kernel.org> |
| 16 | * |
| 17 | * This is the generic async execution mechanism. Work items as are |
| 18 | * executed in process context. The worker pool is shared and |
Libin | b11895c | 2013-08-21 08:50:39 +0800 | [diff] [blame] | 19 | * automatically managed. There are two worker pools for each CPU (one for |
| 20 | * normal work items and the other for high priority ones) and some extra |
| 21 | * pools for workqueues which are not bound to any specific CPU - the |
| 22 | * number of these backing pools is dynamic. |
Tejun Heo | c54fce6 | 2010-09-10 16:51:36 +0200 | [diff] [blame] | 23 | * |
Benjamin Peterson | 9a26149 | 2017-08-06 19:33:22 -0700 | [diff] [blame] | 24 | * Please read Documentation/core-api/workqueue.rst for details. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 25 | */ |
| 26 | |
Paul Gortmaker | 9984de1 | 2011-05-23 14:51:41 -0400 | [diff] [blame] | 27 | #include <linux/export.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 28 | #include <linux/kernel.h> |
| 29 | #include <linux/sched.h> |
| 30 | #include <linux/init.h> |
| 31 | #include <linux/signal.h> |
| 32 | #include <linux/completion.h> |
| 33 | #include <linux/workqueue.h> |
| 34 | #include <linux/slab.h> |
| 35 | #include <linux/cpu.h> |
| 36 | #include <linux/notifier.h> |
| 37 | #include <linux/kthread.h> |
James Bottomley | 1fa44ec | 2006-02-23 12:43:43 -0600 | [diff] [blame] | 38 | #include <linux/hardirq.h> |
Christoph Lameter | 4693402 | 2006-10-11 01:21:26 -0700 | [diff] [blame] | 39 | #include <linux/mempolicy.h> |
Rafael J. Wysocki | 341a595 | 2006-12-06 20:34:49 -0800 | [diff] [blame] | 40 | #include <linux/freezer.h> |
Peter Zijlstra | d5abe66 | 2006-12-06 20:37:26 -0800 | [diff] [blame] | 41 | #include <linux/kallsyms.h> |
| 42 | #include <linux/debug_locks.h> |
Johannes Berg | 4e6045f | 2007-10-18 23:39:55 -0700 | [diff] [blame] | 43 | #include <linux/lockdep.h> |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 44 | #include <linux/idr.h> |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 45 | #include <linux/jhash.h> |
Sasha Levin | 42f8570 | 2012-12-17 10:01:23 -0500 | [diff] [blame] | 46 | #include <linux/hashtable.h> |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 47 | #include <linux/rculist.h> |
Tejun Heo | bce9038 | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 48 | #include <linux/nodemask.h> |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 49 | #include <linux/moduleparam.h> |
Tejun Heo | 3d1cb20 | 2013-04-30 15:27:22 -0700 | [diff] [blame] | 50 | #include <linux/uaccess.h> |
Sergey Senozhatsky | a6d5930 | 2018-01-11 09:53:35 +0900 | [diff] [blame] | 51 | #include <linux/nmi.h> |
Sergey Senozhatsky | ec32c8a | 2021-05-20 19:14:22 +0900 | [diff] [blame] | 52 | #include <linux/kvm_para.h> |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 53 | |
Tejun Heo | ea13844 | 2013-01-18 14:05:55 -0800 | [diff] [blame] | 54 | #include "workqueue_internal.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 55 | |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 56 | enum { |
Tejun Heo | bc2ae0f | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 57 | /* |
Tejun Heo | 2464757 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 58 | * worker_pool flags |
Tejun Heo | bc2ae0f | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 59 | * |
Tejun Heo | 2464757 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 60 | * A bound pool is either associated or disassociated with its CPU. |
Tejun Heo | bc2ae0f | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 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 |
Tejun Heo | 2464757 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 67 | * be executing on any CPU. The pool behaves as an unbound one. |
Tejun Heo | bc2ae0f | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 68 | * |
Tejun Heo | bc3a1af | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 69 | * Note that DISASSOCIATED should be flipped only while holding |
Lai Jiangshan | 92f9c5c | 2014-05-20 17:46:34 +0800 | [diff] [blame] | 70 | * attach_mutex to avoid changing binding state while |
Lai Jiangshan | 4736cbf | 2014-05-20 17:46:35 +0800 | [diff] [blame] | 71 | * worker_attach_to_pool() is in progress. |
Tejun Heo | bc2ae0f | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 72 | */ |
Tejun Heo | 692b482 | 2017-10-09 08:04:13 -0700 | [diff] [blame] | 73 | POOL_MANAGER_ACTIVE = 1 << 0, /* being managed */ |
Tejun Heo | 2464757 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 74 | POOL_DISASSOCIATED = 1 << 2, /* cpu can't serve workers */ |
Tejun Heo | db7bccf | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 75 | |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 76 | /* worker flags */ |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 77 | WORKER_DIE = 1 << 1, /* die die die */ |
| 78 | WORKER_IDLE = 1 << 2, /* is idle */ |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 79 | WORKER_PREP = 1 << 3, /* preparing to run works */ |
Tejun Heo | fb0e7be | 2010-06-29 10:07:15 +0200 | [diff] [blame] | 80 | WORKER_CPU_INTENSIVE = 1 << 6, /* cpu intensive */ |
Tejun Heo | f342179 | 2010-07-02 10:03:51 +0200 | [diff] [blame] | 81 | WORKER_UNBOUND = 1 << 7, /* worker is unbound */ |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 82 | WORKER_REBOUND = 1 << 8, /* worker was rebound */ |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 83 | |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 84 | WORKER_NOT_RUNNING = WORKER_PREP | WORKER_CPU_INTENSIVE | |
| 85 | WORKER_UNBOUND | WORKER_REBOUND, |
Tejun Heo | db7bccf | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 86 | |
Tejun Heo | e34cdddb | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 87 | NR_STD_WORKER_POOLS = 2, /* # standard pools per cpu */ |
Tejun Heo | 4ce62e9 | 2012-07-13 22:16:44 -0700 | [diff] [blame] | 88 | |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 89 | UNBOUND_POOL_HASH_ORDER = 6, /* hashed by pool->attrs */ |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 90 | BUSY_WORKER_HASH_ORDER = 6, /* 64 pointers */ |
Tejun Heo | db7bccf | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 91 | |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 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 | |
Tejun Heo | 3233cdb | 2011-02-16 18:10:19 +0100 | [diff] [blame] | 95 | MAYDAY_INITIAL_TIMEOUT = HZ / 100 >= 2 ? HZ / 100 : 2, |
| 96 | /* call for help after 10ms |
| 97 | (min two ticks) */ |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 98 | MAYDAY_INTERVAL = HZ / 10, /* and then every 100ms */ |
| 99 | CREATE_COOLDOWN = HZ, /* time to breath after fail */ |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 100 | |
| 101 | /* |
| 102 | * Rescue workers are used only on emergencies and shared by |
Dongsheng Yang | 8698a74 | 2014-03-11 18:09:12 +0800 | [diff] [blame] | 103 | * all cpus. Give MIN_NICE. |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 104 | */ |
Dongsheng Yang | 8698a74 | 2014-03-11 18:09:12 +0800 | [diff] [blame] | 105 | RESCUER_NICE_LEVEL = MIN_NICE, |
| 106 | HIGHPRI_NICE_LEVEL = MIN_NICE, |
Tejun Heo | ecf6881 | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 107 | |
| 108 | WQ_NAME_LEN = 24, |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 109 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 110 | |
| 111 | /* |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 112 | * Structure fields follow one of the following exclusion rules. |
| 113 | * |
Tejun Heo | e41e704 | 2010-08-24 14:22:47 +0200 | [diff] [blame] | 114 | * I: Modifiable by initialization/destruction paths and read-only for |
| 115 | * everyone else. |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 116 | * |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 117 | * P: Preemption protected. Disabling preemption is enough and should |
| 118 | * only be modified and accessed from the local cpu. |
| 119 | * |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 120 | * L: pool->lock protected. Access with pool->lock held. |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 121 | * |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 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. |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 126 | * |
Lai Jiangshan | 92f9c5c | 2014-05-20 17:46:34 +0800 | [diff] [blame] | 127 | * A: pool->attach_mutex protected. |
Tejun Heo | 822d840 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 128 | * |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 129 | * PL: wq_pool_mutex protected. |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 130 | * |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 131 | * PR: wq_pool_mutex protected for writes. Sched-RCU protected for reads. |
Tejun Heo | 5bcab33 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 132 | * |
Lai Jiangshan | 5b95e1a | 2015-05-12 20:32:29 +0800 | [diff] [blame] | 133 | * PW: wq_pool_mutex and wq->mutex protected for writes. Either for reads. |
| 134 | * |
| 135 | * PWR: wq_pool_mutex and wq->mutex protected for writes. Either or |
| 136 | * sched-RCU for reads. |
| 137 | * |
Lai Jiangshan | 3c25a55 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 138 | * WQ: wq->mutex protected. |
| 139 | * |
Lai Jiangshan | b592760 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 140 | * WR: wq->mutex protected for writes. Sched-RCU protected for reads. |
Tejun Heo | 2e109a2 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 141 | * |
| 142 | * MD: wq_mayday_lock protected. |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 143 | */ |
| 144 | |
Tejun Heo | 2eaebdb | 2013-01-18 14:05:55 -0800 | [diff] [blame] | 145 | /* struct worker is defined in workqueue_internal.h */ |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 146 | |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 147 | struct worker_pool { |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 148 | spinlock_t lock; /* the pool lock */ |
Tejun Heo | d84ff05 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 149 | int cpu; /* I: the associated cpu */ |
Tejun Heo | f3f90ad | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 150 | int node; /* I: the associated node ID */ |
Tejun Heo | 9daf9e6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 151 | int id; /* I: pool ID */ |
Tejun Heo | 11ebea5 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 152 | unsigned int flags; /* X: flags */ |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 153 | |
Tejun Heo | 82607adc | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 154 | unsigned long watchdog_ts; /* L: watchdog timestamp */ |
| 155 | |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 156 | struct list_head worklist; /* L: list of pending works */ |
| 157 | int nr_workers; /* L: total number of workers */ |
Lai Jiangshan | ea1abd6 | 2012-09-18 09:59:22 -0700 | [diff] [blame] | 158 | |
| 159 | /* nr_idle includes the ones off idle_list for rebinding */ |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 160 | int nr_idle; /* L: currently idle ones */ |
| 161 | |
| 162 | struct list_head idle_list; /* X: list of idle workers */ |
| 163 | struct timer_list idle_timer; /* L: worker idle timeout */ |
| 164 | struct timer_list mayday_timer; /* L: SOS timer for workers */ |
| 165 | |
Tejun Heo | c5aa87b | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 166 | /* a workers is either on busy_hash or idle_list, or the manager */ |
Tejun Heo | c9e7cf2 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 167 | DECLARE_HASHTABLE(busy_hash, BUSY_WORKER_HASH_ORDER); |
| 168 | /* L: hash of busy workers */ |
| 169 | |
Tejun Heo | bc3a1af | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 170 | /* see manage_workers() for details on the two manager mutexes */ |
Tejun Heo | 2607d7a | 2015-03-09 09:22:28 -0400 | [diff] [blame] | 171 | struct worker *manager; /* L: purely informational */ |
Lai Jiangshan | 92f9c5c | 2014-05-20 17:46:34 +0800 | [diff] [blame] | 172 | struct mutex attach_mutex; /* attach/detach exclusion */ |
| 173 | struct list_head workers; /* A: attached workers */ |
Lai Jiangshan | 60f5a4b | 2014-05-20 17:46:29 +0800 | [diff] [blame] | 174 | struct completion *detach_completion; /* all workers detached */ |
Tejun Heo | e19e397 | 2013-01-24 11:39:44 -0800 | [diff] [blame] | 175 | |
Lai Jiangshan | 7cda9aa | 2014-05-20 17:46:32 +0800 | [diff] [blame] | 176 | struct ida worker_ida; /* worker IDs for task name */ |
Tejun Heo | e19e397 | 2013-01-24 11:39:44 -0800 | [diff] [blame] | 177 | |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 178 | struct workqueue_attrs *attrs; /* I: worker attributes */ |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 179 | struct hlist_node hash_node; /* PL: unbound_pool_hash node */ |
| 180 | int refcnt; /* PL: refcnt for unbound pools */ |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 181 | |
Tejun Heo | e19e397 | 2013-01-24 11:39:44 -0800 | [diff] [blame] | 182 | /* |
| 183 | * The current concurrency level. As it's likely to be accessed |
| 184 | * from other CPUs during try_to_wake_up(), put it in a separate |
| 185 | * cacheline. |
| 186 | */ |
| 187 | atomic_t nr_running ____cacheline_aligned_in_smp; |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 188 | |
| 189 | /* |
| 190 | * Destruction of pool is sched-RCU protected to allow dereferences |
| 191 | * from get_work_pool(). |
| 192 | */ |
| 193 | struct rcu_head rcu; |
Tejun Heo | 8b03ae3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 194 | } ____cacheline_aligned_in_smp; |
| 195 | |
| 196 | /* |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 197 | * The per-pool workqueue. While queued, the lower WORK_STRUCT_FLAG_BITS |
| 198 | * of work_struct->data are used for flags and the remaining high bits |
| 199 | * point to the pwq; thus, pwqs need to be aligned at two's power of the |
| 200 | * number of flag bits. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 201 | */ |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 202 | struct pool_workqueue { |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 203 | struct worker_pool *pool; /* I: the associated pool */ |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 204 | struct workqueue_struct *wq; /* I: the owning workqueue */ |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 205 | int work_color; /* L: current color */ |
| 206 | int flush_color; /* L: flushing color */ |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 207 | int refcnt; /* L: reference count */ |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 208 | int nr_in_flight[WORK_NR_COLORS]; |
| 209 | /* L: nr of in_flight works */ |
Tejun Heo | 1e19ffc | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 210 | int nr_active; /* L: nr of active works */ |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 211 | int max_active; /* L: max active works */ |
Tejun Heo | 1e19ffc | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 212 | struct list_head delayed_works; /* L: delayed works */ |
Lai Jiangshan | 3c25a55 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 213 | struct list_head pwqs_node; /* WR: node on wq->pwqs */ |
Tejun Heo | 2e109a2 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 214 | struct list_head mayday_node; /* MD: node on wq->maydays */ |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 215 | |
| 216 | /* |
| 217 | * Release of unbound pwq is punted to system_wq. See put_pwq() |
| 218 | * and pwq_unbound_release_workfn() for details. pool_workqueue |
| 219 | * itself is also sched-RCU protected so that the first pwq can be |
Lai Jiangshan | b09f4fd | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 220 | * determined without grabbing wq->mutex. |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 221 | */ |
| 222 | struct work_struct unbound_release_work; |
| 223 | struct rcu_head rcu; |
Tejun Heo | e904e6c | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 224 | } __aligned(1 << WORK_STRUCT_FLAG_BITS); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 225 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 226 | /* |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 227 | * Structure used to wait for workqueue flush. |
| 228 | */ |
| 229 | struct wq_flusher { |
Lai Jiangshan | 3c25a55 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 230 | struct list_head list; /* WQ: list of flushers */ |
| 231 | int flush_color; /* WQ: flush color waiting for */ |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 232 | struct completion done; /* flush completion */ |
| 233 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 234 | |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 235 | struct wq_device; |
| 236 | |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 237 | /* |
Tejun Heo | c5aa87b | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 238 | * The externally visible workqueue. It relays the issued work items to |
| 239 | * the appropriate worker_pool through its pool_workqueues. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 240 | */ |
| 241 | struct workqueue_struct { |
Lai Jiangshan | 3c25a55 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 242 | struct list_head pwqs; /* WR: all pwqs of this wq */ |
Tejun Heo | e2dca7a | 2015-03-09 09:22:28 -0400 | [diff] [blame] | 243 | struct list_head list; /* PR: list of all workqueues */ |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 244 | |
Lai Jiangshan | 3c25a55 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 245 | struct mutex mutex; /* protects this wq */ |
| 246 | int work_color; /* WQ: current work color */ |
| 247 | int flush_color; /* WQ: current flush color */ |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 248 | atomic_t nr_pwqs_to_flush; /* flush in progress */ |
Lai Jiangshan | 3c25a55 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 249 | struct wq_flusher *first_flusher; /* WQ: first flusher */ |
| 250 | struct list_head flusher_queue; /* WQ: flush waiters */ |
| 251 | struct list_head flusher_overflow; /* WQ: flush overflow list */ |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 252 | |
Tejun Heo | 2e109a2 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 253 | struct list_head maydays; /* MD: pwqs requesting rescue */ |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 254 | struct worker *rescuer; /* I: rescue worker */ |
| 255 | |
Lai Jiangshan | 87fc741 | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 256 | int nr_drainers; /* WQ: drain in progress */ |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 257 | int saved_max_active; /* WQ: saved pwq max_active */ |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 258 | |
Lai Jiangshan | 5b95e1a | 2015-05-12 20:32:29 +0800 | [diff] [blame] | 259 | struct workqueue_attrs *unbound_attrs; /* PW: only for unbound wqs */ |
| 260 | struct pool_workqueue *dfl_pwq; /* PW: only for unbound wqs */ |
Tejun Heo | 6029a91 | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 261 | |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 262 | #ifdef CONFIG_SYSFS |
| 263 | struct wq_device *wq_dev; /* I: for sysfs interface */ |
| 264 | #endif |
Johannes Berg | 4e6045f | 2007-10-18 23:39:55 -0700 | [diff] [blame] | 265 | #ifdef CONFIG_LOCKDEP |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 266 | struct lockdep_map lockdep_map; |
Johannes Berg | 4e6045f | 2007-10-18 23:39:55 -0700 | [diff] [blame] | 267 | #endif |
Tejun Heo | ecf6881 | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 268 | char name[WQ_NAME_LEN]; /* I: workqueue name */ |
Tejun Heo | 2728fd2 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 269 | |
Tejun Heo | e2dca7a | 2015-03-09 09:22:28 -0400 | [diff] [blame] | 270 | /* |
| 271 | * Destruction of workqueue_struct is sched-RCU protected to allow |
| 272 | * walking the workqueues list without grabbing wq_pool_mutex. |
| 273 | * This is used to dump all workqueues from sysrq. |
| 274 | */ |
| 275 | struct rcu_head rcu; |
| 276 | |
Tejun Heo | 2728fd2 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 277 | /* hot fields used during command issue, aligned to cacheline */ |
| 278 | unsigned int flags ____cacheline_aligned; /* WQ: WQ_* flags */ |
| 279 | struct pool_workqueue __percpu *cpu_pwqs; /* I: per-cpu pwqs */ |
Lai Jiangshan | 5b95e1a | 2015-05-12 20:32:29 +0800 | [diff] [blame] | 280 | struct pool_workqueue __rcu *numa_pwq_tbl[]; /* PWR: unbound pwqs indexed by node */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 281 | }; |
| 282 | |
Tejun Heo | e904e6c | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 283 | static struct kmem_cache *pwq_cache; |
| 284 | |
Tejun Heo | bce9038 | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 285 | static cpumask_var_t *wq_numa_possible_cpumask; |
| 286 | /* possible CPUs of each node */ |
| 287 | |
Tejun Heo | d55262c | 2013-04-01 11:23:38 -0700 | [diff] [blame] | 288 | static bool wq_disable_numa; |
| 289 | module_param_named(disable_numa, wq_disable_numa, bool, 0444); |
| 290 | |
Viresh Kumar | cee22a1 | 2013-04-08 16:45:40 +0530 | [diff] [blame] | 291 | /* see the comment above the definition of WQ_POWER_EFFICIENT */ |
Luis R. Rodriguez | 552f530 | 2015-05-27 11:09:39 +0930 | [diff] [blame] | 292 | static bool wq_power_efficient = IS_ENABLED(CONFIG_WQ_POWER_EFFICIENT_DEFAULT); |
Viresh Kumar | cee22a1 | 2013-04-08 16:45:40 +0530 | [diff] [blame] | 293 | module_param_named(power_efficient, wq_power_efficient, bool, 0444); |
| 294 | |
Tejun Heo | 863b710 | 2016-09-16 15:49:34 -0400 | [diff] [blame] | 295 | static bool wq_online; /* can kworkers be created yet? */ |
Tejun Heo | 3347fa0 | 2016-09-16 15:49:32 -0400 | [diff] [blame] | 296 | |
Tejun Heo | bce9038 | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 297 | static bool wq_numa_enabled; /* unbound NUMA affinity enabled */ |
| 298 | |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 299 | /* buf for wq_update_unbound_numa_attrs(), protected by CPU hotplug exclusion */ |
| 300 | static struct workqueue_attrs *wq_update_unbound_numa_attrs_buf; |
| 301 | |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 302 | static DEFINE_MUTEX(wq_pool_mutex); /* protects pools and workqueues list */ |
Tejun Heo | 2e109a2 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 303 | static DEFINE_SPINLOCK(wq_mayday_lock); /* protects wq->maydays list */ |
Tejun Heo | 692b482 | 2017-10-09 08:04:13 -0700 | [diff] [blame] | 304 | static DECLARE_WAIT_QUEUE_HEAD(wq_manager_wait); /* wait for manager to go away */ |
Tejun Heo | 5bcab33 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 305 | |
Tejun Heo | e2dca7a | 2015-03-09 09:22:28 -0400 | [diff] [blame] | 306 | static LIST_HEAD(workqueues); /* PR: list of all workqueues */ |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 307 | static bool workqueue_freezing; /* PL: have wqs started freezing? */ |
Tejun Heo | 7d19c5c | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 308 | |
Mike Galbraith | ef557180 | 2016-02-09 17:59:38 -0500 | [diff] [blame] | 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); |
Frederic Weisbecker | b05a792 | 2015-04-27 17:58:39 +0800 | [diff] [blame] | 314 | |
Tejun Heo | f303fcc | 2016-02-09 17:59:38 -0500 | [diff] [blame] | 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 | |
Tejun Heo | 7d19c5c | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 327 | /* the per-cpu worker pools */ |
Peter Zijlstra | 2552821 | 2016-03-15 14:52:49 -0700 | [diff] [blame] | 328 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct worker_pool [NR_STD_WORKER_POOLS], cpu_worker_pools); |
Tejun Heo | 7d19c5c | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 329 | |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 330 | static DEFINE_IDR(worker_pool_idr); /* PR: idr of all pools */ |
Tejun Heo | 7d19c5c | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 331 | |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 332 | /* PL: hash of all unbound pools keyed by pool->attrs */ |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 333 | static DEFINE_HASHTABLE(unbound_pool_hash, UNBOUND_POOL_HASH_ORDER); |
| 334 | |
Tejun Heo | c5aa87b | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 335 | /* I: attributes used when instantiating standard unbound pools on demand */ |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 336 | static struct workqueue_attrs *unbound_std_wq_attrs[NR_STD_WORKER_POOLS]; |
| 337 | |
Tejun Heo | 8a2b753 | 2013-09-05 12:30:04 -0400 | [diff] [blame] | 338 | /* I: attributes used when instantiating ordered pools on demand */ |
| 339 | static struct workqueue_attrs *ordered_wq_attrs[NR_STD_WORKER_POOLS]; |
| 340 | |
Tejun Heo | d320c03 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 341 | struct workqueue_struct *system_wq __read_mostly; |
Marc Dionne | ad7b1f8 | 2013-05-06 17:44:55 -0400 | [diff] [blame] | 342 | EXPORT_SYMBOL(system_wq); |
Valentin Ilie | 044c782 | 2012-08-19 00:52:42 +0300 | [diff] [blame] | 343 | struct workqueue_struct *system_highpri_wq __read_mostly; |
Joonsoo Kim | 1aabe90 | 2012-08-15 23:25:39 +0900 | [diff] [blame] | 344 | EXPORT_SYMBOL_GPL(system_highpri_wq); |
Valentin Ilie | 044c782 | 2012-08-19 00:52:42 +0300 | [diff] [blame] | 345 | struct workqueue_struct *system_long_wq __read_mostly; |
Tejun Heo | d320c03 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 346 | EXPORT_SYMBOL_GPL(system_long_wq); |
Valentin Ilie | 044c782 | 2012-08-19 00:52:42 +0300 | [diff] [blame] | 347 | struct workqueue_struct *system_unbound_wq __read_mostly; |
Tejun Heo | f342179 | 2010-07-02 10:03:51 +0200 | [diff] [blame] | 348 | EXPORT_SYMBOL_GPL(system_unbound_wq); |
Valentin Ilie | 044c782 | 2012-08-19 00:52:42 +0300 | [diff] [blame] | 349 | struct workqueue_struct *system_freezable_wq __read_mostly; |
Tejun Heo | 24d51ad | 2011-02-21 09:52:50 +0100 | [diff] [blame] | 350 | EXPORT_SYMBOL_GPL(system_freezable_wq); |
Viresh Kumar | 0668106 | 2013-04-24 17:12:54 +0530 | [diff] [blame] | 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); |
Tejun Heo | d320c03 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 355 | |
Tejun Heo | 7d19c5c | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 356 | static int worker_thread(void *__worker); |
Frederic Weisbecker | 6ba9442 | 2015-04-02 19:14:39 +0800 | [diff] [blame] | 357 | static void workqueue_sysfs_unregister(struct workqueue_struct *wq); |
Tejun Heo | 7d19c5c | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 358 | |
Tejun Heo | 97bd234 | 2010-10-05 10:41:14 +0200 | [diff] [blame] | 359 | #define CREATE_TRACE_POINTS |
| 360 | #include <trace/events/workqueue.h> |
| 361 | |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 362 | #define assert_rcu_or_pool_mutex() \ |
Paul E. McKenney | f78f5b9 | 2015-06-18 15:50:02 -0700 | [diff] [blame] | 363 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held() && \ |
| 364 | !lockdep_is_held(&wq_pool_mutex), \ |
| 365 | "sched RCU or wq_pool_mutex should be held") |
Tejun Heo | 5bcab33 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 366 | |
Lai Jiangshan | b09f4fd | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 367 | #define assert_rcu_or_wq_mutex(wq) \ |
Paul E. McKenney | f78f5b9 | 2015-06-18 15:50:02 -0700 | [diff] [blame] | 368 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held() && \ |
| 369 | !lockdep_is_held(&wq->mutex), \ |
| 370 | "sched RCU or wq->mutex should be held") |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 371 | |
Lai Jiangshan | 5b95e1a | 2015-05-12 20:32:29 +0800 | [diff] [blame] | 372 | #define assert_rcu_or_wq_mutex_or_pool_mutex(wq) \ |
Paul E. McKenney | f78f5b9 | 2015-06-18 15:50:02 -0700 | [diff] [blame] | 373 | RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held() && \ |
| 374 | !lockdep_is_held(&wq->mutex) && \ |
| 375 | !lockdep_is_held(&wq_pool_mutex), \ |
| 376 | "sched RCU, wq->mutex or wq_pool_mutex should be held") |
Lai Jiangshan | 5b95e1a | 2015-05-12 20:32:29 +0800 | [diff] [blame] | 377 | |
Tejun Heo | f02ae73 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 378 | #define for_each_cpu_worker_pool(pool, cpu) \ |
| 379 | for ((pool) = &per_cpu(cpu_worker_pools, cpu)[0]; \ |
| 380 | (pool) < &per_cpu(cpu_worker_pools, cpu)[NR_STD_WORKER_POOLS]; \ |
Tejun Heo | 7a62c2c | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 381 | (pool)++) |
Tejun Heo | 4ce62e9 | 2012-07-13 22:16:44 -0700 | [diff] [blame] | 382 | |
Tejun Heo | 49e3cf4 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 383 | /** |
Tejun Heo | 1711696 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 384 | * for_each_pool - iterate through all worker_pools in the system |
| 385 | * @pool: iteration cursor |
Tejun Heo | 611c92a | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 386 | * @pi: integer used for iteration |
Tejun Heo | fa1b54e | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 387 | * |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 388 | * This must be called either with wq_pool_mutex held or sched RCU read |
| 389 | * locked. If the pool needs to be used beyond the locking in effect, the |
| 390 | * caller is responsible for guaranteeing that the pool stays online. |
Tejun Heo | fa1b54e | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 391 | * |
| 392 | * The if/else clause exists only for the lockdep assertion and can be |
| 393 | * ignored. |
Tejun Heo | 1711696 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 394 | */ |
Tejun Heo | 611c92a | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 395 | #define for_each_pool(pool, pi) \ |
| 396 | idr_for_each_entry(&worker_pool_idr, pool, pi) \ |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 397 | if (({ assert_rcu_or_pool_mutex(); false; })) { } \ |
Tejun Heo | fa1b54e | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 398 | else |
Tejun Heo | 1711696 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 399 | |
| 400 | /** |
Tejun Heo | 822d840 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 401 | * for_each_pool_worker - iterate through all workers of a worker_pool |
| 402 | * @worker: iteration cursor |
Tejun Heo | 822d840 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 403 | * @pool: worker_pool to iterate workers of |
| 404 | * |
Lai Jiangshan | 92f9c5c | 2014-05-20 17:46:34 +0800 | [diff] [blame] | 405 | * This must be called with @pool->attach_mutex. |
Tejun Heo | 822d840 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 406 | * |
| 407 | * The if/else clause exists only for the lockdep assertion and can be |
| 408 | * ignored. |
| 409 | */ |
Lai Jiangshan | da02846 | 2014-05-20 17:46:31 +0800 | [diff] [blame] | 410 | #define for_each_pool_worker(worker, pool) \ |
| 411 | list_for_each_entry((worker), &(pool)->workers, node) \ |
Lai Jiangshan | 92f9c5c | 2014-05-20 17:46:34 +0800 | [diff] [blame] | 412 | if (({ lockdep_assert_held(&pool->attach_mutex); false; })) { } \ |
Tejun Heo | 822d840 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 413 | else |
| 414 | |
| 415 | /** |
Tejun Heo | 49e3cf4 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 416 | * for_each_pwq - iterate through all pool_workqueues of the specified workqueue |
| 417 | * @pwq: iteration cursor |
| 418 | * @wq: the target workqueue |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 419 | * |
Lai Jiangshan | b09f4fd | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 420 | * This must be called either with wq->mutex held or sched RCU read locked. |
Tejun Heo | 794b18b | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 421 | * If the pwq needs to be used beyond the locking in effect, the caller is |
| 422 | * responsible for guaranteeing that the pwq stays online. |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 423 | * |
| 424 | * The if/else clause exists only for the lockdep assertion and can be |
| 425 | * ignored. |
Tejun Heo | 49e3cf4 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 426 | */ |
| 427 | #define for_each_pwq(pwq, wq) \ |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 428 | list_for_each_entry_rcu((pwq), &(wq)->pwqs, pwqs_node) \ |
Lai Jiangshan | b09f4fd | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 429 | if (({ assert_rcu_or_wq_mutex(wq); false; })) { } \ |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 430 | else |
Tejun Heo | f342179 | 2010-07-02 10:03:51 +0200 | [diff] [blame] | 431 | |
Thomas Gleixner | dc186ad | 2009-11-16 01:09:48 +0900 | [diff] [blame] | 432 | #ifdef CONFIG_DEBUG_OBJECTS_WORK |
| 433 | |
| 434 | static struct debug_obj_descr work_debug_descr; |
| 435 | |
Stanislaw Gruszka | 9977728 | 2011-03-07 09:58:33 +0100 | [diff] [blame] | 436 | static void *work_debug_hint(void *addr) |
| 437 | { |
| 438 | return ((struct work_struct *) addr)->func; |
| 439 | } |
| 440 | |
Du, Changbin | b9fdac7 | 2016-05-19 17:09:41 -0700 | [diff] [blame] | 441 | static bool work_is_static_object(void *addr) |
| 442 | { |
| 443 | struct work_struct *work = addr; |
| 444 | |
| 445 | return test_bit(WORK_STRUCT_STATIC_BIT, work_data_bits(work)); |
| 446 | } |
| 447 | |
Thomas Gleixner | dc186ad | 2009-11-16 01:09:48 +0900 | [diff] [blame] | 448 | /* |
| 449 | * fixup_init is called when: |
| 450 | * - an active object is initialized |
| 451 | */ |
Du, Changbin | 02a982a | 2016-05-19 17:09:26 -0700 | [diff] [blame] | 452 | static bool work_fixup_init(void *addr, enum debug_obj_state state) |
Thomas Gleixner | dc186ad | 2009-11-16 01:09:48 +0900 | [diff] [blame] | 453 | { |
| 454 | struct work_struct *work = addr; |
| 455 | |
| 456 | switch (state) { |
| 457 | case ODEBUG_STATE_ACTIVE: |
| 458 | cancel_work_sync(work); |
| 459 | debug_object_init(work, &work_debug_descr); |
Du, Changbin | 02a982a | 2016-05-19 17:09:26 -0700 | [diff] [blame] | 460 | return true; |
Thomas Gleixner | dc186ad | 2009-11-16 01:09:48 +0900 | [diff] [blame] | 461 | default: |
Du, Changbin | 02a982a | 2016-05-19 17:09:26 -0700 | [diff] [blame] | 462 | return false; |
Thomas Gleixner | dc186ad | 2009-11-16 01:09:48 +0900 | [diff] [blame] | 463 | } |
| 464 | } |
| 465 | |
| 466 | /* |
Thomas Gleixner | dc186ad | 2009-11-16 01:09:48 +0900 | [diff] [blame] | 467 | * fixup_free is called when: |
| 468 | * - an active object is freed |
| 469 | */ |
Du, Changbin | 02a982a | 2016-05-19 17:09:26 -0700 | [diff] [blame] | 470 | static bool work_fixup_free(void *addr, enum debug_obj_state state) |
Thomas Gleixner | dc186ad | 2009-11-16 01:09:48 +0900 | [diff] [blame] | 471 | { |
| 472 | struct work_struct *work = addr; |
| 473 | |
| 474 | switch (state) { |
| 475 | case ODEBUG_STATE_ACTIVE: |
| 476 | cancel_work_sync(work); |
| 477 | debug_object_free(work, &work_debug_descr); |
Du, Changbin | 02a982a | 2016-05-19 17:09:26 -0700 | [diff] [blame] | 478 | return true; |
Thomas Gleixner | dc186ad | 2009-11-16 01:09:48 +0900 | [diff] [blame] | 479 | default: |
Du, Changbin | 02a982a | 2016-05-19 17:09:26 -0700 | [diff] [blame] | 480 | return false; |
Thomas Gleixner | dc186ad | 2009-11-16 01:09:48 +0900 | [diff] [blame] | 481 | } |
| 482 | } |
| 483 | |
| 484 | static struct debug_obj_descr work_debug_descr = { |
| 485 | .name = "work_struct", |
Stanislaw Gruszka | 9977728 | 2011-03-07 09:58:33 +0100 | [diff] [blame] | 486 | .debug_hint = work_debug_hint, |
Du, Changbin | b9fdac7 | 2016-05-19 17:09:41 -0700 | [diff] [blame] | 487 | .is_static_object = work_is_static_object, |
Thomas Gleixner | dc186ad | 2009-11-16 01:09:48 +0900 | [diff] [blame] | 488 | .fixup_init = work_fixup_init, |
Thomas Gleixner | dc186ad | 2009-11-16 01:09:48 +0900 | [diff] [blame] | 489 | .fixup_free = work_fixup_free, |
| 490 | }; |
| 491 | |
| 492 | static inline void debug_work_activate(struct work_struct *work) |
| 493 | { |
| 494 | debug_object_activate(work, &work_debug_descr); |
| 495 | } |
| 496 | |
| 497 | static inline void debug_work_deactivate(struct work_struct *work) |
| 498 | { |
| 499 | debug_object_deactivate(work, &work_debug_descr); |
| 500 | } |
| 501 | |
| 502 | void __init_work(struct work_struct *work, int onstack) |
| 503 | { |
| 504 | if (onstack) |
| 505 | debug_object_init_on_stack(work, &work_debug_descr); |
| 506 | else |
| 507 | debug_object_init(work, &work_debug_descr); |
| 508 | } |
| 509 | EXPORT_SYMBOL_GPL(__init_work); |
| 510 | |
| 511 | void destroy_work_on_stack(struct work_struct *work) |
| 512 | { |
| 513 | debug_object_free(work, &work_debug_descr); |
| 514 | } |
| 515 | EXPORT_SYMBOL_GPL(destroy_work_on_stack); |
| 516 | |
Thomas Gleixner | ea2e64f | 2014-03-23 14:20:44 +0000 | [diff] [blame] | 517 | void destroy_delayed_work_on_stack(struct delayed_work *work) |
| 518 | { |
| 519 | destroy_timer_on_stack(&work->timer); |
| 520 | debug_object_free(&work->work, &work_debug_descr); |
| 521 | } |
| 522 | EXPORT_SYMBOL_GPL(destroy_delayed_work_on_stack); |
| 523 | |
Thomas Gleixner | dc186ad | 2009-11-16 01:09:48 +0900 | [diff] [blame] | 524 | #else |
| 525 | static inline void debug_work_activate(struct work_struct *work) { } |
| 526 | static inline void debug_work_deactivate(struct work_struct *work) { } |
| 527 | #endif |
| 528 | |
Li Bin | 4e8b22b | 2013-09-10 09:52:35 +0800 | [diff] [blame] | 529 | /** |
| 530 | * worker_pool_assign_id - allocate ID and assing it to @pool |
| 531 | * @pool: the pool pointer of interest |
| 532 | * |
| 533 | * Returns 0 if ID in [0, WORK_OFFQ_POOL_NONE) is allocated and assigned |
| 534 | * successfully, -errno on failure. |
| 535 | */ |
Tejun Heo | 9daf9e6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 536 | static int worker_pool_assign_id(struct worker_pool *pool) |
| 537 | { |
| 538 | int ret; |
| 539 | |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 540 | lockdep_assert_held(&wq_pool_mutex); |
Tejun Heo | 5bcab33 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 541 | |
Li Bin | 4e8b22b | 2013-09-10 09:52:35 +0800 | [diff] [blame] | 542 | ret = idr_alloc(&worker_pool_idr, pool, 0, WORK_OFFQ_POOL_NONE, |
| 543 | GFP_KERNEL); |
Tejun Heo | 229641a | 2013-04-01 17:08:13 -0700 | [diff] [blame] | 544 | if (ret >= 0) { |
Tejun Heo | e68035f | 2013-03-13 14:59:38 -0700 | [diff] [blame] | 545 | pool->id = ret; |
Tejun Heo | 229641a | 2013-04-01 17:08:13 -0700 | [diff] [blame] | 546 | return 0; |
| 547 | } |
Tejun Heo | 9daf9e6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 548 | return ret; |
| 549 | } |
| 550 | |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 551 | /** |
Tejun Heo | df2d5ae | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 552 | * unbound_pwq_by_node - return the unbound pool_workqueue for the given node |
| 553 | * @wq: the target workqueue |
| 554 | * @node: the node ID |
| 555 | * |
Lai Jiangshan | 5b95e1a | 2015-05-12 20:32:29 +0800 | [diff] [blame] | 556 | * This must be called with any of wq_pool_mutex, wq->mutex or sched RCU |
| 557 | * read locked. |
Tejun Heo | df2d5ae | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 558 | * If the pwq needs to be used beyond the locking in effect, the caller is |
| 559 | * responsible for guaranteeing that the pwq stays online. |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 560 | * |
| 561 | * Return: The unbound pool_workqueue for @node. |
Tejun Heo | df2d5ae | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 562 | */ |
| 563 | static struct pool_workqueue *unbound_pwq_by_node(struct workqueue_struct *wq, |
| 564 | int node) |
| 565 | { |
Lai Jiangshan | 5b95e1a | 2015-05-12 20:32:29 +0800 | [diff] [blame] | 566 | assert_rcu_or_wq_mutex_or_pool_mutex(wq); |
Tejun Heo | d6e022f | 2016-02-03 13:54:25 -0500 | [diff] [blame] | 567 | |
| 568 | /* |
| 569 | * XXX: @node can be NUMA_NO_NODE if CPU goes offline while a |
| 570 | * delayed item is pending. The plan is to keep CPU -> NODE |
| 571 | * mapping valid and stable across CPU on/offlines. Once that |
| 572 | * happens, this workaround can be removed. |
| 573 | */ |
| 574 | if (unlikely(node == NUMA_NO_NODE)) |
| 575 | return wq->dfl_pwq; |
| 576 | |
Tejun Heo | df2d5ae | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 577 | return rcu_dereference_raw(wq->numa_pwq_tbl[node]); |
| 578 | } |
| 579 | |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 580 | static unsigned int work_color_to_flags(int color) |
| 581 | { |
| 582 | return color << WORK_STRUCT_COLOR_SHIFT; |
| 583 | } |
| 584 | |
| 585 | static int get_work_color(struct work_struct *work) |
| 586 | { |
| 587 | return (*work_data_bits(work) >> WORK_STRUCT_COLOR_SHIFT) & |
| 588 | ((1 << WORK_STRUCT_COLOR_BITS) - 1); |
| 589 | } |
| 590 | |
| 591 | static int work_next_color(int color) |
| 592 | { |
| 593 | return (color + 1) % WORK_NR_COLORS; |
Oleg Nesterov | a848e3b | 2007-05-09 02:34:17 -0700 | [diff] [blame] | 594 | } |
| 595 | |
David Howells | 4594bf1 | 2006-12-07 11:33:26 +0000 | [diff] [blame] | 596 | /* |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 597 | * While queued, %WORK_STRUCT_PWQ is set and non flag bits of a work's data |
| 598 | * contain the pointer to the queued pwq. Once execution starts, the flag |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 599 | * is cleared and the high bits contain OFFQ flags and pool ID. |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 600 | * |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 601 | * set_work_pwq(), set_work_pool_and_clear_pending(), mark_work_canceling() |
| 602 | * and clear_work_data() can be used to set the pwq, pool or clear |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 603 | * work->data. These functions should only be called while the work is |
| 604 | * owned - ie. while the PENDING bit is set. |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 605 | * |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 606 | * get_work_pool() and get_work_pwq() can be used to obtain the pool or pwq |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 607 | * corresponding to a work. Pool is available once the work has been |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 608 | * queued anywhere after initialization until it is sync canceled. pwq is |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 609 | * available only while the work item is queued. |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 610 | * |
| 611 | * %WORK_OFFQ_CANCELING is used to mark a work item which is being |
| 612 | * canceled. While being canceled, a work item may have its PENDING set |
| 613 | * but stay off timer and worklist for arbitrarily long and nobody should |
| 614 | * try to steal the PENDING bit. |
David Howells | 4594bf1 | 2006-12-07 11:33:26 +0000 | [diff] [blame] | 615 | */ |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 616 | static inline void set_work_data(struct work_struct *work, unsigned long data, |
| 617 | unsigned long flags) |
David Howells | 365970a | 2006-11-22 14:54:49 +0000 | [diff] [blame] | 618 | { |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 619 | WARN_ON_ONCE(!work_pending(work)); |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 620 | atomic_long_set(&work->data, data | flags | work_static(work)); |
David Howells | 365970a | 2006-11-22 14:54:49 +0000 | [diff] [blame] | 621 | } |
David Howells | 365970a | 2006-11-22 14:54:49 +0000 | [diff] [blame] | 622 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 623 | static void set_work_pwq(struct work_struct *work, struct pool_workqueue *pwq, |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 624 | unsigned long extra_flags) |
Oleg Nesterov | 4d707b9 | 2010-04-23 17:40:40 +0200 | [diff] [blame] | 625 | { |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 626 | set_work_data(work, (unsigned long)pwq, |
| 627 | WORK_STRUCT_PENDING | WORK_STRUCT_PWQ | extra_flags); |
Oleg Nesterov | 4d707b9 | 2010-04-23 17:40:40 +0200 | [diff] [blame] | 628 | } |
| 629 | |
Lai Jiangshan | 4468a00 | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 630 | static void set_work_pool_and_keep_pending(struct work_struct *work, |
| 631 | int pool_id) |
| 632 | { |
| 633 | set_work_data(work, (unsigned long)pool_id << WORK_OFFQ_POOL_SHIFT, |
