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 | * |
| 24 | * Please read Documentation/workqueue.txt 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> |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 51 | |
Tejun Heo | ea13844 | 2013-01-18 14:05:55 -0800 | [diff] [blame] | 52 | #include "workqueue_internal.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 53 | |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 54 | enum { |
Tejun Heo | bc2ae0f | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 55 | /* |
Tejun Heo | 2464757 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 56 | * worker_pool flags |
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 | * A bound pool is either associated or disassociated with its CPU. |
Tejun Heo | bc2ae0f | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 59 | * While associated (!DISASSOCIATED), all workers are bound to the |
| 60 | * CPU and none has %WORKER_UNBOUND set and concurrency management |
| 61 | * is in effect. |
| 62 | * |
| 63 | * While DISASSOCIATED, the cpu may be offline and all workers have |
| 64 | * %WORKER_UNBOUND set and concurrency management disabled, and may |
Tejun Heo | 2464757 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 65 | * be executing on any CPU. The pool behaves as an unbound one. |
Tejun Heo | bc2ae0f | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 66 | * |
Tejun Heo | bc3a1af | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 67 | * Note that DISASSOCIATED should be flipped only while holding |
| 68 | * manager_mutex to avoid changing binding state while |
Tejun Heo | 2464757 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 69 | * create_worker() is in progress. |
Tejun Heo | bc2ae0f | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 70 | */ |
Tejun Heo | 11ebea5 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 71 | POOL_MANAGE_WORKERS = 1 << 0, /* need to manage workers */ |
Tejun Heo | 2464757 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 72 | POOL_DISASSOCIATED = 1 << 2, /* cpu can't serve workers */ |
Tejun Heo | 35b6bb6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 73 | POOL_FREEZING = 1 << 3, /* freeze in progress */ |
Tejun Heo | db7bccf | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 74 | |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 75 | /* worker flags */ |
| 76 | WORKER_STARTED = 1 << 0, /* started */ |
| 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 | * |
Tejun Heo | 822d840 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 127 | * MG: pool->manager_mutex and pool->lock protected. Writes require both |
| 128 | * locks. Reads can happen under either lock. |
| 129 | * |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 130 | * PL: wq_pool_mutex protected. |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 131 | * |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 132 | * PR: wq_pool_mutex protected for writes. Sched-RCU protected for reads. |
Tejun Heo | 5bcab33 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 133 | * |
Lai Jiangshan | 3c25a55 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 134 | * WQ: wq->mutex protected. |
| 135 | * |
Lai Jiangshan | b592760 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 136 | * WR: wq->mutex protected for writes. Sched-RCU protected for reads. |
Tejun Heo | 2e109a2 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 137 | * |
| 138 | * MD: wq_mayday_lock protected. |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 139 | */ |
| 140 | |
Tejun Heo | 2eaebdb | 2013-01-18 14:05:55 -0800 | [diff] [blame] | 141 | /* struct worker is defined in workqueue_internal.h */ |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 142 | |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 143 | struct worker_pool { |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 144 | spinlock_t lock; /* the pool lock */ |
Tejun Heo | d84ff05 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 145 | int cpu; /* I: the associated cpu */ |
Tejun Heo | f3f90ad | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 146 | int node; /* I: the associated node ID */ |
Tejun Heo | 9daf9e6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 147 | int id; /* I: pool ID */ |
Tejun Heo | 11ebea5 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 148 | unsigned int flags; /* X: flags */ |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 149 | |
| 150 | struct list_head worklist; /* L: list of pending works */ |
| 151 | int nr_workers; /* L: total number of workers */ |
Lai Jiangshan | ea1abd6 | 2012-09-18 09:59:22 -0700 | [diff] [blame] | 152 | |
| 153 | /* nr_idle includes the ones off idle_list for rebinding */ |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 154 | int nr_idle; /* L: currently idle ones */ |
| 155 | |
| 156 | struct list_head idle_list; /* X: list of idle workers */ |
| 157 | struct timer_list idle_timer; /* L: worker idle timeout */ |
| 158 | struct timer_list mayday_timer; /* L: SOS timer for workers */ |
| 159 | |
Tejun Heo | c5aa87b | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 160 | /* 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] | 161 | DECLARE_HASHTABLE(busy_hash, BUSY_WORKER_HASH_ORDER); |
| 162 | /* L: hash of busy workers */ |
| 163 | |
Tejun Heo | bc3a1af | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 164 | /* see manage_workers() for details on the two manager mutexes */ |
Tejun Heo | 34a06bd | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 165 | struct mutex manager_arb; /* manager arbitration */ |
Tejun Heo | bc3a1af | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 166 | struct mutex manager_mutex; /* manager exclusion */ |
Tejun Heo | 822d840 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 167 | struct idr worker_idr; /* MG: worker IDs and iteration */ |
Tejun Heo | e19e397 | 2013-01-24 11:39:44 -0800 | [diff] [blame] | 168 | |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 169 | struct workqueue_attrs *attrs; /* I: worker attributes */ |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 170 | struct hlist_node hash_node; /* PL: unbound_pool_hash node */ |
| 171 | int refcnt; /* PL: refcnt for unbound pools */ |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 172 | |
Tejun Heo | e19e397 | 2013-01-24 11:39:44 -0800 | [diff] [blame] | 173 | /* |
| 174 | * The current concurrency level. As it's likely to be accessed |
| 175 | * from other CPUs during try_to_wake_up(), put it in a separate |
| 176 | * cacheline. |
| 177 | */ |
| 178 | atomic_t nr_running ____cacheline_aligned_in_smp; |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 179 | |
| 180 | /* |
| 181 | * Destruction of pool is sched-RCU protected to allow dereferences |
| 182 | * from get_work_pool(). |
| 183 | */ |
| 184 | struct rcu_head rcu; |
Tejun Heo | 8b03ae3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 185 | } ____cacheline_aligned_in_smp; |
| 186 | |
| 187 | /* |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 188 | * The per-pool workqueue. While queued, the lower WORK_STRUCT_FLAG_BITS |
| 189 | * of work_struct->data are used for flags and the remaining high bits |
| 190 | * point to the pwq; thus, pwqs need to be aligned at two's power of the |
| 191 | * number of flag bits. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 192 | */ |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 193 | struct pool_workqueue { |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 194 | struct worker_pool *pool; /* I: the associated pool */ |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 195 | struct workqueue_struct *wq; /* I: the owning workqueue */ |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 196 | int work_color; /* L: current color */ |
| 197 | int flush_color; /* L: flushing color */ |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 198 | int refcnt; /* L: reference count */ |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 199 | int nr_in_flight[WORK_NR_COLORS]; |
| 200 | /* L: nr of in_flight works */ |
Tejun Heo | 1e19ffc | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 201 | int nr_active; /* L: nr of active works */ |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 202 | int max_active; /* L: max active works */ |
Tejun Heo | 1e19ffc | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 203 | struct list_head delayed_works; /* L: delayed works */ |
Lai Jiangshan | 3c25a55 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 204 | struct list_head pwqs_node; /* WR: node on wq->pwqs */ |
Tejun Heo | 2e109a2 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 205 | struct list_head mayday_node; /* MD: node on wq->maydays */ |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 206 | |
| 207 | /* |
| 208 | * Release of unbound pwq is punted to system_wq. See put_pwq() |
| 209 | * and pwq_unbound_release_workfn() for details. pool_workqueue |
| 210 | * itself is also sched-RCU protected so that the first pwq can be |
Lai Jiangshan | b09f4fd | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 211 | * determined without grabbing wq->mutex. |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 212 | */ |
| 213 | struct work_struct unbound_release_work; |
| 214 | struct rcu_head rcu; |
Tejun Heo | e904e6c | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 215 | } __aligned(1 << WORK_STRUCT_FLAG_BITS); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 216 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 217 | /* |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 218 | * Structure used to wait for workqueue flush. |
| 219 | */ |
| 220 | struct wq_flusher { |
Lai Jiangshan | 3c25a55 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 221 | struct list_head list; /* WQ: list of flushers */ |
| 222 | int flush_color; /* WQ: flush color waiting for */ |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 223 | struct completion done; /* flush completion */ |
| 224 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 225 | |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 226 | struct wq_device; |
| 227 | |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 228 | /* |
Tejun Heo | c5aa87b | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 229 | * The externally visible workqueue. It relays the issued work items to |