| 634 | WORK_STRUCT_PENDING); |
| 635 | } |
| 636 | |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 637 | static void set_work_pool_and_clear_pending(struct work_struct *work, |
| 638 | int pool_id) |
David Howells | 365970a | 2006-11-22 14:54:49 +0000 | [diff] [blame] | 639 | { |
Tejun Heo | 23657bb | 2012-08-13 17:08:19 -0700 | [diff] [blame] | 640 | /* |
| 641 | * The following wmb is paired with the implied mb in |
| 642 | * test_and_set_bit(PENDING) and ensures all updates to @work made |
| 643 | * here are visible to and precede any updates by the next PENDING |
| 644 | * owner. |
| 645 | */ |
| 646 | smp_wmb(); |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 647 | set_work_data(work, (unsigned long)pool_id << WORK_OFFQ_POOL_SHIFT, 0); |
Roman Pen | 346c09f | 2016-04-26 13:15:35 +0200 | [diff] [blame] | 648 | /* |
| 649 | * The following mb guarantees that previous clear of a PENDING bit |
| 650 | * will not be reordered with any speculative LOADS or STORES from |
| 651 | * work->current_func, which is executed afterwards. This possible |
| 652 | * reordering can lead to a missed execution on attempt to qeueue |
| 653 | * the same @work. E.g. consider this case: |
| 654 | * |
| 655 | * CPU#0 CPU#1 |
| 656 | * ---------------------------- -------------------------------- |
| 657 | * |
| 658 | * 1 STORE event_indicated |
| 659 | * 2 queue_work_on() { |
| 660 | * 3 test_and_set_bit(PENDING) |
| 661 | * 4 } set_..._and_clear_pending() { |
| 662 | * 5 set_work_data() # clear bit |
| 663 | * 6 smp_mb() |
| 664 | * 7 work->current_func() { |
| 665 | * 8 LOAD event_indicated |
| 666 | * } |
| 667 | * |
| 668 | * Without an explicit full barrier speculative LOAD on line 8 can |
| 669 | * be executed before CPU#0 does STORE on line 1. If that happens, |
| 670 | * CPU#0 observes the PENDING bit is still set and new execution of |
| 671 | * a @work is not queued in a hope, that CPU#1 will eventually |
| 672 | * finish the queued @work. Meanwhile CPU#1 does not see |
| 673 | * event_indicated is set, because speculative LOAD was executed |
| 674 | * before actual STORE. |
| 675 | */ |
| 676 | smp_mb(); |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 677 | } |
| 678 | |
| 679 | static void clear_work_data(struct work_struct *work) |
| 680 | { |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 681 | smp_wmb(); /* see set_work_pool_and_clear_pending() */ |
| 682 | set_work_data(work, WORK_STRUCT_NO_POOL, 0); |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 683 | } |
| 684 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 685 | static struct pool_workqueue *get_work_pwq(struct work_struct *work) |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 686 | { |
Tejun Heo | e120153 | 2010-07-22 14:14:25 +0200 | [diff] [blame] | 687 | unsigned long data = atomic_long_read(&work->data); |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 688 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 689 | if (data & WORK_STRUCT_PWQ) |
Tejun Heo | e120153 | 2010-07-22 14:14:25 +0200 | [diff] [blame] | 690 | return (void *)(data & WORK_STRUCT_WQ_DATA_MASK); |
| 691 | else |
| 692 | return NULL; |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 693 | } |
| 694 | |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 695 | /** |
| 696 | * get_work_pool - return the worker_pool a given work was associated with |
| 697 | * @work: the work item of interest |
| 698 | * |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 699 | * Pools are created and destroyed under wq_pool_mutex, and allows read |
| 700 | * access under sched-RCU read lock. As such, this function should be |
| 701 | * called under wq_pool_mutex or with preemption disabled. |
Tejun Heo | fa1b54e | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 702 | * |
| 703 | * All fields of the returned pool are accessible as long as the above |
| 704 | * mentioned locking is in effect. If the returned pool needs to be used |
| 705 | * beyond the critical section, the caller is responsible for ensuring the |
| 706 | * returned pool is and stays online. |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 707 | * |
| 708 | * Return: The worker_pool @work was last associated with. %NULL if none. |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 709 | */ |
| 710 | static struct worker_pool *get_work_pool(struct work_struct *work) |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 711 | { |
Tejun Heo | e120153 | 2010-07-22 14:14:25 +0200 | [diff] [blame] | 712 | unsigned long data = atomic_long_read(&work->data); |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 713 | int pool_id; |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 714 | |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 715 | assert_rcu_or_pool_mutex(); |
Tejun Heo | fa1b54e | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 716 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 717 | if (data & WORK_STRUCT_PWQ) |
| 718 | return ((struct pool_workqueue *) |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 719 | (data & WORK_STRUCT_WQ_DATA_MASK))->pool; |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 720 | |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 721 | pool_id = data >> WORK_OFFQ_POOL_SHIFT; |
| 722 | if (pool_id == WORK_OFFQ_POOL_NONE) |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 723 | return NULL; |
| 724 | |
Tejun Heo | fa1b54e | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 725 | return idr_find(&worker_pool_idr, pool_id); |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 726 | } |
| 727 | |
| 728 | /** |
| 729 | * get_work_pool_id - return the worker pool ID a given work is associated with |
| 730 | * @work: the work item of interest |
| 731 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 732 | * Return: The worker_pool ID @work was last associated with. |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 733 | * %WORK_OFFQ_POOL_NONE if none. |
| 734 | */ |
| 735 | static int get_work_pool_id(struct work_struct *work) |
| 736 | { |
Lai Jiangshan | 54d5b7d | 2013-02-07 13:14:20 -0800 | [diff] [blame] | 737 | unsigned long data = atomic_long_read(&work->data); |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 738 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 739 | if (data & WORK_STRUCT_PWQ) |
| 740 | return ((struct pool_workqueue *) |
Lai Jiangshan | 54d5b7d | 2013-02-07 13:14:20 -0800 | [diff] [blame] | 741 | (data & WORK_STRUCT_WQ_DATA_MASK))->pool->id; |
| 742 | |
| 743 | return data >> WORK_OFFQ_POOL_SHIFT; |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 744 | } |
| 745 | |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 746 | static void mark_work_canceling(struct work_struct *work) |
| 747 | { |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 748 | unsigned long pool_id = get_work_pool_id(work); |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 749 | |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 750 | pool_id <<= WORK_OFFQ_POOL_SHIFT; |
| 751 | set_work_data(work, pool_id | WORK_OFFQ_CANCELING, WORK_STRUCT_PENDING); |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 752 | } |
| 753 | |
| 754 | static bool work_is_canceling(struct work_struct *work) |
| 755 | { |
| 756 | unsigned long data = atomic_long_read(&work->data); |
| 757 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 758 | return !(data & WORK_STRUCT_PWQ) && (data & WORK_OFFQ_CANCELING); |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 759 | } |
| 760 | |
David Howells | 365970a | 2006-11-22 14:54:49 +0000 | [diff] [blame] | 761 | /* |
Tejun Heo | 3270476 | 2012-07-13 22:16:45 -0700 | [diff] [blame] | 762 | * Policy functions. These define the policies on how the global worker |
| 763 | * pools are managed. Unless noted otherwise, these functions assume that |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 764 | * they're being called with pool->lock held. |
David Howells | 365970a | 2006-11-22 14:54:49 +0000 | [diff] [blame] | 765 | */ |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 766 | |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 767 | static bool __need_more_worker(struct worker_pool *pool) |
David Howells | 365970a | 2006-11-22 14:54:49 +0000 | [diff] [blame] | 768 | { |
Tejun Heo | e19e397 | 2013-01-24 11:39:44 -0800 | [diff] [blame] | 769 | return !atomic_read(&pool->nr_running); |
David Howells | 365970a | 2006-11-22 14:54:49 +0000 | [diff] [blame] | 770 | } |
| 771 | |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 772 | /* |
| 773 | * Need to wake up a worker? Called from anything but currently |
| 774 | * running workers. |
Tejun Heo | 974271c | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 775 | * |
| 776 | * Note that, because unbound workers never contribute to nr_running, this |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 777 | * function will always return %true for unbound pools as long as the |
Tejun Heo | 974271c | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 778 | * worklist isn't empty. |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 779 | */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 780 | static bool need_more_worker(struct worker_pool *pool) |
David Howells | 365970a | 2006-11-22 14:54:49 +0000 | [diff] [blame] | 781 | { |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 782 | return !list_empty(&pool->worklist) && __need_more_worker(pool); |
David Howells | 365970a | 2006-11-22 14:54:49 +0000 | [diff] [blame] | 783 | } |
| 784 | |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 785 | /* Can I start working? Called from busy but !running workers. */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 786 | static bool may_start_working(struct worker_pool *pool) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 787 | { |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 788 | return pool->nr_idle; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 789 | } |
| 790 | |
| 791 | /* Do I need to keep working? Called from currently running workers. */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 792 | static bool keep_working(struct worker_pool *pool) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 793 | { |
Tejun Heo | e19e397 | 2013-01-24 11:39:44 -0800 | [diff] [blame] | 794 | return !list_empty(&pool->worklist) && |
| 795 | atomic_read(&pool->nr_running) <= 1; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 796 | } |
| 797 | |
| 798 | /* Do we need a new worker? Called from manager. */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 799 | static bool need_to_create_worker(struct worker_pool *pool) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 800 | { |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 801 | return need_more_worker(pool) && !may_start_working(pool); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 802 | } |
| 803 | |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 804 | /* Do we have too many workers and should some go away? */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 805 | static bool too_many_workers(struct worker_pool *pool) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 806 | { |
Tejun Heo | 692b482 | 2017-10-09 08:04:13 -0700 | [diff] [blame] | 807 | bool managing = pool->flags & POOL_MANAGER_ACTIVE; |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 808 | int nr_idle = pool->nr_idle + managing; /* manager is considered idle */ |
| 809 | int nr_busy = pool->nr_workers - nr_idle; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 810 | |
| 811 | return nr_idle > 2 && (nr_idle - 2) * MAX_IDLE_WORKERS_RATIO >= nr_busy; |
| 812 | } |
| 813 | |
| 814 | /* |
| 815 | * Wake up functions. |
| 816 | */ |
| 817 | |
Lai Jiangshan | 1037de3 | 2014-05-22 16:44:07 +0800 | [diff] [blame] | 818 | /* Return the first idle worker. Safe with preemption disabled */ |
| 819 | static struct worker *first_idle_worker(struct worker_pool *pool) |
Tejun Heo | 7e11629 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 820 | { |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 821 | if (unlikely(list_empty(&pool->idle_list))) |
Tejun Heo | 7e11629 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 822 | return NULL; |
| 823 | |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 824 | return list_first_entry(&pool->idle_list, struct worker, entry); |
Tejun Heo | 7e11629 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 825 | } |
| 826 | |
| 827 | /** |
| 828 | * wake_up_worker - wake up an idle worker |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 829 | * @pool: worker pool to wake worker from |
Tejun Heo | 7e11629 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 830 | * |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 831 | * Wake up the first idle worker of @pool. |
Tejun Heo | 7e11629 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 832 | * |
| 833 | * CONTEXT: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 834 | * spin_lock_irq(pool->lock). |
Tejun Heo | 7e11629 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 835 | */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 836 | static void wake_up_worker(struct worker_pool *pool) |
Tejun Heo | 7e11629 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 837 | { |
Lai Jiangshan | 1037de3 | 2014-05-22 16:44:07 +0800 | [diff] [blame] | 838 | struct worker *worker = first_idle_worker(pool); |
Tejun Heo | 7e11629 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 839 | |
| 840 | if (likely(worker)) |
| 841 | wake_up_process(worker->task); |
| 842 | } |
| 843 | |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 844 | /** |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 845 | * wq_worker_waking_up - a worker is waking up |
| 846 | * @task: task waking up |
| 847 | * @cpu: CPU @task is waking up to |
| 848 | * |
| 849 | * This function is called during try_to_wake_up() when a worker is |
| 850 | * being awoken. |
| 851 | * |
| 852 | * CONTEXT: |
| 853 | * spin_lock_irq(rq->lock) |
| 854 | */ |
Tejun Heo | d84ff05 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 855 | void wq_worker_waking_up(struct task_struct *task, int cpu) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 856 | { |
| 857 | struct worker *worker = kthread_data(task); |
| 858 | |
Joonsoo Kim | 3657600 | 2012-10-26 23:03:49 +0900 | [diff] [blame] | 859 | if (!(worker->flags & WORKER_NOT_RUNNING)) { |
Tejun Heo | ec22ca5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 860 | WARN_ON_ONCE(worker->pool->cpu != cpu); |
Tejun Heo | e19e397 | 2013-01-24 11:39:44 -0800 | [diff] [blame] | 861 | atomic_inc(&worker->pool->nr_running); |
Joonsoo Kim | 3657600 | 2012-10-26 23:03:49 +0900 | [diff] [blame] | 862 | } |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 863 | } |
| 864 | |
| 865 | /** |
| 866 | * wq_worker_sleeping - a worker is going to sleep |
| 867 | * @task: task going to sleep |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 868 | * |
| 869 | * This function is called during schedule() when a busy worker is |
| 870 | * going to sleep. Worker on the same cpu can be woken up by |
| 871 | * returning pointer to its task. |
| 872 | * |
| 873 | * CONTEXT: |
| 874 | * spin_lock_irq(rq->lock) |
| 875 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 876 | * Return: |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 877 | * Worker task on @cpu to wake up, %NULL if none. |
| 878 | */ |
Alexander Gordeev | 9b7f659 | 2016-03-02 12:53:31 +0100 | [diff] [blame] | 879 | struct task_struct *wq_worker_sleeping(struct task_struct *task) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 880 | { |
| 881 | struct worker *worker = kthread_data(task), *to_wakeup = NULL; |
Tejun Heo | 111c225 | 2013-01-17 17:16:24 -0800 | [diff] [blame] | 882 | struct worker_pool *pool; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 883 | |
Tejun Heo | 111c225 | 2013-01-17 17:16:24 -0800 | [diff] [blame] | 884 | /* |
| 885 | * Rescuers, which may not have all the fields set up like normal |
| 886 | * workers, also reach here, let's not access anything before |
| 887 | * checking NOT_RUNNING. |
| 888 | */ |
Steven Rostedt | 2d64672 | 2010-12-03 23:12:33 -0500 | [diff] [blame] | 889 | if (worker->flags & WORKER_NOT_RUNNING) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 890 | return NULL; |
| 891 | |
Tejun Heo | 111c225 | 2013-01-17 17:16:24 -0800 | [diff] [blame] | 892 | pool = worker->pool; |
Tejun Heo | 111c225 | 2013-01-17 17:16:24 -0800 | [diff] [blame] | 893 | |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 894 | /* this can only happen on the local cpu */ |
Alexander Gordeev | 9b7f659 | 2016-03-02 12:53:31 +0100 | [diff] [blame] | 895 | if (WARN_ON_ONCE(pool->cpu != raw_smp_processor_id())) |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 896 | return NULL; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 897 | |
| 898 | /* |
| 899 | * The counterpart of the following dec_and_test, implied mb, |
| 900 | * worklist not empty test sequence is in insert_work(). |
| 901 | * Please read comment there. |
| 902 | * |
Tejun Heo | 628c78e | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 903 | * NOT_RUNNING is clear. This means that we're bound to and |
| 904 | * running on the local cpu w/ rq lock held and preemption |
| 905 | * disabled, which in turn means that none else could be |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 906 | * manipulating idle_list, so dereferencing idle_list without pool |
Tejun Heo | 628c78e | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 907 | * lock is safe. |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 908 | */ |
Tejun Heo | e19e397 | 2013-01-24 11:39:44 -0800 | [diff] [blame] | 909 | if (atomic_dec_and_test(&pool->nr_running) && |
| 910 | !list_empty(&pool->worklist)) |
Lai Jiangshan | 1037de3 | 2014-05-22 16:44:07 +0800 | [diff] [blame] | 911 | to_wakeup = first_idle_worker(pool); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 912 | return to_wakeup ? to_wakeup->task : NULL; |
| 913 | } |
| 914 | |
| 915 | /** |
| 916 | * worker_set_flags - set worker flags and adjust nr_running accordingly |
Tejun Heo | cb44476 | 2010-07-02 10:03:50 +0200 | [diff] [blame] | 917 | * @worker: self |
Tejun Heo | d302f01 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 918 | * @flags: flags to set |
Tejun Heo | d302f01 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 919 | * |
Lai Jiangshan | 228f1d0 | 2014-07-22 13:02:00 +0800 | [diff] [blame] | 920 | * Set @flags in @worker->flags and adjust nr_running accordingly. |
Tejun Heo | d302f01 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 921 | * |
Tejun Heo | cb44476 | 2010-07-02 10:03:50 +0200 | [diff] [blame] | 922 | * CONTEXT: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 923 | * spin_lock_irq(pool->lock) |
Tejun Heo | d302f01 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 924 | */ |
Lai Jiangshan | 228f1d0 | 2014-07-22 13:02:00 +0800 | [diff] [blame] | 925 | static inline void worker_set_flags(struct worker *worker, unsigned int flags) |
Tejun Heo | d302f01 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 926 | { |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 927 | struct worker_pool *pool = worker->pool; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 928 | |
Tejun Heo | cb44476 | 2010-07-02 10:03:50 +0200 | [diff] [blame] | 929 | WARN_ON_ONCE(worker->task != current); |
| 930 | |
Lai Jiangshan | 228f1d0 | 2014-07-22 13:02:00 +0800 | [diff] [blame] | 931 | /* If transitioning into NOT_RUNNING, adjust nr_running. */ |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 932 | if ((flags & WORKER_NOT_RUNNING) && |
| 933 | !(worker->flags & WORKER_NOT_RUNNING)) { |
Lai Jiangshan | 228f1d0 | 2014-07-22 13:02:00 +0800 | [diff] [blame] | 934 | atomic_dec(&pool->nr_running); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 935 | } |
| 936 | |
Tejun Heo | d302f01 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 937 | worker->flags |= flags; |
| 938 | } |
| 939 | |
| 940 | /** |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 941 | * worker_clr_flags - clear worker flags and adjust nr_running accordingly |
Tejun Heo | cb44476 | 2010-07-02 10:03:50 +0200 | [diff] [blame] | 942 | * @worker: self |
Tejun Heo | d302f01 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 943 | * @flags: flags to clear |
| 944 | * |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 945 | * Clear @flags in @worker->flags and adjust nr_running accordingly. |
Tejun Heo | d302f01 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 946 | * |
Tejun Heo | cb44476 | 2010-07-02 10:03:50 +0200 | [diff] [blame] | 947 | * CONTEXT: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 948 | * spin_lock_irq(pool->lock) |
Tejun Heo | d302f01 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 949 | */ |
| 950 | static inline void worker_clr_flags(struct worker *worker, unsigned int flags) |
| 951 | { |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 952 | struct worker_pool *pool = worker->pool; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 953 | unsigned int oflags = worker->flags; |
| 954 | |
Tejun Heo | cb44476 | 2010-07-02 10:03:50 +0200 | [diff] [blame] | 955 | WARN_ON_ONCE(worker->task != current); |
| 956 | |
Tejun Heo | d302f01 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 957 | worker->flags &= ~flags; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 958 | |
Tejun Heo | 42c025f | 2011-01-11 15:58:49 +0100 | [diff] [blame] | 959 | /* |
| 960 | * If transitioning out of NOT_RUNNING, increment nr_running. Note |
| 961 | * that the nested NOT_RUNNING is not a noop. NOT_RUNNING is mask |
| 962 | * of multiple flags, not a single flag. |
| 963 | */ |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 964 | if ((flags & WORKER_NOT_RUNNING) && (oflags & WORKER_NOT_RUNNING)) |
| 965 | if (!(worker->flags & WORKER_NOT_RUNNING)) |
Tejun Heo | e19e397 | 2013-01-24 11:39:44 -0800 | [diff] [blame] | 966 | atomic_inc(&pool->nr_running); |
Tejun Heo | d302f01 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 967 | } |
| 968 | |
| 969 | /** |
Tejun Heo | 8cca0ee | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 970 | * find_worker_executing_work - find worker which is executing a work |
Tejun Heo | c9e7cf2 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 971 | * @pool: pool of interest |
Tejun Heo | 8cca0ee | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 972 | * @work: work to find worker for |
| 973 | * |
Tejun Heo | c9e7cf2 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 974 | * Find a worker which is executing @work on @pool by searching |
| 975 | * @pool->busy_hash which is keyed by the address of @work. For a worker |
Tejun Heo | a2c1c57 | 2012-12-18 10:35:02 -0800 | [diff] [blame] | 976 | * to match, its current execution should match the address of @work and |
| 977 | * its work function. This is to avoid unwanted dependency between |
| 978 | * unrelated work executions through a work item being recycled while still |
| 979 | * being executed. |
| 980 | * |
| 981 | * This is a bit tricky. A work item may be freed once its execution |
| 982 | * starts and nothing prevents the freed area from being recycled for |
| 983 | * another work item. If the same work item address ends up being reused |
| 984 | * before the original execution finishes, workqueue will identify the |
| 985 | * recycled work item as currently executing and make it wait until the |
| 986 | * current execution finishes, introducing an unwanted dependency. |
| 987 | * |
Tejun Heo | c5aa87b | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 988 | * This function checks the work item address and work function to avoid |
| 989 | * false positives. Note that this isn't complete as one may construct a |
| 990 | * work function which can introduce dependency onto itself through a |
| 991 | * recycled work item. Well, if somebody wants to shoot oneself in the |
| 992 | * foot that badly, there's only so much we can do, and if such deadlock |
| 993 | * actually occurs, it should be easy to locate the culprit work function. |
Tejun Heo | 8cca0ee | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 994 | * |
| 995 | * CONTEXT: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 996 | * spin_lock_irq(pool->lock). |
Tejun Heo | 8cca0ee | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 997 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 998 | * Return: |
| 999 | * Pointer to worker which is executing @work if found, %NULL |
Tejun Heo | 8cca0ee | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 1000 | * otherwise. |
| 1001 | */ |
Tejun Heo | c9e7cf2 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1002 | static struct worker *find_worker_executing_work(struct worker_pool *pool, |
Tejun Heo | 8cca0ee | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 1003 | struct work_struct *work) |
| 1004 | { |
Sasha Levin | 42f8570 | 2012-12-17 10:01:23 -0500 | [diff] [blame] | 1005 | struct worker *worker; |
Sasha Levin | 42f8570 | 2012-12-17 10:01:23 -0500 | [diff] [blame] | 1006 | |
Sasha Levin | b67bfe0 | 2013-02-27 17:06:00 -0800 | [diff] [blame] | 1007 | hash_for_each_possible(pool->busy_hash, worker, hentry, |
Tejun Heo | a2c1c57 | 2012-12-18 10:35:02 -0800 | [diff] [blame] | 1008 | (unsigned long)work) |
| 1009 | if (worker->current_work == work && |
| 1010 | worker->current_func == work->func) |
Sasha Levin | 42f8570 | 2012-12-17 10:01:23 -0500 | [diff] [blame] | 1011 | return worker; |
| 1012 | |
| 1013 | return NULL; |
Tejun Heo | 8cca0ee | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 1014 | } |
| 1015 | |
| 1016 | /** |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1017 | * move_linked_works - move linked works to a list |
| 1018 | * @work: start of series of works to be scheduled |
| 1019 | * @head: target list to append @work to |
Shailendra Verma | 402dd89 | 2015-05-23 10:38:14 +0530 | [diff] [blame] | 1020 | * @nextp: out parameter for nested worklist walking |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1021 | * |
| 1022 | * Schedule linked works starting from @work to @head. Work series to |
| 1023 | * be scheduled starts at @work and includes any consecutive work with |
| 1024 | * WORK_STRUCT_LINKED set in its predecessor. |
| 1025 | * |
| 1026 | * If @nextp is not NULL, it's updated to point to the next work of |
| 1027 | * the last scheduled work. This allows move_linked_works() to be |
| 1028 | * nested inside outer list_for_each_entry_safe(). |
| 1029 | * |
| 1030 | * CONTEXT: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1031 | * spin_lock_irq(pool->lock). |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1032 | */ |
| 1033 | static void move_linked_works(struct work_struct *work, struct list_head *head, |
| 1034 | struct work_struct **nextp) |
| 1035 | { |
| 1036 | struct work_struct *n; |
| 1037 | |
| 1038 | /* |
| 1039 | * Linked worklist will always end before the end of the list, |
| 1040 | * use NULL for list head. |
| 1041 | */ |
| 1042 | list_for_each_entry_safe_from(work, n, NULL, entry) { |
| 1043 | list_move_tail(&work->entry, head); |
| 1044 | if (!(*work_data_bits(work) & WORK_STRUCT_LINKED)) |
| 1045 | break; |
| 1046 | } |
| 1047 | |
| 1048 | /* |
| 1049 | * If we're already inside safe list traversal and have moved |
| 1050 | * multiple works to the scheduled queue, the next position |
| 1051 | * needs to be updated. |
| 1052 | */ |
| 1053 | if (nextp) |
| 1054 | *nextp = n; |
| 1055 | } |
| 1056 | |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1057 | /** |
| 1058 | * get_pwq - get an extra reference on the specified pool_workqueue |
| 1059 | * @pwq: pool_workqueue to get |
| 1060 | * |
| 1061 | * Obtain an extra reference on @pwq. The caller should guarantee that |
| 1062 | * @pwq has positive refcnt and be holding the matching pool->lock. |
| 1063 | */ |
| 1064 | static void get_pwq(struct pool_workqueue *pwq) |
| 1065 | { |
| 1066 | lockdep_assert_held(&pwq->pool->lock); |
| 1067 | WARN_ON_ONCE(pwq->refcnt <= 0); |
| 1068 | pwq->refcnt++; |
| 1069 | } |
| 1070 | |
| 1071 | /** |
| 1072 | * put_pwq - put a pool_workqueue reference |
| 1073 | * @pwq: pool_workqueue to put |
| 1074 | * |
| 1075 | * Drop a reference of @pwq. If its refcnt reaches zero, schedule its |
| 1076 | * destruction. The caller should be holding the matching pool->lock. |
| 1077 | */ |
| 1078 | static void put_pwq(struct pool_workqueue *pwq) |
| 1079 | { |
| 1080 | lockdep_assert_held(&pwq->pool->lock); |
| 1081 | if (likely(--pwq->refcnt)) |
| 1082 | return; |
| 1083 | if (WARN_ON_ONCE(!(pwq->wq->flags & WQ_UNBOUND))) |
| 1084 | return; |
| 1085 | /* |
| 1086 | * @pwq can't be released under pool->lock, bounce to |
| 1087 | * pwq_unbound_release_workfn(). This never recurses on the same |
| 1088 | * pool->lock as this path is taken only for unbound workqueues and |
| 1089 | * the release work item is scheduled on a per-cpu workqueue. To |
| 1090 | * avoid lockdep warning, unbound pool->locks are given lockdep |
| 1091 | * subclass of 1 in get_unbound_pool(). |
| 1092 | */ |
| 1093 | schedule_work(&pwq->unbound_release_work); |
| 1094 | } |
| 1095 | |
Tejun Heo | dce90d4 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 1096 | /** |
| 1097 | * put_pwq_unlocked - put_pwq() with surrounding pool lock/unlock |
| 1098 | * @pwq: pool_workqueue to put (can be %NULL) |
| 1099 | * |
| 1100 | * put_pwq() with locking. This function also allows %NULL @pwq. |
| 1101 | */ |
| 1102 | static void put_pwq_unlocked(struct pool_workqueue *pwq) |
| 1103 | { |
| 1104 | if (pwq) { |
| 1105 | /* |
| 1106 | * As both pwqs and pools are sched-RCU protected, the |
| 1107 | * following lock operations are safe. |
| 1108 | */ |
| 1109 | spin_lock_irq(&pwq->pool->lock); |
| 1110 | put_pwq(pwq); |
| 1111 | spin_unlock_irq(&pwq->pool->lock); |
| 1112 | } |
| 1113 | } |
| 1114 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1115 | static void pwq_activate_delayed_work(struct work_struct *work) |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1116 | { |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1117 | struct pool_workqueue *pwq = get_work_pwq(work); |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1118 | |
| 1119 | trace_workqueue_activate_work(work); |
Tejun Heo | 82607adc | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 1120 | if (list_empty(&pwq->pool->worklist)) |
| 1121 | pwq->pool->watchdog_ts = jiffies; |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1122 | move_linked_works(work, &pwq->pool->worklist, NULL); |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1123 | __clear_bit(WORK_STRUCT_DELAYED_BIT, work_data_bits(work)); |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1124 | pwq->nr_active++; |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1125 | } |
| 1126 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1127 | static void pwq_activate_first_delayed(struct pool_workqueue *pwq) |
Lai Jiangshan | 3aa6249 | 2012-09-18 10:40:00 -0700 | [diff] [blame] | 1128 | { |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1129 | struct work_struct *work = list_first_entry(&pwq->delayed_works, |
Lai Jiangshan | 3aa6249 | 2012-09-18 10:40:00 -0700 | [diff] [blame] | 1130 | struct work_struct, entry); |
| 1131 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1132 | pwq_activate_delayed_work(work); |
Lai Jiangshan | 3aa6249 | 2012-09-18 10:40:00 -0700 | [diff] [blame] | 1133 | } |
| 1134 | |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1135 | /** |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1136 | * pwq_dec_nr_in_flight - decrement pwq's nr_in_flight |
| 1137 | * @pwq: pwq of interest |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1138 | * @color: color of work which left the queue |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1139 | * |
| 1140 | * A work either has completed or is removed from pending queue, |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1141 | * decrement nr_in_flight of its pwq and handle workqueue flushing. |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1142 | * |
| 1143 | * CONTEXT: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1144 | * spin_lock_irq(pool->lock). |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1145 | */ |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1146 | static void pwq_dec_nr_in_flight(struct pool_workqueue *pwq, int color) |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1147 | { |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1148 | /* uncolored work items don't participate in flushing or nr_active */ |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1149 | if (color == WORK_NO_COLOR) |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1150 | goto out_put; |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1151 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1152 | pwq->nr_in_flight[color]--; |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1153 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1154 | pwq->nr_active--; |
| 1155 | if (!list_empty(&pwq->delayed_works)) { |
Lai Jiangshan | b3f9f40 | 2012-09-18 10:40:00 -0700 | [diff] [blame] | 1156 | /* one down, submit a delayed one */ |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1157 | if (pwq->nr_active < pwq->max_active) |
| 1158 | pwq_activate_first_delayed(pwq); |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1159 | } |
| 1160 | |
| 1161 | /* is flush in progress and are we at the flushing tip? */ |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1162 | if (likely(pwq->flush_color != color)) |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1163 | goto out_put; |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1164 | |
| 1165 | /* are there still in-flight works? */ |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1166 | if (pwq->nr_in_flight[color]) |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1167 | goto out_put; |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1168 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1169 | /* this pwq is done, clear flush_color */ |
| 1170 | pwq->flush_color = -1; |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1171 | |
| 1172 | /* |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1173 | * If this was the last pwq, wake up the first flusher. It |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1174 | * will handle the rest. |
| 1175 | */ |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1176 | if (atomic_dec_and_test(&pwq->wq->nr_pwqs_to_flush)) |
| 1177 | complete(&pwq->wq->first_flusher->done); |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1178 | out_put: |
| 1179 | put_pwq(pwq); |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1180 | } |
| 1181 | |
Tejun Heo | 36e227d | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1182 | /** |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1183 | * try_to_grab_pending - steal work item from worklist and disable irq |
Tejun Heo | 36e227d | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1184 | * @work: work item to steal |
| 1185 | * @is_dwork: @work is a delayed_work |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1186 | * @flags: place to store irq state |
Tejun Heo | 36e227d | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1187 | * |
| 1188 | * Try to grab PENDING bit of @work. This function can handle @work in any |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 1189 | * stable state - idle, on timer or on worklist. |
Tejun Heo | 36e227d | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1190 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 1191 | * Return: |
Tejun Heo | 36e227d | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1192 | * 1 if @work was pending and we successfully stole PENDING |
| 1193 | * 0 if @work was idle and we claimed PENDING |
| 1194 | * -EAGAIN if PENDING couldn't be grabbed at the moment, safe to busy-retry |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1195 | * -ENOENT if someone else is canceling @work, this state may persist |
| 1196 | * for arbitrarily long |
Tejun Heo | 36e227d | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1197 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 1198 | * Note: |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1199 | * On >= 0 return, the caller owns @work's PENDING bit. To avoid getting |
Tejun Heo | e0aecdd | 2012-08-21 13:18:24 -0700 | [diff] [blame] | 1200 | * interrupted while holding PENDING and @work off queue, irq must be |
| 1201 | * disabled on entry. This, combined with delayed_work->timer being |
| 1202 | * irqsafe, ensures that we return -EAGAIN for finite short period of time. |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1203 | * |
| 1204 | * On successful return, >= 0, irq is disabled and the caller is |
| 1205 | * responsible for releasing it using local_irq_restore(*@flags). |
| 1206 | * |
Tejun Heo | e0aecdd | 2012-08-21 13:18:24 -0700 | [diff] [blame] | 1207 | * This function is safe to call from any context including IRQ handler. |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1208 | */ |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1209 | static int try_to_grab_pending(struct work_struct *work, bool is_dwork, |
| 1210 | unsigned long *flags) |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1211 | { |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1212 | struct worker_pool *pool; |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1213 | struct pool_workqueue *pwq; |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1214 | |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1215 | local_irq_save(*flags); |
| 1216 | |
Tejun Heo | 36e227d | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1217 | /* try to steal the timer if it exists */ |
| 1218 | if (is_dwork) { |
| 1219 | struct delayed_work *dwork = to_delayed_work(work); |
| 1220 | |
Tejun Heo | e0aecdd | 2012-08-21 13:18:24 -0700 | [diff] [blame] | 1221 | /* |
| 1222 | * dwork->timer is irqsafe. If del_timer() fails, it's |
| 1223 | * guaranteed that the timer is not queued anywhere and not |
| 1224 | * running on the local CPU. |
| 1225 | */ |
Tejun Heo | 36e227d | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1226 | if (likely(del_timer(&dwork->timer))) |
| 1227 | return 1; |
| 1228 | } |
| 1229 | |
| 1230 | /* try to claim PENDING the normal way */ |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1231 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) |
| 1232 | return 0; |
| 1233 | |
| 1234 | /* |
| 1235 | * The queueing is in progress, or it is already queued. Try to |
| 1236 | * steal it from ->worklist without clearing WORK_STRUCT_PENDING. |
| 1237 | */ |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1238 | pool = get_work_pool(work); |
| 1239 | if (!pool) |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1240 | goto fail; |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1241 | |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1242 | spin_lock(&pool->lock); |
Lai Jiangshan | 0b3dae6 | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 1243 | /* |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1244 | * work->data is guaranteed to point to pwq only while the work |
| 1245 | * item is queued on pwq->wq, and both updating work->data to point |
| 1246 | * to pwq on queueing and to pool on dequeueing are done under |
| 1247 | * pwq->pool->lock. This in turn guarantees that, if work->data |
| 1248 | * points to pwq which is associated with a locked pool, the work |
Lai Jiangshan | 0b3dae6 | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 1249 | * item is currently queued on that pool. |
| 1250 | */ |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1251 | pwq = get_work_pwq(work); |
| 1252 | if (pwq && pwq->pool == pool) { |
Tejun Heo | 1606283 | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 1253 | debug_work_deactivate(work); |
Lai Jiangshan | 3aa6249 | 2012-09-18 10:40:00 -0700 | [diff] [blame] | 1254 | |
Tejun Heo | 1606283 | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 1255 | /* |
| 1256 | * A delayed work item cannot be grabbed directly because |
| 1257 | * it might have linked NO_COLOR work items which, if left |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1258 | * on the delayed_list, will confuse pwq->nr_active |
Tejun Heo | 1606283 | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 1259 | * management later on and cause stall. Make sure the work |
| 1260 | * item is activated before grabbing. |
| 1261 | */ |
| 1262 | if (*work_data_bits(work) & WORK_STRUCT_DELAYED) |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1263 | pwq_activate_delayed_work(work); |
Lai Jiangshan | 3aa6249 | 2012-09-18 10:40:00 -0700 | [diff] [blame] | 1264 | |
Tejun Heo | 1606283 | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 1265 | list_del_init(&work->entry); |
Lai Jiangshan | 9c34a70 | 2014-07-11 00:11:13 +0800 | [diff] [blame] | 1266 | pwq_dec_nr_in_flight(pwq, get_work_color(work)); |
Tejun Heo | 36e227d | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1267 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1268 | /* work->data points to pwq iff queued, point to pool */ |
Tejun Heo | 1606283 | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 1269 | set_work_pool_and_keep_pending(work, pool->id); |
Lai Jiangshan | 4468a00 | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 1270 | |
Tejun Heo | 1606283 | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 1271 | spin_unlock(&pool->lock); |
| 1272 | return 1; |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1273 | } |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1274 | spin_unlock(&pool->lock); |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1275 | fail: |
| 1276 | local_irq_restore(*flags); |
| 1277 | if (work_is_canceling(work)) |
| 1278 | return -ENOENT; |
| 1279 | cpu_relax(); |
Tejun Heo | 36e227d | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1280 | return -EAGAIN; |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1281 | } |
| 1282 | |
| 1283 | /** |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 1284 | * insert_work - insert a work into a pool |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1285 | * @pwq: pwq @work belongs to |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 1286 | * @work: work to insert |
| 1287 | * @head: insertion point |
| 1288 | * @extra_flags: extra WORK_STRUCT_* flags to set |
| 1289 | * |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1290 | * Insert @work which belongs to @pwq after @head. @extra_flags is or'd to |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 1291 | * work_struct flags. |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 1292 | * |
| 1293 | * CONTEXT: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1294 | * spin_lock_irq(pool->lock). |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 1295 | */ |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1296 | static void insert_work(struct pool_workqueue *pwq, struct work_struct *work, |
| 1297 | struct list_head *head, unsigned int extra_flags) |
Oleg Nesterov | b89deed | 2007-05-09 02:33:52 -0700 | [diff] [blame] | 1298 | { |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1299 | struct worker_pool *pool = pwq->pool; |
Frederic Weisbecker | e1d8aa9 | 2009-01-12 23:15:46 +0100 | [diff] [blame] | 1300 | |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 1301 | /* we own @work, set data and link */ |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1302 | set_work_pwq(work, pwq, extra_flags); |
Oleg Nesterov | 1a4d9b0 | 2008-07-25 01:47:47 -0700 | [diff] [blame] | 1303 | list_add_tail(&work->entry, head); |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1304 | get_pwq(pwq); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1305 | |
| 1306 | /* |
Tejun Heo | c5aa87b | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 1307 | * Ensure either wq_worker_sleeping() sees the above |