| 230 | * the appropriate worker_pool through its pool_workqueues. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 231 | */ |
| 232 | struct workqueue_struct { |
Lai Jiangshan | 3c25a55 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 233 | struct list_head pwqs; /* WR: all pwqs of this wq */ |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 234 | struct list_head list; /* PL: list of all workqueues */ |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 235 | |
Lai Jiangshan | 3c25a55 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 236 | struct mutex mutex; /* protects this wq */ |
| 237 | int work_color; /* WQ: current work color */ |
| 238 | int flush_color; /* WQ: current flush color */ |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 239 | atomic_t nr_pwqs_to_flush; /* flush in progress */ |
Lai Jiangshan | 3c25a55 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 240 | struct wq_flusher *first_flusher; /* WQ: first flusher */ |
| 241 | struct list_head flusher_queue; /* WQ: flush waiters */ |
| 242 | struct list_head flusher_overflow; /* WQ: flush overflow list */ |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 243 | |
Tejun Heo | 2e109a2 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 244 | struct list_head maydays; /* MD: pwqs requesting rescue */ |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 245 | struct worker *rescuer; /* I: rescue worker */ |
| 246 | |
Lai Jiangshan | 87fc741 | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 247 | int nr_drainers; /* WQ: drain in progress */ |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 248 | int saved_max_active; /* WQ: saved pwq max_active */ |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 249 | |
Tejun Heo | 6029a91 | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 250 | struct workqueue_attrs *unbound_attrs; /* WQ: only for unbound wqs */ |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 251 | struct pool_workqueue *dfl_pwq; /* WQ: only for unbound wqs */ |
Tejun Heo | 6029a91 | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 252 | |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 253 | #ifdef CONFIG_SYSFS |
| 254 | struct wq_device *wq_dev; /* I: for sysfs interface */ |
| 255 | #endif |
Johannes Berg | 4e6045f | 2007-10-18 23:39:55 -0700 | [diff] [blame] | 256 | #ifdef CONFIG_LOCKDEP |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 257 | struct lockdep_map lockdep_map; |
Johannes Berg | 4e6045f | 2007-10-18 23:39:55 -0700 | [diff] [blame] | 258 | #endif |
Tejun Heo | ecf6881 | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 259 | char name[WQ_NAME_LEN]; /* I: workqueue name */ |
Tejun Heo | 2728fd2 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 260 | |
| 261 | /* hot fields used during command issue, aligned to cacheline */ |
| 262 | unsigned int flags ____cacheline_aligned; /* WQ: WQ_* flags */ |
| 263 | struct pool_workqueue __percpu *cpu_pwqs; /* I: per-cpu pwqs */ |
Tejun Heo | df2d5ae | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 264 | struct pool_workqueue __rcu *numa_pwq_tbl[]; /* FR: unbound pwqs indexed by node */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 265 | }; |
| 266 | |
Tejun Heo | e904e6c | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 267 | static struct kmem_cache *pwq_cache; |
| 268 | |
Tejun Heo | bce9038 | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 269 | static int wq_numa_tbl_len; /* highest possible NUMA node id + 1 */ |
| 270 | static cpumask_var_t *wq_numa_possible_cpumask; |
| 271 | /* possible CPUs of each node */ |
| 272 | |
Tejun Heo | d55262c | 2013-04-01 11:23:38 -0700 | [diff] [blame] | 273 | static bool wq_disable_numa; |
| 274 | module_param_named(disable_numa, wq_disable_numa, bool, 0444); |
| 275 | |
Viresh Kumar | cee22a1 | 2013-04-08 16:45:40 +0530 | [diff] [blame] | 276 | /* see the comment above the definition of WQ_POWER_EFFICIENT */ |
| 277 | #ifdef CONFIG_WQ_POWER_EFFICIENT_DEFAULT |
| 278 | static bool wq_power_efficient = true; |
| 279 | #else |
| 280 | static bool wq_power_efficient; |
| 281 | #endif |
| 282 | |
| 283 | module_param_named(power_efficient, wq_power_efficient, bool, 0444); |
| 284 | |
Tejun Heo | bce9038 | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 285 | static bool wq_numa_enabled; /* unbound NUMA affinity enabled */ |
| 286 | |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 287 | /* buf for wq_update_unbound_numa_attrs(), protected by CPU hotplug exclusion */ |
| 288 | static struct workqueue_attrs *wq_update_unbound_numa_attrs_buf; |
| 289 | |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 290 | static DEFINE_MUTEX(wq_pool_mutex); /* protects pools and workqueues list */ |
Tejun Heo | 2e109a2 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 291 | static DEFINE_SPINLOCK(wq_mayday_lock); /* protects wq->maydays list */ |
Tejun Heo | 5bcab33 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 292 | |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 293 | static LIST_HEAD(workqueues); /* PL: list of all workqueues */ |
| 294 | static bool workqueue_freezing; /* PL: have wqs started freezing? */ |
Tejun Heo | 7d19c5c | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 295 | |
| 296 | /* the per-cpu worker pools */ |
| 297 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct worker_pool [NR_STD_WORKER_POOLS], |
| 298 | cpu_worker_pools); |
| 299 | |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 300 | static DEFINE_IDR(worker_pool_idr); /* PR: idr of all pools */ |
Tejun Heo | 7d19c5c | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 301 | |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 302 | /* PL: hash of all unbound pools keyed by pool->attrs */ |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 303 | static DEFINE_HASHTABLE(unbound_pool_hash, UNBOUND_POOL_HASH_ORDER); |
| 304 | |
Tejun Heo | c5aa87b | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 305 | /* I: attributes used when instantiating standard unbound pools on demand */ |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 306 | static struct workqueue_attrs *unbound_std_wq_attrs[NR_STD_WORKER_POOLS]; |
| 307 | |
Tejun Heo | 8a2b753 | 2013-09-05 12:30:04 -0400 | [diff] [blame] | 308 | /* I: attributes used when instantiating ordered pools on demand */ |
| 309 | static struct workqueue_attrs *ordered_wq_attrs[NR_STD_WORKER_POOLS]; |
| 310 | |
Tejun Heo | d320c03 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 311 | struct workqueue_struct *system_wq __read_mostly; |
Marc Dionne | ad7b1f8 | 2013-05-06 17:44:55 -0400 | [diff] [blame] | 312 | EXPORT_SYMBOL(system_wq); |
Valentin Ilie | 044c782 | 2012-08-19 00:52:42 +0300 | [diff] [blame] | 313 | struct workqueue_struct *system_highpri_wq __read_mostly; |
Joonsoo Kim | 1aabe90 | 2012-08-15 23:25:39 +0900 | [diff] [blame] | 314 | EXPORT_SYMBOL_GPL(system_highpri_wq); |
Valentin Ilie | 044c782 | 2012-08-19 00:52:42 +0300 | [diff] [blame] | 315 | struct workqueue_struct *system_long_wq __read_mostly; |
Tejun Heo | d320c03 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 316 | EXPORT_SYMBOL_GPL(system_long_wq); |
Valentin Ilie | 044c782 | 2012-08-19 00:52:42 +0300 | [diff] [blame] | 317 | struct workqueue_struct *system_unbound_wq __read_mostly; |
Tejun Heo | f342179 | 2010-07-02 10:03:51 +0200 | [diff] [blame] | 318 | EXPORT_SYMBOL_GPL(system_unbound_wq); |
Valentin Ilie | 044c782 | 2012-08-19 00:52:42 +0300 | [diff] [blame] | 319 | struct workqueue_struct *system_freezable_wq __read_mostly; |
Tejun Heo | 24d51ad | 2011-02-21 09:52:50 +0100 | [diff] [blame] | 320 | EXPORT_SYMBOL_GPL(system_freezable_wq); |
Viresh Kumar | 0668106 | 2013-04-24 17:12:54 +0530 | [diff] [blame] | 321 | struct workqueue_struct *system_power_efficient_wq __read_mostly; |
| 322 | EXPORT_SYMBOL_GPL(system_power_efficient_wq); |
| 323 | struct workqueue_struct *system_freezable_power_efficient_wq __read_mostly; |
| 324 | EXPORT_SYMBOL_GPL(system_freezable_power_efficient_wq); |
Tejun Heo | d320c03 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 325 | |
Tejun Heo | 7d19c5c | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 326 | static int worker_thread(void *__worker); |
| 327 | static void copy_workqueue_attrs(struct workqueue_attrs *to, |
| 328 | const struct workqueue_attrs *from); |
| 329 | |
Tejun Heo | 97bd234 | 2010-10-05 10:41:14 +0200 | [diff] [blame] | 330 | #define CREATE_TRACE_POINTS |
| 331 | #include <trace/events/workqueue.h> |
| 332 | |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 333 | #define assert_rcu_or_pool_mutex() \ |
Tejun Heo | 5bcab33 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 334 | rcu_lockdep_assert(rcu_read_lock_sched_held() || \ |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 335 | lockdep_is_held(&wq_pool_mutex), \ |
| 336 | "sched RCU or wq_pool_mutex should be held") |
Tejun Heo | 5bcab33 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 337 | |
Lai Jiangshan | b09f4fd | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 338 | #define assert_rcu_or_wq_mutex(wq) \ |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 339 | rcu_lockdep_assert(rcu_read_lock_sched_held() || \ |
Lai Jiangshan | b592760 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 340 | lockdep_is_held(&wq->mutex), \ |
Lai Jiangshan | b09f4fd | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 341 | "sched RCU or wq->mutex should be held") |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 342 | |
Tejun Heo | 822d840 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 343 | #ifdef CONFIG_LOCKDEP |
| 344 | #define assert_manager_or_pool_lock(pool) \ |
Lai Jiangshan | 519e3c1 | 2013-03-20 03:28:21 +0800 | [diff] [blame] | 345 | WARN_ONCE(debug_locks && \ |
| 346 | !lockdep_is_held(&(pool)->manager_mutex) && \ |
Tejun Heo | 822d840 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 347 | !lockdep_is_held(&(pool)->lock), \ |
| 348 | "pool->manager_mutex or ->lock should be held") |
| 349 | #else |
| 350 | #define assert_manager_or_pool_lock(pool) do { } while (0) |
| 351 | #endif |
| 352 | |
Tejun Heo | f02ae73 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 353 | #define for_each_cpu_worker_pool(pool, cpu) \ |
| 354 | for ((pool) = &per_cpu(cpu_worker_pools, cpu)[0]; \ |
| 355 | (pool) < &per_cpu(cpu_worker_pools, cpu)[NR_STD_WORKER_POOLS]; \ |
Tejun Heo | 7a62c2c | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 356 | (pool)++) |
Tejun Heo | 4ce62e9 | 2012-07-13 22:16:44 -0700 | [diff] [blame] | 357 | |
Tejun Heo | 49e3cf4 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 358 | /** |
Tejun Heo | 1711696 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 359 | * for_each_pool - iterate through all worker_pools in the system |
| 360 | * @pool: iteration cursor |
Tejun Heo | 611c92a | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 361 | * @pi: integer used for iteration |
Tejun Heo | fa1b54e | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 362 | * |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 363 | * This must be called either with wq_pool_mutex held or sched RCU read |
| 364 | * locked. If the pool needs to be used beyond the locking in effect, the |
| 365 | * caller is responsible for guaranteeing that the pool stays online. |
Tejun Heo | fa1b54e | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 366 | * |
| 367 | * The if/else clause exists only for the lockdep assertion and can be |
| 368 | * ignored. |
Tejun Heo | 1711696 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 369 | */ |
Tejun Heo | 611c92a | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 370 | #define for_each_pool(pool, pi) \ |
| 371 | idr_for_each_entry(&worker_pool_idr, pool, pi) \ |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 372 | if (({ assert_rcu_or_pool_mutex(); false; })) { } \ |
Tejun Heo | fa1b54e | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 373 | else |
Tejun Heo | 1711696 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 374 | |
| 375 | /** |
Tejun Heo | 822d840 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 376 | * for_each_pool_worker - iterate through all workers of a worker_pool |
| 377 | * @worker: iteration cursor |
| 378 | * @wi: integer used for iteration |
| 379 | * @pool: worker_pool to iterate workers of |
| 380 | * |
| 381 | * This must be called with either @pool->manager_mutex or ->lock held. |
| 382 | * |
| 383 | * The if/else clause exists only for the lockdep assertion and can be |
| 384 | * ignored. |
| 385 | */ |
| 386 | #define for_each_pool_worker(worker, wi, pool) \ |
| 387 | idr_for_each_entry(&(pool)->worker_idr, (worker), (wi)) \ |
| 388 | if (({ assert_manager_or_pool_lock((pool)); false; })) { } \ |
| 389 | else |
| 390 | |
| 391 | /** |
Tejun Heo | 49e3cf4 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 392 | * for_each_pwq - iterate through all pool_workqueues of the specified workqueue |
| 393 | * @pwq: iteration cursor |
| 394 | * @wq: the target workqueue |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 395 | * |
Lai Jiangshan | b09f4fd | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 396 | * 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] | 397 | * If the pwq needs to be used beyond the locking in effect, the caller is |
| 398 | * responsible for guaranteeing that the pwq stays online. |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 399 | * |
| 400 | * The if/else clause exists only for the lockdep assertion and can be |
| 401 | * ignored. |
Tejun Heo | 49e3cf4 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 402 | */ |
| 403 | #define for_each_pwq(pwq, wq) \ |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 404 | list_for_each_entry_rcu((pwq), &(wq)->pwqs, pwqs_node) \ |
Lai Jiangshan | b09f4fd | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 405 | if (({ assert_rcu_or_wq_mutex(wq); false; })) { } \ |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 406 | else |
Tejun Heo | f342179 | 2010-07-02 10:03:51 +0200 | [diff] [blame] | 407 | |
Thomas Gleixner | dc186ad | 2009-11-16 01:09:48 +0900 | [diff] [blame] | 408 | #ifdef CONFIG_DEBUG_OBJECTS_WORK |
| 409 | |
| 410 | static struct debug_obj_descr work_debug_descr; |
| 411 | |
Stanislaw Gruszka | 9977728 | 2011-03-07 09:58:33 +0100 | [diff] [blame] | 412 | static void *work_debug_hint(void *addr) |
| 413 | { |
| 414 | return ((struct work_struct *) addr)->func; |
| 415 | } |
| 416 | |
Thomas Gleixner | dc186ad | 2009-11-16 01:09:48 +0900 | [diff] [blame] | 417 | /* |
| 418 | * fixup_init is called when: |
| 419 | * - an active object is initialized |
| 420 | */ |
| 421 | static int work_fixup_init(void *addr, enum debug_obj_state state) |
| 422 | { |
| 423 | struct work_struct *work = addr; |
| 424 | |
| 425 | switch (state) { |
| 426 | case ODEBUG_STATE_ACTIVE: |
| 427 | cancel_work_sync(work); |
| 428 | debug_object_init(work, &work_debug_descr); |
| 429 | return 1; |
| 430 | default: |
| 431 | return 0; |
| 432 | } |
| 433 | } |
| 434 | |
| 435 | /* |
| 436 | * fixup_activate is called when: |
| 437 | * - an active object is activated |
| 438 | * - an unknown object is activated (might be a statically initialized object) |
| 439 | */ |
| 440 | static int work_fixup_activate(void *addr, enum debug_obj_state state) |
| 441 | { |
| 442 | struct work_struct *work = addr; |
| 443 | |
| 444 | switch (state) { |
| 445 | |
| 446 | case ODEBUG_STATE_NOTAVAILABLE: |
| 447 | /* |
| 448 | * This is not really a fixup. The work struct was |
| 449 | * statically initialized. We just make sure that it |
| 450 | * is tracked in the object tracker. |
| 451 | */ |
Tejun Heo | 22df02b | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 452 | if (test_bit(WORK_STRUCT_STATIC_BIT, work_data_bits(work))) { |
Thomas Gleixner | dc186ad | 2009-11-16 01:09:48 +0900 | [diff] [blame] | 453 | debug_object_init(work, &work_debug_descr); |
| 454 | debug_object_activate(work, &work_debug_descr); |
| 455 | return 0; |
| 456 | } |
| 457 | WARN_ON_ONCE(1); |
| 458 | return 0; |
| 459 | |
| 460 | case ODEBUG_STATE_ACTIVE: |
| 461 | WARN_ON(1); |
| 462 | |
| 463 | default: |
| 464 | return 0; |
| 465 | } |
| 466 | } |
| 467 | |
| 468 | /* |
| 469 | * fixup_free is called when: |
| 470 | * - an active object is freed |
| 471 | */ |
| 472 | static int work_fixup_free(void *addr, enum debug_obj_state state) |
| 473 | { |
| 474 | struct work_struct *work = addr; |
| 475 | |
| 476 | switch (state) { |
| 477 | case ODEBUG_STATE_ACTIVE: |
| 478 | cancel_work_sync(work); |
| 479 | debug_object_free(work, &work_debug_descr); |
| 480 | return 1; |
| 481 | default: |
| 482 | return 0; |
| 483 | } |
| 484 | } |
| 485 | |
| 486 | static struct debug_obj_descr work_debug_descr = { |
| 487 | .name = "work_struct", |
Stanislaw Gruszka | 9977728 | 2011-03-07 09:58:33 +0100 | [diff] [blame] | 488 | .debug_hint = work_debug_hint, |
Thomas Gleixner | dc186ad | 2009-11-16 01:09:48 +0900 | [diff] [blame] | 489 | .fixup_init = work_fixup_init, |
| 490 | .fixup_activate = work_fixup_activate, |
| 491 | .fixup_free = work_fixup_free, |
| 492 | }; |
| 493 | |
| 494 | static inline void debug_work_activate(struct work_struct *work) |
| 495 | { |
| 496 | debug_object_activate(work, &work_debug_descr); |
| 497 | } |
| 498 | |
| 499 | static inline void debug_work_deactivate(struct work_struct *work) |
| 500 | { |
| 501 | debug_object_deactivate(work, &work_debug_descr); |
| 502 | } |
| 503 | |
| 504 | void __init_work(struct work_struct *work, int onstack) |
| 505 | { |
| 506 | if (onstack) |
| 507 | debug_object_init_on_stack(work, &work_debug_descr); |
| 508 | else |
| 509 | debug_object_init(work, &work_debug_descr); |
| 510 | } |
| 511 | EXPORT_SYMBOL_GPL(__init_work); |
| 512 | |
| 513 | void destroy_work_on_stack(struct work_struct *work) |
| 514 | { |
| 515 | debug_object_free(work, &work_debug_descr); |
| 516 | } |
| 517 | EXPORT_SYMBOL_GPL(destroy_work_on_stack); |
| 518 | |
Thomas Gleixner | ea2e64f | 2014-03-23 14:20:44 +0000 | [diff] [blame] | 519 | void destroy_delayed_work_on_stack(struct delayed_work *work) |
| 520 | { |
| 521 | destroy_timer_on_stack(&work->timer); |
| 522 | debug_object_free(&work->work, &work_debug_descr); |
| 523 | } |
| 524 | EXPORT_SYMBOL_GPL(destroy_delayed_work_on_stack); |
| 525 | |
Thomas Gleixner | dc186ad | 2009-11-16 01:09:48 +0900 | [diff] [blame] | 526 | #else |
| 527 | static inline void debug_work_activate(struct work_struct *work) { } |
| 528 | static inline void debug_work_deactivate(struct work_struct *work) { } |
| 529 | #endif |
| 530 | |
Li Bin | 4e8b22b | 2013-09-10 09:52:35 +0800 | [diff] [blame] | 531 | /** |
| 532 | * worker_pool_assign_id - allocate ID and assing it to @pool |
| 533 | * @pool: the pool pointer of interest |
| 534 | * |
| 535 | * Returns 0 if ID in [0, WORK_OFFQ_POOL_NONE) is allocated and assigned |
| 536 | * successfully, -errno on failure. |
| 537 | */ |
Tejun Heo | 9daf9e6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 538 | static int worker_pool_assign_id(struct worker_pool *pool) |
| 539 | { |
| 540 | int ret; |
| 541 | |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 542 | lockdep_assert_held(&wq_pool_mutex); |
Tejun Heo | 5bcab33 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 543 | |
Li Bin | 4e8b22b | 2013-09-10 09:52:35 +0800 | [diff] [blame] | 544 | ret = idr_alloc(&worker_pool_idr, pool, 0, WORK_OFFQ_POOL_NONE, |
| 545 | GFP_KERNEL); |
Tejun Heo | 229641a | 2013-04-01 17:08:13 -0700 | [diff] [blame] | 546 | if (ret >= 0) { |
Tejun Heo | e68035f | 2013-03-13 14:59:38 -0700 | [diff] [blame] | 547 | pool->id = ret; |
Tejun Heo | 229641a | 2013-04-01 17:08:13 -0700 | [diff] [blame] | 548 | return 0; |
| 549 | } |
Tejun Heo | 9daf9e6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 550 | return ret; |
| 551 | } |
| 552 | |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 553 | /** |
Tejun Heo | df2d5ae | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 554 | * unbound_pwq_by_node - return the unbound pool_workqueue for the given node |
| 555 | * @wq: the target workqueue |
| 556 | * @node: the node ID |
| 557 | * |
| 558 | * This must be called either with pwq_lock held or sched RCU read locked. |
| 559 | * If the pwq needs to be used beyond the locking in effect, the caller is |
| 560 | * responsible for guaranteeing that the pwq stays online. |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 561 | * |
| 562 | * Return: The unbound pool_workqueue for @node. |