| 1308 | * list_add_tail() or we see zero nr_running to avoid workers lying |
| 1309 | * around lazily while there are works to be processed. |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1310 | */ |
| 1311 | smp_mb(); |
| 1312 | |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1313 | if (__need_more_worker(pool)) |
| 1314 | wake_up_worker(pool); |
Oleg Nesterov | b89deed | 2007-05-09 02:33:52 -0700 | [diff] [blame] | 1315 | } |
| 1316 | |
Tejun Heo | c8efcc2 | 2010-12-20 19:32:04 +0100 | [diff] [blame] | 1317 | /* |
| 1318 | * Test whether @work is being queued from another work executing on the |
Tejun Heo | 8d03ecf | 2013-02-13 19:29:10 -0800 | [diff] [blame] | 1319 | * same workqueue. |
Tejun Heo | c8efcc2 | 2010-12-20 19:32:04 +0100 | [diff] [blame] | 1320 | */ |
| 1321 | static bool is_chained_work(struct workqueue_struct *wq) |
| 1322 | { |
Tejun Heo | 8d03ecf | 2013-02-13 19:29:10 -0800 | [diff] [blame] | 1323 | struct worker *worker; |
Tejun Heo | c8efcc2 | 2010-12-20 19:32:04 +0100 | [diff] [blame] | 1324 | |
Tejun Heo | 8d03ecf | 2013-02-13 19:29:10 -0800 | [diff] [blame] | 1325 | worker = current_wq_worker(); |
| 1326 | /* |
| 1327 | * Return %true iff I'm a worker execuing a work item on @wq. If |
| 1328 | * I'm @worker, it's safe to dereference it without locking. |
| 1329 | */ |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1330 | return worker && worker->current_pwq->wq == wq; |
Tejun Heo | c8efcc2 | 2010-12-20 19:32:04 +0100 | [diff] [blame] | 1331 | } |
| 1332 | |
Mike Galbraith | ef557180 | 2016-02-09 17:59:38 -0500 | [diff] [blame] | 1333 | /* |
| 1334 | * When queueing an unbound work item to a wq, prefer local CPU if allowed |
| 1335 | * by wq_unbound_cpumask. Otherwise, round robin among the allowed ones to |
| 1336 | * avoid perturbing sensitive tasks. |
| 1337 | */ |
| 1338 | static int wq_select_unbound_cpu(int cpu) |
| 1339 | { |
Tejun Heo | f303fcc | 2016-02-09 17:59:38 -0500 | [diff] [blame] | 1340 | static bool printed_dbg_warning; |
Mike Galbraith | ef557180 | 2016-02-09 17:59:38 -0500 | [diff] [blame] | 1341 | int new_cpu; |
| 1342 | |
Tejun Heo | f303fcc | 2016-02-09 17:59:38 -0500 | [diff] [blame] | 1343 | if (likely(!wq_debug_force_rr_cpu)) { |
| 1344 | if (cpumask_test_cpu(cpu, wq_unbound_cpumask)) |
| 1345 | return cpu; |
| 1346 | } else if (!printed_dbg_warning) { |
| 1347 | pr_warn("workqueue: round-robin CPU selection forced, expect performance impact\n"); |
| 1348 | printed_dbg_warning = true; |
| 1349 | } |
| 1350 | |
Mike Galbraith | ef557180 | 2016-02-09 17:59:38 -0500 | [diff] [blame] | 1351 | if (cpumask_empty(wq_unbound_cpumask)) |
| 1352 | return cpu; |
| 1353 | |
| 1354 | new_cpu = __this_cpu_read(wq_rr_cpu_last); |
| 1355 | new_cpu = cpumask_next_and(new_cpu, wq_unbound_cpumask, cpu_online_mask); |
| 1356 | if (unlikely(new_cpu >= nr_cpu_ids)) { |
| 1357 | new_cpu = cpumask_first_and(wq_unbound_cpumask, cpu_online_mask); |
| 1358 | if (unlikely(new_cpu >= nr_cpu_ids)) |
| 1359 | return cpu; |
| 1360 | } |
| 1361 | __this_cpu_write(wq_rr_cpu_last, new_cpu); |
| 1362 | |
| 1363 | return new_cpu; |
| 1364 | } |
| 1365 | |
Tejun Heo | d84ff05 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 1366 | static void __queue_work(int cpu, struct workqueue_struct *wq, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1367 | struct work_struct *work) |
| 1368 | { |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1369 | struct pool_workqueue *pwq; |
Tejun Heo | c917808 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1370 | struct worker_pool *last_pool; |
Tejun Heo | 1e19ffc | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1371 | struct list_head *worklist; |
Tejun Heo | 8a2e8e5d | 2010-08-25 10:33:56 +0200 | [diff] [blame] | 1372 | unsigned int work_flags; |
Joonsoo Kim | b75cac9 | 2012-08-15 23:25:37 +0900 | [diff] [blame] | 1373 | unsigned int req_cpu = cpu; |
Tejun Heo | 8930cab | 2012-08-03 10:30:45 -0700 | [diff] [blame] | 1374 | |
| 1375 | /* |
| 1376 | * While a work item is PENDING && off queue, a task trying to |
| 1377 | * steal the PENDING will busy-loop waiting for it to either get |
| 1378 | * queued or lose PENDING. Grabbing PENDING and queueing should |
| 1379 | * happen with IRQ disabled. |
| 1380 | */ |
| 1381 | WARN_ON_ONCE(!irqs_disabled()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1382 | |
Tejun Heo | 1e19ffc | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1383 | |
Li Bin | 9ef28a7 | 2013-09-09 13:13:58 +0800 | [diff] [blame] | 1384 | /* if draining, only works from the same workqueue are allowed */ |
Tejun Heo | 618b01e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1385 | if (unlikely(wq->flags & __WQ_DRAINING) && |
Tejun Heo | c8efcc2 | 2010-12-20 19:32:04 +0100 | [diff] [blame] | 1386 | WARN_ON_ONCE(!is_chained_work(wq))) |
Tejun Heo | e41e704 | 2010-08-24 14:22:47 +0200 | [diff] [blame] | 1387 | return; |
Tejun Heo | 9e8cd2f | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1388 | retry: |
Tejun Heo | c917808 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1389 | /* pwq which will be used unless @work is executing elsewhere */ |
Hillf Danton | 48c3362 | 2020-01-24 20:14:45 -0500 | [diff] [blame] | 1390 | if (wq->flags & WQ_UNBOUND) { |
| 1391 | if (req_cpu == WORK_CPU_UNBOUND) |
| 1392 | cpu = wq_select_unbound_cpu(raw_smp_processor_id()); |
Tejun Heo | df2d5ae | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 1393 | pwq = unbound_pwq_by_node(wq, cpu_to_node(cpu)); |
Hillf Danton | 48c3362 | 2020-01-24 20:14:45 -0500 | [diff] [blame] | 1394 | } else { |
| 1395 | if (req_cpu == WORK_CPU_UNBOUND) |
| 1396 | cpu = raw_smp_processor_id(); |
| 1397 | pwq = per_cpu_ptr(wq->cpu_pwqs, cpu); |
| 1398 | } |
Tejun Heo | dbf2576 | 2012-08-20 14:51:23 -0700 | [diff] [blame] | 1399 | |
Tejun Heo | c917808 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1400 | /* |
| 1401 | * If @work was previously on a different pool, it might still be |
| 1402 | * running there, in which case the work needs to be queued on that |
| 1403 | * pool to guarantee non-reentrancy. |
| 1404 | */ |
| 1405 | last_pool = get_work_pool(work); |
| 1406 | if (last_pool && last_pool != pwq->pool) { |
| 1407 | struct worker *worker; |
Tejun Heo | 18aa9ef | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 1408 | |
Tejun Heo | c917808 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1409 | spin_lock(&last_pool->lock); |
Tejun Heo | 18aa9ef | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 1410 | |
Tejun Heo | c917808 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1411 | worker = find_worker_executing_work(last_pool, work); |
Tejun Heo | 18aa9ef | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 1412 | |
Tejun Heo | c917808 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1413 | if (worker && worker->current_pwq->wq == wq) { |
| 1414 | pwq = worker->current_pwq; |
Tejun Heo | 8930cab | 2012-08-03 10:30:45 -0700 | [diff] [blame] | 1415 | } else { |
Tejun Heo | c917808 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1416 | /* meh... not running there, queue here */ |
| 1417 | spin_unlock(&last_pool->lock); |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1418 | spin_lock(&pwq->pool->lock); |
Tejun Heo | 8930cab | 2012-08-03 10:30:45 -0700 | [diff] [blame] | 1419 | } |
Tejun Heo | f342179 | 2010-07-02 10:03:51 +0200 | [diff] [blame] | 1420 | } else { |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1421 | spin_lock(&pwq->pool->lock); |
Tejun Heo | 502ca9d | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 1422 | } |
| 1423 | |
Tejun Heo | 9e8cd2f | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1424 | /* |
| 1425 | * pwq is determined and locked. For unbound pools, we could have |
| 1426 | * raced with pwq release and it could already be dead. If its |
| 1427 | * refcnt is zero, repeat pwq selection. Note that pwqs never die |
Tejun Heo | df2d5ae | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 1428 | * without another pwq replacing it in the numa_pwq_tbl or while |
| 1429 | * work items are executing on it, so the retrying is guaranteed to |
Tejun Heo | 9e8cd2f | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1430 | * make forward-progress. |
| 1431 | */ |
| 1432 | if (unlikely(!pwq->refcnt)) { |
| 1433 | if (wq->flags & WQ_UNBOUND) { |
| 1434 | spin_unlock(&pwq->pool->lock); |
| 1435 | cpu_relax(); |
| 1436 | goto retry; |
| 1437 | } |
| 1438 | /* oops */ |
| 1439 | WARN_ONCE(true, "workqueue: per-cpu pwq for %s on cpu%d has 0 refcnt", |
| 1440 | wq->name, cpu); |
| 1441 | } |
| 1442 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1443 | /* pwq determined, queue */ |
| 1444 | trace_workqueue_queue_work(req_cpu, pwq, work); |
Tejun Heo | 502ca9d | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 1445 | |
Dan Carpenter | f5b2552 | 2012-04-13 22:06:58 +0300 | [diff] [blame] | 1446 | if (WARN_ON(!list_empty(&work->entry))) { |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1447 | spin_unlock(&pwq->pool->lock); |
Dan Carpenter | f5b2552 | 2012-04-13 22:06:58 +0300 | [diff] [blame] | 1448 | return; |
| 1449 | } |
Tejun Heo | 1e19ffc | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1450 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1451 | pwq->nr_in_flight[pwq->work_color]++; |
| 1452 | work_flags = work_color_to_flags(pwq->work_color); |
Tejun Heo | 1e19ffc | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1453 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1454 | if (likely(pwq->nr_active < pwq->max_active)) { |
Tejun Heo | cdadf00 | 2010-10-05 10:49:55 +0200 | [diff] [blame] | 1455 | trace_workqueue_activate_work(work); |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1456 | pwq->nr_active++; |
| 1457 | worklist = &pwq->pool->worklist; |
Tejun Heo | 82607adc | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 1458 | if (list_empty(worklist)) |
| 1459 | pwq->pool->watchdog_ts = jiffies; |
Tejun Heo | 8a2e8e5d | 2010-08-25 10:33:56 +0200 | [diff] [blame] | 1460 | } else { |
| 1461 | work_flags |= WORK_STRUCT_DELAYED; |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1462 | worklist = &pwq->delayed_works; |
Tejun Heo | 8a2e8e5d | 2010-08-25 10:33:56 +0200 | [diff] [blame] | 1463 | } |
Tejun Heo | 1e19ffc | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1464 | |
Zqiang | 9948ff5 | 2021-02-18 11:16:49 +0800 | [diff] [blame] | 1465 | debug_work_activate(work); |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1466 | insert_work(pwq, work, worklist, work_flags); |
Tejun Heo | 1e19ffc | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1467 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1468 | spin_unlock(&pwq->pool->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1469 | } |
| 1470 | |
Rolf Eike Beer | 0fcb78c | 2006-07-30 03:03:42 -0700 | [diff] [blame] | 1471 | /** |
Zhang Rui | c1a220e | 2008-07-23 21:28:39 -0700 | [diff] [blame] | 1472 | * queue_work_on - queue work on specific cpu |
| 1473 | * @cpu: CPU number to execute work on |
| 1474 | * @wq: workqueue to use |
| 1475 | * @work: work to queue |
| 1476 | * |
Zhang Rui | c1a220e | 2008-07-23 21:28:39 -0700 | [diff] [blame] | 1477 | * We queue the work to a specific CPU, the caller must ensure it |
| 1478 | * can't go away. |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 1479 | * |
| 1480 | * Return: %false if @work was already on a queue, %true otherwise. |
Zhang Rui | c1a220e | 2008-07-23 21:28:39 -0700 | [diff] [blame] | 1481 | */ |
Tejun Heo | d4283e9 | 2012-08-03 10:30:44 -0700 | [diff] [blame] | 1482 | bool queue_work_on(int cpu, struct workqueue_struct *wq, |
| 1483 | struct work_struct *work) |
Zhang Rui | c1a220e | 2008-07-23 21:28:39 -0700 | [diff] [blame] | 1484 | { |
Tejun Heo | d4283e9 | 2012-08-03 10:30:44 -0700 | [diff] [blame] | 1485 | bool ret = false; |
Tejun Heo | 8930cab | 2012-08-03 10:30:45 -0700 | [diff] [blame] | 1486 | unsigned long flags; |
| 1487 | |
| 1488 | local_irq_save(flags); |
Zhang Rui | c1a220e | 2008-07-23 21:28:39 -0700 | [diff] [blame] | 1489 | |
Tejun Heo | 22df02b | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 1490 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 1491 | __queue_work(cpu, wq, work); |
Tejun Heo | d4283e9 | 2012-08-03 10:30:44 -0700 | [diff] [blame] | 1492 | ret = true; |
Zhang Rui | c1a220e | 2008-07-23 21:28:39 -0700 | [diff] [blame] | 1493 | } |
Tejun Heo | 8930cab | 2012-08-03 10:30:45 -0700 | [diff] [blame] | 1494 | |
| 1495 | local_irq_restore(flags); |
Zhang Rui | c1a220e | 2008-07-23 21:28:39 -0700 | [diff] [blame] | 1496 | return ret; |
| 1497 | } |
Marc Dionne | ad7b1f8 | 2013-05-06 17:44:55 -0400 | [diff] [blame] | 1498 | EXPORT_SYMBOL(queue_work_on); |
Zhang Rui | c1a220e | 2008-07-23 21:28:39 -0700 | [diff] [blame] | 1499 | |
Tejun Heo | d8e794d | 2012-08-03 10:30:45 -0700 | [diff] [blame] | 1500 | void delayed_work_timer_fn(unsigned long __data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1501 | { |
David Howells | 52bad64 | 2006-11-22 14:54:01 +0000 | [diff] [blame] | 1502 | struct delayed_work *dwork = (struct delayed_work *)__data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1503 | |
Tejun Heo | e0aecdd | 2012-08-21 13:18:24 -0700 | [diff] [blame] | 1504 | /* should have been called from irqsafe timer with irq already off */ |
Lai Jiangshan | 60c057b | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 1505 | __queue_work(dwork->cpu, dwork->wq, &dwork->work); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1506 | } |
Konstantin Khlebnikov | 1438ade5 | 2013-01-24 16:36:31 +0400 | [diff] [blame] | 1507 | EXPORT_SYMBOL(delayed_work_timer_fn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1508 | |
Tejun Heo | 7beb2ed | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1509 | static void __queue_delayed_work(int cpu, struct workqueue_struct *wq, |
| 1510 | struct delayed_work *dwork, unsigned long delay) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1511 | { |
Tejun Heo | 7beb2ed | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1512 | struct timer_list *timer = &dwork->timer; |
| 1513 | struct work_struct *work = &dwork->work; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1514 | |
Tejun Heo | 637fdba | 2017-03-06 15:33:42 -0500 | [diff] [blame] | 1515 | WARN_ON_ONCE(!wq); |
Tejun Heo | 7beb2ed | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1516 | WARN_ON_ONCE(timer->function != delayed_work_timer_fn || |
| 1517 | timer->data != (unsigned long)dwork); |
Tejun Heo | fc4b514 | 2012-12-04 07:40:39 -0800 | [diff] [blame] | 1518 | WARN_ON_ONCE(timer_pending(timer)); |
| 1519 | WARN_ON_ONCE(!list_empty(&work->entry)); |
Tejun Heo | 7beb2ed | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1520 | |
Tejun Heo | 8852aac | 2012-12-01 16:23:42 -0800 | [diff] [blame] | 1521 | /* |
| 1522 | * If @delay is 0, queue @dwork->work immediately. This is for |
| 1523 | * both optimization and correctness. The earliest @timer can |
| 1524 | * expire is on the closest next tick and delayed_work users depend |
| 1525 | * on that there's no such delay when @delay is 0. |
| 1526 | */ |
| 1527 | if (!delay) { |
| 1528 | __queue_work(cpu, wq, &dwork->work); |
| 1529 | return; |
| 1530 | } |
| 1531 | |
Lai Jiangshan | 60c057b | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 1532 | dwork->wq = wq; |
Tejun Heo | 1265057 | 2012-08-08 09:38:42 -0700 | [diff] [blame] | 1533 | dwork->cpu = cpu; |
Tejun Heo | 7beb2ed | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1534 | timer->expires = jiffies + delay; |
| 1535 | |
Tejun Heo | 041bd12 | 2016-02-09 16:11:26 -0500 | [diff] [blame] | 1536 | if (unlikely(cpu != WORK_CPU_UNBOUND)) |
| 1537 | add_timer_on(timer, cpu); |
| 1538 | else |
| 1539 | add_timer(timer); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1540 | } |
| 1541 | |
Rolf Eike Beer | 0fcb78c | 2006-07-30 03:03:42 -0700 | [diff] [blame] | 1542 | /** |
| 1543 | * queue_delayed_work_on - queue work on specific CPU after delay |
| 1544 | * @cpu: CPU number to execute work on |
| 1545 | * @wq: workqueue to use |
Randy Dunlap | af9997e | 2006-12-22 01:06:52 -0800 | [diff] [blame] | 1546 | * @dwork: work to queue |
Rolf Eike Beer | 0fcb78c | 2006-07-30 03:03:42 -0700 | [diff] [blame] | 1547 | * @delay: number of jiffies to wait before queueing |
| 1548 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 1549 | * Return: %false if @work was already on a queue, %true otherwise. If |
Tejun Heo | 715f130 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1550 | * @delay is zero and @dwork is idle, it will be scheduled for immediate |
| 1551 | * execution. |
Rolf Eike Beer | 0fcb78c | 2006-07-30 03:03:42 -0700 | [diff] [blame] | 1552 | */ |
Tejun Heo | d4283e9 | 2012-08-03 10:30:44 -0700 | [diff] [blame] | 1553 | bool queue_delayed_work_on(int cpu, struct workqueue_struct *wq, |
| 1554 | struct delayed_work *dwork, unsigned long delay) |
Venkatesh Pallipadi | 7a6bc1c | 2006-06-28 13:50:33 -0700 | [diff] [blame] | 1555 | { |
David Howells | 52bad64 | 2006-11-22 14:54:01 +0000 | [diff] [blame] | 1556 | struct work_struct *work = &dwork->work; |
Tejun Heo | d4283e9 | 2012-08-03 10:30:44 -0700 | [diff] [blame] | 1557 | bool ret = false; |
Tejun Heo | 8930cab | 2012-08-03 10:30:45 -0700 | [diff] [blame] | 1558 | unsigned long flags; |
| 1559 | |
| 1560 | /* read the comment in __queue_work() */ |
| 1561 | local_irq_save(flags); |
Venkatesh Pallipadi | 7a6bc1c | 2006-06-28 13:50:33 -0700 | [diff] [blame] | 1562 | |
Tejun Heo | 22df02b | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 1563 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
Tejun Heo | 7beb2ed | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1564 | __queue_delayed_work(cpu, wq, dwork, delay); |
Tejun Heo | d4283e9 | 2012-08-03 10:30:44 -0700 | [diff] [blame] | 1565 | ret = true; |
Venkatesh Pallipadi | 7a6bc1c | 2006-06-28 13:50:33 -0700 | [diff] [blame] | 1566 | } |
Tejun Heo | 8930cab | 2012-08-03 10:30:45 -0700 | [diff] [blame] | 1567 | |
| 1568 | local_irq_restore(flags); |
Venkatesh Pallipadi | 7a6bc1c | 2006-06-28 13:50:33 -0700 | [diff] [blame] | 1569 | return ret; |
| 1570 | } |
Marc Dionne | ad7b1f8 | 2013-05-06 17:44:55 -0400 | [diff] [blame] | 1571 | EXPORT_SYMBOL(queue_delayed_work_on); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1572 | |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1573 | /** |
Tejun Heo | 8376fe2 | 2012-08-03 10:30:47 -0700 | [diff] [blame] | 1574 | * mod_delayed_work_on - modify delay of or queue a delayed work on specific CPU |
| 1575 | * @cpu: CPU number to execute work on |
| 1576 | * @wq: workqueue to use |
| 1577 | * @dwork: work to queue |
| 1578 | * @delay: number of jiffies to wait before queueing |
| 1579 | * |
| 1580 | * If @dwork is idle, equivalent to queue_delayed_work_on(); otherwise, |
| 1581 | * modify @dwork's timer so that it expires after @delay. If @delay is |
| 1582 | * zero, @work is guaranteed to be scheduled immediately regardless of its |
| 1583 | * current state. |
| 1584 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 1585 | * Return: %false if @dwork was idle and queued, %true if @dwork was |
Tejun Heo | 8376fe2 | 2012-08-03 10:30:47 -0700 | [diff] [blame] | 1586 | * pending and its timer was modified. |
| 1587 | * |
Tejun Heo | e0aecdd | 2012-08-21 13:18:24 -0700 | [diff] [blame] | 1588 | * This function is safe to call from any context including IRQ handler. |
Tejun Heo | 8376fe2 | 2012-08-03 10:30:47 -0700 | [diff] [blame] | 1589 | * See try_to_grab_pending() for details. |
| 1590 | */ |
| 1591 | bool mod_delayed_work_on(int cpu, struct workqueue_struct *wq, |
| 1592 | struct delayed_work *dwork, unsigned long delay) |
| 1593 | { |
| 1594 | unsigned long flags; |
| 1595 | int ret; |
| 1596 | |
| 1597 | do { |
| 1598 | ret = try_to_grab_pending(&dwork->work, true, &flags); |
| 1599 | } while (unlikely(ret == -EAGAIN)); |
| 1600 | |
| 1601 | if (likely(ret >= 0)) { |
| 1602 | __queue_delayed_work(cpu, wq, dwork, delay); |
| 1603 | local_irq_restore(flags); |
| 1604 | } |
| 1605 | |
| 1606 | /* -ENOENT from try_to_grab_pending() becomes %true */ |
| 1607 | return ret; |
| 1608 | } |
| 1609 | EXPORT_SYMBOL_GPL(mod_delayed_work_on); |
| 1610 | |
| 1611 | /** |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1612 | * worker_enter_idle - enter idle state |
| 1613 | * @worker: worker which is entering idle state |
| 1614 | * |
| 1615 | * @worker is entering idle state. Update stats and idle timer if |
| 1616 | * necessary. |
| 1617 | * |
| 1618 | * LOCKING: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1619 | * spin_lock_irq(pool->lock). |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1620 | */ |
| 1621 | static void worker_enter_idle(struct worker *worker) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1622 | { |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1623 | struct worker_pool *pool = worker->pool; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1624 | |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 1625 | if (WARN_ON_ONCE(worker->flags & WORKER_IDLE) || |
| 1626 | WARN_ON_ONCE(!list_empty(&worker->entry) && |
| 1627 | (worker->hentry.next || worker->hentry.pprev))) |
| 1628 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1629 | |
Lai Jiangshan | 051e185 | 2014-07-22 13:03:02 +0800 | [diff] [blame] | 1630 | /* can't use worker_set_flags(), also called from create_worker() */ |
Tejun Heo | cb44476 | 2010-07-02 10:03:50 +0200 | [diff] [blame] | 1631 | worker->flags |= WORKER_IDLE; |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1632 | pool->nr_idle++; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1633 | worker->last_active = jiffies; |
Peter Zijlstra | d5abe66 | 2006-12-06 20:37:26 -0800 | [diff] [blame] | 1634 | |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1635 | /* idle_list is LIFO */ |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1636 | list_add(&worker->entry, &pool->idle_list); |
Tejun Heo | db7bccf | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1637 | |
Tejun Heo | 628c78e | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 1638 | if (too_many_workers(pool) && !timer_pending(&pool->idle_timer)) |
| 1639 | mod_timer(&pool->idle_timer, jiffies + IDLE_WORKER_TIMEOUT); |
Tejun Heo | cb44476 | 2010-07-02 10:03:50 +0200 | [diff] [blame] | 1640 | |
Tejun Heo | 544ecf3 | 2012-05-14 15:04:50 -0700 | [diff] [blame] | 1641 | /* |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 1642 | * Sanity check nr_running. Because wq_unbind_fn() releases |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1643 | * pool->lock between setting %WORKER_UNBOUND and zapping |
Tejun Heo | 628c78e | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 1644 | * nr_running, the warning may trigger spuriously. Check iff |
| 1645 | * unbind is not in progress. |
Tejun Heo | 544ecf3 | 2012-05-14 15:04:50 -0700 | [diff] [blame] | 1646 | */ |
Tejun Heo | 2464757 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1647 | WARN_ON_ONCE(!(pool->flags & POOL_DISASSOCIATED) && |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1648 | pool->nr_workers == pool->nr_idle && |
Tejun Heo | e19e397 | 2013-01-24 11:39:44 -0800 | [diff] [blame] | 1649 | atomic_read(&pool->nr_running)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1650 | } |
| 1651 | |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1652 | /** |
| 1653 | * worker_leave_idle - leave idle state |
| 1654 | * @worker: worker which is leaving idle state |
| 1655 | * |
| 1656 | * @worker is leaving idle state. Update stats. |
| 1657 | * |
| 1658 | * LOCKING: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1659 | * spin_lock_irq(pool->lock). |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1660 | */ |
| 1661 | static void worker_leave_idle(struct worker *worker) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1662 | { |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1663 | struct worker_pool *pool = worker->pool; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1664 | |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 1665 | if (WARN_ON_ONCE(!(worker->flags & WORKER_IDLE))) |
| 1666 | return; |
Tejun Heo | d302f01 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 1667 | worker_clr_flags(worker, WORKER_IDLE); |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1668 | pool->nr_idle--; |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1669 | list_del_init(&worker->entry); |
| 1670 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1671 | |
Lai Jiangshan | f7537df | 2014-07-15 17:24:15 +0800 | [diff] [blame] | 1672 | static struct worker *alloc_worker(int node) |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1673 | { |
| 1674 | struct worker *worker; |
| 1675 | |
Lai Jiangshan | f7537df | 2014-07-15 17:24:15 +0800 | [diff] [blame] | 1676 | worker = kzalloc_node(sizeof(*worker), GFP_KERNEL, node); |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1677 | if (worker) { |
| 1678 | INIT_LIST_HEAD(&worker->entry); |
Tejun Heo | affee4b | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1679 | INIT_LIST_HEAD(&worker->scheduled); |
Lai Jiangshan | da02846 | 2014-05-20 17:46:31 +0800 | [diff] [blame] | 1680 | INIT_LIST_HEAD(&worker->node); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1681 | /* on creation a worker is in !idle && prep state */ |
| 1682 | worker->flags = WORKER_PREP; |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1683 | } |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1684 | return worker; |
| 1685 | } |
| 1686 | |
| 1687 | /** |
Lai Jiangshan | 4736cbf | 2014-05-20 17:46:35 +0800 | [diff] [blame] | 1688 | * worker_attach_to_pool() - attach a worker to a pool |
| 1689 | * @worker: worker to be attached |
| 1690 | * @pool: the target pool |
| 1691 | * |
| 1692 | * Attach @worker to @pool. Once attached, the %WORKER_UNBOUND flag and |
| 1693 | * cpu-binding of @worker are kept coordinated with the pool across |
| 1694 | * cpu-[un]hotplugs. |
| 1695 | */ |
| 1696 | static void worker_attach_to_pool(struct worker *worker, |
| 1697 | struct worker_pool *pool) |
| 1698 | { |
| 1699 | mutex_lock(&pool->attach_mutex); |
| 1700 | |
| 1701 | /* |
| 1702 | * set_cpus_allowed_ptr() will fail if the cpumask doesn't have any |
| 1703 | * online CPUs. It'll be re-applied when any of the CPUs come up. |
| 1704 | */ |
| 1705 | set_cpus_allowed_ptr(worker->task, pool->attrs->cpumask); |
| 1706 | |
| 1707 | /* |
| 1708 | * The pool->attach_mutex ensures %POOL_DISASSOCIATED remains |
| 1709 | * stable across this function. See the comments above the |
| 1710 | * flag definition for details. |
| 1711 | */ |
| 1712 | if (pool->flags & POOL_DISASSOCIATED) |
| 1713 | worker->flags |= WORKER_UNBOUND; |
| 1714 | |
| 1715 | list_add_tail(&worker->node, &pool->workers); |
| 1716 | |
| 1717 | mutex_unlock(&pool->attach_mutex); |
| 1718 | } |
| 1719 | |
| 1720 | /** |
Lai Jiangshan | 60f5a4b | 2014-05-20 17:46:29 +0800 | [diff] [blame] | 1721 | * worker_detach_from_pool() - detach a worker from its pool |
| 1722 | * @worker: worker which is attached to its pool |
| 1723 | * @pool: the pool @worker is attached to |
| 1724 | * |
Lai Jiangshan | 4736cbf | 2014-05-20 17:46:35 +0800 | [diff] [blame] | 1725 | * Undo the attaching which had been done in worker_attach_to_pool(). The |
| 1726 | * caller worker shouldn't access to the pool after detached except it has |
| 1727 | * other reference to the pool. |
Lai Jiangshan | 60f5a4b | 2014-05-20 17:46:29 +0800 | [diff] [blame] | 1728 | */ |
| 1729 | static void worker_detach_from_pool(struct worker *worker, |
| 1730 | struct worker_pool *pool) |
| 1731 | { |
| 1732 | struct completion *detach_completion = NULL; |
| 1733 | |
Lai Jiangshan | 92f9c5c | 2014-05-20 17:46:34 +0800 | [diff] [blame] | 1734 | mutex_lock(&pool->attach_mutex); |
Lai Jiangshan | da02846 | 2014-05-20 17:46:31 +0800 | [diff] [blame] | 1735 | list_del(&worker->node); |
| 1736 | if (list_empty(&pool->workers)) |
Lai Jiangshan | 60f5a4b | 2014-05-20 17:46:29 +0800 | [diff] [blame] | 1737 | detach_completion = pool->detach_completion; |
Lai Jiangshan | 92f9c5c | 2014-05-20 17:46:34 +0800 | [diff] [blame] | 1738 | mutex_unlock(&pool->attach_mutex); |
Lai Jiangshan | 60f5a4b | 2014-05-20 17:46:29 +0800 | [diff] [blame] | 1739 | |
Lai Jiangshan | b62c075 | 2014-06-03 15:32:52 +0800 | [diff] [blame] | 1740 | /* clear leftover flags without pool->lock after it is detached */ |
| 1741 | worker->flags &= ~(WORKER_UNBOUND | WORKER_REBOUND); |
| 1742 | |
Lai Jiangshan | 60f5a4b | 2014-05-20 17:46:29 +0800 | [diff] [blame] | 1743 | if (detach_completion) |
| 1744 | complete(detach_completion); |
| 1745 | } |
| 1746 | |
| 1747 | /** |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1748 | * create_worker - create a new workqueue worker |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1749 | * @pool: pool the new worker will belong to |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1750 | * |
Lai Jiangshan | 051e185 | 2014-07-22 13:03:02 +0800 | [diff] [blame] | 1751 | * Create and start a new worker which is attached to @pool. |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1752 | * |
| 1753 | * CONTEXT: |
| 1754 | * Might sleep. Does GFP_KERNEL allocations. |
| 1755 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 1756 | * Return: |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1757 | * Pointer to the newly created worker. |
| 1758 | */ |
Tejun Heo | bc2ae0f | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 1759 | static struct worker *create_worker(struct worker_pool *pool) |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1760 | { |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1761 | struct worker *worker = NULL; |
Tejun Heo | f342179 | 2010-07-02 10:03:51 +0200 | [diff] [blame] | 1762 | int id = -1; |
Tejun Heo | e3c916a | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 1763 | char id_buf[16]; |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1764 | |
Lai Jiangshan | 7cda9aa | 2014-05-20 17:46:32 +0800 | [diff] [blame] | 1765 | /* ID is needed to determine kthread name */ |
| 1766 | id = ida_simple_get(&pool->worker_ida, 0, 0, GFP_KERNEL); |
Tejun Heo | 822d840 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 1767 | if (id < 0) |
| 1768 | goto fail; |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1769 | |
Lai Jiangshan | f7537df | 2014-07-15 17:24:15 +0800 | [diff] [blame] | 1770 | worker = alloc_worker(pool->node); |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1771 | if (!worker) |
| 1772 | goto fail; |
| 1773 | |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1774 | worker->pool = pool; |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1775 | worker->id = id; |
| 1776 | |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 1777 | if (pool->cpu >= 0) |
Tejun Heo | e3c916a | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 1778 | snprintf(id_buf, sizeof(id_buf), "%d:%d%s", pool->cpu, id, |
| 1779 | pool->attrs->nice < 0 ? "H" : ""); |
Tejun Heo | f342179 | 2010-07-02 10:03:51 +0200 | [diff] [blame] | 1780 | else |
Tejun Heo | e3c916a | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 1781 | snprintf(id_buf, sizeof(id_buf), "u%d:%d", pool->id, id); |
| 1782 | |
Tejun Heo | f3f90ad | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 1783 | worker->task = kthread_create_on_node(worker_thread, worker, pool->node, |
Tejun Heo | e3c916a | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 1784 | "kworker/%s", id_buf); |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1785 | if (IS_ERR(worker->task)) |
| 1786 | goto fail; |
| 1787 | |
Oleg Nesterov | 9115122 | 2013-11-14 12:56:18 +0100 | [diff] [blame] | 1788 | set_user_nice(worker->task, pool->attrs->nice); |
Peter Zijlstra | 25834c7 | 2015-05-15 17:43:34 +0200 | [diff] [blame] | 1789 | kthread_bind_mask(worker->task, pool->attrs->cpumask); |
Oleg Nesterov | 9115122 | 2013-11-14 12:56:18 +0100 | [diff] [blame] | 1790 | |
Lai Jiangshan | da02846 | 2014-05-20 17:46:31 +0800 | [diff] [blame] | 1791 | /* successful, attach the worker to the pool */ |
Lai Jiangshan | 4736cbf | 2014-05-20 17:46:35 +0800 | [diff] [blame] | 1792 | worker_attach_to_pool(worker, pool); |
Tejun Heo | 822d840 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 1793 | |
Lai Jiangshan | 051e185 | 2014-07-22 13:03:02 +0800 | [diff] [blame] | 1794 | /* start the newly created worker */ |
| 1795 | spin_lock_irq(&pool->lock); |
| 1796 | worker->pool->nr_workers++; |
| 1797 | worker_enter_idle(worker); |
| 1798 | wake_up_process(worker->task); |
| 1799 | spin_unlock_irq(&pool->lock); |
| 1800 | |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1801 | return worker; |
Tejun Heo | 822d840 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 1802 | |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1803 | fail: |
Lai Jiangshan | 9625ab1 | 2014-05-20 17:46:27 +0800 | [diff] [blame] | 1804 | if (id >= 0) |
Lai Jiangshan | 7cda9aa | 2014-05-20 17:46:32 +0800 | [diff] [blame] | 1805 | ida_simple_remove(&pool->worker_ida, id); |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1806 | kfree(worker); |
| 1807 | return NULL; |
| 1808 | } |
| 1809 | |
| 1810 | /** |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1811 | * destroy_worker - destroy a workqueue worker |
| 1812 | * @worker: worker to be destroyed |
| 1813 | * |
Lai Jiangshan | 73eb7fe | 2014-05-20 17:46:28 +0800 | [diff] [blame] | 1814 | * Destroy @worker and adjust @pool stats accordingly. The worker should |
| 1815 | * be idle. |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1816 | * |
| 1817 | * CONTEXT: |
Lai Jiangshan | 60f5a4b | 2014-05-20 17:46:29 +0800 | [diff] [blame] | 1818 | * spin_lock_irq(pool->lock). |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1819 | */ |
| 1820 | static void destroy_worker(struct worker *worker) |
| 1821 | { |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1822 | struct worker_pool *pool = worker->pool; |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1823 | |
Tejun Heo | cd54968 | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 1824 | lockdep_assert_held(&pool->lock); |
| 1825 | |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1826 | /* sanity check frenzy */ |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 1827 | if (WARN_ON(worker->current_work) || |
Lai Jiangshan | 73eb7fe | 2014-05-20 17:46:28 +0800 | [diff] [blame] | 1828 | WARN_ON(!list_empty(&worker->scheduled)) || |
| 1829 | WARN_ON(!(worker->flags & WORKER_IDLE))) |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 1830 | return; |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1831 | |
Lai Jiangshan | 73eb7fe | 2014-05-20 17:46:28 +0800 | [diff] [blame] | 1832 | pool->nr_workers--; |
| 1833 | pool->nr_idle--; |
Lai Jiangshan | 5bdfff9 | 2014-02-15 22:02:28 +0800 | [diff] [blame] | 1834 | |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1835 | list_del_init(&worker->entry); |
Tejun Heo | cb44476 | 2010-07-02 10:03:50 +0200 | [diff] [blame] | 1836 | worker->flags |= WORKER_DIE; |
Lai Jiangshan | 60f5a4b | 2014-05-20 17:46:29 +0800 | [diff] [blame] | 1837 | wake_up_process(worker->task); |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1838 | } |
| 1839 | |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1840 | static void idle_worker_timeout(unsigned long __pool) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1841 | { |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1842 | struct worker_pool *pool = (void *)__pool; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1843 | |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1844 | spin_lock_irq(&pool->lock); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1845 | |
Lai Jiangshan | 3347fc9 | 2014-05-20 17:46:30 +0800 | [diff] [blame] | 1846 | while (too_many_workers(pool)) { |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1847 | struct worker *worker; |
| 1848 | unsigned long expires; |
| 1849 | |
| 1850 | /* idle_list is kept in LIFO order, check the last one */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1851 | worker = list_entry(pool->idle_list.prev, struct worker, entry); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1852 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; |
| 1853 | |
Lai Jiangshan | 3347fc9 | 2014-05-20 17:46:30 +0800 | [diff] [blame] | 1854 | if (time_before(jiffies, expires)) { |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1855 | mod_timer(&pool->idle_timer, expires); |
Lai Jiangshan | 3347fc9 | 2014-05-20 17:46:30 +0800 | [diff] [blame] | 1856 | break; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1857 | } |
Lai Jiangshan | 3347fc9 | 2014-05-20 17:46:30 +0800 | [diff] [blame] | 1858 | |
| 1859 | destroy_worker(worker); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1860 | } |
| 1861 | |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1862 | spin_unlock_irq(&pool->lock); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1863 | } |
| 1864 | |
Tejun Heo | 493a172 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 1865 | static void send_mayday(struct work_struct *work) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1866 | { |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1867 | struct pool_workqueue *pwq = get_work_pwq(work); |
| 1868 | struct workqueue_struct *wq = pwq->wq; |
Tejun Heo | 493a172 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 1869 | |
Tejun Heo | 2e109a2 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 1870 | lockdep_assert_held(&wq_mayday_lock); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1871 | |
Tejun Heo | 493008a | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 1872 | if (!wq->rescuer) |
Tejun Heo | 493a172 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 1873 | return; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1874 | |
| 1875 | /* mayday mayday mayday */ |
Tejun Heo | 493a172 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 1876 | if (list_empty(&pwq->mayday_node)) { |
Lai Jiangshan | 77668c8 | 2014-04-18 11:04:16 -0400 | [diff] [blame] | 1877 | /* |
| 1878 | * If @pwq is for an unbound wq, its base ref may be put at |
| 1879 | * any time due to an attribute change. Pin @pwq until the |
| 1880 | * rescuer is done with it. |
| 1881 | */ |
| 1882 | get_pwq(pwq); |
Tejun Heo | 493a172 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 1883 | list_add_tail(&pwq->mayday_node, &wq->maydays); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1884 | wake_up_process(wq->rescuer->task); |
Tejun Heo | 493a172 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 1885 | } |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1886 | } |
| 1887 | |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 1888 | static void pool_mayday_timeout(unsigned long __pool) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1889 | { |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1890 | struct worker_pool *pool = (void *)__pool; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1891 | struct work_struct *work; |
| 1892 | |
Tejun Heo | b2d8290 | 2014-12-08 12:39:16 -0500 | [diff] [blame] | 1893 | spin_lock_irq(&pool->lock); |
| 1894 | spin_lock(&wq_mayday_lock); /* for wq->maydays */ |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1895 | |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1896 | if (need_to_create_worker(pool)) { |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1897 | /* |
| 1898 | * We've been trying to create a new worker but |
| 1899 | * haven't been successful. We might be hitting an |
| 1900 | * allocation deadlock. Send distress signals to |
| 1901 | * rescuers. |
| 1902 | */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1903 | list_for_each_entry(work, &pool->worklist, entry) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1904 | send_mayday(work); |
| 1905 | } |
| 1906 | |
Tejun Heo | b2d8290 | 2014-12-08 12:39:16 -0500 | [diff] [blame] | 1907 | spin_unlock(&wq_mayday_lock); |
| 1908 | spin_unlock_irq(&pool->lock); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1909 | |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1910 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INTERVAL); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1911 | } |
| 1912 | |
| 1913 | /** |
| 1914 | * maybe_create_worker - create a new worker if necessary |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1915 | * @pool: pool to create a new worker for |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1916 | * |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1917 | * Create a new worker for @pool if necessary. @pool is guaranteed to |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1918 | * have at least one idle worker on return from this function. If |
| 1919 | * creating a new worker takes longer than MAYDAY_INTERVAL, mayday is |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1920 | * sent to all rescuers with works scheduled on @pool to resolve |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1921 | * possible allocation deadlock. |
| 1922 | * |
Tejun Heo | c5aa87b | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 1923 | * On return, need_to_create_worker() is guaranteed to be %false and |
| 1924 | * may_start_working() %true. |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1925 | * |
| 1926 | * LOCKING: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1927 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1928 | * multiple times. Does GFP_KERNEL allocations. Called only from |
| 1929 | * manager. |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1930 | */ |
Tejun Heo | 29187a9 | 2015-01-16 14:21:16 -0500 | [diff] [blame] | 1931 | static void maybe_create_worker(struct worker_pool *pool) |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1932 | __releases(&pool->lock) |
| 1933 | __acquires(&pool->lock) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1934 | { |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1935 | restart: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1936 | spin_unlock_irq(&pool->lock); |
Tejun Heo | 9f9c236 | 2010-07-14 11:31:20 +0200 | [diff] [blame] | 1937 | |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1938 | /* if we don't make progress in MAYDAY_INITIAL_TIMEOUT, call for help */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1939 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INITIAL_TIMEOUT); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1940 | |
| 1941 | while (true) { |
Lai Jiangshan | 051e185 | 2014-07-22 13:03:02 +0800 | [diff] [blame] | 1942 | if (create_worker(pool) || !need_to_create_worker(pool)) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1943 | break; |
| 1944 | |
Lai Jiangshan | e212f36 | 2014-06-03 15:32:17 +0800 | [diff] [blame] | 1945 | schedule_timeout_interruptible(CREATE_COOLDOWN); |
Tejun Heo | 9f9c236 | 2010-07-14 11:31:20 +0200 | [diff] [blame] | 1946 | |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1947 | if (!need_to_create_worker(pool)) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1948 | break; |
| 1949 | } |
| 1950 | |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1951 | del_timer_sync(&pool->mayday_timer); |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1952 | spin_lock_irq(&pool->lock); |
Lai Jiangshan | 051e185 | 2014-07-22 13:03:02 +0800 | [diff] [blame] | 1953 | /* |
| 1954 | * This is necessary even after a new worker was just successfully |
| 1955 | * created as @pool->lock was dropped and the new worker might have |
| 1956 | * already become busy. |
| 1957 | */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1958 | if (need_to_create_worker(pool)) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1959 | goto restart; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1960 | } |
| 1961 | |
| 1962 | /** |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1963 | * manage_workers - manage worker pool |
| 1964 | * @worker: self |