Tejun Heo | df2d5ae | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 563 | */ |
| 564 | static struct pool_workqueue *unbound_pwq_by_node(struct workqueue_struct *wq, |
| 565 | int node) |
| 566 | { |
| 567 | assert_rcu_or_wq_mutex(wq); |
| 568 | return rcu_dereference_raw(wq->numa_pwq_tbl[node]); |
| 569 | } |
| 570 | |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 571 | static unsigned int work_color_to_flags(int color) |
| 572 | { |
| 573 | return color << WORK_STRUCT_COLOR_SHIFT; |
| 574 | } |
| 575 | |
| 576 | static int get_work_color(struct work_struct *work) |
| 577 | { |
| 578 | return (*work_data_bits(work) >> WORK_STRUCT_COLOR_SHIFT) & |
| 579 | ((1 << WORK_STRUCT_COLOR_BITS) - 1); |
| 580 | } |
| 581 | |
| 582 | static int work_next_color(int color) |
| 583 | { |
| 584 | return (color + 1) % WORK_NR_COLORS; |
Oleg Nesterov | a848e3b | 2007-05-09 02:34:17 -0700 | [diff] [blame] | 585 | } |
| 586 | |
David Howells | 4594bf1 | 2006-12-07 11:33:26 +0000 | [diff] [blame] | 587 | /* |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 588 | * While queued, %WORK_STRUCT_PWQ is set and non flag bits of a work's data |
| 589 | * contain the pointer to the queued pwq. Once execution starts, the flag |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 590 | * is cleared and the high bits contain OFFQ flags and pool ID. |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 591 | * |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 592 | * set_work_pwq(), set_work_pool_and_clear_pending(), mark_work_canceling() |
| 593 | * 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] | 594 | * work->data. These functions should only be called while the work is |
| 595 | * owned - ie. while the PENDING bit is set. |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 596 | * |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 597 | * 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] | 598 | * corresponding to a work. Pool is available once the work has been |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 599 | * queued anywhere after initialization until it is sync canceled. pwq is |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 600 | * available only while the work item is queued. |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 601 | * |
| 602 | * %WORK_OFFQ_CANCELING is used to mark a work item which is being |
| 603 | * canceled. While being canceled, a work item may have its PENDING set |
| 604 | * but stay off timer and worklist for arbitrarily long and nobody should |
| 605 | * try to steal the PENDING bit. |
David Howells | 4594bf1 | 2006-12-07 11:33:26 +0000 | [diff] [blame] | 606 | */ |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 607 | static inline void set_work_data(struct work_struct *work, unsigned long data, |
| 608 | unsigned long flags) |
David Howells | 365970a | 2006-11-22 14:54:49 +0000 | [diff] [blame] | 609 | { |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 610 | WARN_ON_ONCE(!work_pending(work)); |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 611 | atomic_long_set(&work->data, data | flags | work_static(work)); |
David Howells | 365970a | 2006-11-22 14:54:49 +0000 | [diff] [blame] | 612 | } |
David Howells | 365970a | 2006-11-22 14:54:49 +0000 | [diff] [blame] | 613 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 614 | 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] | 615 | unsigned long extra_flags) |
Oleg Nesterov | 4d707b9 | 2010-04-23 17:40:40 +0200 | [diff] [blame] | 616 | { |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 617 | set_work_data(work, (unsigned long)pwq, |
| 618 | WORK_STRUCT_PENDING | WORK_STRUCT_PWQ | extra_flags); |
Oleg Nesterov | 4d707b9 | 2010-04-23 17:40:40 +0200 | [diff] [blame] | 619 | } |
| 620 | |
Lai Jiangshan | 4468a00 | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 621 | static void set_work_pool_and_keep_pending(struct work_struct *work, |
| 622 | int pool_id) |
| 623 | { |
| 624 | set_work_data(work, (unsigned long)pool_id << WORK_OFFQ_POOL_SHIFT, |
| 625 | WORK_STRUCT_PENDING); |
| 626 | } |
| 627 | |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 628 | static void set_work_pool_and_clear_pending(struct work_struct *work, |
| 629 | int pool_id) |
David Howells | 365970a | 2006-11-22 14:54:49 +0000 | [diff] [blame] | 630 | { |
Tejun Heo | 23657bb | 2012-08-13 17:08:19 -0700 | [diff] [blame] | 631 | /* |
| 632 | * The following wmb is paired with the implied mb in |
| 633 | * test_and_set_bit(PENDING) and ensures all updates to @work made |
| 634 | * here are visible to and precede any updates by the next PENDING |
| 635 | * owner. |
| 636 | */ |
| 637 | smp_wmb(); |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 638 | set_work_data(work, (unsigned long)pool_id << WORK_OFFQ_POOL_SHIFT, 0); |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 639 | } |
| 640 | |
| 641 | static void clear_work_data(struct work_struct *work) |
| 642 | { |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 643 | smp_wmb(); /* see set_work_pool_and_clear_pending() */ |
| 644 | set_work_data(work, WORK_STRUCT_NO_POOL, 0); |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 645 | } |
| 646 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 647 | static struct pool_workqueue *get_work_pwq(struct work_struct *work) |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 648 | { |
Tejun Heo | e120153 | 2010-07-22 14:14:25 +0200 | [diff] [blame] | 649 | unsigned long data = atomic_long_read(&work->data); |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 650 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 651 | if (data & WORK_STRUCT_PWQ) |
Tejun Heo | e120153 | 2010-07-22 14:14:25 +0200 | [diff] [blame] | 652 | return (void *)(data & WORK_STRUCT_WQ_DATA_MASK); |
| 653 | else |
| 654 | return NULL; |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 655 | } |
| 656 | |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 657 | /** |
| 658 | * get_work_pool - return the worker_pool a given work was associated with |
| 659 | * @work: the work item of interest |
| 660 | * |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 661 | * Pools are created and destroyed under wq_pool_mutex, and allows read |
| 662 | * access under sched-RCU read lock. As such, this function should be |
| 663 | * called under wq_pool_mutex or with preemption disabled. |
Tejun Heo | fa1b54e | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 664 | * |
| 665 | * All fields of the returned pool are accessible as long as the above |
| 666 | * mentioned locking is in effect. If the returned pool needs to be used |
| 667 | * beyond the critical section, the caller is responsible for ensuring the |
| 668 | * returned pool is and stays online. |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 669 | * |
| 670 | * Return: The worker_pool @work was last associated with. %NULL if none. |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 671 | */ |
| 672 | static struct worker_pool *get_work_pool(struct work_struct *work) |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 673 | { |
Tejun Heo | e120153 | 2010-07-22 14:14:25 +0200 | [diff] [blame] | 674 | unsigned long data = atomic_long_read(&work->data); |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 675 | int pool_id; |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 676 | |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 677 | assert_rcu_or_pool_mutex(); |
Tejun Heo | fa1b54e | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 678 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 679 | if (data & WORK_STRUCT_PWQ) |
| 680 | return ((struct pool_workqueue *) |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 681 | (data & WORK_STRUCT_WQ_DATA_MASK))->pool; |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 682 | |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 683 | pool_id = data >> WORK_OFFQ_POOL_SHIFT; |
| 684 | if (pool_id == WORK_OFFQ_POOL_NONE) |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 685 | return NULL; |
| 686 | |
Tejun Heo | fa1b54e | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 687 | return idr_find(&worker_pool_idr, pool_id); |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 688 | } |
| 689 | |
| 690 | /** |
| 691 | * get_work_pool_id - return the worker pool ID a given work is associated with |
| 692 | * @work: the work item of interest |
| 693 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 694 | * Return: The worker_pool ID @work was last associated with. |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 695 | * %WORK_OFFQ_POOL_NONE if none. |
| 696 | */ |
| 697 | static int get_work_pool_id(struct work_struct *work) |
| 698 | { |
Lai Jiangshan | 54d5b7d | 2013-02-07 13:14:20 -0800 | [diff] [blame] | 699 | unsigned long data = atomic_long_read(&work->data); |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 700 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 701 | if (data & WORK_STRUCT_PWQ) |
| 702 | return ((struct pool_workqueue *) |
Lai Jiangshan | 54d5b7d | 2013-02-07 13:14:20 -0800 | [diff] [blame] | 703 | (data & WORK_STRUCT_WQ_DATA_MASK))->pool->id; |
| 704 | |
| 705 | return data >> WORK_OFFQ_POOL_SHIFT; |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 706 | } |
| 707 | |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 708 | static void mark_work_canceling(struct work_struct *work) |
| 709 | { |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 710 | unsigned long pool_id = get_work_pool_id(work); |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 711 | |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 712 | pool_id <<= WORK_OFFQ_POOL_SHIFT; |
| 713 | set_work_data(work, pool_id | WORK_OFFQ_CANCELING, WORK_STRUCT_PENDING); |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 714 | } |
| 715 | |
| 716 | static bool work_is_canceling(struct work_struct *work) |
| 717 | { |
| 718 | unsigned long data = atomic_long_read(&work->data); |
| 719 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 720 | return !(data & WORK_STRUCT_PWQ) && (data & WORK_OFFQ_CANCELING); |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 721 | } |
| 722 | |
David Howells | 365970a | 2006-11-22 14:54:49 +0000 | [diff] [blame] | 723 | /* |
Tejun Heo | 3270476 | 2012-07-13 22:16:45 -0700 | [diff] [blame] | 724 | * Policy functions. These define the policies on how the global worker |
| 725 | * pools are managed. Unless noted otherwise, these functions assume that |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 726 | * they're being called with pool->lock held. |
David Howells | 365970a | 2006-11-22 14:54:49 +0000 | [diff] [blame] | 727 | */ |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 728 | |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 729 | static bool __need_more_worker(struct worker_pool *pool) |
David Howells | 365970a | 2006-11-22 14:54:49 +0000 | [diff] [blame] | 730 | { |
Tejun Heo | e19e397 | 2013-01-24 11:39:44 -0800 | [diff] [blame] | 731 | return !atomic_read(&pool->nr_running); |
David Howells | 365970a | 2006-11-22 14:54:49 +0000 | [diff] [blame] | 732 | } |
| 733 | |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 734 | /* |
| 735 | * Need to wake up a worker? Called from anything but currently |
| 736 | * running workers. |
Tejun Heo | 974271c | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 737 | * |
| 738 | * Note that, because unbound workers never contribute to nr_running, this |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 739 | * function will always return %true for unbound pools as long as the |
Tejun Heo | 974271c | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 740 | * worklist isn't empty. |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 741 | */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 742 | static bool need_more_worker(struct worker_pool *pool) |
David Howells | 365970a | 2006-11-22 14:54:49 +0000 | [diff] [blame] | 743 | { |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 744 | return !list_empty(&pool->worklist) && __need_more_worker(pool); |
David Howells | 365970a | 2006-11-22 14:54:49 +0000 | [diff] [blame] | 745 | } |
| 746 | |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 747 | /* Can I start working? Called from busy but !running workers. */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 748 | static bool may_start_working(struct worker_pool *pool) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 749 | { |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 750 | return pool->nr_idle; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 751 | } |
| 752 | |
| 753 | /* Do I need to keep working? Called from currently running workers. */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 754 | static bool keep_working(struct worker_pool *pool) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 755 | { |
Tejun Heo | e19e397 | 2013-01-24 11:39:44 -0800 | [diff] [blame] | 756 | return !list_empty(&pool->worklist) && |
| 757 | atomic_read(&pool->nr_running) <= 1; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 758 | } |
| 759 | |
| 760 | /* Do we need a new worker? Called from manager. */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 761 | static bool need_to_create_worker(struct worker_pool *pool) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 762 | { |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 763 | return need_more_worker(pool) && !may_start_working(pool); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 764 | } |
| 765 | |
| 766 | /* Do I need to be the manager? */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 767 | static bool need_to_manage_workers(struct worker_pool *pool) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 768 | { |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 769 | return need_to_create_worker(pool) || |
Tejun Heo | 11ebea5 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 770 | (pool->flags & POOL_MANAGE_WORKERS); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 771 | } |
| 772 | |
| 773 | /* Do we have too many workers and should some go away? */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 774 | static bool too_many_workers(struct worker_pool *pool) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 775 | { |
Tejun Heo | 34a06bd | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 776 | bool managing = mutex_is_locked(&pool->manager_arb); |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 777 | int nr_idle = pool->nr_idle + managing; /* manager is considered idle */ |
| 778 | int nr_busy = pool->nr_workers - nr_idle; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 779 | |
Lai Jiangshan | ea1abd6 | 2012-09-18 09:59:22 -0700 | [diff] [blame] | 780 | /* |
| 781 | * nr_idle and idle_list may disagree if idle rebinding is in |
| 782 | * progress. Never return %true if idle_list is empty. |
| 783 | */ |
| 784 | if (list_empty(&pool->idle_list)) |
| 785 | return false; |
| 786 | |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 787 | return nr_idle > 2 && (nr_idle - 2) * MAX_IDLE_WORKERS_RATIO >= nr_busy; |
| 788 | } |
| 789 | |
| 790 | /* |
| 791 | * Wake up functions. |
| 792 | */ |
| 793 | |
Tejun Heo | 7e11629 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 794 | /* Return the first worker. Safe with preemption disabled */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 795 | static struct worker *first_worker(struct worker_pool *pool) |
Tejun Heo | 7e11629 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 796 | { |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 797 | if (unlikely(list_empty(&pool->idle_list))) |
Tejun Heo | 7e11629 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 798 | return NULL; |
| 799 | |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 800 | return list_first_entry(&pool->idle_list, struct worker, entry); |
Tejun Heo | 7e11629 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 801 | } |
| 802 | |
| 803 | /** |
| 804 | * wake_up_worker - wake up an idle worker |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 805 | * @pool: worker pool to wake worker from |
Tejun Heo | 7e11629 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 806 | * |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 807 | * Wake up the first idle worker of @pool. |
Tejun Heo | 7e11629 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 808 | * |
| 809 | * CONTEXT: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 810 | * spin_lock_irq(pool->lock). |
Tejun Heo | 7e11629 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 811 | */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 812 | static void wake_up_worker(struct worker_pool *pool) |
Tejun Heo | 7e11629 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 813 | { |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 814 | struct worker *worker = first_worker(pool); |
Tejun Heo | 7e11629 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 815 | |
| 816 | if (likely(worker)) |
| 817 | wake_up_process(worker->task); |
| 818 | } |
| 819 | |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 820 | /** |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 821 | * wq_worker_waking_up - a worker is waking up |
| 822 | * @task: task waking up |
| 823 | * @cpu: CPU @task is waking up to |
| 824 | * |
| 825 | * This function is called during try_to_wake_up() when a worker is |
| 826 | * being awoken. |
| 827 | * |
| 828 | * CONTEXT: |
| 829 | * spin_lock_irq(rq->lock) |
| 830 | */ |
Tejun Heo | d84ff05 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 831 | void wq_worker_waking_up(struct task_struct *task, int cpu) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 832 | { |
| 833 | struct worker *worker = kthread_data(task); |
| 834 | |
Joonsoo Kim | 3657600 | 2012-10-26 23:03:49 +0900 | [diff] [blame] | 835 | if (!(worker->flags & WORKER_NOT_RUNNING)) { |
Tejun Heo | ec22ca5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 836 | WARN_ON_ONCE(worker->pool->cpu != cpu); |
Tejun Heo | e19e397 | 2013-01-24 11:39:44 -0800 | [diff] [blame] | 837 | atomic_inc(&worker->pool->nr_running); |
Joonsoo Kim | 3657600 | 2012-10-26 23:03:49 +0900 | [diff] [blame] | 838 | } |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 839 | } |
| 840 | |
| 841 | /** |
| 842 | * wq_worker_sleeping - a worker is going to sleep |
| 843 | * @task: task going to sleep |
| 844 | * @cpu: CPU in question, must be the current CPU number |
| 845 | * |
| 846 | * This function is called during schedule() when a busy worker is |
| 847 | * going to sleep. Worker on the same cpu can be woken up by |
| 848 | * returning pointer to its task. |
| 849 | * |
| 850 | * CONTEXT: |
| 851 | * spin_lock_irq(rq->lock) |
| 852 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 853 | * Return: |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 854 | * Worker task on @cpu to wake up, %NULL if none. |
| 855 | */ |
Tejun Heo | d84ff05 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 856 | struct task_struct *wq_worker_sleeping(struct task_struct *task, int cpu) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 857 | { |
| 858 | struct worker *worker = kthread_data(task), *to_wakeup = NULL; |
Tejun Heo | 111c225 | 2013-01-17 17:16:24 -0800 | [diff] [blame] | 859 | struct worker_pool *pool; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 860 | |
Tejun Heo | 111c225 | 2013-01-17 17:16:24 -0800 | [diff] [blame] | 861 | /* |
| 862 | * Rescuers, which may not have all the fields set up like normal |
| 863 | * workers, also reach here, let's not access anything before |
| 864 | * checking NOT_RUNNING. |
| 865 | */ |
Steven Rostedt | 2d64672 | 2010-12-03 23:12:33 -0500 | [diff] [blame] | 866 | if (worker->flags & WORKER_NOT_RUNNING) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 867 | return NULL; |
| 868 | |
Tejun Heo | 111c225 | 2013-01-17 17:16:24 -0800 | [diff] [blame] | 869 | pool = worker->pool; |
Tejun Heo | 111c225 | 2013-01-17 17:16:24 -0800 | [diff] [blame] | 870 | |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 871 | /* this can only happen on the local cpu */ |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 872 | if (WARN_ON_ONCE(cpu != raw_smp_processor_id())) |
| 873 | return NULL; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 874 | |
| 875 | /* |
| 876 | * The counterpart of the following dec_and_test, implied mb, |
| 877 | * worklist not empty test sequence is in insert_work(). |
| 878 | * Please read comment there. |
| 879 | * |