| 1965 | * |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 1966 | * Assume the manager role and manage the worker pool @worker belongs |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1967 | * to. At any given time, there can be only zero or one manager per |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 1968 | * pool. The exclusion is handled automatically by this function. |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1969 | * |
| 1970 | * The caller can safely start processing works on false return. On |
| 1971 | * true return, it's guaranteed that need_to_create_worker() is false |
| 1972 | * and may_start_working() is true. |
| 1973 | * |
| 1974 | * CONTEXT: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1975 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1976 | * multiple times. Does GFP_KERNEL allocations. |
| 1977 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 1978 | * Return: |
Tejun Heo | 29187a9 | 2015-01-16 14:21:16 -0500 | [diff] [blame] | 1979 | * %false if the pool doesn't need management and the caller can safely |
| 1980 | * start processing works, %true if management function was performed and |
| 1981 | * the conditions that the caller verified before calling the function may |
| 1982 | * no longer be true. |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1983 | */ |
| 1984 | static bool manage_workers(struct worker *worker) |
| 1985 | { |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1986 | struct worker_pool *pool = worker->pool; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1987 | |
Tejun Heo | 692b482 | 2017-10-09 08:04:13 -0700 | [diff] [blame] | 1988 | if (pool->flags & POOL_MANAGER_ACTIVE) |
Tejun Heo | 29187a9 | 2015-01-16 14:21:16 -0500 | [diff] [blame] | 1989 | return false; |
Tejun Heo | 692b482 | 2017-10-09 08:04:13 -0700 | [diff] [blame] | 1990 | |
| 1991 | pool->flags |= POOL_MANAGER_ACTIVE; |
Tejun Heo | 2607d7a | 2015-03-09 09:22:28 -0400 | [diff] [blame] | 1992 | pool->manager = worker; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1993 | |
Tejun Heo | 29187a9 | 2015-01-16 14:21:16 -0500 | [diff] [blame] | 1994 | maybe_create_worker(pool); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1995 | |
Tejun Heo | 2607d7a | 2015-03-09 09:22:28 -0400 | [diff] [blame] | 1996 | pool->manager = NULL; |
Tejun Heo | 692b482 | 2017-10-09 08:04:13 -0700 | [diff] [blame] | 1997 | pool->flags &= ~POOL_MANAGER_ACTIVE; |
| 1998 | wake_up(&wq_manager_wait); |
Tejun Heo | 29187a9 | 2015-01-16 14:21:16 -0500 | [diff] [blame] | 1999 | return true; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2000 | } |
| 2001 | |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2002 | /** |
| 2003 | * process_one_work - process single work |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2004 | * @worker: self |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2005 | * @work: work to process |
| 2006 | * |
| 2007 | * Process @work. This function contains all the logics necessary to |
| 2008 | * process a single work including synchronization against and |
| 2009 | * interaction with other workers on the same cpu, queueing and |
| 2010 | * flushing. As long as context requirement is met, any worker can |
| 2011 | * call this function to process a work. |
| 2012 | * |
| 2013 | * CONTEXT: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2014 | * spin_lock_irq(pool->lock) which is released and regrabbed. |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2015 | */ |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2016 | static void process_one_work(struct worker *worker, struct work_struct *work) |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2017 | __releases(&pool->lock) |
| 2018 | __acquires(&pool->lock) |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2019 | { |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2020 | struct pool_workqueue *pwq = get_work_pwq(work); |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2021 | struct worker_pool *pool = worker->pool; |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2022 | bool cpu_intensive = pwq->wq->flags & WQ_CPU_INTENSIVE; |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2023 | int work_color; |
Tejun Heo | 7e11629 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 2024 | struct worker *collision; |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2025 | #ifdef CONFIG_LOCKDEP |
| 2026 | /* |
| 2027 | * It is permissible to free the struct work_struct from |
| 2028 | * inside the function that is called from it, this we need to |
| 2029 | * take into account for lockdep too. To avoid bogus "held |
| 2030 | * lock freed" warnings as well as problems when looking into |
| 2031 | * work->lockdep_map, make a copy and use that here. |
| 2032 | */ |
Peter Zijlstra | 4d82a1d | 2012-05-15 08:06:19 -0700 | [diff] [blame] | 2033 | struct lockdep_map lockdep_map; |
| 2034 | |
| 2035 | lockdep_copy_map(&lockdep_map, &work->lockdep_map); |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2036 | #endif |
Lai Jiangshan | 807407c | 2014-06-03 15:33:28 +0800 | [diff] [blame] | 2037 | /* ensure we're on the correct CPU */ |
Lai Jiangshan | 85327af | 2014-06-03 15:33:28 +0800 | [diff] [blame] | 2038 | WARN_ON_ONCE(!(pool->flags & POOL_DISASSOCIATED) && |
Tejun Heo | ec22ca5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2039 | raw_smp_processor_id() != pool->cpu); |
Tejun Heo | 25511a4 | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 2040 | |
Tejun Heo | 7e11629 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 2041 | /* |
| 2042 | * A single work shouldn't be executed concurrently by |
| 2043 | * multiple workers on a single cpu. Check whether anyone is |
| 2044 | * already processing the work. If so, defer the work to the |
| 2045 | * currently executing one. |
| 2046 | */ |
Tejun Heo | c9e7cf2 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2047 | collision = find_worker_executing_work(pool, work); |
Tejun Heo | 7e11629 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 2048 | if (unlikely(collision)) { |
| 2049 | move_linked_works(work, &collision->scheduled, NULL); |
| 2050 | return; |
| 2051 | } |
| 2052 | |
Tejun Heo | 8930cab | 2012-08-03 10:30:45 -0700 | [diff] [blame] | 2053 | /* claim and dequeue */ |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2054 | debug_work_deactivate(work); |
Tejun Heo | c9e7cf2 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2055 | hash_add(pool->busy_hash, &worker->hentry, (unsigned long)work); |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2056 | worker->current_work = work; |
Tejun Heo | a2c1c57 | 2012-12-18 10:35:02 -0800 | [diff] [blame] | 2057 | worker->current_func = work->func; |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2058 | worker->current_pwq = pwq; |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2059 | work_color = get_work_color(work); |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 2060 | |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2061 | list_del_init(&work->entry); |
| 2062 | |
Tejun Heo | 649027d | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2063 | /* |
Lai Jiangshan | 228f1d0 | 2014-07-22 13:02:00 +0800 | [diff] [blame] | 2064 | * CPU intensive works don't participate in concurrency management. |
| 2065 | * They're the scheduler's responsibility. This takes @worker out |
| 2066 | * of concurrency management and the next code block will chain |
| 2067 | * execution of the pending work items. |
Tejun Heo | fb0e7be | 2010-06-29 10:07:15 +0200 | [diff] [blame] | 2068 | */ |
| 2069 | if (unlikely(cpu_intensive)) |
Lai Jiangshan | 228f1d0 | 2014-07-22 13:02:00 +0800 | [diff] [blame] | 2070 | worker_set_flags(worker, WORKER_CPU_INTENSIVE); |
Tejun Heo | fb0e7be | 2010-06-29 10:07:15 +0200 | [diff] [blame] | 2071 | |
Tejun Heo | 974271c | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2072 | /* |
Lai Jiangshan | a489a03 | 2014-07-22 13:01:59 +0800 | [diff] [blame] | 2073 | * Wake up another worker if necessary. The condition is always |
| 2074 | * false for normal per-cpu workers since nr_running would always |
| 2075 | * be >= 1 at this point. This is used to chain execution of the |
| 2076 | * pending work items for WORKER_NOT_RUNNING workers such as the |
Lai Jiangshan | 228f1d0 | 2014-07-22 13:02:00 +0800 | [diff] [blame] | 2077 | * UNBOUND and CPU_INTENSIVE ones. |
Tejun Heo | 974271c | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2078 | */ |
Lai Jiangshan | a489a03 | 2014-07-22 13:01:59 +0800 | [diff] [blame] | 2079 | if (need_more_worker(pool)) |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2080 | wake_up_worker(pool); |
Tejun Heo | 974271c | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2081 | |
Tejun Heo | 8930cab | 2012-08-03 10:30:45 -0700 | [diff] [blame] | 2082 | /* |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2083 | * Record the last pool and clear PENDING which should be the last |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2084 | * update to @work. Also, do this inside @pool->lock so that |
Tejun Heo | 23657bb | 2012-08-13 17:08:19 -0700 | [diff] [blame] | 2085 | * PENDING and queued state changes happen together while IRQ is |
| 2086 | * disabled. |
Tejun Heo | 8930cab | 2012-08-03 10:30:45 -0700 | [diff] [blame] | 2087 | */ |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2088 | set_work_pool_and_clear_pending(work, pool->id); |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2089 | |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2090 | spin_unlock_irq(&pool->lock); |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2091 | |
Peter Zijlstra | a1d1493 | 2017-08-23 12:52:32 +0200 | [diff] [blame] | 2092 | lock_map_acquire(&pwq->wq->lockdep_map); |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2093 | lock_map_acquire(&lockdep_map); |
Peter Zijlstra | e6f3faa | 2017-08-23 13:23:30 +0200 | [diff] [blame] | 2094 | /* |
Peter Zijlstra | f52be57 | 2017-08-29 10:59:39 +0200 | [diff] [blame] | 2095 | * Strictly speaking we should mark the invariant state without holding |
| 2096 | * any locks, that is, before these two lock_map_acquire()'s. |
Peter Zijlstra | e6f3faa | 2017-08-23 13:23:30 +0200 | [diff] [blame] | 2097 | * |
| 2098 | * However, that would result in: |
| 2099 | * |
| 2100 | * A(W1) |
| 2101 | * WFC(C) |
| 2102 | * A(W1) |
| 2103 | * C(C) |
| 2104 | * |
| 2105 | * Which would create W1->C->W1 dependencies, even though there is no |
| 2106 | * actual deadlock possible. There are two solutions, using a |
| 2107 | * read-recursive acquire on the work(queue) 'locks', but this will then |
Peter Zijlstra | f52be57 | 2017-08-29 10:59:39 +0200 | [diff] [blame] | 2108 | * hit the lockdep limitation on recursive locks, or simply discard |
Peter Zijlstra | e6f3faa | 2017-08-23 13:23:30 +0200 | [diff] [blame] | 2109 | * these locks. |
| 2110 | * |
| 2111 | * AFAICT there is no possible deadlock scenario between the |
| 2112 | * flush_work() and complete() primitives (except for single-threaded |
| 2113 | * workqueues), so hiding them isn't a problem. |
| 2114 | */ |
Peter Zijlstra | f52be57 | 2017-08-29 10:59:39 +0200 | [diff] [blame] | 2115 | lockdep_invariant_state(true); |
Arjan van de Ven | e36c886 | 2010-08-21 13:07:26 -0700 | [diff] [blame] | 2116 | trace_workqueue_execute_start(work); |
Tejun Heo | a2c1c57 | 2012-12-18 10:35:02 -0800 | [diff] [blame] | 2117 | worker->current_func(work); |
Arjan van de Ven | e36c886 | 2010-08-21 13:07:26 -0700 | [diff] [blame] | 2118 | /* |
| 2119 | * While we must be careful to not use "work" after this, the trace |
| 2120 | * point will only record its address. |
| 2121 | */ |
| 2122 | trace_workqueue_execute_end(work); |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2123 | lock_map_release(&lockdep_map); |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2124 | lock_map_release(&pwq->wq->lockdep_map); |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2125 | |
| 2126 | if (unlikely(in_atomic() || lockdep_depth(current) > 0)) { |
Valentin Ilie | 044c782 | 2012-08-19 00:52:42 +0300 | [diff] [blame] | 2127 | pr_err("BUG: workqueue leaked lock or atomic: %s/0x%08x/%d\n" |
| 2128 | " last function: %pf\n", |
Tejun Heo | a2c1c57 | 2012-12-18 10:35:02 -0800 | [diff] [blame] | 2129 | current->comm, preempt_count(), task_pid_nr(current), |
| 2130 | worker->current_func); |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2131 | debug_show_held_locks(current); |
| 2132 | dump_stack(); |
| 2133 | } |
| 2134 | |
Tejun Heo | b22ce27 | 2013-08-28 17:33:37 -0400 | [diff] [blame] | 2135 | /* |
| 2136 | * The following prevents a kworker from hogging CPU on !PREEMPT |
| 2137 | * kernels, where a requeueing work item waiting for something to |
| 2138 | * happen could deadlock with stop_machine as such work item could |
| 2139 | * indefinitely requeue itself while all other CPUs are trapped in |
Joe Lawrence | 789cbbe | 2014-10-05 13:24:21 -0400 | [diff] [blame] | 2140 | * stop_machine. At the same time, report a quiescent RCU state so |
| 2141 | * the same condition doesn't freeze RCU. |
Tejun Heo | b22ce27 | 2013-08-28 17:33:37 -0400 | [diff] [blame] | 2142 | */ |
Joe Lawrence | 3e28e37 | 2014-10-05 13:24:22 -0400 | [diff] [blame] | 2143 | cond_resched_rcu_qs(); |
Tejun Heo | b22ce27 | 2013-08-28 17:33:37 -0400 | [diff] [blame] | 2144 | |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2145 | spin_lock_irq(&pool->lock); |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2146 | |
Tejun Heo | fb0e7be | 2010-06-29 10:07:15 +0200 | [diff] [blame] | 2147 | /* clear cpu intensive status */ |
| 2148 | if (unlikely(cpu_intensive)) |
| 2149 | worker_clr_flags(worker, WORKER_CPU_INTENSIVE); |
| 2150 | |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2151 | /* we're done with it, release */ |
Sasha Levin | 42f8570 | 2012-12-17 10:01:23 -0500 | [diff] [blame] | 2152 | hash_del(&worker->hentry); |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2153 | worker->current_work = NULL; |
Tejun Heo | a2c1c57 | 2012-12-18 10:35:02 -0800 | [diff] [blame] | 2154 | worker->current_func = NULL; |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2155 | worker->current_pwq = NULL; |
Tejun Heo | 3d1cb20 | 2013-04-30 15:27:22 -0700 | [diff] [blame] | 2156 | worker->desc_valid = false; |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2157 | pwq_dec_nr_in_flight(pwq, work_color); |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2158 | } |
| 2159 | |
Tejun Heo | affee4b | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2160 | /** |
| 2161 | * process_scheduled_works - process scheduled works |
| 2162 | * @worker: self |
| 2163 | * |
| 2164 | * Process all scheduled works. Please note that the scheduled list |
| 2165 | * may change while processing a work, so this function repeatedly |
| 2166 | * fetches a work from the top and executes it. |
| 2167 | * |
| 2168 | * CONTEXT: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2169 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
Tejun Heo | affee4b | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2170 | * multiple times. |
| 2171 | */ |
| 2172 | static void process_scheduled_works(struct worker *worker) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2173 | { |
Tejun Heo | affee4b | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2174 | while (!list_empty(&worker->scheduled)) { |
| 2175 | struct work_struct *work = list_first_entry(&worker->scheduled, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2176 | struct work_struct, entry); |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2177 | process_one_work(worker, work); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2178 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2179 | } |
| 2180 | |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2181 | /** |
| 2182 | * worker_thread - the worker thread function |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2183 | * @__worker: self |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2184 | * |
Tejun Heo | c5aa87b | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 2185 | * The worker thread function. All workers belong to a worker_pool - |
| 2186 | * either a per-cpu one or dynamic unbound one. These workers process all |
| 2187 | * work items regardless of their specific target workqueue. The only |
| 2188 | * exception is work items which belong to workqueues with a rescuer which |
| 2189 | * will be explained in rescuer_thread(). |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 2190 | * |
| 2191 | * Return: 0 |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2192 | */ |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2193 | static int worker_thread(void *__worker) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2194 | { |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2195 | struct worker *worker = __worker; |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2196 | struct worker_pool *pool = worker->pool; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2197 | |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2198 | /* tell the scheduler that this is a workqueue worker */ |
| 2199 | worker->task->flags |= PF_WQ_WORKER; |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2200 | woke_up: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2201 | spin_lock_irq(&pool->lock); |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 2202 | |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 2203 | /* am I supposed to die? */ |
| 2204 | if (unlikely(worker->flags & WORKER_DIE)) { |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2205 | spin_unlock_irq(&pool->lock); |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 2206 | WARN_ON_ONCE(!list_empty(&worker->entry)); |
| 2207 | worker->task->flags &= ~PF_WQ_WORKER; |
Lai Jiangshan | 60f5a4b | 2014-05-20 17:46:29 +0800 | [diff] [blame] | 2208 | |
| 2209 | set_task_comm(worker->task, "kworker/dying"); |
Lai Jiangshan | 7cda9aa | 2014-05-20 17:46:32 +0800 | [diff] [blame] | 2210 | ida_simple_remove(&pool->worker_ida, worker->id); |
Lai Jiangshan | 60f5a4b | 2014-05-20 17:46:29 +0800 | [diff] [blame] | 2211 | worker_detach_from_pool(worker, pool); |
| 2212 | kfree(worker); |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 2213 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2214 | } |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 2215 | |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2216 | worker_leave_idle(worker); |
Tejun Heo | db7bccf | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2217 | recheck: |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2218 | /* no more worker necessary? */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2219 | if (!need_more_worker(pool)) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2220 | goto sleep; |
| 2221 | |
| 2222 | /* do we need to manage? */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2223 | if (unlikely(!may_start_working(pool)) && manage_workers(worker)) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2224 | goto recheck; |
| 2225 | |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2226 | /* |
| 2227 | * ->scheduled list can only be filled while a worker is |
| 2228 | * preparing to process a work or actually processing it. |
| 2229 | * Make sure nobody diddled with it while I was sleeping. |
| 2230 | */ |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 2231 | WARN_ON_ONCE(!list_empty(&worker->scheduled)); |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2232 | |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2233 | /* |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 2234 | * Finish PREP stage. We're guaranteed to have at least one idle |
| 2235 | * worker or that someone else has already assumed the manager |
| 2236 | * role. This is where @worker starts participating in concurrency |
| 2237 | * management if applicable and concurrency management is restored |
| 2238 | * after being rebound. See rebind_workers() for details. |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2239 | */ |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 2240 | worker_clr_flags(worker, WORKER_PREP | WORKER_REBOUND); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2241 | |
| 2242 | do { |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2243 | struct work_struct *work = |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2244 | list_first_entry(&pool->worklist, |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2245 | struct work_struct, entry); |
| 2246 | |
Tejun Heo | 82607adc | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 2247 | pool->watchdog_ts = jiffies; |
| 2248 | |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2249 | if (likely(!(*work_data_bits(work) & WORK_STRUCT_LINKED))) { |
| 2250 | /* optimization path, not strictly necessary */ |
| 2251 | process_one_work(worker, work); |
| 2252 | if (unlikely(!list_empty(&worker->scheduled))) |
| 2253 | process_scheduled_works(worker); |
| 2254 | } else { |
| 2255 | move_linked_works(work, &worker->scheduled, NULL); |
| 2256 | process_scheduled_works(worker); |
| 2257 | } |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2258 | } while (keep_working(pool)); |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2259 | |
Lai Jiangshan | 228f1d0 | 2014-07-22 13:02:00 +0800 | [diff] [blame] | 2260 | worker_set_flags(worker, WORKER_PREP); |
Tejun Heo | d313dd8 | 2010-07-02 10:03:51 +0200 | [diff] [blame] | 2261 | sleep: |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2262 | /* |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2263 | * pool->lock is held and there's no work to process and no need to |
| 2264 | * manage, sleep. Workers are woken up only while holding |
| 2265 | * pool->lock or from local cpu, so setting the current state |
| 2266 | * before releasing pool->lock is enough to prevent losing any |
| 2267 | * event. |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2268 | */ |
| 2269 | worker_enter_idle(worker); |
Peter Zijlstra | c5a94a6 | 2017-08-23 13:58:44 +0200 | [diff] [blame] | 2270 | __set_current_state(TASK_IDLE); |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2271 | spin_unlock_irq(&pool->lock); |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2272 | schedule(); |
| 2273 | goto woke_up; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2274 | } |
| 2275 | |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2276 | /** |
| 2277 | * rescuer_thread - the rescuer thread function |
Tejun Heo | 111c225 | 2013-01-17 17:16:24 -0800 | [diff] [blame] | 2278 | * @__rescuer: self |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2279 | * |
| 2280 | * Workqueue rescuer thread function. There's one rescuer for each |
Tejun Heo | 493008a | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 2281 | * workqueue which has WQ_MEM_RECLAIM set. |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2282 | * |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 2283 | * Regular work processing on a pool may block trying to create a new |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2284 | * worker which uses GFP_KERNEL allocation which has slight chance of |
| 2285 | * developing into deadlock if some works currently on the same queue |
| 2286 | * need to be processed to satisfy the GFP_KERNEL allocation. This is |
| 2287 | * the problem rescuer solves. |
| 2288 | * |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 2289 | * When such condition is possible, the pool summons rescuers of all |
| 2290 | * workqueues which have works queued on the pool and let them process |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2291 | * those works so that forward progress can be guaranteed. |
| 2292 | * |
| 2293 | * This should happen rarely. |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 2294 | * |
| 2295 | * Return: 0 |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2296 | */ |
Tejun Heo | 111c225 | 2013-01-17 17:16:24 -0800 | [diff] [blame] | 2297 | static int rescuer_thread(void *__rescuer) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2298 | { |
Tejun Heo | 111c225 | 2013-01-17 17:16:24 -0800 | [diff] [blame] | 2299 | struct worker *rescuer = __rescuer; |
| 2300 | struct workqueue_struct *wq = rescuer->rescue_wq; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2301 | struct list_head *scheduled = &rescuer->scheduled; |
Lai Jiangshan | 4d595b8 | 2014-04-18 11:04:16 -0400 | [diff] [blame] | 2302 | bool should_stop; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2303 | |
| 2304 | set_user_nice(current, RESCUER_NICE_LEVEL); |
Tejun Heo | 111c225 | 2013-01-17 17:16:24 -0800 | [diff] [blame] | 2305 | |
| 2306 | /* |
| 2307 | * Mark rescuer as worker too. As WORKER_PREP is never cleared, it |
| 2308 | * doesn't participate in concurrency management. |
| 2309 | */ |
| 2310 | rescuer->task->flags |= PF_WQ_WORKER; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2311 | repeat: |
Peter Zijlstra | c5a94a6 | 2017-08-23 13:58:44 +0200 | [diff] [blame] | 2312 | set_current_state(TASK_IDLE); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2313 | |
Lai Jiangshan | 4d595b8 | 2014-04-18 11:04:16 -0400 | [diff] [blame] | 2314 | /* |
| 2315 | * By the time the rescuer is requested to stop, the workqueue |
| 2316 | * shouldn't have any work pending, but @wq->maydays may still have |
| 2317 | * pwq(s) queued. This can happen by non-rescuer workers consuming |
| 2318 | * all the work items before the rescuer got to them. Go through |
| 2319 | * @wq->maydays processing before acting on should_stop so that the |
| 2320 | * list is always empty on exit. |
| 2321 | */ |
| 2322 | should_stop = kthread_should_stop(); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2323 | |
Tejun Heo | 493a172 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 2324 | /* see whether any pwq is asking for help */ |
Tejun Heo | 2e109a2 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 2325 | spin_lock_irq(&wq_mayday_lock); |
Tejun Heo | 493a172 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 2326 | |
| 2327 | while (!list_empty(&wq->maydays)) { |
| 2328 | struct pool_workqueue *pwq = list_first_entry(&wq->maydays, |
| 2329 | struct pool_workqueue, mayday_node); |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2330 | struct worker_pool *pool = pwq->pool; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2331 | struct work_struct *work, *n; |
Tejun Heo | 82607adc | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 2332 | bool first = true; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2333 | |
| 2334 | __set_current_state(TASK_RUNNING); |
Tejun Heo | 493a172 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 2335 | list_del_init(&pwq->mayday_node); |
| 2336 | |
Tejun Heo | 2e109a2 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 2337 | spin_unlock_irq(&wq_mayday_lock); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2338 | |
Lai Jiangshan | 51697d39 | 2014-05-20 17:46:36 +0800 | [diff] [blame] | 2339 | worker_attach_to_pool(rescuer, pool); |
| 2340 | |
| 2341 | spin_lock_irq(&pool->lock); |
Lai Jiangshan | b310410 | 2013-02-19 12:17:02 -0800 | [diff] [blame] | 2342 | rescuer->pool = pool; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2343 | |
| 2344 | /* |
| 2345 | * Slurp in all works issued via this workqueue and |
| 2346 | * process'em. |
| 2347 | */ |
Tejun Heo | 0479c8c | 2014-12-04 10:14:13 -0500 | [diff] [blame] | 2348 | WARN_ON_ONCE(!list_empty(scheduled)); |
Tejun Heo | 82607adc | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 2349 | list_for_each_entry_safe(work, n, &pool->worklist, entry) { |
| 2350 | if (get_work_pwq(work) == pwq) { |
| 2351 | if (first) |
| 2352 | pool->watchdog_ts = jiffies; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2353 | move_linked_works(work, scheduled, &n); |
Tejun Heo | 82607adc | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 2354 | } |
| 2355 | first = false; |
| 2356 | } |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2357 | |
NeilBrown | 008847f | 2014-12-08 12:39:16 -0500 | [diff] [blame] | 2358 | if (!list_empty(scheduled)) { |
| 2359 | process_scheduled_works(rescuer); |
| 2360 | |
| 2361 | /* |
| 2362 | * The above execution of rescued work items could |
| 2363 | * have created more to rescue through |
| 2364 | * pwq_activate_first_delayed() or chained |
| 2365 | * queueing. Let's put @pwq back on mayday list so |
| 2366 | * that such back-to-back work items, which may be |
| 2367 | * being used to relieve memory pressure, don't |
| 2368 | * incur MAYDAY_INTERVAL delay inbetween. |
| 2369 | */ |
| 2370 | if (need_to_create_worker(pool)) { |
| 2371 | spin_lock(&wq_mayday_lock); |
Tejun Heo | 9e0f33d | 2019-09-25 06:59:15 -0700 | [diff] [blame] | 2372 | /* |
| 2373 | * Queue iff we aren't racing destruction |
| 2374 | * and somebody else hasn't queued it already. |
| 2375 | */ |
| 2376 | if (wq->rescuer && list_empty(&pwq->mayday_node)) { |
| 2377 | get_pwq(pwq); |
| 2378 | list_add_tail(&pwq->mayday_node, &wq->maydays); |
| 2379 | } |
NeilBrown | 008847f | 2014-12-08 12:39:16 -0500 | [diff] [blame] | 2380 | spin_unlock(&wq_mayday_lock); |
| 2381 | } |
| 2382 | } |
Tejun Heo | 7576958 | 2011-02-14 14:04:46 +0100 | [diff] [blame] | 2383 | |
| 2384 | /* |
Lai Jiangshan | 77668c8 | 2014-04-18 11:04:16 -0400 | [diff] [blame] | 2385 | * Put the reference grabbed by send_mayday(). @pool won't |
Lai Jiangshan | 13b1d62 | 2014-07-22 13:03:47 +0800 | [diff] [blame] | 2386 | * go away while we're still attached to it. |
Lai Jiangshan | 77668c8 | 2014-04-18 11:04:16 -0400 | [diff] [blame] | 2387 | */ |
| 2388 | put_pwq(pwq); |
| 2389 | |
| 2390 | /* |
Lai Jiangshan | d8ca83e | 2014-07-16 14:56:36 +0800 | [diff] [blame] | 2391 | * Leave this pool. If need_more_worker() is %true, notify a |
Tejun Heo | 7576958 | 2011-02-14 14:04:46 +0100 | [diff] [blame] | 2392 | * regular worker; otherwise, we end up with 0 concurrency |
| 2393 | * and stalling the execution. |
| 2394 | */ |
Lai Jiangshan | d8ca83e | 2014-07-16 14:56:36 +0800 | [diff] [blame] | 2395 | if (need_more_worker(pool)) |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2396 | wake_up_worker(pool); |
Tejun Heo | 7576958 | 2011-02-14 14:04:46 +0100 | [diff] [blame] | 2397 | |
Lai Jiangshan | b310410 | 2013-02-19 12:17:02 -0800 | [diff] [blame] | 2398 | rescuer->pool = NULL; |
Lai Jiangshan | 13b1d62 | 2014-07-22 13:03:47 +0800 | [diff] [blame] | 2399 | spin_unlock_irq(&pool->lock); |
| 2400 | |
| 2401 | worker_detach_from_pool(rescuer, pool); |
| 2402 | |
| 2403 | spin_lock_irq(&wq_mayday_lock); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2404 | } |
| 2405 | |
Tejun Heo | 2e109a2 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 2406 | spin_unlock_irq(&wq_mayday_lock); |
Tejun Heo | 493a172 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 2407 | |
Lai Jiangshan | 4d595b8 | 2014-04-18 11:04:16 -0400 | [diff] [blame] | 2408 | if (should_stop) { |
| 2409 | __set_current_state(TASK_RUNNING); |
| 2410 | rescuer->task->flags &= ~PF_WQ_WORKER; |
| 2411 | return 0; |
| 2412 | } |
| 2413 | |
Tejun Heo | 111c225 | 2013-01-17 17:16:24 -0800 | [diff] [blame] | 2414 | /* rescuers should never participate in concurrency management */ |
| 2415 | WARN_ON_ONCE(!(rescuer->flags & WORKER_NOT_RUNNING)); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2416 | schedule(); |
| 2417 | goto repeat; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2418 | } |
| 2419 | |
Tejun Heo | fca839c | 2015-12-07 10:58:57 -0500 | [diff] [blame] | 2420 | /** |
| 2421 | * check_flush_dependency - check for flush dependency sanity |
| 2422 | * @target_wq: workqueue being flushed |
| 2423 | * @target_work: work item being flushed (NULL for workqueue flushes) |
| 2424 | * |
| 2425 | * %current is trying to flush the whole @target_wq or @target_work on it. |
| 2426 | * If @target_wq doesn't have %WQ_MEM_RECLAIM, verify that %current is not |
| 2427 | * reclaiming memory or running on a workqueue which doesn't have |
| 2428 | * %WQ_MEM_RECLAIM as that can break forward-progress guarantee leading to |
| 2429 | * a deadlock. |
| 2430 | */ |
| 2431 | static void check_flush_dependency(struct workqueue_struct *target_wq, |
| 2432 | struct work_struct *target_work) |
| 2433 | { |
| 2434 | work_func_t target_func = target_work ? target_work->func : NULL; |
| 2435 | struct worker *worker; |
| 2436 | |
| 2437 | if (target_wq->flags & WQ_MEM_RECLAIM) |
| 2438 | return; |
| 2439 | |
| 2440 | worker = current_wq_worker(); |
| 2441 | |
| 2442 | WARN_ONCE(current->flags & PF_MEMALLOC, |
| 2443 | "workqueue: PF_MEMALLOC task %d(%s) is flushing !WQ_MEM_RECLAIM %s:%pf", |
| 2444 | current->pid, current->comm, target_wq->name, target_func); |
Tejun Heo | 23d11a5 | 2016-01-29 05:59:46 -0500 | [diff] [blame] | 2445 | WARN_ONCE(worker && ((worker->current_pwq->wq->flags & |
| 2446 | (WQ_MEM_RECLAIM | __WQ_LEGACY)) == WQ_MEM_RECLAIM), |
Tejun Heo | fca839c | 2015-12-07 10:58:57 -0500 | [diff] [blame] | 2447 | "workqueue: WQ_MEM_RECLAIM %s:%pf is flushing !WQ_MEM_RECLAIM %s:%pf", |
| 2448 | worker->current_pwq->wq->name, worker->current_func, |
| 2449 | target_wq->name, target_func); |
| 2450 | } |
| 2451 | |
Oleg Nesterov | fc2e4d7 | 2007-05-09 02:33:51 -0700 | [diff] [blame] | 2452 | struct wq_barrier { |
| 2453 | struct work_struct work; |
| 2454 | struct completion done; |
Tejun Heo | 2607d7a | 2015-03-09 09:22:28 -0400 | [diff] [blame] | 2455 | struct task_struct *task; /* purely informational */ |
Oleg Nesterov | fc2e4d7 | 2007-05-09 02:33:51 -0700 | [diff] [blame] | 2456 | }; |
| 2457 | |
| 2458 | static void wq_barrier_func(struct work_struct *work) |
| 2459 | { |
| 2460 | struct wq_barrier *barr = container_of(work, struct wq_barrier, work); |
| 2461 | complete(&barr->done); |
| 2462 | } |
| 2463 | |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2464 | /** |
| 2465 | * insert_wq_barrier - insert a barrier work |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2466 | * @pwq: pwq to insert barrier into |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2467 | * @barr: wq_barrier to insert |
Tejun Heo | affee4b | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2468 | * @target: target work to attach @barr to |
| 2469 | * @worker: worker currently executing @target, NULL if @target is not executing |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2470 | * |
Tejun Heo | affee4b | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2471 | * @barr is linked to @target such that @barr is completed only after |
| 2472 | * @target finishes execution. Please note that the ordering |
| 2473 | * guarantee is observed only with respect to @target and on the local |
| 2474 | * cpu. |
| 2475 | * |
| 2476 | * Currently, a queued barrier can't be canceled. This is because |
| 2477 | * try_to_grab_pending() can't determine whether the work to be |
| 2478 | * grabbed is at the head of the queue and thus can't clear LINKED |
| 2479 | * flag of the previous work while there must be a valid next work |
| 2480 | * after a work with LINKED flag set. |
| 2481 | * |
| 2482 | * Note that when @worker is non-NULL, @target may be modified |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2483 | * underneath us, so we can't reliably determine pwq from @target. |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2484 | * |
| 2485 | * CONTEXT: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2486 | * spin_lock_irq(pool->lock). |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2487 | */ |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2488 | static void insert_wq_barrier(struct pool_workqueue *pwq, |
Tejun Heo | affee4b | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2489 | struct wq_barrier *barr, |
| 2490 | struct work_struct *target, struct worker *worker) |
Oleg Nesterov | fc2e4d7 | 2007-05-09 02:33:51 -0700 | [diff] [blame] | 2491 | { |
Tejun Heo | affee4b | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2492 | struct list_head *head; |
| 2493 | unsigned int linked = 0; |
| 2494 | |
Thomas Gleixner | dc186ad | 2009-11-16 01:09:48 +0900 | [diff] [blame] | 2495 | /* |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2496 | * debugobject calls are safe here even with pool->lock locked |
Thomas Gleixner | dc186ad | 2009-11-16 01:09:48 +0900 | [diff] [blame] | 2497 | * as we know for sure that this will not trigger any of the |
| 2498 | * checks and call back into the fixup functions where we |
| 2499 | * might deadlock. |
| 2500 | */ |
Andrew Morton | ca1cab3 | 2010-10-26 14:22:34 -0700 | [diff] [blame] | 2501 | INIT_WORK_ONSTACK(&barr->work, wq_barrier_func); |
Tejun Heo | 22df02b | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2502 | __set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(&barr->work)); |
Boqun Feng | 52fa5bc | 2017-08-17 17:46:12 +0800 | [diff] [blame] | 2503 | |
| 2504 | /* |
| 2505 | * Explicitly init the crosslock for wq_barrier::done, make its lock |
| 2506 | * key a subkey of the corresponding work. As a result we won't |
| 2507 | * build a dependency between wq_barrier::done and unrelated work. |
| 2508 | */ |
| 2509 | lockdep_init_map_crosslock((struct lockdep_map *)&barr->done.map, |
| 2510 | "(complete)wq_barr::done", |
| 2511 | target->lockdep_map.key, 1); |
| 2512 | __init_completion(&barr->done); |
Tejun Heo | 2607d7a | 2015-03-09 09:22:28 -0400 | [diff] [blame] | 2513 | barr->task = current; |
Oleg Nesterov | 83c2252 | 2007-05-09 02:33:54 -0700 | [diff] [blame] | 2514 | |
Tejun Heo | affee4b | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2515 | /* |
| 2516 | * If @target is currently being executed, schedule the |
| 2517 | * barrier to the worker; otherwise, put it after @target. |
| 2518 | */ |
| 2519 | if (worker) |
| 2520 | head = worker->scheduled.next; |
| 2521 | else { |
| 2522 | unsigned long *bits = work_data_bits(target); |
| 2523 | |
| 2524 | head = target->entry.next; |
| 2525 | /* there can already be other linked works, inherit and set */ |
| 2526 | linked = *bits & WORK_STRUCT_LINKED; |
| 2527 | __set_bit(WORK_STRUCT_LINKED_BIT, bits); |
| 2528 | } |
| 2529 | |
Thomas Gleixner | dc186ad | 2009-11-16 01:09:48 +0900 | [diff] [blame] | 2530 | debug_work_activate(&barr->work); |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2531 | insert_work(pwq, &barr->work, head, |
Tejun Heo | affee4b | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2532 | work_color_to_flags(WORK_NO_COLOR) | linked); |
Oleg Nesterov | fc2e4d7 | 2007-05-09 02:33:51 -0700 | [diff] [blame] | 2533 | } |
| 2534 | |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2535 | /** |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2536 | * flush_workqueue_prep_pwqs - prepare pwqs for workqueue flushing |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2537 | * @wq: workqueue being flushed |
| 2538 | * @flush_color: new flush color, < 0 for no-op |
| 2539 | * @work_color: new work color, < 0 for no-op |
| 2540 | * |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2541 | * Prepare pwqs for workqueue flushing. |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2542 | * |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2543 | * If @flush_color is non-negative, flush_color on all pwqs should be |
| 2544 | * -1. If no pwq has in-flight commands at the specified color, all |
| 2545 | * pwq->flush_color's stay at -1 and %false is returned. If any pwq |
| 2546 | * has in flight commands, its pwq->flush_color is set to |
| 2547 | * @flush_color, @wq->nr_pwqs_to_flush is updated accordingly, pwq |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2548 | * wakeup logic is armed and %true is returned. |
| 2549 | * |
| 2550 | * The caller should have initialized @wq->first_flusher prior to |
| 2551 | * calling this function with non-negative @flush_color. If |
| 2552 | * @flush_color is negative, no flush color update is done and %false |
| 2553 | * is returned. |
| 2554 | * |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2555 | * If @work_color is non-negative, all pwqs should have the same |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2556 | * work_color which is previous to @work_color and all will be |
| 2557 | * advanced to @work_color. |
| 2558 | * |
| 2559 | * CONTEXT: |
Lai Jiangshan | 3c25a55 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 2560 | * mutex_lock(wq->mutex). |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2561 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 2562 | * Return: |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2563 | * %true if @flush_color >= 0 and there's something to flush. %false |
| 2564 | * otherwise. |
| 2565 | */ |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2566 | static bool flush_workqueue_prep_pwqs(struct workqueue_struct *wq, |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2567 | int flush_color, int work_color) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2568 | { |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2569 | bool wait = false; |
Tejun Heo | 49e3cf4 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 2570 | struct pool_workqueue *pwq; |
Oleg Nesterov | 1444196 | 2007-05-23 13:57:57 -0700 | [diff] [blame] | 2571 | |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2572 | if (flush_color >= 0) { |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 2573 | WARN_ON_ONCE(atomic_read(&wq->nr_pwqs_to_flush)); |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2574 | atomic_set(&wq->nr_pwqs_to_flush, 1); |
Thomas Gleixner | dc186ad | 2009-11-16 01:09:48 +0900 | [diff] [blame] | 2575 | } |
Oleg Nesterov | 1444196 | 2007-05-23 13:57:57 -0700 | [diff] [blame] | 2576 | |
Tejun Heo | 49e3cf4 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 2577 | for_each_pwq(pwq, wq) { |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2578 | struct worker_pool *pool = pwq->pool; |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2579 | |
Lai Jiangshan | b09f4fd | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 2580 | spin_lock_irq(&pool->lock); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2581 | |
| 2582 | if (flush_color >= 0) { |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 2583 | WARN_ON_ONCE(pwq->flush_color != -1); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2584 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2585 | if (pwq->nr_in_flight[flush_color]) { |
| 2586 | pwq->flush_color = flush_color; |
| 2587 | atomic_inc(&wq->nr_pwqs_to_flush); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2588 | wait = true; |
| 2589 | } |
| 2590 | } |
| 2591 | |
| 2592 | if (work_color >= 0) { |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 2593 | WARN_ON_ONCE(work_color != work_next_color(pwq->work_color)); |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2594 | pwq->work_color = work_color; |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2595 | } |
| 2596 | |
Lai Jiangshan | b09f4fd | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 2597 | spin_unlock_irq(&pool->lock); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2598 | } |
| 2599 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2600 | if (flush_color >= 0 && atomic_dec_and_test(&wq->nr_pwqs_to_flush)) |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2601 | complete(&wq->first_flusher->done); |
| 2602 | |
| 2603 | return wait; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2604 | } |
| 2605 | |
Rolf Eike Beer | 0fcb78c | 2006-07-30 03:03:42 -0700 | [diff] [blame] | 2606 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2607 | * flush_workqueue - ensure that any scheduled work has run to completion. |
Rolf Eike Beer | 0fcb78c | 2006-07-30 03:03:42 -0700 | [diff] [blame] | 2608 | * @wq: workqueue to flush |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2609 | * |
Tejun Heo | c5aa87b | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 2610 | * This function sleeps until all work items which were queued on entry |
| 2611 | * have finished execution, but it is not livelocked by new incoming ones. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2612 | */ |
Harvey Harrison | 7ad5b3a | 2008-02-08 04:19:53 -0800 | [diff] [blame] | 2613 | void flush_workqueue(struct workqueue_struct *wq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2614 | { |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2615 | struct wq_flusher this_flusher = { |
| 2616 | .list = LIST_HEAD_INIT(this_flusher.list), |
| 2617 | .flush_color = -1, |
| 2618 | .done = COMPLETION_INITIALIZER_ONSTACK(this_flusher.done), |
| 2619 | }; |
| 2620 | int next_color; |
Oleg Nesterov | b1f4ec1 | 2007-05-09 02:34:12 -0700 | [diff] [blame] | 2621 | |
Tejun Heo | 3347fa0 | 2016-09-16 15:49:32 -0400 | [diff] [blame] | 2622 | if (WARN_ON(!wq_online)) |
| 2623 | return; |
| 2624 | |
Ingo Molnar | 3295f0e | 2008-08-11 10:30:30 +0200 | [diff] [blame] | 2625 | lock_map_acquire(&wq->lockdep_map); |
| 2626 | lock_map_release(&wq->lockdep_map); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2627 | |
Lai Jiangshan | 3c25a55 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 2628 | mutex_lock(&wq->mutex); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2629 | |
| 2630 | /* |
| 2631 | * Start-to-wait phase |
| 2632 | */ |
| 2633 | next_color = work_next_color(wq->work_color); |
| 2634 | |
| 2635 | if (next_color != wq->flush_color) { |
| 2636 | /* |
| 2637 | * Color space is not full. The current work_color |
| 2638 | * becomes our flush_color and work_color is advanced |
| 2639 | * by one. |
| 2640 | */ |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 2641 | WARN_ON_ONCE(!list_empty(&wq->flusher_overflow)); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2642 | this_flusher.flush_color = wq->work_color; |
| 2643 | wq->work_color = next_color; |
| 2644 | |
| 2645 | if (!wq->first_flusher) { |
| 2646 | /* no flush in progress, become the first flusher */ |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 2647 | WARN_ON_ONCE(wq->flush_color != this_flusher.flush_color); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2648 | |
| 2649 | wq->first_flusher = &this_flusher; |
| 2650 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2651 | if (!flush_workqueue_prep_pwqs(wq, wq->flush_color, |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2652 | wq->work_color)) { |
| 2653 | /* nothing to flush, done */ |
| 2654 | wq->flush_color = next_color; |
| 2655 | wq->first_flusher = NULL; |
| 2656 | goto out_unlock; |
| 2657 | } |
| 2658 | } else { |
| 2659 | /* wait in queue */ |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 2660 | WARN_ON_ONCE(wq->flush_color == this_flusher.flush_color); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2661 | list_add_tail(&this_flusher.list, &wq->flusher_queue); |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2662 | flush_workqueue_prep_pwqs(wq, -1, wq->work_color); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2663 | } |
| 2664 | } else { |
| 2665 | /* |
| 2666 | * Oops, color space is full, wait on overflow queue. |
| 2667 | * The next flush completion will assign us |
| 2668 | * flush_color and transfer to flusher_queue. |
| 2669 | */ |
| 2670 | list_add_tail(&this_flusher.list, &wq->flusher_overflow); |
| 2671 | } |
| 2672 | |
Tejun Heo | fca839c | 2015-12-07 10:58:57 -0500 | [diff] [blame] | 2673 | check_flush_dependency(wq, NULL); |
| 2674 | |
Lai Jiangshan | 3c25a55 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 2675 | mutex_unlock(&wq->mutex); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2676 | |
| 2677 | wait_for_completion(&this_flusher.done); |
| 2678 | |
| 2679 | /* |
| 2680 | * Wake-up-and-cascade phase |
| 2681 | * |
| 2682 | * First flushers are responsible for cascading flushes and |
| 2683 | * handling overflow. Non-first flushers can simply return. |
| 2684 | */ |
| 2685 | if (wq->first_flusher != &this_flusher) |
| 2686 | return; |
| 2687 | |
Lai Jiangshan | 3c25a55 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 2688 | mutex_lock(&wq->mutex); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2689 | |
Tejun Heo | 4ce48b3 | 2010-07-02 10:03:51 +0200 | [diff] [blame] | 2690 | /* we might have raced, check again with mutex held */ |
| 2691 | if (wq->first_flusher != &this_flusher) |
| 2692 | goto out_unlock; |
| 2693 | |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2694 | wq->first_flusher = NULL; |
| 2695 | |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 2696 | WARN_ON_ONCE(!list_empty(&this_flusher.list)); |
| 2697 | WARN_ON_ONCE(wq->flush_color != this_flusher.flush_color); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2698 | |
| 2699 | while (true) { |
| 2700 | struct wq_flusher *next, *tmp; |
| 2701 | |
| 2702 | /* complete all the flushers sharing the current flush color */ |
| 2703 | list_for_each_entry_safe(next, tmp, &wq->flusher_queue, list) { |
| 2704 | if (next->flush_color != wq->flush_color) |
| 2705 | break; |
| 2706 | list_del_init(&next->list); |
| 2707 | complete(&next->done); |
| 2708 | } |
| 2709 | |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 2710 | WARN_ON_ONCE(!list_empty(&wq->flusher_overflow) && |
| 2711 | wq->flush_color != work_next_color(wq->work_color)); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2712 | |
| 2713 | /* this flush_color is finished, advance by one */ |
| 2714 | wq->flush_color = work_next_color(wq->flush_color); |
| 2715 | |
| 2716 | /* one color has been freed, handle overflow queue */ |
| 2717 | if (!list_empty(&wq->flusher_overflow)) { |
| 2718 | /* |
| 2719 | * Assign the same color to all overflowed |
| 2720 | * flushers, advance work_color and append to |
| 2721 | * flusher_queue. This is the start-to-wait |
| 2722 | * phase for these overflowed flushers. |
| 2723 | */ |
| 2724 | list_for_each_entry(tmp, &wq->flusher_overflow, list) |
| 2725 | tmp->flush_color = wq->work_color; |
| 2726 | |
| 2727 | wq->work_color = work_next_color(wq->work_color); |
| 2728 | |
| 2729 | list_splice_tail_init(&wq->flusher_overflow, |
| 2730 | &wq->flusher_queue); |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2731 | flush_workqueue_prep_pwqs(wq, -1, wq->work_color); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2732 | } |
| 2733 | |
| 2734 | if (list_empty(&wq->flusher_queue)) { |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 2735 | WARN_ON_ONCE(wq->flush_color != wq->work_color); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2736 | break; |
| 2737 | } |
| 2738 | |
| 2739 | /* |
| 2740 | * Need to flush more colors. Make the next flusher |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2741 | * the new first flusher and arm pwqs. |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2742 | */ |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 2743 | WARN_ON_ONCE(wq->flush_color == wq->work_color); |
| 2744 | WARN_ON_ONCE(wq->flush_color != next->flush_color); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2745 | |
| 2746 | list_del_init(&next->list); |
| 2747 | wq->first_flusher = next; |
| 2748 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2749 | if (flush_workqueue_prep_pwqs(wq, wq->flush_color, -1)) |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2750 | break; |
| 2751 | |
| 2752 | /* |
| 2753 | * Meh... this color is already done, clear first |
| 2754 | * flusher and repeat cascading. |
| 2755 | */ |
| 2756 | wq->first_flusher = NULL; |
| 2757 | } |
| 2758 | |
| 2759 | out_unlock: |
Lai Jiangshan | 3c25a55 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 2760 | mutex_unlock(&wq->mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2761 | } |
Tim Gardner | 1dadafa | 2015-08-04 11:26:04 -0600 | [diff] [blame] | 2762 | EXPORT_SYMBOL(flush_workqueue); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2763 | |
Tejun Heo | 9c5a2ba | 2011-04-05 18:01:44 +0200 | [diff] [blame] | 2764 | /** |
| 2765 | * drain_workqueue - drain a workqueue |
| 2766 | * @wq: workqueue to drain |
| 2767 | * |
| 2768 | * Wait until the workqueue becomes empty. While draining is in progress, |
| 2769 | * only chain queueing is allowed. IOW, only currently pending or running |
| 2770 | * work items on @wq can queue further work items on it. @wq is flushed |
Chen Hanxiao | b749b1b | 2015-05-13 06:10:05 -0400 | [diff] [blame] | 2771 | * repeatedly until it becomes empty. The number of flushing is determined |
Tejun Heo | 9c5a2ba | 2011-04-05 18:01:44 +0200 | [diff] [blame] | 2772 | * by the depth of chaining and should be relatively short. Whine if it |
| 2773 | * takes too long. |
| 2774 | */ |
| 2775 | void drain_workqueue(struct workqueue_struct *wq) |
| 2776 | { |
| 2777 | unsigned int flush_cnt = 0; |
Tejun Heo | 49e3cf4 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 2778 | struct pool_workqueue *pwq; |
Tejun Heo | 9c5a2ba | 2011-04-05 18:01:44 +0200 | [diff] [blame] | 2779 | |
| 2780 | /* |
| 2781 | * __queue_work() needs to test whether there are drainers, is much |
| 2782 | * hotter than drain_workqueue() and already looks at @wq->flags. |
Tejun Heo | 618b01e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 2783 | * Use __WQ_DRAINING so that queue doesn't have to check nr_drainers. |
Tejun Heo | 9c5a2ba | 2011-04-05 18:01:44 +0200 | [diff] [blame] | 2784 | */ |
Lai Jiangshan | 87fc741 | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 2785 | mutex_lock(&wq->mutex); |
Tejun Heo | 9c5a2ba | 2011-04-05 18:01:44 +0200 | [diff] [blame] | 2786 | if (!wq->nr_drainers++) |
Tejun Heo | 618b01e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 2787 | wq->flags |= __WQ_DRAINING; |
Lai Jiangshan | 87fc741 | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 2788 | mutex_unlock(&wq->mutex); |
Tejun Heo | 9c5a2ba | 2011-04-05 18:01:44 +0200 | [diff] [blame] | 2789 | reflush: |
| 2790 | flush_workqueue(wq); |
| 2791 | |
Lai Jiangshan | b09f4fd | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 2792 | mutex_lock(&wq->mutex); |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 2793 | |
Tejun Heo | 49e3cf4 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 2794 | for_each_pwq(pwq, wq) { |
Thomas Tuttle | fa2563e | 2011-09-14 16:22:28 -0700 | [diff] [blame] | 2795 | bool drained; |
Tejun Heo | 9c5a2ba | 2011-04-05 18:01:44 +0200 | [diff] [blame] | 2796 | |
Lai Jiangshan | b09f4fd | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 2797 | spin_lock_irq(&pwq->pool->lock); |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2798 | drained = !pwq->nr_active && list_empty(&pwq->delayed_works); |
Lai Jiangshan | b09f4fd | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 2799 | spin_unlock_irq(&pwq->pool->lock); |
Thomas Tuttle | fa2563e | 2011-09-14 16:22:28 -0700 | [diff] [blame] | 2800 | |
| 2801 | if (drained) |
Tejun Heo | 9c5a2ba | 2011-04-05 18:01:44 +0200 | [diff] [blame] | 2802 | continue; |
| 2803 | |
| 2804 | if (++flush_cnt == 10 || |
| 2805 | (flush_cnt % 100 == 0 && flush_cnt <= 1000)) |
Tejun Heo | c5aa87b | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 2806 | pr_warn("workqueue %s: drain_workqueue() isn't complete after %u tries\n", |
Valentin Ilie | 044c782 | 2012-08-19 00:52:42 +0300 | [diff] [blame] | 2807 | wq->name, flush_cnt); |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 2808 | |
Lai Jiangshan | b09f4fd | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 2809 | mutex_unlock(&wq->mutex); |
Tejun Heo | 9c5a2ba | 2011-04-05 18:01:44 +0200 | [diff] [blame] | 2810 | goto reflush; |
| 2811 | } |
| 2812 | |
Tejun Heo | 9c5a2ba | 2011-04-05 18:01:44 +0200 | [diff] [blame] | 2813 | if (!--wq->nr_drainers) |
Tejun Heo | 618b01e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 2814 | wq->flags &= ~__WQ_DRAINING; |
Lai Jiangshan | 87fc741 | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 2815 | mutex_unlock(&wq->mutex); |
Tejun Heo | 9c5a2ba | 2011-04-05 18:01:44 +0200 | [diff] [blame] | 2816 | } |
| 2817 | EXPORT_SYMBOL_GPL(drain_workqueue); |
| 2818 | |
Tejun Heo | 606a502 | 2012-08-20 14:51:23 -0700 | [diff] [blame] | 2819 | static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr) |
Tejun Heo | baf5902 | 2010-09-16 10:42:16 +0200 | [diff] [blame] | 2820 | { |
| 2821 | struct worker *worker = NULL; |
Tejun Heo | c9e7cf2 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2822 | struct worker_pool *pool; |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2823 | struct pool_workqueue *pwq; |
Tejun Heo | baf5902 | 2010-09-16 10:42:16 +0200 | [diff] [blame] | 2824 | |
| 2825 | might_sleep(); |
Tejun Heo | baf5902 | 2010-09-16 10:42:16 +0200 | [diff] [blame] | 2826 | |
Tejun Heo | fa1b54e | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 2827 | local_irq_disable(); |
| 2828 | pool = get_work_pool(work); |
| 2829 | if (!pool) { |
| 2830 | local_irq_enable(); |
| 2831 | return false; |
| 2832 | } |
| 2833 | |
| 2834 | spin_lock(&pool->lock); |
Lai Jiangshan | 0b3dae6 | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 2835 | /* see the comment in try_to_grab_pending() with the same code */ |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2836 | pwq = get_work_pwq(work); |
| 2837 | if (pwq) { |
| 2838 | if (unlikely(pwq->pool != pool)) |
Tejun Heo | baf5902 | 2010-09-16 10:42:16 +0200 | [diff] [blame] | 2839 | goto already_gone; |
Tejun Heo | 606a502 | 2012-08-20 14:51:23 -0700 | [diff] [blame] | 2840 | } else { |
Tejun Heo | c9e7cf2 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2841 | worker = find_worker_executing_work(pool, work); |
Tejun Heo | baf5902 | 2010-09-16 10:42:16 +0200 | [diff] [blame] | 2842 | if (!worker) |
| 2843 | goto already_gone; |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2844 | pwq = worker->current_pwq; |
Tejun Heo | 606a502 | 2012-08-20 14:51:23 -0700 | [diff] [blame] | 2845 | } |
Tejun Heo | baf5902 | 2010-09-16 10:42:16 +0200 | [diff] [blame] | 2846 | |
Tejun Heo | fca839c | 2015-12-07 10:58:57 -0500 | [diff] [blame] | 2847 | check_flush_dependency(pwq->wq, work); |
| 2848 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2849 | insert_wq_barrier(pwq, barr, work, worker); |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2850 | spin_unlock_irq(&pool->lock); |
Tejun Heo | baf5902 | 2010-09-16 10:42:16 +0200 | [diff] [blame] | 2851 | |
Tejun Heo | e159489 | 2011-01-09 23:32:15 +0100 | [diff] [blame] | 2852 | /* |
Peter Zijlstra | a1d1493 | 2017-08-23 12:52:32 +0200 | [diff] [blame] | 2853 | * Force a lock recursion deadlock when using flush_work() inside a |
| 2854 | * single-threaded or rescuer equipped workqueue. |
| 2855 | * |
| 2856 | * For single threaded workqueues the deadlock happens when the work |
| 2857 | * is after the work issuing the flush_work(). For rescuer equipped |
| 2858 | * workqueues the deadlock happens when the rescuer stalls, blocking |
| 2859 | * forward progress. |
Tejun Heo | e159489 | 2011-01-09 23:32:15 +0100 | [diff] [blame] | 2860 | */ |
Peter Zijlstra | a1d1493 | 2017-08-23 12:52:32 +0200 | [diff] [blame] | 2861 | if (pwq->wq->saved_max_active == 1 || pwq->wq->rescuer) { |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2862 | lock_map_acquire(&pwq->wq->lockdep_map); |
Peter Zijlstra | a1d1493 | 2017-08-23 12:52:32 +0200 | [diff] [blame] | 2863 | lock_map_release(&pwq->wq->lockdep_map); |
| 2864 | } |
Tejun Heo | e159489 | 2011-01-09 23:32:15 +0100 | [diff] [blame] | 2865 | |
Tejun Heo | baf5902 | 2010-09-16 10:42:16 +0200 | [diff] [blame] | 2866 | return true; |
| 2867 | already_gone: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2868 | spin_unlock_irq(&pool->lock); |
Tejun Heo | baf5902 | 2010-09-16 10:42:16 +0200 | [diff] [blame] | 2869 | return false; |
| 2870 | } |
| 2871 | |
Oleg Nesterov | db70089 | 2008-07-25 01:47:49 -0700 | [diff] [blame] | 2872 | /** |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2873 | * flush_work - wait for a work to finish executing the last queueing instance |
| 2874 | * @work: the work to flush |
Oleg Nesterov | db70089 | 2008-07-25 01:47:49 -0700 | [diff] [blame] | 2875 | * |
Tejun Heo | 606a502 | 2012-08-20 14:51:23 -0700 | [diff] [blame] | 2876 | * Wait until @work has finished execution. @work is guaranteed to be idle |
| 2877 | * on return if it hasn't been requeued since flush started. |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2878 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 2879 | * Return: |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2880 | * %true if flush_work() waited for the work to finish execution, |
| 2881 | * %false if it was already idle. |
Oleg Nesterov | db70089 | 2008-07-25 01:47:49 -0700 | [diff] [blame] | 2882 | */ |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2883 | bool flush_work(struct work_struct *work) |
Oleg Nesterov | db70089 | 2008-07-25 01:47:49 -0700 | [diff] [blame] | 2884 | { |
Bjorn Helgaas | 12997d1 | 2013-11-18 11:00:29 -0700 | [diff] [blame] | 2885 | struct wq_barrier barr; |
| 2886 | |
Tejun Heo | 3347fa0 | 2016-09-16 15:49:32 -0400 | [diff] [blame] | 2887 | if (WARN_ON(!wq_online)) |
| 2888 | return false; |
| 2889 | |
Stephen Boyd | 0976dfc | 2012-04-20 17:28:50 -0700 | [diff] [blame] | 2890 | lock_map_acquire(&work->lockdep_map); |
| 2891 | lock_map_release(&work->lockdep_map); |
| 2892 | |
Bjorn Helgaas | 12997d1 | 2013-11-18 11:00:29 -0700 | [diff] [blame] | 2893 | if (start_flush_work(work, &barr)) { |
| 2894 | wait_for_completion(&barr.done); |
| 2895 | destroy_work_on_stack(&barr.work); |
| 2896 | return true; |
| 2897 | } else { |
| 2898 | return false; |
| 2899 | } |
Oleg Nesterov | db70089 | 2008-07-25 01:47:49 -0700 | [diff] [blame] | 2900 | } |
| 2901 | EXPORT_SYMBOL_GPL(flush_work); |
| 2902 | |
Tejun Heo | 8603e1b3 | 2015-03-05 08:04:13 -0500 | [diff] [blame] | 2903 | struct cwt_wait { |
Ingo Molnar | ac6424b | 2017-06-20 12:06:13 +0200 | [diff] [blame] | 2904 | wait_queue_entry_t wait; |
Tejun Heo | 8603e1b3 | 2015-03-05 08:04:13 -0500 | [diff] [blame] | 2905 | struct work_struct *work; |
| 2906 | }; |
| 2907 | |
Ingo Molnar | ac6424b | 2017-06-20 12:06:13 +0200 | [diff] [blame] | 2908 | static int cwt_wakefn(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) |
Tejun Heo | 8603e1b3 | 2015-03-05 08:04:13 -0500 | [diff] [blame] | 2909 | { |
| 2910 | struct cwt_wait *cwait = container_of(wait, struct cwt_wait, wait); |
| 2911 | |
| 2912 | if (cwait->work != key) |
| 2913 | return 0; |
| 2914 | return autoremove_wake_function(wait, mode, sync, key); |
| 2915 | } |
| 2916 | |
Tejun Heo | 36e227d | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 2917 | static bool __cancel_work_timer(struct work_struct *work, bool is_dwork) |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2918 | { |
Tejun Heo | 8603e1b3 | 2015-03-05 08:04:13 -0500 | [diff] [blame] | 2919 | static DECLARE_WAIT_QUEUE_HEAD(cancel_waitq); |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 2920 | unsigned long flags; |
Oleg Nesterov | 1f1f642 | 2007-07-15 23:41:44 -0700 | [diff] [blame] | 2921 | int ret; |
| 2922 | |
| 2923 | do { |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 2924 | ret = try_to_grab_pending(work, is_dwork, &flags); |
| 2925 | /* |
Tejun Heo | 8603e1b3 | 2015-03-05 08:04:13 -0500 | [diff] [blame] | 2926 | * If someone else is already canceling, wait for it to |
| 2927 | * finish. flush_work() doesn't work for PREEMPT_NONE |
| 2928 | * because we may get scheduled between @work's completion |
| 2929 | * and the other canceling task resuming and clearing |
| 2930 | * CANCELING - flush_work() will return false immediately |
| 2931 | * as @work is no longer busy, try_to_grab_pending() will |
| 2932 | * return -ENOENT as @work is still being canceled and the |
| 2933 | * other canceling task won't be able to clear CANCELING as |
| 2934 | * we're hogging the CPU. |
| 2935 | * |
| 2936 | * Let's wait for completion using a waitqueue. As this |
| 2937 | * may lead to the thundering herd problem, use a custom |
| 2938 | * wake function which matches @work along with exclusive |
| 2939 | * wait and wakeup. |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 2940 | */ |
Tejun Heo | 8603e1b3 | 2015-03-05 08:04:13 -0500 | [diff] [blame] | 2941 | if (unlikely(ret == -ENOENT)) { |
| 2942 | struct cwt_wait cwait; |
| 2943 | |
| 2944 | init_wait(&cwait.wait); |
| 2945 | cwait.wait.func = cwt_wakefn; |
| 2946 | cwait.work = work; |
| 2947 | |
| 2948 | prepare_to_wait_exclusive(&cancel_waitq, &cwait.wait, |
| 2949 | TASK_UNINTERRUPTIBLE); |
| 2950 | if (work_is_canceling(work)) |
| 2951 | schedule(); |
| 2952 | finish_wait(&cancel_waitq, &cwait.wait); |
| 2953 | } |
Oleg Nesterov | 1f1f642 | 2007-07-15 23:41:44 -0700 | [diff] [blame] | 2954 | } while (unlikely(ret < 0)); |
| 2955 | |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 2956 | /* tell other tasks trying to grab @work to back off */ |
| 2957 | mark_work_canceling(work); |
| 2958 | local_irq_restore(flags); |
| 2959 | |
Tejun Heo | 3347fa0 | 2016-09-16 15:49:32 -0400 | [diff] [blame] | 2960 | /* |
| 2961 | * This allows canceling during early boot. We know that @work |
| 2962 | * isn't executing. |
| 2963 | */ |
| 2964 | if (wq_online) |
| 2965 | flush_work(work); |
| 2966 | |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 2967 | clear_work_data(work); |
Tejun Heo | 8603e1b3 | 2015-03-05 08:04:13 -0500 | [diff] [blame] | 2968 | |
| 2969 | /* |
| 2970 | * Paired with prepare_to_wait() above so that either |
| 2971 | * waitqueue_active() is visible here or !work_is_canceling() is |
| 2972 | * visible there. |
| 2973 | */ |
| 2974 | smp_mb(); |
| 2975 | if (waitqueue_active(&cancel_waitq)) |
| 2976 | __wake_up(&cancel_waitq, TASK_NORMAL, 1, work); |
| 2977 | |
Oleg Nesterov | 1f1f642 | 2007-07-15 23:41:44 -0700 | [diff] [blame] | 2978 | return ret; |
| 2979 | } |
| 2980 | |
Oleg Nesterov | 6e84d64 | 2007-05-09 02:34:46 -0700 | [diff] [blame] | 2981 | /** |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2982 | * cancel_work_sync - cancel a work and wait for it to finish |
| 2983 | * @work: the work to cancel |
Oleg Nesterov | 6e84d64 | 2007-05-09 02:34:46 -0700 | [diff] [blame] | 2984 | * |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2985 | * Cancel @work and wait for its execution to finish. This function |
| 2986 | * can be used even if the work re-queues itself or migrates to |
| 2987 | * another workqueue. On return from this function, @work is |
| 2988 | * guaranteed to be not pending or executing on any CPU. |
Oleg Nesterov | 1f1f642 | 2007-07-15 23:41:44 -0700 | [diff] [blame] | 2989 | * |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2990 | * cancel_work_sync(&delayed_work->work) must not be used for |
| 2991 | * delayed_work's. Use cancel_delayed_work_sync() instead. |
Oleg Nesterov | 6e84d64 | 2007-05-09 02:34:46 -0700 | [diff] [blame] | 2992 | * |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2993 | * The caller must ensure that the workqueue on which @work was last |
Oleg Nesterov | 6e84d64 | 2007-05-09 02:34:46 -0700 | [diff] [blame] | 2994 | * queued can't be destroyed before this function returns. |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2995 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 2996 | * Return: |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2997 | * %true if @work was pending, %false otherwise. |
Oleg Nesterov | 6e84d64 | 2007-05-09 02:34:46 -0700 | [diff] [blame] | 2998 | */ |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2999 | bool cancel_work_sync(struct work_struct *work) |
Oleg Nesterov | 6e84d64 | 2007-05-09 02:34:46 -0700 | [diff] [blame] | 3000 | { |
Tejun Heo | 36e227d | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 3001 | return __cancel_work_timer(work, false); |
Oleg Nesterov | b89deed | 2007-05-09 02:33:52 -0700 | [diff] [blame] | 3002 | } |
Oleg Nesterov | 28e53bd | 2007-05-09 02:34:22 -0700 | [diff] [blame] | 3003 | EXPORT_SYMBOL_GPL(cancel_work_sync); |
Oleg Nesterov | b89deed | 2007-05-09 02:33:52 -0700 | [diff] [blame] | 3004 | |
Oleg Nesterov | 6e84d64 | 2007-05-09 02:34:46 -0700 | [diff] [blame] | 3005 | /** |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 3006 | * flush_delayed_work - wait for a dwork to finish executing the last queueing |
| 3007 | * @dwork: the delayed work to flush |
Oleg Nesterov | 6e84d64 | 2007-05-09 02:34:46 -0700 | [diff] [blame] | 3008 | * |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 3009 | * Delayed timer is cancelled and the pending work is queued for |
| 3010 | * immediate execution. Like flush_work(), this function only |
| 3011 | * considers the last queueing instance of @dwork. |
Oleg Nesterov | 1f1f642 | 2007-07-15 23:41:44 -0700 | [diff] [blame] | 3012 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 3013 | * Return: |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 3014 | * %true if flush_work() waited for the work to finish execution, |
| 3015 | * %false if it was already idle. |
Oleg Nesterov | 6e84d64 | 2007-05-09 02:34:46 -0700 | [diff] [blame] | 3016 | */ |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 3017 | bool flush_delayed_work(struct delayed_work *dwork) |
| 3018 | { |
Tejun Heo | 8930cab | 2012-08-03 10:30:45 -0700 | [diff] [blame] | 3019 | local_irq_disable(); |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 3020 | if (del_timer_sync(&dwork->timer)) |
Lai Jiangshan | 60c057b | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 3021 | __queue_work(dwork->cpu, dwork->wq, &dwork->work); |
Tejun Heo | 8930cab | 2012-08-03 10:30:45 -0700 | [diff] [blame] | 3022 | local_irq_enable(); |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 3023 | return flush_work(&dwork->work); |
| 3024 | } |
| 3025 | EXPORT_SYMBOL(flush_delayed_work); |
| 3026 | |
Jens Axboe | f72b879 | 2016-08-24 15:51:50 -0600 | [diff] [blame] | 3027 | static bool __cancel_work(struct work_struct *work, bool is_dwork) |
| 3028 | { |
| 3029 | unsigned long flags; |
| 3030 | int ret; |
| 3031 | |
| 3032 | do { |
| 3033 | ret = try_to_grab_pending(work, is_dwork, &flags); |
| 3034 | } while (unlikely(ret == -EAGAIN)); |
| 3035 | |
| 3036 | if (unlikely(ret < 0)) |
| 3037 | return false; |
| 3038 | |
| 3039 | set_work_pool_and_clear_pending(work, get_work_pool_id(work)); |
| 3040 | local_irq_restore(flags); |
| 3041 | return ret; |
| 3042 | } |
| 3043 | |
| 3044 | /* |
| 3045 | * See cancel_delayed_work() |
| 3046 | */ |
| 3047 | bool cancel_work(struct work_struct *work) |
| 3048 | { |
| 3049 | return __cancel_work(work, false); |
| 3050 | } |
| 3051 | |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 3052 | /** |
Tejun Heo | 57b30ae | 2012-08-21 13:18:24 -0700 | [diff] [blame] | 3053 | * cancel_delayed_work - cancel a delayed work |
| 3054 | * @dwork: delayed_work to cancel |
Tejun Heo | 0938349 | 2010-09-16 10:48:29 +0200 | [diff] [blame] | 3055 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 3056 | * Kill off a pending delayed_work. |
| 3057 | * |
| 3058 | * Return: %true if @dwork was pending and canceled; %false if it wasn't |
| 3059 | * pending. |
| 3060 | * |
| 3061 | * Note: |
| 3062 | * The work callback function may still be running on return, unless |
| 3063 | * it returns %true and the work doesn't re-arm itself. Explicitly flush or |
| 3064 | * use cancel_delayed_work_sync() to wait on it. |
Tejun Heo | 0938349 | 2010-09-16 10:48:29 +0200 | [diff] [blame] | 3065 | * |
Tejun Heo | 57b30ae | 2012-08-21 13:18:24 -0700 | [diff] [blame] | 3066 | * This function is safe to call from any context including IRQ handler. |
Tejun Heo | 0938349 | 2010-09-16 10:48:29 +0200 | [diff] [blame] | 3067 | */ |
Tejun Heo | 57b30ae | 2012-08-21 13:18:24 -0700 | [diff] [blame] | 3068 | bool cancel_delayed_work(struct delayed_work *dwork) |
Tejun Heo | 0938349 | 2010-09-16 10:48:29 +0200 | [diff] [blame] | 3069 | { |
Jens Axboe | f72b879 | 2016-08-24 15:51:50 -0600 | [diff] [blame] | 3070 | return __cancel_work(&dwork->work, true); |
Tejun Heo | 0938349 | 2010-09-16 10:48:29 +0200 | [diff] [blame] | 3071 | } |
Tejun Heo | 57b30ae | 2012-08-21 13:18:24 -0700 | [diff] [blame] | 3072 | EXPORT_SYMBOL(cancel_delayed_work); |
Tejun Heo | 0938349 | 2010-09-16 10:48:29 +0200 | [diff] [blame] | 3073 | |
| 3074 | /** |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 3075 | * cancel_delayed_work_sync - cancel a delayed work and wait for it to finish |
| 3076 | * @dwork: the delayed work cancel |
| 3077 | * |
| 3078 | * This is cancel_work_sync() for delayed works. |
| 3079 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 3080 | * Return: |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 3081 | * %true if @dwork was pending, %false otherwise. |
| 3082 | */ |
| 3083 | bool cancel_delayed_work_sync(struct delayed_work *dwork) |
Oleg Nesterov | 6e84d64 | 2007-05-09 02:34:46 -0700 | [diff] [blame] | 3084 | { |
Tejun Heo | 36e227d | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 3085 | return __cancel_work_timer(&dwork->work, true); |
Oleg Nesterov | 6e84d64 | 2007-05-09 02:34:46 -0700 | [diff] [blame] | 3086 | } |
Oleg Nesterov | f5a421a | 2007-07-15 23:41:44 -0700 | [diff] [blame] | 3087 | EXPORT_SYMBOL(cancel_delayed_work_sync); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3088 | |
Rolf Eike Beer | 0fcb78c | 2006-07-30 03:03:42 -0700 | [diff] [blame] | 3089 | /** |
Tejun Heo | 31ddd87 | 2010-10-19 11:14:49 +0200 | [diff] [blame] | 3090 | * schedule_on_each_cpu - execute a function synchronously on each online CPU |
Andrew Morton | b613677 | 2006-06-25 05:47:49 -0700 | [diff] [blame] | 3091 | * @func: the function to call |
Andrew Morton | b613677 | 2006-06-25 05:47:49 -0700 | [diff] [blame] | 3092 | * |
Tejun Heo | 31ddd87 | 2010-10-19 11:14:49 +0200 | [diff] [blame] | 3093 | * schedule_on_each_cpu() executes @func on each online CPU using the |
| 3094 | * system workqueue and blocks until all CPUs have completed. |
Andrew Morton | b613677 | 2006-06-25 05:47:49 -0700 | [diff] [blame] | 3095 | * schedule_on_each_cpu() is very slow. |
Tejun Heo | 31ddd87 | 2010-10-19 11:14:49 +0200 | [diff] [blame] | 3096 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 3097 | * Return: |
Tejun Heo | 31ddd87 | 2010-10-19 11:14:49 +0200 | [diff] [blame] | 3098 | * 0 on success, -errno on failure. |
Andrew Morton | b613677 | 2006-06-25 05:47:49 -0700 | [diff] [blame] | 3099 | */ |
David Howells | 65f27f3 | 2006-11-22 14:55:48 +0000 | [diff] [blame] | 3100 | int schedule_on_each_cpu(work_func_t func) |
Christoph Lameter | 15316ba | 2006-01-08 01:00:43 -0800 | [diff] [blame] | 3101 | { |
| 3102 | int cpu; |
Namhyung Kim | 38f5156 | 2010-08-08 14:24:09 +0200 | [diff] [blame] | 3103 | struct work_struct __percpu *works; |
Christoph Lameter | 15316ba | 2006-01-08 01:00:43 -0800 | [diff] [blame] | 3104 | |
Andrew Morton | b613677 | 2006-06-25 05:47:49 -0700 | [diff] [blame] | 3105 | works = alloc_percpu(struct work_struct); |
| 3106 | if (!works) |
Christoph Lameter | 15316ba | 2006-01-08 01:00:43 -0800 | [diff] [blame] | 3107 | return -ENOMEM; |
Andrew Morton | b613677 | 2006-06-25 05:47:49 -0700 | [diff] [blame] | 3108 | |
Gautham R Shenoy | 95402b3 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 3109 | get_online_cpus(); |
Tejun Heo | 9398180 | 2009-11-17 14:06:20 -0800 | [diff] [blame] | 3110 | |
Christoph Lameter | 15316ba | 2006-01-08 01:00:43 -0800 | [diff] [blame] | 3111 | for_each_online_cpu(cpu) { |
Ingo Molnar | 9bfb183 | 2006-12-18 20:05:09 +0100 | [diff] [blame] | 3112 | struct work_struct *work = per_cpu_ptr(works, cpu); |
| 3113 | |
| 3114 | INIT_WORK(work, func); |
Tejun Heo | b71ab8c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 3115 | schedule_work_on(cpu, work); |
Andi Kleen | 65a6446 | 2009-10-14 06:22:47 +0200 | [diff] [blame] | 3116 | } |
Tejun Heo | 9398180 | 2009-11-17 14:06:20 -0800 | [diff] [blame] | 3117 | |
| 3118 | for_each_online_cpu(cpu) |
| 3119 | flush_work(per_cpu_ptr(works, cpu)); |
| 3120 | |
Gautham R Shenoy | 95402b3 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 3121 | put_online_cpus(); |
Andrew Morton | b613677 | 2006-06-25 05:47:49 -0700 | [diff] [blame] | 3122 | free_percpu(works); |
Christoph Lameter | 15316ba | 2006-01-08 01:00:43 -0800 | [diff] [blame] | 3123 | return 0; |
| 3124 | } |
| 3125 | |
Alan Stern | eef6a7d | 2010-02-12 17:39:21 +0900 | [diff] [blame] | 3126 | /** |
James Bottomley | 1fa44ec | 2006-02-23 12:43:43 -0600 | [diff] [blame] | 3127 | * execute_in_process_context - reliably execute the routine with user context |
| 3128 | * @fn: the function to execute |
James Bottomley | 1fa44ec | 2006-02-23 12:43:43 -0600 | [diff] [blame] | 3129 | * @ew: guaranteed storage for the execute work structure (must |
| 3130 | * be available when the work executes) |
| 3131 | * |
| 3132 | * Executes the function immediately if process context is available, |
| 3133 | * otherwise schedules the function for delayed execution. |
| 3134 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 3135 | * Return: 0 - function was executed |
James Bottomley | 1fa44ec | 2006-02-23 12:43:43 -0600 | [diff] [blame] | 3136 | * 1 - function was scheduled for execution |
| 3137 | */ |
David Howells | 65f27f3 | 2006-11-22 14:55:48 +0000 | [diff] [blame] | 3138 | int execute_in_process_context(work_func_t fn, struct execute_work *ew) |
James Bottomley | 1fa44ec | 2006-02-23 12:43:43 -0600 | [diff] [blame] | 3139 | { |
| 3140 | if (!in_interrupt()) { |
David Howells | 65f27f3 | 2006-11-22 14:55:48 +0000 | [diff] [blame] | 3141 | fn(&ew->work); |
James Bottomley | 1fa44ec | 2006-02-23 12:43:43 -0600 | [diff] [blame] | 3142 | return 0; |
| 3143 | } |
| 3144 | |
David Howells | 65f27f3 | 2006-11-22 14:55:48 +0000 | [diff] [blame] | 3145 | INIT_WORK(&ew->work, fn); |
James Bottomley | 1fa44ec | 2006-02-23 12:43:43 -0600 | [diff] [blame] | 3146 | schedule_work(&ew->work); |
| 3147 | |
| 3148 | return 1; |
| 3149 | } |
| 3150 | EXPORT_SYMBOL_GPL(execute_in_process_context); |
| 3151 | |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 3152 | /** |
| 3153 | * free_workqueue_attrs - free a workqueue_attrs |
| 3154 | * @attrs: workqueue_attrs to free |
| 3155 | * |
| 3156 | * Undo alloc_workqueue_attrs(). |
| 3157 | */ |
| 3158 | void free_workqueue_attrs(struct workqueue_attrs *attrs) |
| 3159 | { |
| 3160 | if (attrs) { |
| 3161 | free_cpumask_var(attrs->cpumask); |
| 3162 | kfree(attrs); |
| 3163 | } |
| 3164 | } |
| 3165 | |
| 3166 | /** |
| 3167 | * alloc_workqueue_attrs - allocate a workqueue_attrs |
| 3168 | * @gfp_mask: allocation mask to use |
| 3169 | * |
| 3170 | * Allocate a new workqueue_attrs, initialize with default settings and |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 3171 | * return it. |
| 3172 | * |
| 3173 | * Return: The allocated new workqueue_attr on success. %NULL on failure. |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 3174 | */ |
| 3175 | struct workqueue_attrs *alloc_workqueue_attrs(gfp_t gfp_mask) |
| 3176 | { |
| 3177 | struct workqueue_attrs *attrs; |
| 3178 | |
| 3179 | attrs = kzalloc(sizeof(*attrs), gfp_mask); |
| 3180 | if (!attrs) |
| 3181 | goto fail; |
| 3182 | if (!alloc_cpumask_var(&attrs->cpumask, gfp_mask)) |
| 3183 | goto fail; |
| 3184 | |
Tejun Heo | 13e2e55 | 2013-04-01 11:23:31 -0700 | [diff] [blame] | 3185 | cpumask_copy(attrs->cpumask, cpu_possible_mask); |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 3186 | return attrs; |
| 3187 | fail: |
| 3188 | free_workqueue_attrs(attrs); |
| 3189 | return NULL; |
| 3190 | } |
| 3191 | |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3192 | static void copy_workqueue_attrs(struct workqueue_attrs *to, |
| 3193 | const struct workqueue_attrs *from) |
| 3194 | { |
| 3195 | to->nice = from->nice; |
| 3196 | cpumask_copy(to->cpumask, from->cpumask); |
Shaohua Li | 2865a8f | 2013-08-01 09:56:36 +0800 | [diff] [blame] | 3197 | /* |
| 3198 | * Unlike hash and equality test, this function doesn't ignore |
| 3199 | * ->no_numa as it is used for both pool and wq attrs. Instead, |
| 3200 | * get_unbound_pool() explicitly clears ->no_numa after copying. |
| 3201 | */ |
| 3202 | to->no_numa = from->no_numa; |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3203 | } |
| 3204 | |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3205 | /* hash value of the content of @attr */ |
| 3206 | static u32 wqattrs_hash(const struct workqueue_attrs *attrs) |
| 3207 | { |
| 3208 | u32 hash = 0; |
| 3209 | |
| 3210 | hash = jhash_1word(attrs->nice, hash); |
Tejun Heo | 13e2e55 | 2013-04-01 11:23:31 -0700 | [diff] [blame] | 3211 | hash = jhash(cpumask_bits(attrs->cpumask), |
| 3212 | BITS_TO_LONGS(nr_cpumask_bits) * sizeof(long), hash); |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3213 | return hash; |
| 3214 | } |
| 3215 | |
| 3216 | /* content equality test */ |
| 3217 | static bool wqattrs_equal(const struct workqueue_attrs *a, |
| 3218 | const struct workqueue_attrs *b) |
| 3219 | { |
| 3220 | if (a->nice != b->nice) |
| 3221 | return false; |
| 3222 | if (!cpumask_equal(a->cpumask, b->cpumask)) |
| 3223 | return false; |
| 3224 | return true; |
| 3225 | } |
| 3226 | |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 3227 | /** |
| 3228 | * init_worker_pool - initialize a newly zalloc'd worker_pool |
| 3229 | * @pool: worker_pool to initialize |
| 3230 | * |
Shailendra Verma | 402dd89 | 2015-05-23 10:38:14 +0530 | [diff] [blame] | 3231 | * Initialize a newly zalloc'd @pool. It also allocates @pool->attrs. |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 3232 | * |
| 3233 | * Return: 0 on success, -errno on failure. Even on failure, all fields |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3234 | * inside @pool proper are initialized and put_unbound_pool() can be called |
| 3235 | * on @pool safely to release it. |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 3236 | */ |
| 3237 | static int init_worker_pool(struct worker_pool *pool) |
Tejun Heo | 4e1a1f9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 3238 | { |
| 3239 | spin_lock_init(&pool->lock); |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3240 | pool->id = -1; |
| 3241 | pool->cpu = -1; |
Tejun Heo | f3f90ad | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 3242 | pool->node = NUMA_NO_NODE; |
Tejun Heo | 4e1a1f9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 3243 | pool->flags |= POOL_DISASSOCIATED; |
Tejun Heo | 82607adc | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 3244 | pool->watchdog_ts = jiffies; |
Tejun Heo | 4e1a1f9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 3245 | INIT_LIST_HEAD(&pool->worklist); |
| 3246 | INIT_LIST_HEAD(&pool->idle_list); |
| 3247 | hash_init(pool->busy_hash); |
| 3248 | |
Geliang Tang | c30fb26 | 2017-02-26 08:58:21 +0800 | [diff] [blame] | 3249 | setup_deferrable_timer(&pool->idle_timer, idle_worker_timeout, |
| 3250 | (unsigned long)pool); |
Tejun Heo | 4e1a1f9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 3251 | |
| 3252 | setup_timer(&pool->mayday_timer, pool_mayday_timeout, |
| 3253 | (unsigned long)pool); |
| 3254 | |
Lai Jiangshan | 92f9c5c | 2014-05-20 17:46:34 +0800 | [diff] [blame] | 3255 | mutex_init(&pool->attach_mutex); |
Lai Jiangshan | da02846 | 2014-05-20 17:46:31 +0800 | [diff] [blame] | 3256 | INIT_LIST_HEAD(&pool->workers); |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 3257 | |
Lai Jiangshan | 7cda9aa | 2014-05-20 17:46:32 +0800 | [diff] [blame] | 3258 | ida_init(&pool->worker_ida); |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3259 | INIT_HLIST_NODE(&pool->hash_node); |
| 3260 | pool->refcnt = 1; |
| 3261 | |
| 3262 | /* shouldn't fail above this point */ |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 3263 | pool->attrs = alloc_workqueue_attrs(GFP_KERNEL); |
| 3264 | if (!pool->attrs) |
| 3265 | return -ENOMEM; |
| 3266 | return 0; |
Tejun Heo | 4e1a1f9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 3267 | } |
| 3268 | |
Tejun Heo | e2dca7a | 2015-03-09 09:22:28 -0400 | [diff] [blame] | 3269 | static void rcu_free_wq(struct rcu_head *rcu) |
| 3270 | { |
| 3271 | struct workqueue_struct *wq = |
| 3272 | container_of(rcu, struct workqueue_struct, rcu); |
| 3273 | |
| 3274 | if (!(wq->flags & WQ_UNBOUND)) |
| 3275 | free_percpu(wq->cpu_pwqs); |
| 3276 | else |
| 3277 | free_workqueue_attrs(wq->unbound_attrs); |
| 3278 | |
| 3279 | kfree(wq->rescuer); |
| 3280 | kfree(wq); |
| 3281 | } |
| 3282 | |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3283 | static void rcu_free_pool(struct rcu_head *rcu) |
| 3284 | { |
| 3285 | struct worker_pool *pool = container_of(rcu, struct worker_pool, rcu); |
| 3286 | |
Lai Jiangshan | 7cda9aa | 2014-05-20 17:46:32 +0800 | [diff] [blame] | 3287 | ida_destroy(&pool->worker_ida); |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3288 | free_workqueue_attrs(pool->attrs); |
| 3289 | kfree(pool); |
| 3290 | } |
| 3291 | |
| 3292 | /** |
| 3293 | * put_unbound_pool - put a worker_pool |
| 3294 | * @pool: worker_pool to put |
| 3295 | * |
| 3296 | * Put @pool. If its refcnt reaches zero, it gets destroyed in sched-RCU |
Tejun Heo | c5aa87b | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 3297 | * safe manner. get_unbound_pool() calls this function on its failure path |
| 3298 | * and this function should be able to release pools which went through, |
| 3299 | * successfully or not, init_worker_pool(). |
Tejun Heo | a892cac | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 3300 | * |
| 3301 | * Should be called with wq_pool_mutex held. |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3302 | */ |
| 3303 | static void put_unbound_pool(struct worker_pool *pool) |
| 3304 | { |
Lai Jiangshan | 60f5a4b | 2014-05-20 17:46:29 +0800 | [diff] [blame] | 3305 | DECLARE_COMPLETION_ONSTACK(detach_completion); |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3306 | struct worker *worker; |
| 3307 | |
Tejun Heo | a892cac | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 3308 | lockdep_assert_held(&wq_pool_mutex); |
| 3309 | |
| 3310 | if (--pool->refcnt) |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3311 | return; |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3312 | |
| 3313 | /* sanity checks */ |
Lai Jiangshan | 61d0fbb | 2014-06-03 15:31:45 +0800 | [diff] [blame] | 3314 | if (WARN_ON(!(pool->cpu < 0)) || |
Tejun Heo | a892cac | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 3315 | WARN_ON(!list_empty(&pool->worklist))) |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3316 | return; |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3317 | |
| 3318 | /* release id and unhash */ |
| 3319 | if (pool->id >= 0) |
| 3320 | idr_remove(&worker_pool_idr, pool->id); |
| 3321 | hash_del(&pool->hash_node); |
| 3322 | |
Tejun Heo | c5aa87b | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 3323 | /* |
Tejun Heo | 692b482 | 2017-10-09 08:04:13 -0700 | [diff] [blame] | 3324 | * Become the manager and destroy all workers. This prevents |
| 3325 | * @pool's workers from blocking on attach_mutex. We're the last |
| 3326 | * manager and @pool gets freed with the flag set. |
Tejun Heo | c5aa87b | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 3327 | */ |
Lai Jiangshan | 60f5a4b | 2014-05-20 17:46:29 +0800 | [diff] [blame] | 3328 | spin_lock_irq(&pool->lock); |
Tejun Heo | 692b482 | 2017-10-09 08:04:13 -0700 | [diff] [blame] | 3329 | wait_event_lock_irq(wq_manager_wait, |
| 3330 | !(pool->flags & POOL_MANAGER_ACTIVE), pool->lock); |
| 3331 | pool->flags |= POOL_MANAGER_ACTIVE; |
| 3332 | |
Lai Jiangshan | 1037de3 | 2014-05-22 16:44:07 +0800 | [diff] [blame] | 3333 | while ((worker = first_idle_worker(pool))) |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3334 | destroy_worker(worker); |
| 3335 | WARN_ON(pool->nr_workers || pool->nr_idle); |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3336 | spin_unlock_irq(&pool->lock); |
Lai Jiangshan | 60f5a4b | 2014-05-20 17:46:29 +0800 | [diff] [blame] | 3337 | |
Lai Jiangshan | 92f9c5c | 2014-05-20 17:46:34 +0800 | [diff] [blame] | 3338 | mutex_lock(&pool->attach_mutex); |
Lai Jiangshan | da02846 | 2014-05-20 17:46:31 +0800 | [diff] [blame] | 3339 | if (!list_empty(&pool->workers)) |
Lai Jiangshan | 60f5a4b | 2014-05-20 17:46:29 +0800 | [diff] [blame] | 3340 | pool->detach_completion = &detach_completion; |
Lai Jiangshan | 92f9c5c | 2014-05-20 17:46:34 +0800 | [diff] [blame] | 3341 | mutex_unlock(&pool->attach_mutex); |