Tejun Heo | 628c78e | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 880 | * NOT_RUNNING is clear. This means that we're bound to and |
| 881 | * running on the local cpu w/ rq lock held and preemption |
| 882 | * disabled, which in turn means that none else could be |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 883 | * manipulating idle_list, so dereferencing idle_list without pool |
Tejun Heo | 628c78e | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 884 | * lock is safe. |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 885 | */ |
Tejun Heo | e19e397 | 2013-01-24 11:39:44 -0800 | [diff] [blame] | 886 | if (atomic_dec_and_test(&pool->nr_running) && |
| 887 | !list_empty(&pool->worklist)) |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 888 | to_wakeup = first_worker(pool); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 889 | return to_wakeup ? to_wakeup->task : NULL; |
| 890 | } |
| 891 | |
| 892 | /** |
| 893 | * worker_set_flags - set worker flags and adjust nr_running accordingly |
Tejun Heo | cb44476 | 2010-07-02 10:03:50 +0200 | [diff] [blame] | 894 | * @worker: self |
Tejun Heo | d302f01 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 895 | * @flags: flags to set |
| 896 | * @wakeup: wakeup an idle worker if necessary |
| 897 | * |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 898 | * Set @flags in @worker->flags and adjust nr_running accordingly. If |
| 899 | * nr_running becomes zero and @wakeup is %true, an idle worker is |
| 900 | * woken up. |
Tejun Heo | d302f01 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 901 | * |
Tejun Heo | cb44476 | 2010-07-02 10:03:50 +0200 | [diff] [blame] | 902 | * CONTEXT: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 903 | * spin_lock_irq(pool->lock) |
Tejun Heo | d302f01 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 904 | */ |
| 905 | static inline void worker_set_flags(struct worker *worker, unsigned int flags, |
| 906 | bool wakeup) |
| 907 | { |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 908 | struct worker_pool *pool = worker->pool; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 909 | |
Tejun Heo | cb44476 | 2010-07-02 10:03:50 +0200 | [diff] [blame] | 910 | WARN_ON_ONCE(worker->task != current); |
| 911 | |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 912 | /* |
| 913 | * If transitioning into NOT_RUNNING, adjust nr_running and |
| 914 | * wake up an idle worker as necessary if requested by |
| 915 | * @wakeup. |
| 916 | */ |
| 917 | if ((flags & WORKER_NOT_RUNNING) && |
| 918 | !(worker->flags & WORKER_NOT_RUNNING)) { |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 919 | if (wakeup) { |
Tejun Heo | e19e397 | 2013-01-24 11:39:44 -0800 | [diff] [blame] | 920 | if (atomic_dec_and_test(&pool->nr_running) && |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 921 | !list_empty(&pool->worklist)) |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 922 | wake_up_worker(pool); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 923 | } else |
Tejun Heo | e19e397 | 2013-01-24 11:39:44 -0800 | [diff] [blame] | 924 | atomic_dec(&pool->nr_running); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 925 | } |
| 926 | |
Tejun Heo | d302f01 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 927 | worker->flags |= flags; |
| 928 | } |
| 929 | |
| 930 | /** |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 931 | * worker_clr_flags - clear worker flags and adjust nr_running accordingly |
Tejun Heo | cb44476 | 2010-07-02 10:03:50 +0200 | [diff] [blame] | 932 | * @worker: self |
Tejun Heo | d302f01 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 933 | * @flags: flags to clear |
| 934 | * |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 935 | * Clear @flags in @worker->flags and adjust nr_running accordingly. |
Tejun Heo | d302f01 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 936 | * |
Tejun Heo | cb44476 | 2010-07-02 10:03:50 +0200 | [diff] [blame] | 937 | * CONTEXT: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 938 | * spin_lock_irq(pool->lock) |
Tejun Heo | d302f01 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 939 | */ |
| 940 | static inline void worker_clr_flags(struct worker *worker, unsigned int flags) |
| 941 | { |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 942 | struct worker_pool *pool = worker->pool; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 943 | unsigned int oflags = worker->flags; |
| 944 | |
Tejun Heo | cb44476 | 2010-07-02 10:03:50 +0200 | [diff] [blame] | 945 | WARN_ON_ONCE(worker->task != current); |
| 946 | |
Tejun Heo | d302f01 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 947 | worker->flags &= ~flags; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 948 | |
Tejun Heo | 42c025f | 2011-01-11 15:58:49 +0100 | [diff] [blame] | 949 | /* |
| 950 | * If transitioning out of NOT_RUNNING, increment nr_running. Note |
| 951 | * that the nested NOT_RUNNING is not a noop. NOT_RUNNING is mask |
| 952 | * of multiple flags, not a single flag. |
| 953 | */ |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 954 | if ((flags & WORKER_NOT_RUNNING) && (oflags & WORKER_NOT_RUNNING)) |
| 955 | if (!(worker->flags & WORKER_NOT_RUNNING)) |
Tejun Heo | e19e397 | 2013-01-24 11:39:44 -0800 | [diff] [blame] | 956 | atomic_inc(&pool->nr_running); |
Tejun Heo | d302f01 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 957 | } |
| 958 | |
| 959 | /** |
Tejun Heo | 8cca0ee | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 960 | * find_worker_executing_work - find worker which is executing a work |
Tejun Heo | c9e7cf2 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 961 | * @pool: pool of interest |
Tejun Heo | 8cca0ee | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 962 | * @work: work to find worker for |
| 963 | * |
Tejun Heo | c9e7cf2 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 964 | * Find a worker which is executing @work on @pool by searching |
| 965 | * @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] | 966 | * to match, its current execution should match the address of @work and |
| 967 | * its work function. This is to avoid unwanted dependency between |
| 968 | * unrelated work executions through a work item being recycled while still |
| 969 | * being executed. |
| 970 | * |
| 971 | * This is a bit tricky. A work item may be freed once its execution |
| 972 | * starts and nothing prevents the freed area from being recycled for |
| 973 | * another work item. If the same work item address ends up being reused |
| 974 | * before the original execution finishes, workqueue will identify the |
| 975 | * recycled work item as currently executing and make it wait until the |
| 976 | * current execution finishes, introducing an unwanted dependency. |
| 977 | * |
Tejun Heo | c5aa87b | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 978 | * This function checks the work item address and work function to avoid |
| 979 | * false positives. Note that this isn't complete as one may construct a |
| 980 | * work function which can introduce dependency onto itself through a |
| 981 | * recycled work item. Well, if somebody wants to shoot oneself in the |
| 982 | * foot that badly, there's only so much we can do, and if such deadlock |
| 983 | * actually occurs, it should be easy to locate the culprit work function. |
Tejun Heo | 8cca0ee | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 984 | * |
| 985 | * CONTEXT: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 986 | * spin_lock_irq(pool->lock). |
Tejun Heo | 8cca0ee | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 987 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 988 | * Return: |
| 989 | * Pointer to worker which is executing @work if found, %NULL |
Tejun Heo | 8cca0ee | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 990 | * otherwise. |
| 991 | */ |
Tejun Heo | c9e7cf2 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 992 | static struct worker *find_worker_executing_work(struct worker_pool *pool, |
Tejun Heo | 8cca0ee | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 993 | struct work_struct *work) |
| 994 | { |
Sasha Levin | 42f8570 | 2012-12-17 10:01:23 -0500 | [diff] [blame] | 995 | struct worker *worker; |
Sasha Levin | 42f8570 | 2012-12-17 10:01:23 -0500 | [diff] [blame] | 996 | |
Sasha Levin | b67bfe0 | 2013-02-27 17:06:00 -0800 | [diff] [blame] | 997 | hash_for_each_possible(pool->busy_hash, worker, hentry, |
Tejun Heo | a2c1c57 | 2012-12-18 10:35:02 -0800 | [diff] [blame] | 998 | (unsigned long)work) |
| 999 | if (worker->current_work == work && |
| 1000 | worker->current_func == work->func) |
Sasha Levin | 42f8570 | 2012-12-17 10:01:23 -0500 | [diff] [blame] | 1001 | return worker; |
| 1002 | |
| 1003 | return NULL; |
Tejun Heo | 8cca0ee | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 1004 | } |
| 1005 | |
| 1006 | /** |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1007 | * move_linked_works - move linked works to a list |
| 1008 | * @work: start of series of works to be scheduled |
| 1009 | * @head: target list to append @work to |
| 1010 | * @nextp: out paramter for nested worklist walking |
| 1011 | * |
| 1012 | * Schedule linked works starting from @work to @head. Work series to |
| 1013 | * be scheduled starts at @work and includes any consecutive work with |
| 1014 | * WORK_STRUCT_LINKED set in its predecessor. |
| 1015 | * |
| 1016 | * If @nextp is not NULL, it's updated to point to the next work of |
| 1017 | * the last scheduled work. This allows move_linked_works() to be |
| 1018 | * nested inside outer list_for_each_entry_safe(). |
| 1019 | * |
| 1020 | * CONTEXT: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1021 | * spin_lock_irq(pool->lock). |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1022 | */ |
| 1023 | static void move_linked_works(struct work_struct *work, struct list_head *head, |
| 1024 | struct work_struct **nextp) |
| 1025 | { |
| 1026 | struct work_struct *n; |
| 1027 | |
| 1028 | /* |
| 1029 | * Linked worklist will always end before the end of the list, |
| 1030 | * use NULL for list head. |
| 1031 | */ |
| 1032 | list_for_each_entry_safe_from(work, n, NULL, entry) { |
| 1033 | list_move_tail(&work->entry, head); |
| 1034 | if (!(*work_data_bits(work) & WORK_STRUCT_LINKED)) |