Lai Jiangshan | 60f5a4b | 2014-05-20 17:46:29 +0800 | [diff] [blame] | 3342 | |
| 3343 | if (pool->detach_completion) |
| 3344 | wait_for_completion(pool->detach_completion); |
| 3345 | |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3346 | /* shut down the timers */ |
| 3347 | del_timer_sync(&pool->idle_timer); |
| 3348 | del_timer_sync(&pool->mayday_timer); |
| 3349 | |
| 3350 | /* sched-RCU protected to allow dereferences from get_work_pool() */ |
| 3351 | call_rcu_sched(&pool->rcu, rcu_free_pool); |
| 3352 | } |
| 3353 | |
| 3354 | /** |
| 3355 | * get_unbound_pool - get a worker_pool with the specified attributes |
| 3356 | * @attrs: the attributes of the worker_pool to get |
| 3357 | * |
| 3358 | * Obtain a worker_pool which has the same attributes as @attrs, bump the |
| 3359 | * reference count and return it. If there already is a matching |
| 3360 | * worker_pool, it will be used; otherwise, this function attempts to |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 3361 | * create a new one. |
Tejun Heo | a892cac | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 3362 | * |
| 3363 | * Should be called with wq_pool_mutex held. |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 3364 | * |
| 3365 | * Return: On success, a worker_pool with the same attributes as @attrs. |
| 3366 | * On failure, %NULL. |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3367 | */ |
| 3368 | static struct worker_pool *get_unbound_pool(const struct workqueue_attrs *attrs) |
| 3369 | { |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3370 | u32 hash = wqattrs_hash(attrs); |
| 3371 | struct worker_pool *pool; |
Tejun Heo | f3f90ad | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 3372 | int node; |
Xunlei Pang | e227358 | 2015-10-09 11:53:12 +0800 | [diff] [blame] | 3373 | int target_node = NUMA_NO_NODE; |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3374 | |
Tejun Heo | a892cac | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 3375 | lockdep_assert_held(&wq_pool_mutex); |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3376 | |
| 3377 | /* do we already have a matching pool? */ |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3378 | hash_for_each_possible(unbound_pool_hash, pool, hash_node, hash) { |
| 3379 | if (wqattrs_equal(pool->attrs, attrs)) { |
| 3380 | pool->refcnt++; |
Lai Jiangshan | 3fb1823 | 2014-07-22 13:04:49 +0800 | [diff] [blame] | 3381 | return pool; |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3382 | } |
| 3383 | } |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3384 | |
Xunlei Pang | e227358 | 2015-10-09 11:53:12 +0800 | [diff] [blame] | 3385 | /* if cpumask is contained inside a NUMA node, we belong to that node */ |
| 3386 | if (wq_numa_enabled) { |
| 3387 | for_each_node(node) { |
| 3388 | if (cpumask_subset(attrs->cpumask, |
| 3389 | wq_numa_possible_cpumask[node])) { |
| 3390 | target_node = node; |
| 3391 | break; |
| 3392 | } |
| 3393 | } |
| 3394 | } |
| 3395 | |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3396 | /* nope, create a new one */ |
Xunlei Pang | e227358 | 2015-10-09 11:53:12 +0800 | [diff] [blame] | 3397 | pool = kzalloc_node(sizeof(*pool), GFP_KERNEL, target_node); |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3398 | if (!pool || init_worker_pool(pool) < 0) |
| 3399 | goto fail; |
| 3400 | |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3401 | lockdep_set_subclass(&pool->lock, 1); /* see put_pwq() */ |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3402 | copy_workqueue_attrs(pool->attrs, attrs); |
Xunlei Pang | e227358 | 2015-10-09 11:53:12 +0800 | [diff] [blame] | 3403 | pool->node = target_node; |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3404 | |
Shaohua Li | 2865a8f | 2013-08-01 09:56:36 +0800 | [diff] [blame] | 3405 | /* |
| 3406 | * no_numa isn't a worker_pool attribute, always clear it. See |
| 3407 | * 'struct workqueue_attrs' comments for detail. |
| 3408 | */ |
| 3409 | pool->attrs->no_numa = false; |
| 3410 | |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3411 | if (worker_pool_assign_id(pool) < 0) |
| 3412 | goto fail; |
| 3413 | |
| 3414 | /* create and start the initial worker */ |
Tejun Heo | 3347fa0 | 2016-09-16 15:49:32 -0400 | [diff] [blame] | 3415 | if (wq_online && !create_worker(pool)) |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3416 | goto fail; |
| 3417 | |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3418 | /* install */ |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3419 | hash_add(unbound_pool_hash, &pool->hash_node, hash); |
Lai Jiangshan | 3fb1823 | 2014-07-22 13:04:49 +0800 | [diff] [blame] | 3420 | |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3421 | return pool; |
| 3422 | fail: |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3423 | if (pool) |
| 3424 | put_unbound_pool(pool); |
| 3425 | return NULL; |
| 3426 | } |
| 3427 | |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3428 | static void rcu_free_pwq(struct rcu_head *rcu) |
| 3429 | { |
| 3430 | kmem_cache_free(pwq_cache, |
| 3431 | container_of(rcu, struct pool_workqueue, rcu)); |
| 3432 | } |
| 3433 | |
| 3434 | /* |
| 3435 | * Scheduled on system_wq by put_pwq() when an unbound pwq hits zero refcnt |
| 3436 | * and needs to be destroyed. |
| 3437 | */ |
| 3438 | static void pwq_unbound_release_workfn(struct work_struct *work) |
| 3439 | { |
| 3440 | struct pool_workqueue *pwq = container_of(work, struct pool_workqueue, |
| 3441 | unbound_release_work); |
| 3442 | struct workqueue_struct *wq = pwq->wq; |
| 3443 | struct worker_pool *pool = pwq->pool; |
Yang Yingliang | ad7ae9e | 2021-07-14 17:19:33 +0800 | [diff] [blame] | 3444 | bool is_last = false; |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3445 | |
Yang Yingliang | ad7ae9e | 2021-07-14 17:19:33 +0800 | [diff] [blame] | 3446 | /* |
| 3447 | * when @pwq is not linked, it doesn't hold any reference to the |
| 3448 | * @wq, and @wq is invalid to access. |
| 3449 | */ |
| 3450 | if (!list_empty(&pwq->pwqs_node)) { |
| 3451 | if (WARN_ON_ONCE(!(wq->flags & WQ_UNBOUND))) |
| 3452 | return; |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3453 | |
Yang Yingliang | ad7ae9e | 2021-07-14 17:19:33 +0800 | [diff] [blame] | 3454 | mutex_lock(&wq->mutex); |
| 3455 | list_del_rcu(&pwq->pwqs_node); |
| 3456 | is_last = list_empty(&wq->pwqs); |
| 3457 | mutex_unlock(&wq->mutex); |
| 3458 | } |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3459 | |
Tejun Heo | a892cac | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 3460 | mutex_lock(&wq_pool_mutex); |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3461 | put_unbound_pool(pool); |
Tejun Heo | a892cac | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 3462 | mutex_unlock(&wq_pool_mutex); |
| 3463 | |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3464 | call_rcu_sched(&pwq->rcu, rcu_free_pwq); |
| 3465 | |
| 3466 | /* |
| 3467 | * If we're the last pwq going away, @wq is already dead and no one |
Tejun Heo | e2dca7a | 2015-03-09 09:22:28 -0400 | [diff] [blame] | 3468 | * is gonna access it anymore. Schedule RCU free. |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3469 | */ |
Tejun Heo | e2dca7a | 2015-03-09 09:22:28 -0400 | [diff] [blame] | 3470 | if (is_last) |
| 3471 | call_rcu_sched(&wq->rcu, rcu_free_wq); |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3472 | } |
| 3473 | |
Tejun Heo | 0fbd95a | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3474 | /** |
Tejun Heo | 699ce09 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3475 | * pwq_adjust_max_active - update a pwq's max_active to the current setting |
Tejun Heo | 0fbd95a | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3476 | * @pwq: target pool_workqueue |
Tejun Heo | 0fbd95a | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3477 | * |
Tejun Heo | 699ce09 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3478 | * If @pwq isn't freezing, set @pwq->max_active to the associated |
| 3479 | * workqueue's saved_max_active and activate delayed work items |
| 3480 | * accordingly. If @pwq is freezing, clear @pwq->max_active to zero. |
Tejun Heo | 0fbd95a | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3481 | */ |
Tejun Heo | 699ce09 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3482 | static void pwq_adjust_max_active(struct pool_workqueue *pwq) |
Tejun Heo | 0fbd95a | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3483 | { |
Tejun Heo | 699ce09 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3484 | struct workqueue_struct *wq = pwq->wq; |
| 3485 | bool freezable = wq->flags & WQ_FREEZABLE; |
Tejun Heo | 3347fa0 | 2016-09-16 15:49:32 -0400 | [diff] [blame] | 3486 | unsigned long flags; |
Tejun Heo | 0fbd95a | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3487 | |
Tejun Heo | 699ce09 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3488 | /* for @wq->saved_max_active */ |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 3489 | lockdep_assert_held(&wq->mutex); |
Tejun Heo | 699ce09 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3490 | |
| 3491 | /* fast exit for non-freezable wqs */ |
| 3492 | if (!freezable && pwq->max_active == wq->saved_max_active) |
| 3493 | return; |
| 3494 | |
Tejun Heo | 3347fa0 | 2016-09-16 15:49:32 -0400 | [diff] [blame] | 3495 | /* this function can be called during early boot w/ irq disabled */ |
| 3496 | spin_lock_irqsave(&pwq->pool->lock, flags); |
Tejun Heo | 699ce09 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3497 | |
Lai Jiangshan | 74b414e | 2014-05-22 19:01:16 +0800 | [diff] [blame] | 3498 | /* |
| 3499 | * During [un]freezing, the caller is responsible for ensuring that |
| 3500 | * this function is called at least once after @workqueue_freezing |
| 3501 | * is updated and visible. |
| 3502 | */ |
| 3503 | if (!freezable || !workqueue_freezing) { |
Yunfeng Ye | f6d739c | 2020-11-19 14:21:25 +0800 | [diff] [blame] | 3504 | bool kick = false; |
| 3505 | |
Tejun Heo | 699ce09 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3506 | pwq->max_active = wq->saved_max_active; |
| 3507 | |
| 3508 | while (!list_empty(&pwq->delayed_works) && |
Yunfeng Ye | f6d739c | 2020-11-19 14:21:25 +0800 | [diff] [blame] | 3509 | pwq->nr_active < pwq->max_active) { |
Tejun Heo | 699ce09 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3510 | pwq_activate_first_delayed(pwq); |
Yunfeng Ye | f6d739c | 2020-11-19 14:21:25 +0800 | [diff] [blame] | 3511 | kick = true; |
| 3512 | } |
Lai Jiangshan | 951a078 | 2013-03-20 10:52:30 -0700 | [diff] [blame] | 3513 | |
| 3514 | /* |
| 3515 | * Need to kick a worker after thawed or an unbound wq's |
Yunfeng Ye | f6d739c | 2020-11-19 14:21:25 +0800 | [diff] [blame] | 3516 | * max_active is bumped. In realtime scenarios, always kicking a |
| 3517 | * worker will cause interference on the isolated cpu cores, so |
| 3518 | * let's kick iff work items were activated. |
Lai Jiangshan | 951a078 | 2013-03-20 10:52:30 -0700 | [diff] [blame] | 3519 | */ |
Yunfeng Ye | f6d739c | 2020-11-19 14:21:25 +0800 | [diff] [blame] | 3520 | if (kick) |
| 3521 | wake_up_worker(pwq->pool); |
Tejun Heo | 699ce09 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3522 | } else { |
| 3523 | pwq->max_active = 0; |
| 3524 | } |
| 3525 | |
Tejun Heo | 3347fa0 | 2016-09-16 15:49:32 -0400 | [diff] [blame] | 3526 | spin_unlock_irqrestore(&pwq->pool->lock, flags); |
Tejun Heo | 0fbd95a | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3527 | } |
| 3528 | |
Tejun Heo | e50aba9 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3529 | /* initialize newly alloced @pwq which is associated with @wq and @pool */ |
Tejun Heo | f147f29 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3530 | static void init_pwq(struct pool_workqueue *pwq, struct workqueue_struct *wq, |
| 3531 | struct worker_pool *pool) |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3532 | { |
| 3533 | BUG_ON((unsigned long)pwq & WORK_STRUCT_FLAG_MASK); |
| 3534 | |
Tejun Heo | e50aba9 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3535 | memset(pwq, 0, sizeof(*pwq)); |
| 3536 | |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3537 | pwq->pool = pool; |
| 3538 | pwq->wq = wq; |
| 3539 | pwq->flush_color = -1; |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3540 | pwq->refcnt = 1; |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3541 | INIT_LIST_HEAD(&pwq->delayed_works); |
Tejun Heo | 1befcf3 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3542 | INIT_LIST_HEAD(&pwq->pwqs_node); |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3543 | INIT_LIST_HEAD(&pwq->mayday_node); |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3544 | INIT_WORK(&pwq->unbound_release_work, pwq_unbound_release_workfn); |
Tejun Heo | f147f29 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3545 | } |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3546 | |
Tejun Heo | f147f29 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3547 | /* sync @pwq with the current state of its associated wq and link it */ |
Tejun Heo | 1befcf3 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3548 | static void link_pwq(struct pool_workqueue *pwq) |
Tejun Heo | f147f29 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3549 | { |
| 3550 | struct workqueue_struct *wq = pwq->wq; |
| 3551 | |
| 3552 | lockdep_assert_held(&wq->mutex); |
Tejun Heo | 75ccf59 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3553 | |
Tejun Heo | 1befcf3 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3554 | /* may be called multiple times, ignore if already linked */ |
| 3555 | if (!list_empty(&pwq->pwqs_node)) |
| 3556 | return; |
| 3557 | |
Lai Jiangshan | 29b1cb4 | 2014-07-22 13:04:27 +0800 | [diff] [blame] | 3558 | /* set the matching work_color */ |
Tejun Heo | 75ccf59 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3559 | pwq->work_color = wq->work_color; |
Tejun Heo | 983ca25 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3560 | |
| 3561 | /* sync max_active to the current setting */ |
| 3562 | pwq_adjust_max_active(pwq); |
| 3563 | |
| 3564 | /* link in @pwq */ |
Tejun Heo | 9e8cd2f | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3565 | list_add_rcu(&pwq->pwqs_node, &wq->pwqs); |
Tejun Heo | f147f29 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3566 | } |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 3567 | |
Tejun Heo | f147f29 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3568 | /* obtain a pool matching @attr and create a pwq associating the pool and @wq */ |
| 3569 | static struct pool_workqueue *alloc_unbound_pwq(struct workqueue_struct *wq, |
| 3570 | const struct workqueue_attrs *attrs) |
| 3571 | { |
| 3572 | struct worker_pool *pool; |
| 3573 | struct pool_workqueue *pwq; |
| 3574 | |
| 3575 | lockdep_assert_held(&wq_pool_mutex); |
| 3576 | |
| 3577 | pool = get_unbound_pool(attrs); |
| 3578 | if (!pool) |
| 3579 | return NULL; |
| 3580 | |
Tejun Heo | e50aba9 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3581 | pwq = kmem_cache_alloc_node(pwq_cache, GFP_KERNEL, pool->node); |
Tejun Heo | f147f29 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3582 | if (!pwq) { |
| 3583 | put_unbound_pool(pool); |
| 3584 | return NULL; |
Tejun Heo | df2d5ae | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3585 | } |
Tejun Heo | 6029a91 | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 3586 | |
Tejun Heo | f147f29 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3587 | init_pwq(pwq, wq, pool); |
| 3588 | return pwq; |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3589 | } |
| 3590 | |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3591 | /** |
Gong Zhaogang | 30186c6 | 2015-05-11 11:02:47 -0400 | [diff] [blame] | 3592 | * wq_calc_node_cpumask - calculate a wq_attrs' cpumask for the specified node |
Lai Jiangshan | 042f7df | 2015-04-30 17:16:12 +0800 | [diff] [blame] | 3593 | * @attrs: the wq_attrs of the default pwq of the target workqueue |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3594 | * @node: the target NUMA node |
| 3595 | * @cpu_going_down: if >= 0, the CPU to consider as offline |
| 3596 | * @cpumask: outarg, the resulting cpumask |
| 3597 | * |
| 3598 | * Calculate the cpumask a workqueue with @attrs should use on @node. If |
| 3599 | * @cpu_going_down is >= 0, that cpu is considered offline during |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 3600 | * calculation. The result is stored in @cpumask. |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3601 | * |
| 3602 | * If NUMA affinity is not enabled, @attrs->cpumask is always used. If |
| 3603 | * enabled and @node has online CPUs requested by @attrs, the returned |
| 3604 | * cpumask is the intersection of the possible CPUs of @node and |
| 3605 | * @attrs->cpumask. |
| 3606 | * |
| 3607 | * The caller is responsible for ensuring that the cpumask of @node stays |
| 3608 | * stable. |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 3609 | * |
| 3610 | * Return: %true if the resulting @cpumask is different from @attrs->cpumask, |
| 3611 | * %false if equal. |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3612 | */ |
| 3613 | static bool wq_calc_node_cpumask(const struct workqueue_attrs *attrs, int node, |
| 3614 | int cpu_going_down, cpumask_t *cpumask) |
| 3615 | { |
Tejun Heo | d55262c | 2013-04-01 11:23:38 -0700 | [diff] [blame] | 3616 | if (!wq_numa_enabled || attrs->no_numa) |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3617 | goto use_dfl; |
| 3618 | |
| 3619 | /* does @node have any online CPUs @attrs wants? */ |
| 3620 | cpumask_and(cpumask, cpumask_of_node(node), attrs->cpumask); |
| 3621 | if (cpu_going_down >= 0) |
| 3622 | cpumask_clear_cpu(cpu_going_down, cpumask); |
| 3623 | |
| 3624 | if (cpumask_empty(cpumask)) |
| 3625 | goto use_dfl; |
| 3626 | |
| 3627 | /* yeap, return possible CPUs in @node that @attrs wants */ |
| 3628 | cpumask_and(cpumask, attrs->cpumask, wq_numa_possible_cpumask[node]); |
Michael Bringmann | 1ad0f0a | 2017-07-27 16:27:14 -0500 | [diff] [blame] | 3629 | |
| 3630 | if (cpumask_empty(cpumask)) { |
| 3631 | pr_warn_once("WARNING: workqueue cpumask: online intersect > " |
| 3632 | "possible intersect\n"); |
| 3633 | return false; |
| 3634 | } |
| 3635 | |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3636 | return !cpumask_equal(cpumask, attrs->cpumask); |
| 3637 | |
| 3638 | use_dfl: |
| 3639 | cpumask_copy(cpumask, attrs->cpumask); |
| 3640 | return false; |
| 3641 | } |
| 3642 | |
Tejun Heo | 1befcf3 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3643 | /* install @pwq into @wq's numa_pwq_tbl[] for @node and return the old pwq */ |
| 3644 | static struct pool_workqueue *numa_pwq_tbl_install(struct workqueue_struct *wq, |
| 3645 | int node, |
| 3646 | struct pool_workqueue *pwq) |
| 3647 | { |
| 3648 | struct pool_workqueue *old_pwq; |
| 3649 | |
Lai Jiangshan | 5b95e1a | 2015-05-12 20:32:29 +0800 | [diff] [blame] | 3650 | lockdep_assert_held(&wq_pool_mutex); |
Tejun Heo | 1befcf3 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3651 | lockdep_assert_held(&wq->mutex); |
| 3652 | |
| 3653 | /* link_pwq() can handle duplicate calls */ |
| 3654 | link_pwq(pwq); |
| 3655 | |
| 3656 | old_pwq = rcu_access_pointer(wq->numa_pwq_tbl[node]); |
| 3657 | rcu_assign_pointer(wq->numa_pwq_tbl[node], pwq); |
| 3658 | return old_pwq; |
| 3659 | } |
| 3660 | |
Lai Jiangshan | 2d5f076 | 2015-04-27 17:58:38 +0800 | [diff] [blame] | 3661 | /* context to store the prepared attrs & pwqs before applying */ |
| 3662 | struct apply_wqattrs_ctx { |
| 3663 | struct workqueue_struct *wq; /* target workqueue */ |
| 3664 | struct workqueue_attrs *attrs; /* attrs to apply */ |
Lai Jiangshan | 042f7df | 2015-04-30 17:16:12 +0800 | [diff] [blame] | 3665 | struct list_head list; /* queued for batching commit */ |
Lai Jiangshan | 2d5f076 | 2015-04-27 17:58:38 +0800 | [diff] [blame] | 3666 | struct pool_workqueue *dfl_pwq; |
| 3667 | struct pool_workqueue *pwq_tbl[]; |
| 3668 | }; |
| 3669 | |
| 3670 | /* free the resources after success or abort */ |
| 3671 | static void apply_wqattrs_cleanup(struct apply_wqattrs_ctx *ctx) |
| 3672 | { |
| 3673 | if (ctx) { |
| 3674 | int node; |
| 3675 | |
| 3676 | for_each_node(node) |
| 3677 | put_pwq_unlocked(ctx->pwq_tbl[node]); |
| 3678 | put_pwq_unlocked(ctx->dfl_pwq); |
| 3679 | |
| 3680 | free_workqueue_attrs(ctx->attrs); |
| 3681 | |
| 3682 | kfree(ctx); |
| 3683 | } |
| 3684 | } |
| 3685 | |
| 3686 | /* allocate the attrs and pwqs for later installation */ |
| 3687 | static struct apply_wqattrs_ctx * |
| 3688 | apply_wqattrs_prepare(struct workqueue_struct *wq, |
| 3689 | const struct workqueue_attrs *attrs) |
| 3690 | { |
| 3691 | struct apply_wqattrs_ctx *ctx; |
| 3692 | struct workqueue_attrs *new_attrs, *tmp_attrs; |
| 3693 | int node; |
| 3694 | |
| 3695 | lockdep_assert_held(&wq_pool_mutex); |
| 3696 | |
| 3697 | ctx = kzalloc(sizeof(*ctx) + nr_node_ids * sizeof(ctx->pwq_tbl[0]), |
| 3698 | GFP_KERNEL); |
| 3699 | |
| 3700 | new_attrs = alloc_workqueue_attrs(GFP_KERNEL); |
| 3701 | tmp_attrs = alloc_workqueue_attrs(GFP_KERNEL); |
| 3702 | if (!ctx || !new_attrs || !tmp_attrs) |
| 3703 | goto out_free; |
| 3704 | |
Lai Jiangshan | 042f7df | 2015-04-30 17:16:12 +0800 | [diff] [blame] | 3705 | /* |
| 3706 | * Calculate the attrs of the default pwq. |
| 3707 | * If the user configured cpumask doesn't overlap with the |
| 3708 | * wq_unbound_cpumask, we fallback to the wq_unbound_cpumask. |
| 3709 | */ |
Lai Jiangshan | 2d5f076 | 2015-04-27 17:58:38 +0800 | [diff] [blame] | 3710 | copy_workqueue_attrs(new_attrs, attrs); |
Frederic Weisbecker | b05a792 | 2015-04-27 17:58:39 +0800 | [diff] [blame] | 3711 | cpumask_and(new_attrs->cpumask, new_attrs->cpumask, wq_unbound_cpumask); |
Lai Jiangshan | 042f7df | 2015-04-30 17:16:12 +0800 | [diff] [blame] | 3712 | if (unlikely(cpumask_empty(new_attrs->cpumask))) |
| 3713 | cpumask_copy(new_attrs->cpumask, wq_unbound_cpumask); |
Lai Jiangshan | 2d5f076 | 2015-04-27 17:58:38 +0800 | [diff] [blame] | 3714 | |
| 3715 | /* |
| 3716 | * We may create multiple pwqs with differing cpumasks. Make a |
| 3717 | * copy of @new_attrs which will be modified and used to obtain |
| 3718 | * pools. |
| 3719 | */ |
| 3720 | copy_workqueue_attrs(tmp_attrs, new_attrs); |
| 3721 | |
| 3722 | /* |
| 3723 | * If something goes wrong during CPU up/down, we'll fall back to |
| 3724 | * the default pwq covering whole @attrs->cpumask. Always create |
| 3725 | * it even if we don't use it immediately. |
| 3726 | */ |
| 3727 | ctx->dfl_pwq = alloc_unbound_pwq(wq, new_attrs); |
| 3728 | if (!ctx->dfl_pwq) |
| 3729 | goto out_free; |
| 3730 | |
| 3731 | for_each_node(node) { |
Lai Jiangshan | 042f7df | 2015-04-30 17:16:12 +0800 | [diff] [blame] | 3732 | if (wq_calc_node_cpumask(new_attrs, node, -1, tmp_attrs->cpumask)) { |
Lai Jiangshan | 2d5f076 | 2015-04-27 17:58:38 +0800 | [diff] [blame] | 3733 | ctx->pwq_tbl[node] = alloc_unbound_pwq(wq, tmp_attrs); |
| 3734 | if (!ctx->pwq_tbl[node]) |
| 3735 | goto out_free; |
| 3736 | } else { |
| 3737 | ctx->dfl_pwq->refcnt++; |
| 3738 | ctx->pwq_tbl[node] = ctx->dfl_pwq; |
| 3739 | } |
| 3740 | } |
| 3741 | |
Lai Jiangshan | 042f7df | 2015-04-30 17:16:12 +0800 | [diff] [blame] | 3742 | /* save the user configured attrs and sanitize it. */ |
| 3743 | copy_workqueue_attrs(new_attrs, attrs); |
| 3744 | cpumask_and(new_attrs->cpumask, new_attrs->cpumask, cpu_possible_mask); |
Lai Jiangshan | 2d5f076 | 2015-04-27 17:58:38 +0800 | [diff] [blame] | 3745 | ctx->attrs = new_attrs; |
Lai Jiangshan | 042f7df | 2015-04-30 17:16:12 +0800 | [diff] [blame] | 3746 | |
Lai Jiangshan | 2d5f076 | 2015-04-27 17:58:38 +0800 | [diff] [blame] | 3747 | ctx->wq = wq; |
| 3748 | free_workqueue_attrs(tmp_attrs); |
| 3749 | return ctx; |
| 3750 | |
| 3751 | out_free: |
| 3752 | free_workqueue_attrs(tmp_attrs); |
| 3753 | free_workqueue_attrs(new_attrs); |
| 3754 | apply_wqattrs_cleanup(ctx); |
| 3755 | return NULL; |
| 3756 | } |
| 3757 | |
| 3758 | /* set attrs and install prepared pwqs, @ctx points to old pwqs on return */ |
| 3759 | static void apply_wqattrs_commit(struct apply_wqattrs_ctx *ctx) |
| 3760 | { |
| 3761 | int node; |
| 3762 | |
| 3763 | /* all pwqs have been created successfully, let's install'em */ |
| 3764 | mutex_lock(&ctx->wq->mutex); |
| 3765 | |
| 3766 | copy_workqueue_attrs(ctx->wq->unbound_attrs, ctx->attrs); |
| 3767 | |
| 3768 | /* save the previous pwq and install the new one */ |
| 3769 | for_each_node(node) |
| 3770 | ctx->pwq_tbl[node] = numa_pwq_tbl_install(ctx->wq, node, |
| 3771 | ctx->pwq_tbl[node]); |
| 3772 | |
| 3773 | /* @dfl_pwq might not have been used, ensure it's linked */ |
| 3774 | link_pwq(ctx->dfl_pwq); |
| 3775 | swap(ctx->wq->dfl_pwq, ctx->dfl_pwq); |
| 3776 | |
| 3777 | mutex_unlock(&ctx->wq->mutex); |
| 3778 | } |
| 3779 | |
Lai Jiangshan | a0111cf | 2015-05-19 18:03:47 +0800 | [diff] [blame] | 3780 | static void apply_wqattrs_lock(void) |
| 3781 | { |
| 3782 | /* CPUs should stay stable across pwq creations and installations */ |
| 3783 | get_online_cpus(); |
| 3784 | mutex_lock(&wq_pool_mutex); |
| 3785 | } |
| 3786 | |
| 3787 | static void apply_wqattrs_unlock(void) |
| 3788 | { |
| 3789 | mutex_unlock(&wq_pool_mutex); |
| 3790 | put_online_cpus(); |
| 3791 | } |
| 3792 | |
| 3793 | static int apply_workqueue_attrs_locked(struct workqueue_struct *wq, |
| 3794 | const struct workqueue_attrs *attrs) |
| 3795 | { |
| 3796 | struct apply_wqattrs_ctx *ctx; |
Lai Jiangshan | a0111cf | 2015-05-19 18:03:47 +0800 | [diff] [blame] | 3797 | |
| 3798 | /* only unbound workqueues can change attributes */ |
| 3799 | if (WARN_ON(!(wq->flags & WQ_UNBOUND))) |
| 3800 | return -EINVAL; |
| 3801 | |
| 3802 | /* creating multiple pwqs breaks ordering guarantee */ |
Tejun Heo | 0a94efb | 2017-07-23 08:36:15 -0400 | [diff] [blame] | 3803 | if (!list_empty(&wq->pwqs)) { |
| 3804 | if (WARN_ON(wq->flags & __WQ_ORDERED_EXPLICIT)) |
| 3805 | return -EINVAL; |
| 3806 | |
| 3807 | wq->flags &= ~__WQ_ORDERED; |
| 3808 | } |
Lai Jiangshan | a0111cf | 2015-05-19 18:03:47 +0800 | [diff] [blame] | 3809 | |
| 3810 | ctx = apply_wqattrs_prepare(wq, attrs); |
wanghaibin | 6201171 | 2016-01-07 20:38:59 +0800 | [diff] [blame] | 3811 | if (!ctx) |
| 3812 | return -ENOMEM; |
Lai Jiangshan | a0111cf | 2015-05-19 18:03:47 +0800 | [diff] [blame] | 3813 | |
| 3814 | /* the ctx has been prepared successfully, let's commit it */ |
wanghaibin | 6201171 | 2016-01-07 20:38:59 +0800 | [diff] [blame] | 3815 | apply_wqattrs_commit(ctx); |
Lai Jiangshan | a0111cf | 2015-05-19 18:03:47 +0800 | [diff] [blame] | 3816 | apply_wqattrs_cleanup(ctx); |
| 3817 | |
wanghaibin | 6201171 | 2016-01-07 20:38:59 +0800 | [diff] [blame] | 3818 | return 0; |
Lai Jiangshan | a0111cf | 2015-05-19 18:03:47 +0800 | [diff] [blame] | 3819 | } |
| 3820 | |
Tejun Heo | 9e8cd2f | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3821 | /** |
| 3822 | * apply_workqueue_attrs - apply new workqueue_attrs to an unbound workqueue |
| 3823 | * @wq: the target workqueue |
| 3824 | * @attrs: the workqueue_attrs to apply, allocated with alloc_workqueue_attrs() |
| 3825 | * |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3826 | * Apply @attrs to an unbound workqueue @wq. Unless disabled, on NUMA |
| 3827 | * machines, this function maps a separate pwq to each NUMA node with |
| 3828 | * possibles CPUs in @attrs->cpumask so that work items are affine to the |
| 3829 | * NUMA node it was issued on. Older pwqs are released as in-flight work |
| 3830 | * items finish. Note that a work item which repeatedly requeues itself |
| 3831 | * back-to-back will stay on its current pwq. |
Tejun Heo | 9e8cd2f | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3832 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 3833 | * Performs GFP_KERNEL allocations. |
| 3834 | * |
| 3835 | * Return: 0 on success and -errno on failure. |
Tejun Heo | 9e8cd2f | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3836 | */ |
| 3837 | int apply_workqueue_attrs(struct workqueue_struct *wq, |
| 3838 | const struct workqueue_attrs *attrs) |
| 3839 | { |
Lai Jiangshan | a0111cf | 2015-05-19 18:03:47 +0800 | [diff] [blame] | 3840 | int ret; |
Tejun Heo | 9e8cd2f | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3841 | |
Lai Jiangshan | a0111cf | 2015-05-19 18:03:47 +0800 | [diff] [blame] | 3842 | apply_wqattrs_lock(); |
| 3843 | ret = apply_workqueue_attrs_locked(wq, attrs); |
| 3844 | apply_wqattrs_unlock(); |
Lai Jiangshan | 2d5f076 | 2015-04-27 17:58:38 +0800 | [diff] [blame] | 3845 | |
Tejun Heo | 4862125 | 2013-04-01 11:23:31 -0700 | [diff] [blame] | 3846 | return ret; |
Tejun Heo | 9e8cd2f | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3847 | } |
| 3848 | |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3849 | /** |
| 3850 | * wq_update_unbound_numa - update NUMA affinity of a wq for CPU hot[un]plug |
| 3851 | * @wq: the target workqueue |
| 3852 | * @cpu: the CPU coming up or going down |
| 3853 | * @online: whether @cpu is coming up or going down |
| 3854 | * |
| 3855 | * This function is to be called from %CPU_DOWN_PREPARE, %CPU_ONLINE and |
| 3856 | * %CPU_DOWN_FAILED. @cpu is being hot[un]plugged, update NUMA affinity of |
| 3857 | * @wq accordingly. |
| 3858 | * |
| 3859 | * If NUMA affinity can't be adjusted due to memory allocation failure, it |
| 3860 | * falls back to @wq->dfl_pwq which may not be optimal but is always |
| 3861 | * correct. |
| 3862 | * |
| 3863 | * Note that when the last allowed CPU of a NUMA node goes offline for a |
| 3864 | * workqueue with a cpumask spanning multiple nodes, the workers which were |
| 3865 | * already executing the work items for the workqueue will lose their CPU |
| 3866 | * affinity and may execute on any CPU. This is similar to how per-cpu |
| 3867 | * workqueues behave on CPU_DOWN. If a workqueue user wants strict |
| 3868 | * affinity, it's the user's responsibility to flush the work item from |
| 3869 | * CPU_DOWN_PREPARE. |
| 3870 | */ |
| 3871 | static void wq_update_unbound_numa(struct workqueue_struct *wq, int cpu, |
| 3872 | bool online) |
| 3873 | { |
| 3874 | int node = cpu_to_node(cpu); |
| 3875 | int cpu_off = online ? -1 : cpu; |
| 3876 | struct pool_workqueue *old_pwq = NULL, *pwq; |
| 3877 | struct workqueue_attrs *target_attrs; |
| 3878 | cpumask_t *cpumask; |
| 3879 | |
| 3880 | lockdep_assert_held(&wq_pool_mutex); |
| 3881 | |
Lai Jiangshan | f7142ed | 2015-05-12 20:32:30 +0800 | [diff] [blame] | 3882 | if (!wq_numa_enabled || !(wq->flags & WQ_UNBOUND) || |
| 3883 | wq->unbound_attrs->no_numa) |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3884 | return; |
| 3885 | |
| 3886 | /* |
| 3887 | * We don't wanna alloc/free wq_attrs for each wq for each CPU. |
| 3888 | * Let's use a preallocated one. The following buf is protected by |
| 3889 | * CPU hotplug exclusion. |
| 3890 | */ |
| 3891 | target_attrs = wq_update_unbound_numa_attrs_buf; |
| 3892 | cpumask = target_attrs->cpumask; |
| 3893 | |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3894 | copy_workqueue_attrs(target_attrs, wq->unbound_attrs); |
| 3895 | pwq = unbound_pwq_by_node(wq, node); |
| 3896 | |
| 3897 | /* |
| 3898 | * Let's determine what needs to be done. If the target cpumask is |
Lai Jiangshan | 042f7df | 2015-04-30 17:16:12 +0800 | [diff] [blame] | 3899 | * different from the default pwq's, we need to compare it to @pwq's |
| 3900 | * and create a new one if they don't match. If the target cpumask |
| 3901 | * equals the default pwq's, the default pwq should be used. |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3902 | */ |
Lai Jiangshan | 042f7df | 2015-04-30 17:16:12 +0800 | [diff] [blame] | 3903 | if (wq_calc_node_cpumask(wq->dfl_pwq->pool->attrs, node, cpu_off, cpumask)) { |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3904 | if (cpumask_equal(cpumask, pwq->pool->attrs->cpumask)) |
Lai Jiangshan | f7142ed | 2015-05-12 20:32:30 +0800 | [diff] [blame] | 3905 | return; |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3906 | } else { |
Daeseok Youn | 534a3fb | 2014-04-18 09:08:14 +0900 | [diff] [blame] | 3907 | goto use_dfl_pwq; |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3908 | } |
| 3909 | |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3910 | /* create a new pwq */ |
| 3911 | pwq = alloc_unbound_pwq(wq, target_attrs); |
| 3912 | if (!pwq) { |
Fabian Frederick | 2d91603 | 2014-05-12 13:59:35 -0400 | [diff] [blame] | 3913 | pr_warn("workqueue: allocation failed while updating NUMA affinity of \"%s\"\n", |
| 3914 | wq->name); |
Daeseok Youn | 77f300b | 2014-04-16 14:32:29 +0900 | [diff] [blame] | 3915 | goto use_dfl_pwq; |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3916 | } |
| 3917 | |
Lai Jiangshan | f7142ed | 2015-05-12 20:32:30 +0800 | [diff] [blame] | 3918 | /* Install the new pwq. */ |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3919 | mutex_lock(&wq->mutex); |
| 3920 | old_pwq = numa_pwq_tbl_install(wq, node, pwq); |
| 3921 | goto out_unlock; |
| 3922 | |
| 3923 | use_dfl_pwq: |
Lai Jiangshan | f7142ed | 2015-05-12 20:32:30 +0800 | [diff] [blame] | 3924 | mutex_lock(&wq->mutex); |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3925 | spin_lock_irq(&wq->dfl_pwq->pool->lock); |
| 3926 | get_pwq(wq->dfl_pwq); |
| 3927 | spin_unlock_irq(&wq->dfl_pwq->pool->lock); |
| 3928 | old_pwq = numa_pwq_tbl_install(wq, node, wq->dfl_pwq); |
| 3929 | out_unlock: |
| 3930 | mutex_unlock(&wq->mutex); |
| 3931 | put_pwq_unlocked(old_pwq); |
| 3932 | } |
| 3933 | |
Tejun Heo | 30cdf24 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 3934 | static int alloc_and_link_pwqs(struct workqueue_struct *wq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3935 | { |
Tejun Heo | 49e3cf4 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 3936 | bool highpri = wq->flags & WQ_HIGHPRI; |
Tejun Heo | 8a2b753 | 2013-09-05 12:30:04 -0400 | [diff] [blame] | 3937 | int cpu, ret; |
Frederic Weisbecker | e1d8aa9 | 2009-01-12 23:15:46 +0100 | [diff] [blame] | 3938 | |
Tejun Heo | 30cdf24 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 3939 | if (!(wq->flags & WQ_UNBOUND)) { |
Tejun Heo | 420c0dd | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 3940 | wq->cpu_pwqs = alloc_percpu(struct pool_workqueue); |
| 3941 | if (!wq->cpu_pwqs) |
Tejun Heo | 30cdf24 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 3942 | return -ENOMEM; |
| 3943 | |
| 3944 | for_each_possible_cpu(cpu) { |
Tejun Heo | 7fb98ea | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 3945 | struct pool_workqueue *pwq = |
| 3946 | per_cpu_ptr(wq->cpu_pwqs, cpu); |
Tejun Heo | 7a62c2c | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3947 | struct worker_pool *cpu_pools = |
Tejun Heo | f02ae73 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3948 | per_cpu(cpu_worker_pools, cpu); |
Tejun Heo | 30cdf24 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 3949 | |
Tejun Heo | f147f29 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3950 | init_pwq(pwq, wq, &cpu_pools[highpri]); |
| 3951 | |
| 3952 | mutex_lock(&wq->mutex); |
Tejun Heo | 1befcf3 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3953 | link_pwq(pwq); |
Tejun Heo | f147f29 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3954 | mutex_unlock(&wq->mutex); |
Tejun Heo | 30cdf24 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 3955 | } |
Tejun Heo | 9e8cd2f | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3956 | return 0; |
Tejun Heo | 8a2b753 | 2013-09-05 12:30:04 -0400 | [diff] [blame] | 3957 | } else if (wq->flags & __WQ_ORDERED) { |
| 3958 | ret = apply_workqueue_attrs(wq, ordered_wq_attrs[highpri]); |
| 3959 | /* there should only be single pwq for ordering guarantee */ |
| 3960 | WARN(!ret && (wq->pwqs.next != &wq->dfl_pwq->pwqs_node || |
| 3961 | wq->pwqs.prev != &wq->dfl_pwq->pwqs_node), |
| 3962 | "ordering guarantee broken for workqueue %s\n", wq->name); |
| 3963 | return ret; |
Tejun Heo | 30cdf24 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 3964 | } else { |
Tejun Heo | 9e8cd2f | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3965 | return apply_workqueue_attrs(wq, unbound_std_wq_attrs[highpri]); |
Tejun Heo | 30cdf24 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 3966 | } |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 3967 | } |
| 3968 | |
Tejun Heo | f342179 | 2010-07-02 10:03:51 +0200 | [diff] [blame] | 3969 | static int wq_clamp_max_active(int max_active, unsigned int flags, |
| 3970 | const char *name) |
Tejun Heo | b71ab8c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 3971 | { |
Tejun Heo | f342179 | 2010-07-02 10:03:51 +0200 | [diff] [blame] | 3972 | int lim = flags & WQ_UNBOUND ? WQ_UNBOUND_MAX_ACTIVE : WQ_MAX_ACTIVE; |
| 3973 | |
| 3974 | if (max_active < 1 || max_active > lim) |
Valentin Ilie | 044c782 | 2012-08-19 00:52:42 +0300 | [diff] [blame] | 3975 | pr_warn("workqueue: max_active %d requested for %s is out of range, clamping between %d and %d\n", |
| 3976 | max_active, name, 1, lim); |
Tejun Heo | b71ab8c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 3977 | |
Tejun Heo | f342179 | 2010-07-02 10:03:51 +0200 | [diff] [blame] | 3978 | return clamp_val(max_active, 1, lim); |
Tejun Heo | b71ab8c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 3979 | } |
| 3980 | |
Tejun Heo | b196be8 | 2012-01-10 15:11:35 -0800 | [diff] [blame] | 3981 | struct workqueue_struct *__alloc_workqueue_key(const char *fmt, |
Tejun Heo | d320c03 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 3982 | unsigned int flags, |
| 3983 | int max_active, |
| 3984 | struct lock_class_key *key, |
Tejun Heo | b196be8 | 2012-01-10 15:11:35 -0800 | [diff] [blame] | 3985 | const char *lock_name, ...) |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 3986 | { |
Tejun Heo | df2d5ae | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3987 | size_t tbl_size = 0; |
Tejun Heo | ecf6881 | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 3988 | va_list args; |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 3989 | struct workqueue_struct *wq; |
Tejun Heo | 49e3cf4 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 3990 | struct pool_workqueue *pwq; |
Tejun Heo | b196be8 | 2012-01-10 15:11:35 -0800 | [diff] [blame] | 3991 | |
Tejun Heo | 5c0338c | 2017-07-18 18:41:52 -0400 | [diff] [blame] | 3992 | /* |
| 3993 | * Unbound && max_active == 1 used to imply ordered, which is no |
| 3994 | * longer the case on NUMA machines due to per-node pools. While |
| 3995 | * alloc_ordered_workqueue() is the right way to create an ordered |
| 3996 | * workqueue, keep the previous behavior to avoid subtle breakages |
| 3997 | * on NUMA. |
| 3998 | */ |
| 3999 | if ((flags & WQ_UNBOUND) && max_active == 1) |
| 4000 | flags |= __WQ_ORDERED; |
| 4001 | |
Viresh Kumar | cee22a1 | 2013-04-08 16:45:40 +0530 | [diff] [blame] | 4002 | /* see the comment above the definition of WQ_POWER_EFFICIENT */ |
| 4003 | if ((flags & WQ_POWER_EFFICIENT) && wq_power_efficient) |
| 4004 | flags |= WQ_UNBOUND; |
| 4005 | |
Tejun Heo | ecf6881 | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 4006 | /* allocate wq and format name */ |
Tejun Heo | df2d5ae | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 4007 | if (flags & WQ_UNBOUND) |
Lai Jiangshan | ddcb57e | 2014-07-22 13:05:40 +0800 | [diff] [blame] | 4008 | tbl_size = nr_node_ids * sizeof(wq->numa_pwq_tbl[0]); |
Tejun Heo | df2d5ae | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 4009 | |
| 4010 | wq = kzalloc(sizeof(*wq) + tbl_size, GFP_KERNEL); |
Tejun Heo | b196be8 | 2012-01-10 15:11:35 -0800 | [diff] [blame] | 4011 | if (!wq) |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4012 | return NULL; |
Tejun Heo | b196be8 | 2012-01-10 15:11:35 -0800 | [diff] [blame] | 4013 | |
Tejun Heo | 6029a91 | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 4014 | if (flags & WQ_UNBOUND) { |
| 4015 | wq->unbound_attrs = alloc_workqueue_attrs(GFP_KERNEL); |
| 4016 | if (!wq->unbound_attrs) |
| 4017 | goto err_free_wq; |
| 4018 | } |
| 4019 | |
Tejun Heo | ecf6881 | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 4020 | va_start(args, lock_name); |
| 4021 | vsnprintf(wq->name, sizeof(wq->name), fmt, args); |
Tejun Heo | b196be8 | 2012-01-10 15:11:35 -0800 | [diff] [blame] | 4022 | va_end(args); |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4023 | |
Tejun Heo | d320c03 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4024 | max_active = max_active ?: WQ_DFL_ACTIVE; |
Tejun Heo | b196be8 | 2012-01-10 15:11:35 -0800 | [diff] [blame] | 4025 | max_active = wq_clamp_max_active(max_active, flags, wq->name); |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4026 | |
Tejun Heo | b196be8 | 2012-01-10 15:11:35 -0800 | [diff] [blame] | 4027 | /* init wq */ |
Tejun Heo | 97e37d7 | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 4028 | wq->flags = flags; |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4029 | wq->saved_max_active = max_active; |
Lai Jiangshan | 3c25a55 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4030 | mutex_init(&wq->mutex); |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 4031 | atomic_set(&wq->nr_pwqs_to_flush, 0); |
Tejun Heo | 30cdf24 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4032 | INIT_LIST_HEAD(&wq->pwqs); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 4033 | INIT_LIST_HEAD(&wq->flusher_queue); |
| 4034 | INIT_LIST_HEAD(&wq->flusher_overflow); |
Tejun Heo | 493a172 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 4035 | INIT_LIST_HEAD(&wq->maydays); |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4036 | |
Johannes Berg | eb13ba8 | 2008-01-16 09:51:58 +0100 | [diff] [blame] | 4037 | lockdep_init_map(&wq->lockdep_map, lock_name, key, 0); |
Oleg Nesterov | cce1a16 | 2007-05-09 02:34:13 -0700 | [diff] [blame] | 4038 | INIT_LIST_HEAD(&wq->list); |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4039 | |
Tejun Heo | 30cdf24 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4040 | if (alloc_and_link_pwqs(wq) < 0) |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4041 | goto err_free_wq; |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4042 | |
Tejun Heo | 493008a | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 4043 | /* |
| 4044 | * Workqueues which may be used during memory reclaim should |
| 4045 | * have a rescuer to guarantee forward progress. |
| 4046 | */ |
| 4047 | if (flags & WQ_MEM_RECLAIM) { |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4048 | struct worker *rescuer; |
| 4049 | |
Lai Jiangshan | f7537df | 2014-07-15 17:24:15 +0800 | [diff] [blame] | 4050 | rescuer = alloc_worker(NUMA_NO_NODE); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4051 | if (!rescuer) |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4052 | goto err_destroy; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4053 | |
Tejun Heo | 111c225 | 2013-01-17 17:16:24 -0800 | [diff] [blame] | 4054 | rescuer->rescue_wq = wq; |
| 4055 | rescuer->task = kthread_create(rescuer_thread, rescuer, "%s", |
Tejun Heo | b196be8 | 2012-01-10 15:11:35 -0800 | [diff] [blame] | 4056 | wq->name); |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4057 | if (IS_ERR(rescuer->task)) { |
| 4058 | kfree(rescuer); |
| 4059 | goto err_destroy; |
| 4060 | } |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4061 | |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4062 | wq->rescuer = rescuer; |
Peter Zijlstra | 25834c7 | 2015-05-15 17:43:34 +0200 | [diff] [blame] | 4063 | kthread_bind_mask(rescuer->task, cpu_possible_mask); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4064 | wake_up_process(rescuer->task); |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4065 | } |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4066 | |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 4067 | if ((wq->flags & WQ_SYSFS) && workqueue_sysfs_register(wq)) |
| 4068 | goto err_destroy; |
| 4069 | |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4070 | /* |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4071 | * wq_pool_mutex protects global freeze state and workqueues list. |
| 4072 | * Grab it, adjust max_active and add the new @wq to workqueues |
| 4073 | * list. |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4074 | */ |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4075 | mutex_lock(&wq_pool_mutex); |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4076 | |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 4077 | mutex_lock(&wq->mutex); |
Tejun Heo | 699ce09 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 4078 | for_each_pwq(pwq, wq) |
| 4079 | pwq_adjust_max_active(pwq); |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 4080 | mutex_unlock(&wq->mutex); |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4081 | |
Tejun Heo | e2dca7a | 2015-03-09 09:22:28 -0400 | [diff] [blame] | 4082 | list_add_tail_rcu(&wq->list, &workqueues); |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4083 | |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4084 | mutex_unlock(&wq_pool_mutex); |
Tejun Heo | 1537663 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 4085 | |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4086 | return wq; |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4087 | |
| 4088 | err_free_wq: |
Tejun Heo | 6029a91 | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 4089 | free_workqueue_attrs(wq->unbound_attrs); |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4090 | kfree(wq); |
| 4091 | return NULL; |
| 4092 | err_destroy: |
| 4093 | destroy_workqueue(wq); |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 4094 | return NULL; |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4095 | } |
Tejun Heo | d320c03 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4096 | EXPORT_SYMBOL_GPL(__alloc_workqueue_key); |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4097 | |
| 4098 | /** |
| 4099 | * destroy_workqueue - safely terminate a workqueue |
| 4100 | * @wq: target workqueue |
| 4101 | * |
| 4102 | * Safely destroy a workqueue. All work currently pending will be done first. |
| 4103 | */ |
| 4104 | void destroy_workqueue(struct workqueue_struct *wq) |
| 4105 | { |
Tejun Heo | 49e3cf4 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4106 | struct pool_workqueue *pwq; |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 4107 | int node; |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4108 | |
Tejun Heo | 05905c2 | 2019-09-18 18:43:40 -0700 | [diff] [blame] | 4109 | /* |
| 4110 | * Remove it from sysfs first so that sanity check failure doesn't |
| 4111 | * lead to sysfs name conflicts. |
| 4112 | */ |
| 4113 | workqueue_sysfs_unregister(wq); |
| 4114 | |
Tejun Heo | 9c5a2ba | 2011-04-05 18:01:44 +0200 | [diff] [blame] | 4115 | /* drain it before proceeding with destruction */ |
| 4116 | drain_workqueue(wq); |
Tejun Heo | c8efcc2 | 2010-12-20 19:32:04 +0100 | [diff] [blame] | 4117 | |
Tejun Heo | 05905c2 | 2019-09-18 18:43:40 -0700 | [diff] [blame] | 4118 | /* kill rescuer, if sanity checks fail, leave it w/o rescuer */ |
| 4119 | if (wq->rescuer) { |
| 4120 | struct worker *rescuer = wq->rescuer; |
| 4121 | |
| 4122 | /* this prevents new queueing */ |
| 4123 | spin_lock_irq(&wq_mayday_lock); |
| 4124 | wq->rescuer = NULL; |
| 4125 | spin_unlock_irq(&wq_mayday_lock); |
| 4126 | |
| 4127 | /* rescuer will empty maydays list before exiting */ |
| 4128 | kthread_stop(rescuer->task); |
Tejun Heo | 80797bd | 2019-09-20 13:39:57 -0700 | [diff] [blame] | 4129 | kfree(rescuer); |
Tejun Heo | 05905c2 | 2019-09-18 18:43:40 -0700 | [diff] [blame] | 4130 | } |
| 4131 | |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4132 | /* sanity checks */ |
Lai Jiangshan | b09f4fd | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 4133 | mutex_lock(&wq->mutex); |
Tejun Heo | 49e3cf4 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4134 | for_each_pwq(pwq, wq) { |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4135 | int i; |
| 4136 | |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4137 | for (i = 0; i < WORK_NR_COLORS; i++) { |
| 4138 | if (WARN_ON(pwq->nr_in_flight[i])) { |
Lai Jiangshan | b09f4fd | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 4139 | mutex_unlock(&wq->mutex); |
Tejun Heo | fa07fb6 | 2016-09-05 08:54:06 -0400 | [diff] [blame] | 4140 | show_workqueue_state(); |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4141 | return; |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4142 | } |
| 4143 | } |
| 4144 | |
Lai Jiangshan | 5c52959 | 2013-04-04 10:05:38 +0800 | [diff] [blame] | 4145 | if (WARN_ON((pwq != wq->dfl_pwq) && (pwq->refcnt > 1)) || |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4146 | WARN_ON(pwq->nr_active) || |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4147 | WARN_ON(!list_empty(&pwq->delayed_works))) { |
Lai Jiangshan | b09f4fd | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 4148 | mutex_unlock(&wq->mutex); |
Tejun Heo | fa07fb6 | 2016-09-05 08:54:06 -0400 | [diff] [blame] | 4149 | show_workqueue_state(); |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4150 | return; |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4151 | } |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4152 | } |
Lai Jiangshan | b09f4fd | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 4153 | mutex_unlock(&wq->mutex); |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4154 | |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4155 | /* |
| 4156 | * wq list is used to freeze wq, remove from list after |
| 4157 | * flushing is complete in case freeze races us. |
| 4158 | */ |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4159 | mutex_lock(&wq_pool_mutex); |
Tejun Heo | e2dca7a | 2015-03-09 09:22:28 -0400 | [diff] [blame] | 4160 | list_del_rcu(&wq->list); |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4161 | mutex_unlock(&wq_pool_mutex); |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4162 | |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4163 | if (!(wq->flags & WQ_UNBOUND)) { |
| 4164 | /* |
| 4165 | * The base ref is never dropped on per-cpu pwqs. Directly |
Tejun Heo | e2dca7a | 2015-03-09 09:22:28 -0400 | [diff] [blame] | 4166 | * schedule RCU free. |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4167 | */ |
Tejun Heo | e2dca7a | 2015-03-09 09:22:28 -0400 | [diff] [blame] | 4168 | call_rcu_sched(&wq->rcu, rcu_free_wq); |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4169 | } else { |
| 4170 | /* |
| 4171 | * We're the sole accessor of @wq at this point. Directly |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 4172 | * access numa_pwq_tbl[] and dfl_pwq to put the base refs. |
| 4173 | * @wq will be freed when the last pwq is released. |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4174 | */ |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 4175 | for_each_node(node) { |
| 4176 | pwq = rcu_access_pointer(wq->numa_pwq_tbl[node]); |
| 4177 | RCU_INIT_POINTER(wq->numa_pwq_tbl[node], NULL); |
| 4178 | put_pwq_unlocked(pwq); |
| 4179 | } |
| 4180 | |
| 4181 | /* |
| 4182 | * Put dfl_pwq. @wq may be freed any time after dfl_pwq is |
| 4183 | * put. Don't access it afterwards. |
| 4184 | */ |
| 4185 | pwq = wq->dfl_pwq; |
| 4186 | wq->dfl_pwq = NULL; |
Tejun Heo | dce90d4 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 4187 | put_pwq_unlocked(pwq); |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 4188 | } |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4189 | } |
| 4190 | EXPORT_SYMBOL_GPL(destroy_workqueue); |
| 4191 | |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4192 | /** |
| 4193 | * workqueue_set_max_active - adjust max_active of a workqueue |
| 4194 | * @wq: target workqueue |
| 4195 | * @max_active: new max_active value. |
| 4196 | * |
| 4197 | * Set max_active of @wq to @max_active. |
| 4198 | * |
| 4199 | * CONTEXT: |
| 4200 | * Don't call from IRQ context. |
| 4201 | */ |
| 4202 | void workqueue_set_max_active(struct workqueue_struct *wq, int max_active) |
| 4203 | { |
Tejun Heo | 49e3cf4 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4204 | struct pool_workqueue *pwq; |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4205 | |
Tejun Heo | 8719dce | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4206 | /* disallow meddling with max_active for ordered workqueues */ |
Tejun Heo | 0a94efb | 2017-07-23 08:36:15 -0400 | [diff] [blame] | 4207 | if (WARN_ON(wq->flags & __WQ_ORDERED_EXPLICIT)) |
Tejun Heo | 8719dce | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4208 | return; |
| 4209 | |
Tejun Heo | f342179 | 2010-07-02 10:03:51 +0200 | [diff] [blame] | 4210 | max_active = wq_clamp_max_active(max_active, wq->flags, wq->name); |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4211 | |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 4212 | mutex_lock(&wq->mutex); |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4213 | |
Tejun Heo | 0a94efb | 2017-07-23 08:36:15 -0400 | [diff] [blame] | 4214 | wq->flags &= ~__WQ_ORDERED; |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4215 | wq->saved_max_active = max_active; |
| 4216 | |
Tejun Heo | 699ce09 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 4217 | for_each_pwq(pwq, wq) |
| 4218 | pwq_adjust_max_active(pwq); |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4219 | |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 4220 | mutex_unlock(&wq->mutex); |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4221 | } |
| 4222 | EXPORT_SYMBOL_GPL(workqueue_set_max_active); |
| 4223 | |
| 4224 | /** |
Lukas Wunner | 363e3fd | 2018-02-11 10:38:28 +0100 | [diff] [blame] | 4225 | * current_work - retrieve %current task's work struct |
| 4226 | * |
| 4227 | * Determine if %current task is a workqueue worker and what it's working on. |
| 4228 | * Useful to find out the context that the %current task is running in. |
| 4229 | * |
| 4230 | * Return: work struct if %current task is a workqueue worker, %NULL otherwise. |
| 4231 | */ |
| 4232 | struct work_struct *current_work(void) |
| 4233 | { |
| 4234 | struct worker *worker = current_wq_worker(); |
| 4235 | |
| 4236 | return worker ? worker->current_work : NULL; |
| 4237 | } |
| 4238 | EXPORT_SYMBOL(current_work); |
| 4239 | |
| 4240 | /** |
Tejun Heo | e6267616 | 2013-03-12 17:41:37 -0700 | [diff] [blame] | 4241 | * current_is_workqueue_rescuer - is %current workqueue rescuer? |
| 4242 | * |
| 4243 | * Determine whether %current is a workqueue rescuer. Can be used from |
| 4244 | * work functions to determine whether it's being run off the rescuer task. |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 4245 | * |
| 4246 | * Return: %true if %current is a workqueue rescuer. %false otherwise. |
Tejun Heo | e6267616 | 2013-03-12 17:41:37 -0700 | [diff] [blame] | 4247 | */ |
| 4248 | bool current_is_workqueue_rescuer(void) |
| 4249 | { |
| 4250 | struct worker *worker = current_wq_worker(); |
| 4251 | |
Lai Jiangshan | 6a092df | 2013-03-20 03:28:03 +0800 | [diff] [blame] | 4252 | return worker && worker->rescue_wq; |
Tejun Heo | e6267616 | 2013-03-12 17:41:37 -0700 | [diff] [blame] | 4253 | } |
| 4254 | |
| 4255 | /** |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4256 | * workqueue_congested - test whether a workqueue is congested |
| 4257 | * @cpu: CPU in question |
| 4258 | * @wq: target workqueue |
| 4259 | * |
| 4260 | * Test whether @wq's cpu workqueue for @cpu is congested. There is |
| 4261 | * no synchronization around this function and the test result is |
| 4262 | * unreliable and only useful as advisory hints or for debugging. |
| 4263 | * |
Tejun Heo | d325185 | 2013-05-10 11:10:17 -0700 | [diff] [blame] | 4264 | * If @cpu is WORK_CPU_UNBOUND, the test is performed on the local CPU. |
| 4265 | * Note that both per-cpu and unbound workqueues may be associated with |
| 4266 | * multiple pool_workqueues which have separate congested states. A |
| 4267 | * workqueue being congested on one CPU doesn't mean the workqueue is also |
| 4268 | * contested on other CPUs / NUMA nodes. |
| 4269 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 4270 | * Return: |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4271 | * %true if congested, %false otherwise. |
| 4272 | */ |
Tejun Heo | d84ff05 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 4273 | bool workqueue_congested(int cpu, struct workqueue_struct *wq) |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4274 | { |
Tejun Heo | 7fb98ea | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4275 | struct pool_workqueue *pwq; |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4276 | bool ret; |
| 4277 | |
Lai Jiangshan | 8810945 | 2013-03-20 03:28:10 +0800 | [diff] [blame] | 4278 | rcu_read_lock_sched(); |
Tejun Heo | 7fb98ea | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4279 | |
Tejun Heo | d325185 | 2013-05-10 11:10:17 -0700 | [diff] [blame] | 4280 | if (cpu == WORK_CPU_UNBOUND) |
| 4281 | cpu = smp_processor_id(); |
| 4282 | |
Tejun Heo | 7fb98ea | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4283 | if (!(wq->flags & WQ_UNBOUND)) |
| 4284 | pwq = per_cpu_ptr(wq->cpu_pwqs, cpu); |
| 4285 | else |
Tejun Heo | df2d5ae | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 4286 | pwq = unbound_pwq_by_node(wq, cpu_to_node(cpu)); |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4287 | |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4288 | ret = !list_empty(&pwq->delayed_works); |
Lai Jiangshan | 8810945 | 2013-03-20 03:28:10 +0800 | [diff] [blame] | 4289 | rcu_read_unlock_sched(); |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4290 | |
| 4291 | return ret; |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4292 | } |
| 4293 | EXPORT_SYMBOL_GPL(workqueue_congested); |
| 4294 | |
| 4295 | /** |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4296 | * work_busy - test whether a work is currently pending or running |
| 4297 | * @work: the work to be tested |
| 4298 | * |
| 4299 | * Test whether @work is currently pending or running. There is no |
| 4300 | * synchronization around this function and the test result is |
| 4301 | * unreliable and only useful as advisory hints or for debugging. |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4302 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 4303 | * Return: |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4304 | * OR'd bitmask of WORK_BUSY_* bits. |
| 4305 | */ |
| 4306 | unsigned int work_busy(struct work_struct *work) |
| 4307 | { |
Tejun Heo | fa1b54e | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4308 | struct worker_pool *pool; |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4309 | unsigned long flags; |
| 4310 | unsigned int ret = 0; |
| 4311 | |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4312 | if (work_pending(work)) |
| 4313 | ret |= WORK_BUSY_PENDING; |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4314 | |
Tejun Heo | fa1b54e | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4315 | local_irq_save(flags); |
| 4316 | pool = get_work_pool(work); |
Lai Jiangshan | 038366c | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 4317 | if (pool) { |
Tejun Heo | fa1b54e | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4318 | spin_lock(&pool->lock); |
Lai Jiangshan | 038366c | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 4319 | if (find_worker_executing_work(pool, work)) |
| 4320 | ret |= WORK_BUSY_RUNNING; |
Tejun Heo | fa1b54e | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4321 | spin_unlock(&pool->lock); |
Lai Jiangshan | 038366c | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 4322 | } |
Tejun Heo | fa1b54e | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4323 | local_irq_restore(flags); |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4324 | |
| 4325 | return ret; |
| 4326 | } |
| 4327 | EXPORT_SYMBOL_GPL(work_busy); |
| 4328 | |
Tejun Heo | 3d1cb20 | 2013-04-30 15:27:22 -0700 | [diff] [blame] | 4329 | /** |
| 4330 | * set_worker_desc - set description for the current work item |
| 4331 | * @fmt: printf-style format string |
| 4332 | * @...: arguments for the format string |
| 4333 | * |
| 4334 | * This function can be called by a running work function to describe what |
| 4335 | * the work item is about. If the worker task gets dumped, this |
| 4336 | * information will be printed out together to help debugging. The |
| 4337 | * description can be at most WORKER_DESC_LEN including the trailing '\0'. |
| 4338 | */ |
| 4339 | void set_worker_desc(const char *fmt, ...) |
| 4340 | { |
| 4341 | struct worker *worker = current_wq_worker(); |
| 4342 | va_list args; |
| 4343 | |
| 4344 | if (worker) { |
| 4345 | va_start(args, fmt); |
| 4346 | vsnprintf(worker->desc, sizeof(worker->desc), fmt, args); |
| 4347 | va_end(args); |
| 4348 | worker->desc_valid = true; |
| 4349 | } |
| 4350 | } |
| 4351 | |
| 4352 | /** |
| 4353 | * print_worker_info - print out worker information and description |
| 4354 | * @log_lvl: the log level to use when printing |
| 4355 | * @task: target task |
| 4356 | * |
| 4357 | * If @task is a worker and currently executing a work item, print out the |
| 4358 | * name of the workqueue being serviced and worker description set with |
| 4359 | * set_worker_desc() by the currently executing work item. |
| 4360 | * |
| 4361 | * This function can be safely called on any task as long as the |
| 4362 | * task_struct itself is accessible. While safe, this function isn't |
| 4363 | * synchronized and may print out mixups or garbages of limited length. |
| 4364 | */ |
| 4365 | void print_worker_info(const char *log_lvl, struct task_struct *task) |
| 4366 | { |
| 4367 | work_func_t *fn = NULL; |
| 4368 | char name[WQ_NAME_LEN] = { }; |
| 4369 | char desc[WORKER_DESC_LEN] = { }; |
| 4370 | struct pool_workqueue *pwq = NULL; |
| 4371 | struct workqueue_struct *wq = NULL; |
| 4372 | bool desc_valid = false; |
| 4373 | struct worker *worker; |
| 4374 | |
| 4375 | if (!(task->flags & PF_WQ_WORKER)) |
| 4376 | return; |
| 4377 | |
| 4378 | /* |
| 4379 | * This function is called without any synchronization and @task |
| 4380 | * could be in any state. Be careful with dereferences. |
| 4381 | */ |
Petr Mladek | e700591 | 2016-10-11 13:55:17 -0700 | [diff] [blame] | 4382 | worker = kthread_probe_data(task); |
Tejun Heo | 3d1cb20 | 2013-04-30 15:27:22 -0700 | [diff] [blame] | 4383 | |
| 4384 | /* |
| 4385 | * Carefully copy the associated workqueue's workfn and name. Keep |
| 4386 | * the original last '\0' in case the original contains garbage. |
| 4387 | */ |
| 4388 | probe_kernel_read(&fn, &worker->current_func, sizeof(fn)); |
| 4389 | probe_kernel_read(&pwq, &worker->current_pwq, sizeof(pwq)); |
| 4390 | probe_kernel_read(&wq, &pwq->wq, sizeof(wq)); |
| 4391 | probe_kernel_read(name, wq->name, sizeof(name) - 1); |
| 4392 | |
| 4393 | /* copy worker description */ |
| 4394 | probe_kernel_read(&desc_valid, &worker->desc_valid, sizeof(desc_valid)); |
| 4395 | if (desc_valid) |
| 4396 | probe_kernel_read(desc, worker->desc, sizeof(desc) - 1); |
| 4397 | |
| 4398 | if (fn || name[0] || desc[0]) { |
| 4399 | printk("%sWorkqueue: %s %pf", log_lvl, name, fn); |
| 4400 | if (desc[0]) |
| 4401 | pr_cont(" (%s)", desc); |
| 4402 | pr_cont("\n"); |
| 4403 | } |
| 4404 | } |
| 4405 | |
Tejun Heo | 3494fc3 | 2015-03-09 09:22:28 -0400 | [diff] [blame] | 4406 | static void pr_cont_pool_info(struct worker_pool *pool) |
| 4407 | { |
| 4408 | pr_cont(" cpus=%*pbl", nr_cpumask_bits, pool->attrs->cpumask); |
| 4409 | if (pool->node != NUMA_NO_NODE) |
| 4410 | pr_cont(" node=%d", pool->node); |
| 4411 | pr_cont(" flags=0x%x nice=%d", pool->flags, pool->attrs->nice); |
| 4412 | } |
| 4413 | |
| 4414 | static void pr_cont_work(bool comma, struct work_struct *work) |
| 4415 | { |
| 4416 | if (work->func == wq_barrier_func) { |
| 4417 | struct wq_barrier *barr; |
| 4418 | |
| 4419 | barr = container_of(work, struct wq_barrier, work); |
| 4420 | |
| 4421 | pr_cont("%s BAR(%d)", comma ? "," : "", |
| 4422 | task_pid_nr(barr->task)); |
| 4423 | } else { |
| 4424 | pr_cont("%s %pf", comma ? "," : "", work->func); |
| 4425 | } |
| 4426 | } |
| 4427 | |
| 4428 | static void show_pwq(struct pool_workqueue *pwq) |
| 4429 | { |
| 4430 | struct worker_pool *pool = pwq->pool; |
| 4431 | struct work_struct *work; |
| 4432 | struct worker *worker; |
| 4433 | bool has_in_flight = false, has_pending = false; |
| 4434 | int bkt; |
| 4435 | |
| 4436 | pr_info(" pwq %d:", pool->id); |
| 4437 | pr_cont_pool_info(pool); |
| 4438 | |
Tejun Heo | 9e0f33d | 2019-09-25 06:59:15 -0700 | [diff] [blame] | 4439 | pr_cont(" active=%d/%d refcnt=%d%s\n", |
| 4440 | pwq->nr_active, pwq->max_active, pwq->refcnt, |
Tejun Heo | 3494fc3 | 2015-03-09 09:22:28 -0400 | [diff] [blame] | 4441 | !list_empty(&pwq->mayday_node) ? " MAYDAY" : ""); |
| 4442 | |
| 4443 | hash_for_each(pool->busy_hash, bkt, worker, hentry) { |
| 4444 | if (worker->current_pwq == pwq) { |
| 4445 | has_in_flight = true; |
| 4446 | break; |
| 4447 | } |
| 4448 | } |
| 4449 | if (has_in_flight) { |
| 4450 | bool comma = false; |
| 4451 | |
| 4452 | pr_info(" in-flight:"); |
| 4453 | hash_for_each(pool->busy_hash, bkt, worker, hentry) { |
| 4454 | if (worker->current_pwq != pwq) |
| 4455 | continue; |
| 4456 | |
| 4457 | pr_cont("%s %d%s:%pf", comma ? "," : "", |
| 4458 | task_pid_nr(worker->task), |
| 4459 | worker == pwq->wq->rescuer ? "(RESCUER)" : "", |
| 4460 | worker->current_func); |
| 4461 | list_for_each_entry(work, &worker->scheduled, entry) |
| 4462 | pr_cont_work(false, work); |
| 4463 | comma = true; |
| 4464 | } |
| 4465 | pr_cont("\n"); |
| 4466 | } |
| 4467 | |
| 4468 | list_for_each_entry(work, &pool->worklist, entry) { |
| 4469 | if (get_work_pwq(work) == pwq) { |
| 4470 | has_pending = true; |
| 4471 | break; |
| 4472 | } |
| 4473 | } |
| 4474 | if (has_pending) { |
| 4475 | bool comma = false; |
| 4476 | |
| 4477 | pr_info(" pending:"); |
| 4478 | list_for_each_entry(work, &pool->worklist, entry) { |
| 4479 | if (get_work_pwq(work) != pwq) |
| 4480 | continue; |
| 4481 | |
| 4482 | pr_cont_work(comma, work); |
| 4483 | comma = !(*work_data_bits(work) & WORK_STRUCT_LINKED); |
| 4484 | } |
| 4485 | pr_cont("\n"); |
| 4486 | } |
| 4487 | |
| 4488 | if (!list_empty(&pwq->delayed_works)) { |
| 4489 | bool comma = false; |
| 4490 | |
| 4491 | pr_info(" delayed:"); |
| 4492 | list_for_each_entry(work, &pwq->delayed_works, entry) { |
| 4493 | pr_cont_work(comma, work); |
| 4494 | comma = !(*work_data_bits(work) & WORK_STRUCT_LINKED); |
| 4495 | } |
| 4496 | pr_cont("\n"); |
| 4497 | } |
| 4498 | } |
| 4499 | |
| 4500 | /** |
| 4501 | * show_workqueue_state - dump workqueue state |
| 4502 | * |
Roger Lu | 7b776af | 2016-07-01 11:05:02 +0800 | [diff] [blame] | 4503 | * Called from a sysrq handler or try_to_freeze_tasks() and prints out |
| 4504 | * all busy workqueues and pools. |
Tejun Heo | 3494fc3 | 2015-03-09 09:22:28 -0400 | [diff] [blame] | 4505 | */ |
| 4506 | void show_workqueue_state(void) |
| 4507 | { |
| 4508 | struct workqueue_struct *wq; |
| 4509 | struct worker_pool *pool; |
| 4510 | unsigned long flags; |
| 4511 | int pi; |
| 4512 | |
| 4513 | rcu_read_lock_sched(); |
| 4514 | |
| 4515 | pr_info("Showing busy workqueues and worker pools:\n"); |
| 4516 | |
| 4517 | list_for_each_entry_rcu(wq, &workqueues, list) { |
| 4518 | struct pool_workqueue *pwq; |
| 4519 | bool idle = true; |
| 4520 | |
| 4521 | for_each_pwq(pwq, wq) { |
| 4522 | if (pwq->nr_active || !list_empty(&pwq->delayed_works)) { |
| 4523 | idle = false; |
| 4524 | break; |
| 4525 | } |
| 4526 | } |
| 4527 | if (idle) |
| 4528 | continue; |
| 4529 | |
| 4530 | pr_info("workqueue %s: flags=0x%x\n", wq->name, wq->flags); |
| 4531 | |
| 4532 | for_each_pwq(pwq, wq) { |
| 4533 | spin_lock_irqsave(&pwq->pool->lock, flags); |
| 4534 | if (pwq->nr_active || !list_empty(&pwq->delayed_works)) |
| 4535 | show_pwq(pwq); |
| 4536 | spin_unlock_irqrestore(&pwq->pool->lock, flags); |
Sergey Senozhatsky | a6d5930 | 2018-01-11 09:53:35 +0900 | [diff] [blame] | 4537 | /* |
| 4538 | * We could be printing a lot from atomic context, e.g. |
| 4539 | * sysrq-t -> show_workqueue_state(). Avoid triggering |
| 4540 | * hard lockup. |
| 4541 | */ |
| 4542 | touch_nmi_watchdog(); |
Tejun Heo | 3494fc3 | 2015-03-09 09:22:28 -0400 | [diff] [blame] | 4543 | } |
| 4544 | } |
| 4545 | |
| 4546 | for_each_pool(pool, pi) { |
| 4547 | struct worker *worker; |
| 4548 | bool first = true; |
| 4549 | |
| 4550 | spin_lock_irqsave(&pool->lock, flags); |
| 4551 | if (pool->nr_workers == pool->nr_idle) |
| 4552 | goto next_pool; |
| 4553 | |
| 4554 | pr_info("pool %d:", pool->id); |
| 4555 | pr_cont_pool_info(pool); |
Tejun Heo | 82607adc | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 4556 | pr_cont(" hung=%us workers=%d", |
| 4557 | jiffies_to_msecs(jiffies - pool->watchdog_ts) / 1000, |
| 4558 | pool->nr_workers); |
Tejun Heo | 3494fc3 | 2015-03-09 09:22:28 -0400 | [diff] [blame] | 4559 | if (pool->manager) |
| 4560 | pr_cont(" manager: %d", |
| 4561 | task_pid_nr(pool->manager->task)); |
| 4562 | list_for_each_entry(worker, &pool->idle_list, entry) { |
| 4563 | pr_cont(" %s%d", first ? "idle: " : "", |
| 4564 | task_pid_nr(worker->task)); |
| 4565 | first = false; |
| 4566 | } |
| 4567 | pr_cont("\n"); |
| 4568 | next_pool: |
| 4569 | spin_unlock_irqrestore(&pool->lock, flags); |
Sergey Senozhatsky | a6d5930 | 2018-01-11 09:53:35 +0900 | [diff] [blame] | 4570 | /* |
| 4571 | * We could be printing a lot from atomic context, e.g. |
| 4572 | * sysrq-t -> show_workqueue_state(). Avoid triggering |
| 4573 | * hard lockup. |
| 4574 | */ |
| 4575 | touch_nmi_watchdog(); |
Tejun Heo | 3494fc3 | 2015-03-09 09:22:28 -0400 | [diff] [blame] | 4576 | } |
| 4577 | |
| 4578 | rcu_read_unlock_sched(); |
| 4579 | } |
| 4580 | |
Tejun Heo | db7bccf | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4581 | /* |
| 4582 | * CPU hotplug. |
| 4583 | * |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4584 | * There are two challenges in supporting CPU hotplug. Firstly, there |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 4585 | * are a lot of assumptions on strong associations among work, pwq and |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 4586 | * pool which make migrating pending and scheduled works very |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4587 | * difficult to implement without impacting hot paths. Secondly, |
Tejun Heo | 94cf58b | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 4588 | * worker pools serve mix of short, long and very long running works making |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4589 | * blocked draining impractical. |
| 4590 | * |
Tejun Heo | 2464757 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 4591 | * This is solved by allowing the pools to be disassociated from the CPU |
Tejun Heo | 628c78e | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 4592 | * running as an unbound one and allowing it to be reattached later if the |
| 4593 | * cpu comes back online. |
Tejun Heo | db7bccf | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4594 | */ |
| 4595 | |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 4596 | static void wq_unbind_fn(struct work_struct *work) |
Tejun Heo | db7bccf | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4597 | { |
Tejun Heo | 38db41d | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 4598 | int cpu = smp_processor_id(); |
Tejun Heo | 4ce62e9 | 2012-07-13 22:16:44 -0700 | [diff] [blame] | 4599 | struct worker_pool *pool; |
Tejun Heo | db7bccf | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4600 | struct worker *worker; |
Tejun Heo | db7bccf | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4601 | |
Tejun Heo | f02ae73 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 4602 | for_each_cpu_worker_pool(pool, cpu) { |
Lai Jiangshan | 92f9c5c | 2014-05-20 17:46:34 +0800 | [diff] [blame] | 4603 | mutex_lock(&pool->attach_mutex); |
Tejun Heo | 94cf58b | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 4604 | spin_lock_irq(&pool->lock); |
| 4605 | |
| 4606 | /* |
Lai Jiangshan | 92f9c5c | 2014-05-20 17:46:34 +0800 | [diff] [blame] | 4607 | * We've blocked all attach/detach operations. Make all workers |
Tejun Heo | 94cf58b | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 4608 | * unbound and set DISASSOCIATED. Before this, all workers |
| 4609 | * except for the ones which are still executing works from |
| 4610 | * before the last CPU down must be on the cpu. After |
| 4611 | * this, they may become diasporas. |
| 4612 | */ |
Lai Jiangshan | da02846 | 2014-05-20 17:46:31 +0800 | [diff] [blame] | 4613 | for_each_pool_worker(worker, pool) |
Tejun Heo | c9e7cf2 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 4614 | worker->flags |= WORKER_UNBOUND; |
Tejun Heo | db7bccf | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4615 | |
Tejun Heo | 2464757 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 4616 | pool->flags |= POOL_DISASSOCIATED; |
Tejun Heo | f2d5a0e | 2012-07-17 12:39:26 -0700 | [diff] [blame] | 4617 | |
Tejun Heo | 94cf58b | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 4618 | spin_unlock_irq(&pool->lock); |
Lai Jiangshan | 92f9c5c | 2014-05-20 17:46:34 +0800 | [diff] [blame] | 4619 | mutex_unlock(&pool->attach_mutex); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4620 | |
Lai Jiangshan | eb28342 | 2013-03-08 15:18:28 -0800 | [diff] [blame] | 4621 | /* |
| 4622 | * Call schedule() so that we cross rq->lock and thus can |
| 4623 | * guarantee sched callbacks see the %WORKER_UNBOUND flag. |
| 4624 | * This is necessary as scheduler callbacks may be invoked |
| 4625 | * from other cpus. |
| 4626 | */ |
| 4627 | schedule(); |
Tejun Heo | 628c78e | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 4628 | |
Lai Jiangshan | eb28342 | 2013-03-08 15:18:28 -0800 | [diff] [blame] | 4629 | /* |
| 4630 | * Sched callbacks are disabled now. Zap nr_running. |
| 4631 | * After this, nr_running stays zero and need_more_worker() |
| 4632 | * and keep_working() are always true as long as the |
| 4633 | * worklist is not empty. This pool now behaves as an |
| 4634 | * unbound (in terms of concurrency management) pool which |
| 4635 | * are served by workers tied to the pool. |
| 4636 | */ |
Tejun Heo | e19e397 | 2013-01-24 11:39:44 -0800 | [diff] [blame] | 4637 | atomic_set(&pool->nr_running, 0); |
Lai Jiangshan | eb28342 | 2013-03-08 15:18:28 -0800 | [diff] [blame] | 4638 | |
| 4639 | /* |
| 4640 | * With concurrency management just turned off, a busy |
| 4641 | * worker blocking could lead to lengthy stalls. Kick off |
| 4642 | * unbound chain execution of currently pending work items. |
| 4643 | */ |
| 4644 | spin_lock_irq(&pool->lock); |
| 4645 | wake_up_worker(pool); |
| 4646 | spin_unlock_irq(&pool->lock); |
| 4647 | } |
Tejun Heo | db7bccf | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4648 | } |
| 4649 | |
Tejun Heo | bd7c089 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4650 | /** |
| 4651 | * rebind_workers - rebind all workers of a pool to the associated CPU |
| 4652 | * @pool: pool of interest |
| 4653 | * |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4654 | * @pool->cpu is coming online. Rebind all workers to the CPU. |
Tejun Heo | bd7c089 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4655 | */ |
| 4656 | static void rebind_workers(struct worker_pool *pool) |
| 4657 | { |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4658 | struct worker *worker; |
Tejun Heo | bd7c089 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4659 | |
Lai Jiangshan | 92f9c5c | 2014-05-20 17:46:34 +0800 | [diff] [blame] | 4660 | lockdep_assert_held(&pool->attach_mutex); |
Tejun Heo | bd7c089 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4661 | |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4662 | /* |
| 4663 | * Restore CPU affinity of all workers. As all idle workers should |
| 4664 | * be on the run-queue of the associated CPU before any local |
Shailendra Verma | 402dd89 | 2015-05-23 10:38:14 +0530 | [diff] [blame] | 4665 | * wake-ups for concurrency management happen, restore CPU affinity |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4666 | * of all workers first and then clear UNBOUND. As we're called |
| 4667 | * from CPU_ONLINE, the following shouldn't fail. |
| 4668 | */ |
Lai Jiangshan | da02846 | 2014-05-20 17:46:31 +0800 | [diff] [blame] | 4669 | for_each_pool_worker(worker, pool) |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4670 | WARN_ON_ONCE(set_cpus_allowed_ptr(worker->task, |
| 4671 | pool->attrs->cpumask) < 0); |
| 4672 | |
| 4673 | spin_lock_irq(&pool->lock); |
Wanpeng Li | f7c17d2 | 2016-05-11 17:55:18 +0800 | [diff] [blame] | 4674 | |
| 4675 | /* |
| 4676 | * XXX: CPU hotplug notifiers are weird and can call DOWN_FAILED |
| 4677 | * w/o preceding DOWN_PREPARE. Work around it. CPU hotplug is |
| 4678 | * being reworked and this can go away in time. |
| 4679 | */ |
| 4680 | if (!(pool->flags & POOL_DISASSOCIATED)) { |
| 4681 | spin_unlock_irq(&pool->lock); |
| 4682 | return; |
| 4683 | } |
| 4684 | |
Lai Jiangshan | 3de5e88 | 2014-06-03 15:33:27 +0800 | [diff] [blame] | 4685 | pool->flags &= ~POOL_DISASSOCIATED; |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4686 | |
Lai Jiangshan | da02846 | 2014-05-20 17:46:31 +0800 | [diff] [blame] | 4687 | for_each_pool_worker(worker, pool) { |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4688 | unsigned int worker_flags = worker->flags; |
Tejun Heo | bd7c089 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4689 | |
| 4690 | /* |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4691 | * A bound idle worker should actually be on the runqueue |
| 4692 | * of the associated CPU for local wake-ups targeting it to |
| 4693 | * work. Kick all idle workers so that they migrate to the |
| 4694 | * associated CPU. Doing this in the same loop as |
| 4695 | * replacing UNBOUND with REBOUND is safe as no worker will |
| 4696 | * be bound before @pool->lock is released. |
Tejun Heo | bd7c089 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4697 | */ |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4698 | if (worker_flags & WORKER_IDLE) |
| 4699 | wake_up_process(worker->task); |
| 4700 | |
| 4701 | /* |
| 4702 | * We want to clear UNBOUND but can't directly call |
| 4703 | * worker_clr_flags() or adjust nr_running. Atomically |
| 4704 | * replace UNBOUND with another NOT_RUNNING flag REBOUND. |
| 4705 | * @worker will clear REBOUND using worker_clr_flags() when |
| 4706 | * it initiates the next execution cycle thus restoring |
| 4707 | * concurrency management. Note that when or whether |
| 4708 | * @worker clears REBOUND doesn't affect correctness. |
| 4709 | * |
| 4710 | * ACCESS_ONCE() is necessary because @worker->flags may be |
| 4711 | * tested without holding any lock in |
| 4712 | * wq_worker_waking_up(). Without it, NOT_RUNNING test may |
| 4713 | * fail incorrectly leading to premature concurrency |
| 4714 | * management operations. |
| 4715 | */ |
| 4716 | WARN_ON_ONCE(!(worker_flags & WORKER_UNBOUND)); |
| 4717 | worker_flags |= WORKER_REBOUND; |
| 4718 | worker_flags &= ~WORKER_UNBOUND; |
| 4719 | ACCESS_ONCE(worker->flags) = worker_flags; |
Tejun Heo | bd7c089 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4720 | } |
| 4721 | |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4722 | spin_unlock_irq(&pool->lock); |
Tejun Heo | bd7c089 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4723 | } |
| 4724 | |
Tejun Heo | 7dbc725 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4725 | /** |
| 4726 | * restore_unbound_workers_cpumask - restore cpumask of unbound workers |
| 4727 | * @pool: unbound pool of interest |
| 4728 | * @cpu: the CPU which is coming up |
| 4729 | * |
| 4730 | * An unbound pool may end up with a cpumask which doesn't have any online |
| 4731 | * CPUs. When a worker of such pool get scheduled, the scheduler resets |
| 4732 | * its cpus_allowed. If @cpu is in @pool's cpumask which didn't have any |
| 4733 | * online CPU before, cpus_allowed of all its workers should be restored. |
| 4734 | */ |
| 4735 | static void restore_unbound_workers_cpumask(struct worker_pool *pool, int cpu) |
| 4736 | { |
| 4737 | static cpumask_t cpumask; |
| 4738 | struct worker *worker; |
Tejun Heo | 7dbc725 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4739 | |
Lai Jiangshan | 92f9c5c | 2014-05-20 17:46:34 +0800 | [diff] [blame] | 4740 | lockdep_assert_held(&pool->attach_mutex); |
Tejun Heo | 7dbc725 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4741 | |
| 4742 | /* is @cpu allowed for @pool? */ |
| 4743 | if (!cpumask_test_cpu(cpu, pool->attrs->cpumask)) |
| 4744 | return; |
| 4745 | |
Tejun Heo | 7dbc725 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4746 | cpumask_and(&cpumask, pool->attrs->cpumask, cpu_online_mask); |
Tejun Heo | 7dbc725 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4747 | |
| 4748 | /* as we're called from CPU_ONLINE, the following shouldn't fail */ |
Lai Jiangshan | da02846 | 2014-05-20 17:46:31 +0800 | [diff] [blame] | 4749 | for_each_pool_worker(worker, pool) |
Peter Zijlstra | d945b5e | 2016-06-16 14:38:42 +0200 | [diff] [blame] | 4750 | WARN_ON_ONCE(set_cpus_allowed_ptr(worker->task, &cpumask) < 0); |
Tejun Heo | 7dbc725 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4751 | } |
| 4752 | |
Thomas Gleixner | 7ee681b | 2016-07-13 17:16:29 +0000 | [diff] [blame] | 4753 | int workqueue_prepare_cpu(unsigned int cpu) |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4754 | { |
Thomas Gleixner | 7ee681b | 2016-07-13 17:16:29 +0000 | [diff] [blame] | 4755 | struct worker_pool *pool; |
| 4756 | |
| 4757 | for_each_cpu_worker_pool(pool, cpu) { |
| 4758 | if (pool->nr_workers) |
| 4759 | continue; |
| 4760 | if (!create_worker(pool)) |
| 4761 | return -ENOMEM; |
| 4762 | } |
| 4763 | return 0; |
| 4764 | } |
| 4765 | |
| 4766 | int workqueue_online_cpu(unsigned int cpu) |
| 4767 | { |
Tejun Heo | 4ce62e9 | 2012-07-13 22:16:44 -0700 | [diff] [blame] | 4768 | struct worker_pool *pool; |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 4769 | struct workqueue_struct *wq; |
Tejun Heo | 7dbc725 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4770 | int pi; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4771 | |
Thomas Gleixner | 7ee681b | 2016-07-13 17:16:29 +0000 | [diff] [blame] | 4772 | mutex_lock(&wq_pool_mutex); |
Oleg Nesterov | 00dfcaf | 2008-04-29 01:00:27 -0700 | [diff] [blame] | 4773 | |
Thomas Gleixner | 7ee681b | 2016-07-13 17:16:29 +0000 | [diff] [blame] | 4774 | for_each_pool(pool, pi) { |
| 4775 | mutex_lock(&pool->attach_mutex); |
Tejun Heo | 7dbc725 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4776 | |
Thomas Gleixner | 7ee681b | 2016-07-13 17:16:29 +0000 | [diff] [blame] | 4777 | if (pool->cpu == cpu) |
| 4778 | rebind_workers(pool); |
| 4779 | else if (pool->cpu < 0) |
| 4780 | restore_unbound_workers_cpumask(pool, cpu); |
Tejun Heo | 94cf58b | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 4781 | |
Thomas Gleixner | 7ee681b | 2016-07-13 17:16:29 +0000 | [diff] [blame] | 4782 | mutex_unlock(&pool->attach_mutex); |
Tejun Heo | 6575820 | 2012-07-17 12:39:26 -0700 | [diff] [blame] | 4783 | } |
Thomas Gleixner | 7ee681b | 2016-07-13 17:16:29 +0000 | [diff] [blame] | 4784 | |
| 4785 | /* update NUMA affinity of unbound workqueues */ |
| 4786 | list_for_each_entry(wq, &workqueues, list) |
| 4787 | wq_update_unbound_numa(wq, cpu, true); |
| 4788 | |
| 4789 | mutex_unlock(&wq_pool_mutex); |
| 4790 | return 0; |
Tejun Heo | 6575820 | 2012-07-17 12:39:26 -0700 | [diff] [blame] | 4791 | } |
| 4792 | |
Thomas Gleixner | 7ee681b | 2016-07-13 17:16:29 +0000 | [diff] [blame] | 4793 | int workqueue_offline_cpu(unsigned int cpu) |
Tejun Heo | 6575820 | 2012-07-17 12:39:26 -0700 | [diff] [blame] | 4794 | { |
Tejun Heo | 8db25e7 | 2012-07-17 12:39:28 -0700 | [diff] [blame] | 4795 | struct work_struct unbind_work; |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 4796 | struct workqueue_struct *wq; |
Tejun Heo | 8db25e7 | 2012-07-17 12:39:28 -0700 | [diff] [blame] | 4797 | |
Thomas Gleixner | 7ee681b | 2016-07-13 17:16:29 +0000 | [diff] [blame] | 4798 | /* unbinding per-cpu workers should happen on the local CPU */ |
| 4799 | INIT_WORK_ONSTACK(&unbind_work, wq_unbind_fn); |
| 4800 | queue_work_on(cpu, system_highpri_wq, &unbind_work); |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 4801 | |
Thomas Gleixner | 7ee681b | 2016-07-13 17:16:29 +0000 | [diff] [blame] | 4802 | /* update NUMA affinity of unbound workqueues */ |
| 4803 | mutex_lock(&wq_pool_mutex); |
| 4804 | list_for_each_entry(wq, &workqueues, list) |
| 4805 | wq_update_unbound_numa(wq, cpu, false); |
| 4806 | mutex_unlock(&wq_pool_mutex); |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 4807 | |
Thomas Gleixner | 7ee681b | 2016-07-13 17:16:29 +0000 | [diff] [blame] | 4808 | /* wait for per-cpu unbinding to finish */ |
| 4809 | flush_work(&unbind_work); |
| 4810 | destroy_work_on_stack(&unbind_work); |
| 4811 | return 0; |
Tejun Heo | 6575820 | 2012-07-17 12:39:26 -0700 | [diff] [blame] | 4812 | } |
| 4813 | |
Rusty Russell | 2d3854a | 2008-11-05 13:39:10 +1100 | [diff] [blame] | 4814 | #ifdef CONFIG_SMP |
Rusty Russell | 8ccad40 | 2009-01-16 15:31:15 -0800 | [diff] [blame] | 4815 | |
Rusty Russell | 2d3854a | 2008-11-05 13:39:10 +1100 | [diff] [blame] | 4816 | struct work_for_cpu { |
Tejun Heo | ed48ece | 2012-09-18 12:48:43 -0700 | [diff] [blame] | 4817 | struct work_struct work; |
Rusty Russell | 2d3854a | 2008-11-05 13:39:10 +1100 | [diff] [blame] | 4818 | long (*fn)(void *); |
| 4819 | void *arg; |
| 4820 | long ret; |
| 4821 | }; |
| 4822 | |
Tejun Heo | ed48ece | 2012-09-18 12:48:43 -0700 | [diff] [blame] | 4823 | static void work_for_cpu_fn(struct work_struct *work) |
Rusty Russell | 2d3854a | 2008-11-05 13:39:10 +1100 | [diff] [blame] | 4824 | { |
Tejun Heo | ed48ece | 2012-09-18 12:48:43 -0700 | [diff] [blame] | 4825 | struct work_for_cpu *wfc = container_of(work, struct work_for_cpu, work); |
| 4826 | |
Rusty Russell | 2d3854a | 2008-11-05 13:39:10 +1100 | [diff] [blame] | 4827 | wfc->ret = wfc->fn(wfc->arg); |
| 4828 | } |
| 4829 | |
| 4830 | /** |
Anna-Maria Gleixner | 22aceb3 | 2016-03-10 12:07:38 +0100 | [diff] [blame] | 4831 | * work_on_cpu - run a function in thread context on a particular cpu |
Rusty Russell | 2d3854a | 2008-11-05 13:39:10 +1100 | [diff] [blame] | 4832 | * @cpu: the cpu to run on |
| 4833 | * @fn: the function to run |
| 4834 | * @arg: the function arg |
| 4835 | * |
Rusty Russell | 31ad908 | 2009-01-16 15:31:15 -0800 | [diff] [blame] | 4836 | * It is up to the caller to ensure that the cpu doesn't go offline. |
Andrew Morton | 6b44003e | 2009-04-09 09:50:37 -0600 | [diff] [blame] | 4837 | * The caller must not hold any locks which would prevent @fn from completing. |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 4838 | * |
| 4839 | * Return: The value @fn returns. |
Rusty Russell | 2d3854a | 2008-11-05 13:39:10 +1100 | [diff] [blame] | 4840 | */ |
Tejun Heo | d84ff05 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 4841 | long work_on_cpu(int cpu, long (*fn)(void *), void *arg) |
Rusty Russell | 2d3854a | 2008-11-05 13:39:10 +1100 | [diff] [blame] | 4842 | { |
Tejun Heo | ed48ece | 2012-09-18 12:48:43 -0700 | [diff] [blame] | 4843 | struct work_for_cpu wfc = { .fn = fn, .arg = arg }; |
Rusty Russell | 2d3854a | 2008-11-05 13:39:10 +1100 | [diff] [blame] | 4844 | |
Tejun Heo | ed48ece | 2012-09-18 12:48:43 -0700 | [diff] [blame] | 4845 | INIT_WORK_ONSTACK(&wfc.work, work_for_cpu_fn); |
| 4846 | schedule_work_on(cpu, &wfc.work); |
Bjorn Helgaas | 12997d1 | 2013-11-18 11:00:29 -0700 | [diff] [blame] | 4847 | flush_work(&wfc.work); |
Chuansheng Liu | 440a113 | 2014-01-11 22:26:33 -0500 | [diff] [blame] | 4848 | destroy_work_on_stack(&wfc.work); |
Rusty Russell | 2d3854a | 2008-11-05 13:39:10 +1100 | [diff] [blame] | 4849 | return wfc.ret; |
| 4850 | } |
| 4851 | EXPORT_SYMBOL_GPL(work_on_cpu); |