| 1035 | break; |
| 1036 | } |
| 1037 | |
| 1038 | /* |
| 1039 | * If we're already inside safe list traversal and have moved |
| 1040 | * multiple works to the scheduled queue, the next position |
| 1041 | * needs to be updated. |
| 1042 | */ |
| 1043 | if (nextp) |
| 1044 | *nextp = n; |
| 1045 | } |
| 1046 | |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1047 | /** |
| 1048 | * get_pwq - get an extra reference on the specified pool_workqueue |
| 1049 | * @pwq: pool_workqueue to get |
| 1050 | * |
| 1051 | * Obtain an extra reference on @pwq. The caller should guarantee that |
| 1052 | * @pwq has positive refcnt and be holding the matching pool->lock. |
| 1053 | */ |
| 1054 | static void get_pwq(struct pool_workqueue *pwq) |
| 1055 | { |
| 1056 | lockdep_assert_held(&pwq->pool->lock); |
| 1057 | WARN_ON_ONCE(pwq->refcnt <= 0); |
| 1058 | pwq->refcnt++; |
| 1059 | } |
| 1060 | |
| 1061 | /** |
| 1062 | * put_pwq - put a pool_workqueue reference |
| 1063 | * @pwq: pool_workqueue to put |
| 1064 | * |
| 1065 | * Drop a reference of @pwq. If its refcnt reaches zero, schedule its |
| 1066 | * destruction. The caller should be holding the matching pool->lock. |
| 1067 | */ |
| 1068 | static void put_pwq(struct pool_workqueue *pwq) |
| 1069 | { |
| 1070 | lockdep_assert_held(&pwq->pool->lock); |
| 1071 | if (likely(--pwq->refcnt)) |
| 1072 | return; |
| 1073 | if (WARN_ON_ONCE(!(pwq->wq->flags & WQ_UNBOUND))) |
| 1074 | return; |
| 1075 | /* |
| 1076 | * @pwq can't be released under pool->lock, bounce to |
| 1077 | * pwq_unbound_release_workfn(). This never recurses on the same |
| 1078 | * pool->lock as this path is taken only for unbound workqueues and |
| 1079 | * the release work item is scheduled on a per-cpu workqueue. To |
| 1080 | * avoid lockdep warning, unbound pool->locks are given lockdep |
| 1081 | * subclass of 1 in get_unbound_pool(). |
| 1082 | */ |
| 1083 | schedule_work(&pwq->unbound_release_work); |
| 1084 | } |
| 1085 | |
Tejun Heo | dce90d4 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 1086 | /** |
| 1087 | * put_pwq_unlocked - put_pwq() with surrounding pool lock/unlock |
| 1088 | * @pwq: pool_workqueue to put (can be %NULL) |
| 1089 | * |
| 1090 | * put_pwq() with locking. This function also allows %NULL @pwq. |
| 1091 | */ |
| 1092 | static void put_pwq_unlocked(struct pool_workqueue *pwq) |
| 1093 | { |
| 1094 | if (pwq) { |
| 1095 | /* |
| 1096 | * As both pwqs and pools are sched-RCU protected, the |
| 1097 | * following lock operations are safe. |
| 1098 | */ |
| 1099 | spin_lock_irq(&pwq->pool->lock); |
| 1100 | put_pwq(pwq); |
| 1101 | spin_unlock_irq(&pwq->pool->lock); |
| 1102 | } |
| 1103 | } |
| 1104 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1105 | static void pwq_activate_delayed_work(struct work_struct *work) |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1106 | { |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1107 | struct pool_workqueue *pwq = get_work_pwq(work); |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1108 | |
| 1109 | trace_workqueue_activate_work(work); |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1110 | move_linked_works(work, &pwq->pool->worklist, NULL); |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1111 | __clear_bit(WORK_STRUCT_DELAYED_BIT, work_data_bits(work)); |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1112 | pwq->nr_active++; |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1113 | } |
| 1114 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1115 | static void pwq_activate_first_delayed(struct pool_workqueue *pwq) |
Lai Jiangshan | 3aa6249 | 2012-09-18 10:40:00 -0700 | [diff] [blame] | 1116 | { |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1117 | struct work_struct *work = list_first_entry(&pwq->delayed_works, |
Lai Jiangshan | 3aa6249 | 2012-09-18 10:40:00 -0700 | [diff] [blame] | 1118 | struct work_struct, entry); |
| 1119 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1120 | pwq_activate_delayed_work(work); |
Lai Jiangshan | 3aa6249 | 2012-09-18 10:40:00 -0700 | [diff] [blame] | 1121 | } |
| 1122 | |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1123 | /** |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1124 | * pwq_dec_nr_in_flight - decrement pwq's nr_in_flight |
| 1125 | * @pwq: pwq of interest |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1126 | * @color: color of work which left the queue |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1127 | * |
| 1128 | * A work either has completed or is removed from pending queue, |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1129 | * decrement nr_in_flight of its pwq and handle workqueue flushing. |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1130 | * |
| 1131 | * CONTEXT: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1132 | * spin_lock_irq(pool->lock). |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1133 | */ |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1134 | 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] | 1135 | { |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1136 | /* uncolored work items don't participate in flushing or nr_active */ |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1137 | if (color == WORK_NO_COLOR) |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1138 | goto out_put; |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1139 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1140 | pwq->nr_in_flight[color]--; |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1141 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1142 | pwq->nr_active--; |
| 1143 | if (!list_empty(&pwq->delayed_works)) { |
Lai Jiangshan | b3f9f40 | 2012-09-18 10:40:00 -0700 | [diff] [blame] | 1144 | /* one down, submit a delayed one */ |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1145 | if (pwq->nr_active < pwq->max_active) |
| 1146 | pwq_activate_first_delayed(pwq); |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1147 | } |
| 1148 | |
| 1149 | /* is flush in progress and are we at the flushing tip? */ |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1150 | if (likely(pwq->flush_color != color)) |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1151 | goto out_put; |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1152 | |
| 1153 | /* are there still in-flight works? */ |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1154 | if (pwq->nr_in_flight[color]) |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1155 | goto out_put; |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1156 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1157 | /* this pwq is done, clear flush_color */ |
| 1158 | pwq->flush_color = -1; |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1159 | |
| 1160 | /* |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1161 | * If this was the last pwq, wake up the first flusher. It |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1162 | * will handle the rest. |
| 1163 | */ |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1164 | if (atomic_dec_and_test(&pwq->wq->nr_pwqs_to_flush)) |
| 1165 | complete(&pwq->wq->first_flusher->done); |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1166 | out_put: |
| 1167 | put_pwq(pwq); |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1168 | } |
| 1169 | |
Tejun Heo | 36e227d | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1170 | /** |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1171 | * try_to_grab_pending - steal work item from worklist and disable irq |
Tejun Heo | 36e227d | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1172 | * @work: work item to steal |
| 1173 | * @is_dwork: @work is a delayed_work |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1174 | * @flags: place to store irq state |
Tejun Heo | 36e227d | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1175 | * |
| 1176 | * 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] | 1177 | * stable state - idle, on timer or on worklist. |
Tejun Heo | 36e227d | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1178 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 1179 | * Return: |
Tejun Heo | 36e227d | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1180 | * 1 if @work was pending and we successfully stole PENDING |
| 1181 | * 0 if @work was idle and we claimed PENDING |
| 1182 | * -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] | 1183 | * -ENOENT if someone else is canceling @work, this state may persist |
| 1184 | * for arbitrarily long |
Tejun Heo | 36e227d | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1185 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 1186 | * Note: |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1187 | * 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] | 1188 | * interrupted while holding PENDING and @work off queue, irq must be |
| 1189 | * disabled on entry. This, combined with delayed_work->timer being |
| 1190 | * irqsafe, ensures that we return -EAGAIN for finite short period of time. |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1191 | * |
| 1192 | * On successful return, >= 0, irq is disabled and the caller is |
| 1193 | * responsible for releasing it using local_irq_restore(*@flags). |
| 1194 | * |
Tejun Heo | e0aecdd | 2012-08-21 13:18:24 -0700 | [diff] [blame] | 1195 | * This function is safe to call from any context including IRQ handler. |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1196 | */ |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1197 | static int try_to_grab_pending(struct work_struct *work, bool is_dwork, |
| 1198 | unsigned long *flags) |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1199 | { |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1200 | struct worker_pool *pool; |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1201 | struct pool_workqueue *pwq; |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1202 | |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1203 | local_irq_save(*flags); |
| 1204 | |
Tejun Heo | 36e227d | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1205 | /* try to steal the timer if it exists */ |
| 1206 | if (is_dwork) { |
| 1207 | struct delayed_work *dwork = to_delayed_work(work); |