Thomas Gleixner | 0e8d6a9 | 2017-04-12 22:07:28 +0200 | [diff] [blame] | 4852 | |
| 4853 | /** |
| 4854 | * work_on_cpu_safe - run a function in thread context on a particular cpu |
| 4855 | * @cpu: the cpu to run on |
| 4856 | * @fn: the function to run |
| 4857 | * @arg: the function argument |
| 4858 | * |
| 4859 | * Disables CPU hotplug and calls work_on_cpu(). The caller must not hold |
| 4860 | * any locks which would prevent @fn from completing. |
| 4861 | * |
| 4862 | * Return: The value @fn returns. |
| 4863 | */ |
| 4864 | long work_on_cpu_safe(int cpu, long (*fn)(void *), void *arg) |
| 4865 | { |
| 4866 | long ret = -ENODEV; |
| 4867 | |
| 4868 | get_online_cpus(); |
| 4869 | if (cpu_online(cpu)) |
| 4870 | ret = work_on_cpu(cpu, fn, arg); |
| 4871 | put_online_cpus(); |
| 4872 | return ret; |
| 4873 | } |
| 4874 | EXPORT_SYMBOL_GPL(work_on_cpu_safe); |
Rusty Russell | 2d3854a | 2008-11-05 13:39:10 +1100 | [diff] [blame] | 4875 | #endif /* CONFIG_SMP */ |
| 4876 | |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4877 | #ifdef CONFIG_FREEZER |
Rusty Russell | e7577c5 | 2009-01-01 10:12:25 +1030 | [diff] [blame] | 4878 | |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4879 | /** |
| 4880 | * freeze_workqueues_begin - begin freezing workqueues |
| 4881 | * |
Tejun Heo | 58a69cb | 2011-02-16 09:25:31 +0100 | [diff] [blame] | 4882 | * Start freezing workqueues. After this function returns, all freezable |
Tejun Heo | c5aa87b | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 4883 | * workqueues will queue new works to their delayed_works list instead of |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 4884 | * pool->worklist. |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4885 | * |
| 4886 | * CONTEXT: |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 4887 | * Grabs and releases wq_pool_mutex, wq->mutex and pool->lock's. |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4888 | */ |
| 4889 | void freeze_workqueues_begin(void) |
| 4890 | { |
Tejun Heo | 24b8a84 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4891 | struct workqueue_struct *wq; |
| 4892 | struct pool_workqueue *pwq; |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4893 | |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4894 | mutex_lock(&wq_pool_mutex); |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4895 | |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4896 | WARN_ON_ONCE(workqueue_freezing); |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4897 | workqueue_freezing = true; |
| 4898 | |
Tejun Heo | 24b8a84 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4899 | list_for_each_entry(wq, &workqueues, list) { |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 4900 | mutex_lock(&wq->mutex); |
Tejun Heo | 699ce09 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 4901 | for_each_pwq(pwq, wq) |
| 4902 | pwq_adjust_max_active(pwq); |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 4903 | mutex_unlock(&wq->mutex); |
Tejun Heo | 24b8a84 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4904 | } |
Tejun Heo | 5bcab33 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 4905 | |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4906 | mutex_unlock(&wq_pool_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4907 | } |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4908 | |
| 4909 | /** |
Tejun Heo | 58a69cb | 2011-02-16 09:25:31 +0100 | [diff] [blame] | 4910 | * freeze_workqueues_busy - are freezable workqueues still busy? |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4911 | * |
| 4912 | * Check whether freezing is complete. This function must be called |
| 4913 | * between freeze_workqueues_begin() and thaw_workqueues(). |
| 4914 | * |
| 4915 | * CONTEXT: |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4916 | * Grabs and releases wq_pool_mutex. |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4917 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 4918 | * Return: |
Tejun Heo | 58a69cb | 2011-02-16 09:25:31 +0100 | [diff] [blame] | 4919 | * %true if some freezable workqueues are still busy. %false if freezing |
| 4920 | * is complete. |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4921 | */ |
| 4922 | bool freeze_workqueues_busy(void) |
| 4923 | { |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4924 | bool busy = false; |
Tejun Heo | 24b8a84 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4925 | struct workqueue_struct *wq; |
| 4926 | struct pool_workqueue *pwq; |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4927 | |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4928 | mutex_lock(&wq_pool_mutex); |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4929 | |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4930 | WARN_ON_ONCE(!workqueue_freezing); |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4931 | |
Tejun Heo | 24b8a84 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4932 | list_for_each_entry(wq, &workqueues, list) { |
| 4933 | if (!(wq->flags & WQ_FREEZABLE)) |
| 4934 | continue; |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4935 | /* |
| 4936 | * nr_active is monotonically decreasing. It's safe |
| 4937 | * to peek without lock. |
| 4938 | */ |
Lai Jiangshan | 8810945 | 2013-03-20 03:28:10 +0800 | [diff] [blame] | 4939 | rcu_read_lock_sched(); |
Tejun Heo | 24b8a84 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4940 | for_each_pwq(pwq, wq) { |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4941 | WARN_ON_ONCE(pwq->nr_active < 0); |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 4942 | if (pwq->nr_active) { |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4943 | busy = true; |
Lai Jiangshan | 8810945 | 2013-03-20 03:28:10 +0800 | [diff] [blame] | 4944 | rcu_read_unlock_sched(); |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4945 | goto out_unlock; |
| 4946 | } |
| 4947 | } |
Lai Jiangshan | 8810945 | 2013-03-20 03:28:10 +0800 | [diff] [blame] | 4948 | rcu_read_unlock_sched(); |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4949 | } |
| 4950 | out_unlock: |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4951 | mutex_unlock(&wq_pool_mutex); |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4952 | return busy; |
| 4953 | } |
| 4954 | |
| 4955 | /** |
| 4956 | * thaw_workqueues - thaw workqueues |
| 4957 | * |
| 4958 | * Thaw workqueues. Normal queueing is restored and all collected |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 4959 | * frozen works are transferred to their respective pool worklists. |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4960 | * |
| 4961 | * CONTEXT: |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 4962 | * Grabs and releases wq_pool_mutex, wq->mutex and pool->lock's. |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4963 | */ |
| 4964 | void thaw_workqueues(void) |
| 4965 | { |
Tejun Heo | 24b8a84 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4966 | struct workqueue_struct *wq; |
| 4967 | struct pool_workqueue *pwq; |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4968 | |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4969 | mutex_lock(&wq_pool_mutex); |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4970 | |
| 4971 | if (!workqueue_freezing) |
| 4972 | goto out_unlock; |
| 4973 | |
Lai Jiangshan | 74b414e | 2014-05-22 19:01:16 +0800 | [diff] [blame] | 4974 | workqueue_freezing = false; |
Tejun Heo | 8b03ae3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4975 | |
Tejun Heo | 24b8a84 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4976 | /* restore max_active and repopulate worklist */ |
| 4977 | list_for_each_entry(wq, &workqueues, list) { |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 4978 | mutex_lock(&wq->mutex); |
Tejun Heo | 699ce09 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 4979 | for_each_pwq(pwq, wq) |
| 4980 | pwq_adjust_max_active(pwq); |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 4981 | mutex_unlock(&wq->mutex); |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4982 | } |
| 4983 | |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4984 | out_unlock: |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4985 | mutex_unlock(&wq_pool_mutex); |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4986 | } |
| 4987 | #endif /* CONFIG_FREEZER */ |
| 4988 | |
Lai Jiangshan | 042f7df | 2015-04-30 17:16:12 +0800 | [diff] [blame] | 4989 | static int workqueue_apply_unbound_cpumask(void) |
| 4990 | { |
| 4991 | LIST_HEAD(ctxs); |
| 4992 | int ret = 0; |
| 4993 | struct workqueue_struct *wq; |
| 4994 | struct apply_wqattrs_ctx *ctx, *n; |
| 4995 | |
| 4996 | lockdep_assert_held(&wq_pool_mutex); |
| 4997 | |
| 4998 | list_for_each_entry(wq, &workqueues, list) { |
| 4999 | if (!(wq->flags & WQ_UNBOUND)) |
| 5000 | continue; |
| 5001 | /* creating multiple pwqs breaks ordering guarantee */ |
| 5002 | if (wq->flags & __WQ_ORDERED) |
| 5003 | continue; |
| 5004 | |
| 5005 | ctx = apply_wqattrs_prepare(wq, wq->unbound_attrs); |
| 5006 | if (!ctx) { |
| 5007 | ret = -ENOMEM; |
| 5008 | break; |
| 5009 | } |
| 5010 | |
| 5011 | list_add_tail(&ctx->list, &ctxs); |
| 5012 | } |
| 5013 | |
| 5014 | list_for_each_entry_safe(ctx, n, &ctxs, list) { |
| 5015 | if (!ret) |
| 5016 | apply_wqattrs_commit(ctx); |
| 5017 | apply_wqattrs_cleanup(ctx); |
| 5018 | } |
| 5019 | |
| 5020 | return ret; |
| 5021 | } |
| 5022 | |
| 5023 | /** |
| 5024 | * workqueue_set_unbound_cpumask - Set the low-level unbound cpumask |
| 5025 | * @cpumask: the cpumask to set |
| 5026 | * |
| 5027 | * The low-level workqueues cpumask is a global cpumask that limits |
| 5028 | * the affinity of all unbound workqueues. This function check the @cpumask |
| 5029 | * and apply it to all unbound workqueues and updates all pwqs of them. |
| 5030 | * |
| 5031 | * Retun: 0 - Success |
| 5032 | * -EINVAL - Invalid @cpumask |
| 5033 | * -ENOMEM - Failed to allocate memory for attrs or pwqs. |
| 5034 | */ |
| 5035 | int workqueue_set_unbound_cpumask(cpumask_var_t cpumask) |
| 5036 | { |
| 5037 | int ret = -EINVAL; |
| 5038 | cpumask_var_t saved_cpumask; |
| 5039 | |
| 5040 | if (!zalloc_cpumask_var(&saved_cpumask, GFP_KERNEL)) |
| 5041 | return -ENOMEM; |
| 5042 | |
Lai Jiangshan | 042f7df | 2015-04-30 17:16:12 +0800 | [diff] [blame] | 5043 | cpumask_and(cpumask, cpumask, cpu_possible_mask); |
| 5044 | if (!cpumask_empty(cpumask)) { |
Lai Jiangshan | a0111cf | 2015-05-19 18:03:47 +0800 | [diff] [blame] | 5045 | apply_wqattrs_lock(); |
Lai Jiangshan | 042f7df | 2015-04-30 17:16:12 +0800 | [diff] [blame] | 5046 | |
| 5047 | /* save the old wq_unbound_cpumask. */ |
| 5048 | cpumask_copy(saved_cpumask, wq_unbound_cpumask); |
| 5049 | |
| 5050 | /* update wq_unbound_cpumask at first and apply it to wqs. */ |
| 5051 | cpumask_copy(wq_unbound_cpumask, cpumask); |
| 5052 | ret = workqueue_apply_unbound_cpumask(); |
| 5053 | |
| 5054 | /* restore the wq_unbound_cpumask when failed. */ |
| 5055 | if (ret < 0) |
| 5056 | cpumask_copy(wq_unbound_cpumask, saved_cpumask); |
| 5057 | |
Lai Jiangshan | a0111cf | 2015-05-19 18:03:47 +0800 | [diff] [blame] | 5058 | apply_wqattrs_unlock(); |
Lai Jiangshan | 042f7df | 2015-04-30 17:16:12 +0800 | [diff] [blame] | 5059 | } |
Lai Jiangshan | 042f7df | 2015-04-30 17:16:12 +0800 | [diff] [blame] | 5060 | |
| 5061 | free_cpumask_var(saved_cpumask); |
| 5062 | return ret; |
| 5063 | } |
| 5064 | |
Frederic Weisbecker | 6ba9442 | 2015-04-02 19:14:39 +0800 | [diff] [blame] | 5065 | #ifdef CONFIG_SYSFS |
| 5066 | /* |
| 5067 | * Workqueues with WQ_SYSFS flag set is visible to userland via |
| 5068 | * /sys/bus/workqueue/devices/WQ_NAME. All visible workqueues have the |
| 5069 | * following attributes. |
| 5070 | * |
| 5071 | * per_cpu RO bool : whether the workqueue is per-cpu or unbound |
| 5072 | * max_active RW int : maximum number of in-flight work items |
| 5073 | * |
| 5074 | * Unbound workqueues have the following extra attributes. |
| 5075 | * |
| 5076 | * id RO int : the associated pool ID |
| 5077 | * nice RW int : nice value of the workers |
| 5078 | * cpumask RW mask : bitmask of allowed CPUs for the workers |
| 5079 | */ |
| 5080 | struct wq_device { |
| 5081 | struct workqueue_struct *wq; |
| 5082 | struct device dev; |
| 5083 | }; |
| 5084 | |
| 5085 | static struct workqueue_struct *dev_to_wq(struct device *dev) |
| 5086 | { |
| 5087 | struct wq_device *wq_dev = container_of(dev, struct wq_device, dev); |
| 5088 | |
| 5089 | return wq_dev->wq; |
| 5090 | } |
| 5091 | |
| 5092 | static ssize_t per_cpu_show(struct device *dev, struct device_attribute *attr, |
| 5093 | char *buf) |
| 5094 | { |
| 5095 | struct workqueue_struct *wq = dev_to_wq(dev); |
| 5096 | |
| 5097 | return scnprintf(buf, PAGE_SIZE, "%d\n", (bool)!(wq->flags & WQ_UNBOUND)); |
| 5098 | } |
| 5099 | static DEVICE_ATTR_RO(per_cpu); |
| 5100 | |
| 5101 | static ssize_t max_active_show(struct device *dev, |
| 5102 | struct device_attribute *attr, char *buf) |
| 5103 | { |
| 5104 | struct workqueue_struct *wq = dev_to_wq(dev); |
| 5105 | |
| 5106 | return scnprintf(buf, PAGE_SIZE, "%d\n", wq->saved_max_active); |
| 5107 | } |
| 5108 | |
| 5109 | static ssize_t max_active_store(struct device *dev, |
| 5110 | struct device_attribute *attr, const char *buf, |
| 5111 | size_t count) |
| 5112 | { |
| 5113 | struct workqueue_struct *wq = dev_to_wq(dev); |
| 5114 | int val; |
| 5115 | |
| 5116 | if (sscanf(buf, "%d", &val) != 1 || val <= 0) |
| 5117 | return -EINVAL; |
| 5118 | |
| 5119 | workqueue_set_max_active(wq, val); |
| 5120 | return count; |
| 5121 | } |
| 5122 | static DEVICE_ATTR_RW(max_active); |
| 5123 | |
| 5124 | static struct attribute *wq_sysfs_attrs[] = { |
| 5125 | &dev_attr_per_cpu.attr, |
| 5126 | &dev_attr_max_active.attr, |
| 5127 | NULL, |
| 5128 | }; |
| 5129 | ATTRIBUTE_GROUPS(wq_sysfs); |
| 5130 | |
| 5131 | static ssize_t wq_pool_ids_show(struct device *dev, |
| 5132 | struct device_attribute *attr, char *buf) |
| 5133 | { |
| 5134 | struct workqueue_struct *wq = dev_to_wq(dev); |
| 5135 | const char *delim = ""; |
| 5136 | int node, written = 0; |
| 5137 | |
| 5138 | rcu_read_lock_sched(); |
| 5139 | for_each_node(node) { |
| 5140 | written += scnprintf(buf + written, PAGE_SIZE - written, |
| 5141 | "%s%d:%d", delim, node, |
| 5142 | unbound_pwq_by_node(wq, node)->pool->id); |
| 5143 | delim = " "; |
| 5144 | } |
| 5145 | written += scnprintf(buf + written, PAGE_SIZE - written, "\n"); |
| 5146 | rcu_read_unlock_sched(); |
| 5147 | |
| 5148 | return written; |
| 5149 | } |
| 5150 | |
| 5151 | static ssize_t wq_nice_show(struct device *dev, struct device_attribute *attr, |
| 5152 | char *buf) |
| 5153 | { |
| 5154 | struct workqueue_struct *wq = dev_to_wq(dev); |
| 5155 | int written; |
| 5156 | |
| 5157 | mutex_lock(&wq->mutex); |
| 5158 | written = scnprintf(buf, PAGE_SIZE, "%d\n", wq->unbound_attrs->nice); |
| 5159 | mutex_unlock(&wq->mutex); |
| 5160 | |
| 5161 | return written; |
| 5162 | } |
| 5163 | |
| 5164 | /* prepare workqueue_attrs for sysfs store operations */ |
| 5165 | static struct workqueue_attrs *wq_sysfs_prep_attrs(struct workqueue_struct *wq) |
| 5166 | { |
| 5167 | struct workqueue_attrs *attrs; |
| 5168 | |
Lai Jiangshan | 899a94f | 2015-05-20 14:41:18 +0800 | [diff] [blame] | 5169 | lockdep_assert_held(&wq_pool_mutex); |
| 5170 | |
Frederic Weisbecker | 6ba9442 | 2015-04-02 19:14:39 +0800 | [diff] [blame] | 5171 | attrs = alloc_workqueue_attrs(GFP_KERNEL); |
| 5172 | if (!attrs) |
| 5173 | return NULL; |
| 5174 | |
Frederic Weisbecker | 6ba9442 | 2015-04-02 19:14:39 +0800 | [diff] [blame] | 5175 | copy_workqueue_attrs(attrs, wq->unbound_attrs); |
Frederic Weisbecker | 6ba9442 | 2015-04-02 19:14:39 +0800 | [diff] [blame] | 5176 | return attrs; |
| 5177 | } |
| 5178 | |
| 5179 | static ssize_t wq_nice_store(struct device *dev, struct device_attribute *attr, |
| 5180 | const char *buf, size_t count) |
| 5181 | { |
| 5182 | struct workqueue_struct *wq = dev_to_wq(dev); |
| 5183 | struct workqueue_attrs *attrs; |
Lai Jiangshan | d4d3e25 | 2015-05-19 18:03:48 +0800 | [diff] [blame] | 5184 | int ret = -ENOMEM; |
| 5185 | |
| 5186 | apply_wqattrs_lock(); |
Frederic Weisbecker | 6ba9442 | 2015-04-02 19:14:39 +0800 | [diff] [blame] | 5187 | |
| 5188 | attrs = wq_sysfs_prep_attrs(wq); |
| 5189 | if (!attrs) |
Lai Jiangshan | d4d3e25 | 2015-05-19 18:03:48 +0800 | [diff] [blame] | 5190 | goto out_unlock; |
Frederic Weisbecker | 6ba9442 | 2015-04-02 19:14:39 +0800 | [diff] [blame] | 5191 | |
| 5192 | if (sscanf(buf, "%d", &attrs->nice) == 1 && |
| 5193 | attrs->nice >= MIN_NICE && attrs->nice <= MAX_NICE) |
Lai Jiangshan | d4d3e25 | 2015-05-19 18:03:48 +0800 | [diff] [blame] | 5194 | ret = apply_workqueue_attrs_locked(wq, attrs); |
Frederic Weisbecker | 6ba9442 | 2015-04-02 19:14:39 +0800 | [diff] [blame] | 5195 | else |
| 5196 | ret = -EINVAL; |
| 5197 | |
Lai Jiangshan | d4d3e25 | 2015-05-19 18:03:48 +0800 | [diff] [blame] | 5198 | out_unlock: |
| 5199 | apply_wqattrs_unlock(); |
Frederic Weisbecker | 6ba9442 | 2015-04-02 19:14:39 +0800 | [diff] [blame] | 5200 | free_workqueue_attrs(attrs); |
| 5201 | return ret ?: count; |
| 5202 | } |
| 5203 | |
| 5204 | static ssize_t wq_cpumask_show(struct device *dev, |
| 5205 | struct device_attribute *attr, char *buf) |
| 5206 | { |
| 5207 | struct workqueue_struct *wq = dev_to_wq(dev); |
| 5208 | int written; |
| 5209 | |
| 5210 | mutex_lock(&wq->mutex); |
| 5211 | written = scnprintf(buf, PAGE_SIZE, "%*pb\n", |
| 5212 | cpumask_pr_args(wq->unbound_attrs->cpumask)); |
| 5213 | mutex_unlock(&wq->mutex); |
| 5214 | return written; |
| 5215 | } |
| 5216 | |
| 5217 | static ssize_t wq_cpumask_store(struct device *dev, |
| 5218 | struct device_attribute *attr, |
| 5219 | const char *buf, size_t count) |
| 5220 | { |
| 5221 | struct workqueue_struct *wq = dev_to_wq(dev); |
| 5222 | struct workqueue_attrs *attrs; |
Lai Jiangshan | d4d3e25 | 2015-05-19 18:03:48 +0800 | [diff] [blame] | 5223 | int ret = -ENOMEM; |
| 5224 | |
| 5225 | apply_wqattrs_lock(); |
Frederic Weisbecker | 6ba9442 | 2015-04-02 19:14:39 +0800 | [diff] [blame] | 5226 | |
| 5227 | attrs = wq_sysfs_prep_attrs(wq); |
| 5228 | if (!attrs) |
Lai Jiangshan | d4d3e25 | 2015-05-19 18:03:48 +0800 | [diff] [blame] | 5229 | goto out_unlock; |
Frederic Weisbecker | 6ba9442 | 2015-04-02 19:14:39 +0800 | [diff] [blame] | 5230 | |
| 5231 | ret = cpumask_parse(buf, attrs->cpumask); |
| 5232 | if (!ret) |
Lai Jiangshan | d4d3e25 | 2015-05-19 18:03:48 +0800 | [diff] [blame] | 5233 | ret = apply_workqueue_attrs_locked(wq, attrs); |
Frederic Weisbecker | 6ba9442 | 2015-04-02 19:14:39 +0800 | [diff] [blame] | 5234 | |
Lai Jiangshan | d4d3e25 | 2015-05-19 18:03:48 +0800 | [diff] [blame] | 5235 | out_unlock: |
| 5236 | apply_wqattrs_unlock(); |
Frederic Weisbecker | 6ba9442 | 2015-04-02 19:14:39 +0800 | [diff] [blame] | 5237 | free_workqueue_attrs(attrs); |
| 5238 | return ret ?: count; |
| 5239 | } |
| 5240 | |
| 5241 | static ssize_t wq_numa_show(struct device *dev, struct device_attribute *attr, |
| 5242 | char *buf) |
| 5243 | { |
| 5244 | struct workqueue_struct *wq = dev_to_wq(dev); |
| 5245 | int written; |
| 5246 | |
| 5247 | mutex_lock(&wq->mutex); |
| 5248 | written = scnprintf(buf, PAGE_SIZE, "%d\n", |
| 5249 | !wq->unbound_attrs->no_numa); |
| 5250 | mutex_unlock(&wq->mutex); |
| 5251 | |
| 5252 | return written; |
| 5253 | } |
| 5254 | |
| 5255 | static ssize_t wq_numa_store(struct device *dev, struct device_attribute *attr, |
| 5256 | const char *buf, size_t count) |
| 5257 | { |
| 5258 | struct workqueue_struct *wq = dev_to_wq(dev); |
| 5259 | struct workqueue_attrs *attrs; |
Lai Jiangshan | d4d3e25 | 2015-05-19 18:03:48 +0800 | [diff] [blame] | 5260 | int v, ret = -ENOMEM; |
| 5261 | |
| 5262 | apply_wqattrs_lock(); |
Frederic Weisbecker | 6ba9442 | 2015-04-02 19:14:39 +0800 | [diff] [blame] | 5263 | |
| 5264 | attrs = wq_sysfs_prep_attrs(wq); |
| 5265 | if (!attrs) |
Lai Jiangshan | d4d3e25 | 2015-05-19 18:03:48 +0800 | [diff] [blame] | 5266 | goto out_unlock; |
Frederic Weisbecker | 6ba9442 | 2015-04-02 19:14:39 +0800 | [diff] [blame] | 5267 | |
| 5268 | ret = -EINVAL; |
| 5269 | if (sscanf(buf, "%d", &v) == 1) { |
| 5270 | attrs->no_numa = !v; |
Lai Jiangshan | d4d3e25 | 2015-05-19 18:03:48 +0800 | [diff] [blame] | 5271 | ret = apply_workqueue_attrs_locked(wq, attrs); |
Frederic Weisbecker | 6ba9442 | 2015-04-02 19:14:39 +0800 | [diff] [blame] | 5272 | } |
| 5273 | |
Lai Jiangshan | d4d3e25 | 2015-05-19 18:03:48 +0800 | [diff] [blame] | 5274 | out_unlock: |
| 5275 | apply_wqattrs_unlock(); |
Frederic Weisbecker | 6ba9442 | 2015-04-02 19:14:39 +0800 | [diff] [blame] | 5276 | free_workqueue_attrs(attrs); |
| 5277 | return ret ?: count; |
| 5278 | } |
| 5279 | |
| 5280 | static struct device_attribute wq_sysfs_unbound_attrs[] = { |
| 5281 | __ATTR(pool_ids, 0444, wq_pool_ids_show, NULL), |
| 5282 | __ATTR(nice, 0644, wq_nice_show, wq_nice_store), |
| 5283 | __ATTR(cpumask, 0644, wq_cpumask_show, wq_cpumask_store), |
| 5284 | __ATTR(numa, 0644, wq_numa_show, wq_numa_store), |
| 5285 | __ATTR_NULL, |
| 5286 | }; |
| 5287 | |
| 5288 | static struct bus_type wq_subsys = { |
| 5289 | .name = "workqueue", |
| 5290 | .dev_groups = wq_sysfs_groups, |
| 5291 | }; |
| 5292 | |
Frederic Weisbecker | b05a792 | 2015-04-27 17:58:39 +0800 | [diff] [blame] | 5293 | static ssize_t wq_unbound_cpumask_show(struct device *dev, |
| 5294 | struct device_attribute *attr, char *buf) |
| 5295 | { |
| 5296 | int written; |
| 5297 | |
Lai Jiangshan | 042f7df | 2015-04-30 17:16:12 +0800 | [diff] [blame] | 5298 | mutex_lock(&wq_pool_mutex); |
Frederic Weisbecker | b05a792 | 2015-04-27 17:58:39 +0800 | [diff] [blame] | 5299 | written = scnprintf(buf, PAGE_SIZE, "%*pb\n", |
| 5300 | cpumask_pr_args(wq_unbound_cpumask)); |
Lai Jiangshan | 042f7df | 2015-04-30 17:16:12 +0800 | [diff] [blame] | 5301 | mutex_unlock(&wq_pool_mutex); |
Frederic Weisbecker | b05a792 | 2015-04-27 17:58:39 +0800 | [diff] [blame] | 5302 | |
| 5303 | return written; |
| 5304 | } |
| 5305 | |
Lai Jiangshan | 042f7df | 2015-04-30 17:16:12 +0800 | [diff] [blame] | 5306 | static ssize_t wq_unbound_cpumask_store(struct device *dev, |
| 5307 | struct device_attribute *attr, const char *buf, size_t count) |
| 5308 | { |
| 5309 | cpumask_var_t cpumask; |
| 5310 | int ret; |
| 5311 | |
| 5312 | if (!zalloc_cpumask_var(&cpumask, GFP_KERNEL)) |
| 5313 | return -ENOMEM; |
| 5314 | |
| 5315 | ret = cpumask_parse(buf, cpumask); |
| 5316 | if (!ret) |
| 5317 | ret = workqueue_set_unbound_cpumask(cpumask); |
| 5318 | |
| 5319 | free_cpumask_var(cpumask); |
| 5320 | return ret ? ret : count; |
| 5321 | } |
| 5322 | |
Frederic Weisbecker | b05a792 | 2015-04-27 17:58:39 +0800 | [diff] [blame] | 5323 | static struct device_attribute wq_sysfs_cpumask_attr = |
Lai Jiangshan | 042f7df | 2015-04-30 17:16:12 +0800 | [diff] [blame] | 5324 | __ATTR(cpumask, 0644, wq_unbound_cpumask_show, |
| 5325 | wq_unbound_cpumask_store); |
Frederic Weisbecker | b05a792 | 2015-04-27 17:58:39 +0800 | [diff] [blame] | 5326 | |
Frederic Weisbecker | 6ba9442 | 2015-04-02 19:14:39 +0800 | [diff] [blame] | 5327 | static int __init wq_sysfs_init(void) |
| 5328 | { |
Frederic Weisbecker | b05a792 | 2015-04-27 17:58:39 +0800 | [diff] [blame] | 5329 | int err; |
| 5330 | |
| 5331 | err = subsys_virtual_register(&wq_subsys, NULL); |
| 5332 | if (err) |
| 5333 | return err; |
| 5334 | |
| 5335 | return device_create_file(wq_subsys.dev_root, &wq_sysfs_cpumask_attr); |
Frederic Weisbecker | 6ba9442 | 2015-04-02 19:14:39 +0800 | [diff] [blame] | 5336 | } |
| 5337 | core_initcall(wq_sysfs_init); |
| 5338 | |
| 5339 | static void wq_device_release(struct device *dev) |
| 5340 | { |
| 5341 | struct wq_device *wq_dev = container_of(dev, struct wq_device, dev); |
| 5342 | |
| 5343 | kfree(wq_dev); |
| 5344 | } |
| 5345 | |
| 5346 | /** |
| 5347 | * workqueue_sysfs_register - make a workqueue visible in sysfs |
| 5348 | * @wq: the workqueue to register |
| 5349 | * |
| 5350 | * Expose @wq in sysfs under /sys/bus/workqueue/devices. |
| 5351 | * alloc_workqueue*() automatically calls this function if WQ_SYSFS is set |
| 5352 | * which is the preferred method. |
| 5353 | * |
| 5354 | * Workqueue user should use this function directly iff it wants to apply |
| 5355 | * workqueue_attrs before making the workqueue visible in sysfs; otherwise, |
| 5356 | * apply_workqueue_attrs() may race against userland updating the |
| 5357 | * attributes. |
| 5358 | * |
| 5359 | * Return: 0 on success, -errno on failure. |
| 5360 | */ |
| 5361 | int workqueue_sysfs_register(struct workqueue_struct *wq) |
| 5362 | { |
| 5363 | struct wq_device *wq_dev; |
| 5364 | int ret; |
| 5365 | |
| 5366 | /* |
Shailendra Verma | 402dd89 | 2015-05-23 10:38:14 +0530 | [diff] [blame] | 5367 | * Adjusting max_active or creating new pwqs by applying |
Frederic Weisbecker | 6ba9442 | 2015-04-02 19:14:39 +0800 | [diff] [blame] | 5368 | * attributes breaks ordering guarantee. Disallow exposing ordered |
| 5369 | * workqueues. |
| 5370 | */ |
Tejun Heo | 0a94efb | 2017-07-23 08:36:15 -0400 | [diff] [blame] | 5371 | if (WARN_ON(wq->flags & __WQ_ORDERED_EXPLICIT)) |
Frederic Weisbecker | 6ba9442 | 2015-04-02 19:14:39 +0800 | [diff] [blame] | 5372 | return -EINVAL; |
| 5373 | |
| 5374 | wq->wq_dev = wq_dev = kzalloc(sizeof(*wq_dev), GFP_KERNEL); |
| 5375 | if (!wq_dev) |
| 5376 | return -ENOMEM; |
| 5377 | |
| 5378 | wq_dev->wq = wq; |
| 5379 | wq_dev->dev.bus = &wq_subsys; |
Frederic Weisbecker | 6ba9442 | 2015-04-02 19:14:39 +0800 | [diff] [blame] | 5380 | wq_dev->dev.release = wq_device_release; |
Lars-Peter Clausen | 23217b4 | 2016-02-17 21:04:41 +0100 | [diff] [blame] | 5381 | dev_set_name(&wq_dev->dev, "%s", wq->name); |
Frederic Weisbecker | 6ba9442 | 2015-04-02 19:14:39 +0800 | [diff] [blame] | 5382 | |
| 5383 | /* |
| 5384 | * unbound_attrs are created separately. Suppress uevent until |
| 5385 | * everything is ready. |
| 5386 | */ |
| 5387 | dev_set_uevent_suppress(&wq_dev->dev, true); |
| 5388 | |
| 5389 | ret = device_register(&wq_dev->dev); |
| 5390 | if (ret) { |
Arvind Yadav | 7c84e5e | 2018-03-06 15:35:43 +0530 | [diff] [blame] | 5391 | put_device(&wq_dev->dev); |
Frederic Weisbecker | 6ba9442 | 2015-04-02 19:14:39 +0800 | [diff] [blame] | 5392 | wq->wq_dev = NULL; |
| 5393 | return ret; |
| 5394 | } |
| 5395 | |
| 5396 | if (wq->flags & WQ_UNBOUND) { |
| 5397 | struct device_attribute *attr; |
| 5398 | |
| 5399 | for (attr = wq_sysfs_unbound_attrs; attr->attr.name; attr++) { |
| 5400 | ret = device_create_file(&wq_dev->dev, attr); |
| 5401 | if (ret) { |
| 5402 | device_unregister(&wq_dev->dev); |
| 5403 | wq->wq_dev = NULL; |
| 5404 | return ret; |
| 5405 | } |
| 5406 | } |
| 5407 | } |
| 5408 | |
| 5409 | dev_set_uevent_suppress(&wq_dev->dev, false); |
| 5410 | kobject_uevent(&wq_dev->dev.kobj, KOBJ_ADD); |
| 5411 | return 0; |
| 5412 | } |
| 5413 | |
| 5414 | /** |
| 5415 | * workqueue_sysfs_unregister - undo workqueue_sysfs_register() |
| 5416 | * @wq: the workqueue to unregister |
| 5417 | * |
| 5418 | * If @wq is registered to sysfs by workqueue_sysfs_register(), unregister. |
| 5419 | */ |
| 5420 | static void workqueue_sysfs_unregister(struct workqueue_struct *wq) |
| 5421 | { |
| 5422 | struct wq_device *wq_dev = wq->wq_dev; |
| 5423 | |
| 5424 | if (!wq->wq_dev) |
| 5425 | return; |
| 5426 | |
| 5427 | wq->wq_dev = NULL; |
| 5428 | device_unregister(&wq_dev->dev); |
| 5429 | } |
| 5430 | #else /* CONFIG_SYSFS */ |
| 5431 | static void workqueue_sysfs_unregister(struct workqueue_struct *wq) { } |
| 5432 | #endif /* CONFIG_SYSFS */ |
| 5433 | |
Tejun Heo | 82607adc | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 5434 | /* |
| 5435 | * Workqueue watchdog. |
| 5436 | * |
| 5437 | * Stall may be caused by various bugs - missing WQ_MEM_RECLAIM, illegal |
| 5438 | * flush dependency, a concurrency managed work item which stays RUNNING |
| 5439 | * indefinitely. Workqueue stalls can be very difficult to debug as the |
| 5440 | * usual warning mechanisms don't trigger and internal workqueue state is |
| 5441 | * largely opaque. |
| 5442 | * |
| 5443 | * Workqueue watchdog monitors all worker pools periodically and dumps |
| 5444 | * state if some pools failed to make forward progress for a while where |
| 5445 | * forward progress is defined as the first item on ->worklist changing. |
| 5446 | * |
| 5447 | * This mechanism is controlled through the kernel parameter |
| 5448 | * "workqueue.watchdog_thresh" which can be updated at runtime through the |
| 5449 | * corresponding sysfs parameter file. |
| 5450 | */ |
| 5451 | #ifdef CONFIG_WQ_WATCHDOG |
| 5452 | |
| 5453 | static void wq_watchdog_timer_fn(unsigned long data); |
| 5454 | |
| 5455 | static unsigned long wq_watchdog_thresh = 30; |
| 5456 | static struct timer_list wq_watchdog_timer = |
| 5457 | TIMER_DEFERRED_INITIALIZER(wq_watchdog_timer_fn, 0, 0); |
| 5458 | |
| 5459 | static unsigned long wq_watchdog_touched = INITIAL_JIFFIES; |
| 5460 | static DEFINE_PER_CPU(unsigned long, wq_watchdog_touched_cpu) = INITIAL_JIFFIES; |
| 5461 | |
| 5462 | static void wq_watchdog_reset_touched(void) |
| 5463 | { |
| 5464 | int cpu; |
| 5465 | |
| 5466 | wq_watchdog_touched = jiffies; |
| 5467 | for_each_possible_cpu(cpu) |
| 5468 | per_cpu(wq_watchdog_touched_cpu, cpu) = jiffies; |
| 5469 | } |
| 5470 | |
| 5471 | static void wq_watchdog_timer_fn(unsigned long data) |
| 5472 | { |
| 5473 | unsigned long thresh = READ_ONCE(wq_watchdog_thresh) * HZ; |
| 5474 | bool lockup_detected = false; |
Sergey Senozhatsky | ec32c8a | 2021-05-20 19:14:22 +0900 | [diff] [blame] | 5475 | unsigned long now = jiffies; |
Tejun Heo | 82607adc | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 5476 | struct worker_pool *pool; |
| 5477 | int pi; |
| 5478 | |
| 5479 | if (!thresh) |
| 5480 | return; |
| 5481 | |
| 5482 | rcu_read_lock(); |
| 5483 | |
| 5484 | for_each_pool(pool, pi) { |
| 5485 | unsigned long pool_ts, touched, ts; |
| 5486 | |
| 5487 | if (list_empty(&pool->worklist)) |
| 5488 | continue; |
| 5489 | |
Sergey Senozhatsky | ec32c8a | 2021-05-20 19:14:22 +0900 | [diff] [blame] | 5490 | /* |
| 5491 | * If a virtual machine is stopped by the host it can look to |
| 5492 | * the watchdog like a stall. |
| 5493 | */ |
| 5494 | kvm_check_and_clear_guest_paused(); |
| 5495 | |
Tejun Heo | 82607adc | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 5496 | /* get the latest of pool and touched timestamps */ |
| 5497 | pool_ts = READ_ONCE(pool->watchdog_ts); |
| 5498 | touched = READ_ONCE(wq_watchdog_touched); |
| 5499 | |
| 5500 | if (time_after(pool_ts, touched)) |
| 5501 | ts = pool_ts; |
| 5502 | else |
| 5503 | ts = touched; |
| 5504 | |
| 5505 | if (pool->cpu >= 0) { |
| 5506 | unsigned long cpu_touched = |
| 5507 | READ_ONCE(per_cpu(wq_watchdog_touched_cpu, |
| 5508 | pool->cpu)); |
| 5509 | if (time_after(cpu_touched, ts)) |
| 5510 | ts = cpu_touched; |
| 5511 | } |
| 5512 | |
| 5513 | /* did we stall? */ |
Sergey Senozhatsky | ec32c8a | 2021-05-20 19:14:22 +0900 | [diff] [blame] | 5514 | if (time_after(now, ts + thresh)) { |
Tejun Heo | 82607adc | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 5515 | lockup_detected = true; |
| 5516 | pr_emerg("BUG: workqueue lockup - pool"); |
| 5517 | pr_cont_pool_info(pool); |
| 5518 | pr_cont(" stuck for %us!\n", |
Sergey Senozhatsky | ec32c8a | 2021-05-20 19:14:22 +0900 | [diff] [blame] | 5519 | jiffies_to_msecs(now - pool_ts) / 1000); |
Tejun Heo | 82607adc | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 5520 | } |
| 5521 | } |
| 5522 | |
| 5523 | rcu_read_unlock(); |
| 5524 | |
| 5525 | if (lockup_detected) |
| 5526 | show_workqueue_state(); |
| 5527 | |
| 5528 | wq_watchdog_reset_touched(); |
| 5529 | mod_timer(&wq_watchdog_timer, jiffies + thresh); |
| 5530 | } |
| 5531 | |
Vincent Whitchurch | 63a0f9d | 2018-08-21 17:25:07 +0200 | [diff] [blame] | 5532 | notrace void wq_watchdog_touch(int cpu) |
Tejun Heo | 82607adc | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 5533 | { |
| 5534 | if (cpu >= 0) |
| 5535 | per_cpu(wq_watchdog_touched_cpu, cpu) = jiffies; |
| 5536 | else |
| 5537 | wq_watchdog_touched = jiffies; |
| 5538 | } |
| 5539 | |
| 5540 | static void wq_watchdog_set_thresh(unsigned long thresh) |
| 5541 | { |
| 5542 | wq_watchdog_thresh = 0; |
| 5543 | del_timer_sync(&wq_watchdog_timer); |
| 5544 | |
| 5545 | if (thresh) { |
| 5546 | wq_watchdog_thresh = thresh; |
| 5547 | wq_watchdog_reset_touched(); |
| 5548 | mod_timer(&wq_watchdog_timer, jiffies + thresh * HZ); |
| 5549 | } |
| 5550 | } |
| 5551 | |
| 5552 | static int wq_watchdog_param_set_thresh(const char *val, |
| 5553 | const struct kernel_param *kp) |
| 5554 | { |
| 5555 | unsigned long thresh; |
| 5556 | int ret; |
| 5557 | |
| 5558 | ret = kstrtoul(val, 0, &thresh); |
| 5559 | if (ret) |
| 5560 | return ret; |
| 5561 | |
| 5562 | if (system_wq) |
| 5563 | wq_watchdog_set_thresh(thresh); |
| 5564 | else |
| 5565 | wq_watchdog_thresh = thresh; |
| 5566 | |
| 5567 | return 0; |
| 5568 | } |
| 5569 | |
| 5570 | static const struct kernel_param_ops wq_watchdog_thresh_ops = { |
| 5571 | .set = wq_watchdog_param_set_thresh, |
| 5572 | .get = param_get_ulong, |
| 5573 | }; |
| 5574 | |
| 5575 | module_param_cb(watchdog_thresh, &wq_watchdog_thresh_ops, &wq_watchdog_thresh, |
| 5576 | 0644); |
| 5577 | |
| 5578 | static void wq_watchdog_init(void) |
| 5579 | { |
| 5580 | wq_watchdog_set_thresh(wq_watchdog_thresh); |
| 5581 | } |
| 5582 | |
| 5583 | #else /* CONFIG_WQ_WATCHDOG */ |
| 5584 | |
| 5585 | static inline void wq_watchdog_init(void) { } |
| 5586 | |
| 5587 | #endif /* CONFIG_WQ_WATCHDOG */ |
| 5588 | |
Tejun Heo | bce9038 | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 5589 | static void __init wq_numa_init(void) |
| 5590 | { |
| 5591 | cpumask_var_t *tbl; |
| 5592 | int node, cpu; |
| 5593 | |
Tejun Heo | bce9038 | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 5594 | if (num_possible_nodes() <= 1) |
| 5595 | return; |
| 5596 | |
Tejun Heo | d55262c | 2013-04-01 11:23:38 -0700 | [diff] [blame] | 5597 | if (wq_disable_numa) { |
| 5598 | pr_info("workqueue: NUMA affinity support disabled\n"); |
| 5599 | return; |
| 5600 | } |
| 5601 | |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 5602 | wq_update_unbound_numa_attrs_buf = alloc_workqueue_attrs(GFP_KERNEL); |
| 5603 | BUG_ON(!wq_update_unbound_numa_attrs_buf); |
| 5604 | |
Tejun Heo | bce9038 | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 5605 | /* |
| 5606 | * We want masks of possible CPUs of each node which isn't readily |
| 5607 | * available. Build one from cpu_to_node() which should have been |
| 5608 | * fully initialized by now. |
| 5609 | */ |
Lai Jiangshan | ddcb57e | 2014-07-22 13:05:40 +0800 | [diff] [blame] | 5610 | tbl = kzalloc(nr_node_ids * sizeof(tbl[0]), GFP_KERNEL); |
Tejun Heo | bce9038 | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 5611 | BUG_ON(!tbl); |
| 5612 | |
| 5613 | for_each_node(node) |
Yasuaki Ishimatsu | 5a6024f | 2014-07-07 09:56:48 -0400 | [diff] [blame] | 5614 | BUG_ON(!zalloc_cpumask_var_node(&tbl[node], GFP_KERNEL, |
Tejun Heo | 1be0c25 | 2013-05-15 14:24:24 -0700 | [diff] [blame] | 5615 | node_online(node) ? node : NUMA_NO_NODE)); |
Tejun Heo | bce9038 | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 5616 | |
| 5617 | for_each_possible_cpu(cpu) { |
| 5618 | node = cpu_to_node(cpu); |
| 5619 | if (WARN_ON(node == NUMA_NO_NODE)) { |
| 5620 | pr_warn("workqueue: NUMA node mapping not available for cpu%d, disabling NUMA support\n", cpu); |
| 5621 | /* happens iff arch is bonkers, let's just proceed */ |
| 5622 | return; |
| 5623 | } |
| 5624 | cpumask_set_cpu(cpu, tbl[node]); |
| 5625 | } |
| 5626 | |
| 5627 | wq_numa_possible_cpumask = tbl; |
| 5628 | wq_numa_enabled = true; |
| 5629 | } |
| 5630 | |
Tejun Heo | 3347fa0 | 2016-09-16 15:49:32 -0400 | [diff] [blame] | 5631 | /** |
| 5632 | * workqueue_init_early - early init for workqueue subsystem |
| 5633 | * |
| 5634 | * This is the first half of two-staged workqueue subsystem initialization |
| 5635 | * and invoked as soon as the bare basics - memory allocation, cpumasks and |
| 5636 | * idr are up. It sets up all the data structures and system workqueues |
| 5637 | * and allows early boot code to create workqueues and queue/cancel work |
| 5638 | * items. Actual work item execution starts only after kthreads can be |
| 5639 | * created and scheduled right before early initcalls. |
| 5640 | */ |
| 5641 | int __init workqueue_init_early(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5642 | { |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 5643 | int std_nice[NR_STD_WORKER_POOLS] = { 0, HIGHPRI_NICE_LEVEL }; |
| 5644 | int i, cpu; |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 5645 | |
Tejun Heo | e904e6c | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 5646 | WARN_ON(__alignof__(struct pool_workqueue) < __alignof__(long long)); |
| 5647 | |
Frederic Weisbecker | b05a792 | 2015-04-27 17:58:39 +0800 | [diff] [blame] | 5648 | BUG_ON(!alloc_cpumask_var(&wq_unbound_cpumask, GFP_KERNEL)); |
| 5649 | cpumask_copy(wq_unbound_cpumask, cpu_possible_mask); |
| 5650 | |
Tejun Heo | e904e6c | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 5651 | pwq_cache = KMEM_CACHE(pool_workqueue, SLAB_PANIC); |
| 5652 | |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 5653 | /* initialize CPU pools */ |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 5654 | for_each_possible_cpu(cpu) { |
Tejun Heo | 4ce62e9 | 2012-07-13 22:16:44 -0700 | [diff] [blame] | 5655 | struct worker_pool *pool; |
Tejun Heo | 8b03ae3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 5656 | |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 5657 | i = 0; |
Tejun Heo | f02ae73 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 5658 | for_each_cpu_worker_pool(pool, cpu) { |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 5659 | BUG_ON(init_worker_pool(pool)); |
Tejun Heo | ec22ca5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 5660 | pool->cpu = cpu; |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 5661 | cpumask_copy(pool->attrs->cpumask, cpumask_of(cpu)); |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 5662 | pool->attrs->nice = std_nice[i++]; |
Tejun Heo | f3f90ad | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 5663 | pool->node = cpu_to_node(cpu); |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 5664 | |
Tejun Heo | 9daf9e6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 5665 | /* alloc pool ID */ |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 5666 | mutex_lock(&wq_pool_mutex); |
Tejun Heo | 9daf9e6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 5667 | BUG_ON(worker_pool_assign_id(pool)); |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 5668 | mutex_unlock(&wq_pool_mutex); |
Tejun Heo | 4ce62e9 | 2012-07-13 22:16:44 -0700 | [diff] [blame] | 5669 | } |
Tejun Heo | 8b03ae3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 5670 | } |
| 5671 | |
Tejun Heo | 8a2b753 | 2013-09-05 12:30:04 -0400 | [diff] [blame] | 5672 | /* create default unbound and ordered wq attrs */ |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 5673 | for (i = 0; i < NR_STD_WORKER_POOLS; i++) { |
| 5674 | struct workqueue_attrs *attrs; |
| 5675 | |
| 5676 | BUG_ON(!(attrs = alloc_workqueue_attrs(GFP_KERNEL))); |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 5677 | attrs->nice = std_nice[i]; |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 5678 | unbound_std_wq_attrs[i] = attrs; |
Tejun Heo | 8a2b753 | 2013-09-05 12:30:04 -0400 | [diff] [blame] | 5679 | |
| 5680 | /* |
| 5681 | * An ordered wq should have only one pwq as ordering is |
| 5682 | * guaranteed by max_active which is enforced by pwqs. |
| 5683 | * Turn off NUMA so that dfl_pwq is used for all nodes. |
| 5684 | */ |
| 5685 | BUG_ON(!(attrs = alloc_workqueue_attrs(GFP_KERNEL))); |
| 5686 | attrs->nice = std_nice[i]; |
| 5687 | attrs->no_numa = true; |
| 5688 | ordered_wq_attrs[i] = attrs; |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 5689 | } |
| 5690 | |
Tejun Heo | d320c03 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 5691 | system_wq = alloc_workqueue("events", 0, 0); |
Joonsoo Kim | 1aabe90 | 2012-08-15 23:25:39 +0900 | [diff] [blame] | 5692 | system_highpri_wq = alloc_workqueue("events_highpri", WQ_HIGHPRI, 0); |
Tejun Heo | d320c03 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 5693 | system_long_wq = alloc_workqueue("events_long", 0, 0); |
Tejun Heo | f342179 | 2010-07-02 10:03:51 +0200 | [diff] [blame] | 5694 | system_unbound_wq = alloc_workqueue("events_unbound", WQ_UNBOUND, |
| 5695 | WQ_UNBOUND_MAX_ACTIVE); |
Tejun Heo | 24d51ad | 2011-02-21 09:52:50 +0100 | [diff] [blame] | 5696 | system_freezable_wq = alloc_workqueue("events_freezable", |
| 5697 | WQ_FREEZABLE, 0); |
Viresh Kumar | 0668106 | 2013-04-24 17:12:54 +0530 | [diff] [blame] | 5698 | system_power_efficient_wq = alloc_workqueue("events_power_efficient", |
| 5699 | WQ_POWER_EFFICIENT, 0); |
| 5700 | system_freezable_power_efficient_wq = alloc_workqueue("events_freezable_power_efficient", |
| 5701 | WQ_FREEZABLE | WQ_POWER_EFFICIENT, |
| 5702 | 0); |
Joonsoo Kim | 1aabe90 | 2012-08-15 23:25:39 +0900 | [diff] [blame] | 5703 | BUG_ON(!system_wq || !system_highpri_wq || !system_long_wq || |
Viresh Kumar | 0668106 | 2013-04-24 17:12:54 +0530 | [diff] [blame] | 5704 | !system_unbound_wq || !system_freezable_wq || |
| 5705 | !system_power_efficient_wq || |
| 5706 | !system_freezable_power_efficient_wq); |
Tejun Heo | 82607adc | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 5707 | |
Tejun Heo | 3347fa0 | 2016-09-16 15:49:32 -0400 | [diff] [blame] | 5708 | return 0; |
| 5709 | } |
| 5710 | |
| 5711 | /** |
| 5712 | * workqueue_init - bring workqueue subsystem fully online |
| 5713 | * |
| 5714 | * This is the latter half of two-staged workqueue subsystem initialization |
| 5715 | * and invoked as soon as kthreads can be created and scheduled. |
| 5716 | * Workqueues have been created and work items queued on them, but there |
| 5717 | * are no kworkers executing the work items yet. Populate the worker pools |
| 5718 | * with the initial workers and enable future kworker creations. |
| 5719 | */ |
| 5720 | int __init workqueue_init(void) |
| 5721 | { |
Tejun Heo | 2186d9f | 2016-10-19 12:01:27 -0400 | [diff] [blame] | 5722 | struct workqueue_struct *wq; |
Tejun Heo | 3347fa0 | 2016-09-16 15:49:32 -0400 | [diff] [blame] | 5723 | struct worker_pool *pool; |
| 5724 | int cpu, bkt; |
| 5725 | |
Tejun Heo | 2186d9f | 2016-10-19 12:01:27 -0400 | [diff] [blame] | 5726 | /* |
| 5727 | * It'd be simpler to initialize NUMA in workqueue_init_early() but |
| 5728 | * CPU to node mapping may not be available that early on some |
| 5729 | * archs such as power and arm64. As per-cpu pools created |
| 5730 | * previously could be missing node hint and unbound pools NUMA |
| 5731 | * affinity, fix them up. |
| 5732 | */ |
| 5733 | wq_numa_init(); |
| 5734 | |
| 5735 | mutex_lock(&wq_pool_mutex); |
| 5736 | |
| 5737 | for_each_possible_cpu(cpu) { |
| 5738 | for_each_cpu_worker_pool(pool, cpu) { |
| 5739 | pool->node = cpu_to_node(cpu); |
| 5740 | } |
| 5741 | } |
| 5742 | |
| 5743 | list_for_each_entry(wq, &workqueues, list) |
| 5744 | wq_update_unbound_numa(wq, smp_processor_id(), true); |
| 5745 | |
| 5746 | mutex_unlock(&wq_pool_mutex); |
| 5747 | |
Tejun Heo | 3347fa0 | 2016-09-16 15:49:32 -0400 | [diff] [blame] | 5748 | /* create the initial workers */ |
| 5749 | for_each_online_cpu(cpu) { |
| 5750 | for_each_cpu_worker_pool(pool, cpu) { |
| 5751 | pool->flags &= ~POOL_DISASSOCIATED; |
| 5752 | BUG_ON(!create_worker(pool)); |
| 5753 | } |
| 5754 | } |
| 5755 | |
| 5756 | hash_for_each(unbound_pool_hash, bkt, pool, hash_node) |
| 5757 | BUG_ON(!create_worker(pool)); |
| 5758 | |
| 5759 | wq_online = true; |
Tejun Heo | 82607adc | 2015-12-08 11:28:04 -0500 | [diff] [blame] | 5760 | wq_watchdog_init(); |
| 5761 | |
Suresh Siddha | 6ee0578 | 2010-07-30 14:57:37 -0700 | [diff] [blame] | 5762 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5763 | } |