| 1208 | |
Tejun Heo | e0aecdd | 2012-08-21 13:18:24 -0700 | [diff] [blame] | 1209 | /* |
| 1210 | * dwork->timer is irqsafe. If del_timer() fails, it's |
| 1211 | * guaranteed that the timer is not queued anywhere and not |
| 1212 | * running on the local CPU. |
| 1213 | */ |
Tejun Heo | 36e227d | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1214 | if (likely(del_timer(&dwork->timer))) |
| 1215 | return 1; |
| 1216 | } |
| 1217 | |
| 1218 | /* try to claim PENDING the normal way */ |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1219 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) |
| 1220 | return 0; |
| 1221 | |
| 1222 | /* |
| 1223 | * The queueing is in progress, or it is already queued. Try to |
| 1224 | * steal it from ->worklist without clearing WORK_STRUCT_PENDING. |
| 1225 | */ |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1226 | pool = get_work_pool(work); |
| 1227 | if (!pool) |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1228 | goto fail; |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1229 | |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1230 | spin_lock(&pool->lock); |
Lai Jiangshan | 0b3dae6 | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 1231 | /* |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1232 | * work->data is guaranteed to point to pwq only while the work |
| 1233 | * item is queued on pwq->wq, and both updating work->data to point |
| 1234 | * to pwq on queueing and to pool on dequeueing are done under |
| 1235 | * pwq->pool->lock. This in turn guarantees that, if work->data |
| 1236 | * points to pwq which is associated with a locked pool, the work |
Lai Jiangshan | 0b3dae6 | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 1237 | * item is currently queued on that pool. |
| 1238 | */ |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1239 | pwq = get_work_pwq(work); |
| 1240 | if (pwq && pwq->pool == pool) { |
Tejun Heo | 1606283 | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 1241 | debug_work_deactivate(work); |
Lai Jiangshan | 3aa6249 | 2012-09-18 10:40:00 -0700 | [diff] [blame] | 1242 | |
Tejun Heo | 1606283 | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 1243 | /* |
| 1244 | * A delayed work item cannot be grabbed directly because |
| 1245 | * it might have linked NO_COLOR work items which, if left |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1246 | * on the delayed_list, will confuse pwq->nr_active |
Tejun Heo | 1606283 | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 1247 | * management later on and cause stall. Make sure the work |
| 1248 | * item is activated before grabbing. |
| 1249 | */ |
| 1250 | if (*work_data_bits(work) & WORK_STRUCT_DELAYED) |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1251 | pwq_activate_delayed_work(work); |
Lai Jiangshan | 3aa6249 | 2012-09-18 10:40:00 -0700 | [diff] [blame] | 1252 | |
Tejun Heo | 1606283 | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 1253 | list_del_init(&work->entry); |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1254 | pwq_dec_nr_in_flight(get_work_pwq(work), get_work_color(work)); |
Tejun Heo | 36e227d | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1255 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1256 | /* work->data points to pwq iff queued, point to pool */ |
Tejun Heo | 1606283 | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 1257 | set_work_pool_and_keep_pending(work, pool->id); |
Lai Jiangshan | 4468a00 | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 1258 | |
Tejun Heo | 1606283 | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 1259 | spin_unlock(&pool->lock); |
| 1260 | return 1; |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1261 | } |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1262 | spin_unlock(&pool->lock); |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1263 | fail: |
| 1264 | local_irq_restore(*flags); |
| 1265 | if (work_is_canceling(work)) |
| 1266 | return -ENOENT; |
| 1267 | cpu_relax(); |
Tejun Heo | 36e227d | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1268 | return -EAGAIN; |
Tejun Heo | bf4ede0 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1269 | } |
| 1270 | |
| 1271 | /** |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 1272 | * insert_work - insert a work into a pool |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1273 | * @pwq: pwq @work belongs to |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 1274 | * @work: work to insert |
| 1275 | * @head: insertion point |
| 1276 | * @extra_flags: extra WORK_STRUCT_* flags to set |
| 1277 | * |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1278 | * 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] | 1279 | * work_struct flags. |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 1280 | * |
| 1281 | * CONTEXT: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1282 | * spin_lock_irq(pool->lock). |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 1283 | */ |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1284 | static void insert_work(struct pool_workqueue *pwq, struct work_struct *work, |
| 1285 | struct list_head *head, unsigned int extra_flags) |
Oleg Nesterov | b89deed | 2007-05-09 02:33:52 -0700 | [diff] [blame] | 1286 | { |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1287 | struct worker_pool *pool = pwq->pool; |
Frederic Weisbecker | e1d8aa9 | 2009-01-12 23:15:46 +0100 | [diff] [blame] | 1288 | |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 1289 | /* we own @work, set data and link */ |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1290 | set_work_pwq(work, pwq, extra_flags); |
Oleg Nesterov | 1a4d9b0 | 2008-07-25 01:47:47 -0700 | [diff] [blame] | 1291 | list_add_tail(&work->entry, head); |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1292 | get_pwq(pwq); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1293 | |
| 1294 | /* |
Tejun Heo | c5aa87b | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 1295 | * Ensure either wq_worker_sleeping() sees the above |
| 1296 | * list_add_tail() or we see zero nr_running to avoid workers lying |
| 1297 | * around lazily while there are works to be processed. |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1298 | */ |
| 1299 | smp_mb(); |
| 1300 | |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1301 | if (__need_more_worker(pool)) |
| 1302 | wake_up_worker(pool); |
Oleg Nesterov | b89deed | 2007-05-09 02:33:52 -0700 | [diff] [blame] | 1303 | } |
| 1304 | |
Tejun Heo | c8efcc2 | 2010-12-20 19:32:04 +0100 | [diff] [blame] | 1305 | /* |
| 1306 | * Test whether @work is being queued from another work executing on the |
Tejun Heo | 8d03ecf | 2013-02-13 19:29:10 -0800 | [diff] [blame] | 1307 | * same workqueue. |
Tejun Heo | c8efcc2 | 2010-12-20 19:32:04 +0100 | [diff] [blame] | 1308 | */ |
| 1309 | static bool is_chained_work(struct workqueue_struct *wq) |
| 1310 | { |
Tejun Heo | 8d03ecf | 2013-02-13 19:29:10 -0800 | [diff] [blame] | 1311 | struct worker *worker; |
Tejun Heo | c8efcc2 | 2010-12-20 19:32:04 +0100 | [diff] [blame] | 1312 | |
Tejun Heo | 8d03ecf | 2013-02-13 19:29:10 -0800 | [diff] [blame] | 1313 | worker = current_wq_worker(); |
| 1314 | /* |
| 1315 | * Return %true iff I'm a worker execuing a work item on @wq. If |
| 1316 | * I'm @worker, it's safe to dereference it without locking. |
| 1317 | */ |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1318 | return worker && worker->current_pwq->wq == wq; |
Tejun Heo | c8efcc2 | 2010-12-20 19:32:04 +0100 | [diff] [blame] | 1319 | } |
| 1320 | |
Tejun Heo | d84ff05 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 1321 | static void __queue_work(int cpu, struct workqueue_struct *wq, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1322 | struct work_struct *work) |
| 1323 | { |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1324 | struct pool_workqueue *pwq; |
Tejun Heo | c917808 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1325 | struct worker_pool *last_pool; |
Tejun Heo | 1e19ffc | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1326 | struct list_head *worklist; |
Tejun Heo | 8a2e8e5d | 2010-08-25 10:33:56 +0200 | [diff] [blame] | 1327 | unsigned int work_flags; |
Joonsoo Kim | b75cac9 | 2012-08-15 23:25:37 +0900 | [diff] [blame] | 1328 | unsigned int req_cpu = cpu; |
Tejun Heo | 8930cab | 2012-08-03 10:30:45 -0700 | [diff] [blame] | 1329 | |
| 1330 | /* |
| 1331 | * While a work item is PENDING && off queue, a task trying to |
| 1332 | * steal the PENDING will busy-loop waiting for it to either get |
| 1333 | * queued or lose PENDING. Grabbing PENDING and queueing should |
| 1334 | * happen with IRQ disabled. |
| 1335 | */ |
| 1336 | WARN_ON_ONCE(!irqs_disabled()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1337 | |
Thomas Gleixner | dc186ad | 2009-11-16 01:09:48 +0900 | [diff] [blame] | 1338 | debug_work_activate(work); |
Tejun Heo | 1e19ffc | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1339 | |
Li Bin | 9ef28a7 | 2013-09-09 13:13:58 +0800 | [diff] [blame] | 1340 | /* if draining, only works from the same workqueue are allowed */ |
Tejun Heo | 618b01e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1341 | if (unlikely(wq->flags & __WQ_DRAINING) && |
Tejun Heo | c8efcc2 | 2010-12-20 19:32:04 +0100 | [diff] [blame] | 1342 | WARN_ON_ONCE(!is_chained_work(wq))) |
Tejun Heo | e41e704 | 2010-08-24 14:22:47 +0200 | [diff] [blame] | 1343 | return; |
Tejun Heo | 9e8cd2f | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1344 | retry: |
Tejun Heo | df2d5ae | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 1345 | if (req_cpu == WORK_CPU_UNBOUND) |
| 1346 | cpu = raw_smp_processor_id(); |
| 1347 | |
Tejun Heo | c917808 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1348 | /* pwq which will be used unless @work is executing elsewhere */ |
Tejun Heo | df2d5ae | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 1349 | if (!(wq->flags & WQ_UNBOUND)) |
Tejun Heo | 7fb98ea | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 1350 | pwq = per_cpu_ptr(wq->cpu_pwqs, cpu); |
Tejun Heo | df2d5ae | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 1351 | else |
| 1352 | pwq = unbound_pwq_by_node(wq, cpu_to_node(cpu)); |
Tejun Heo | dbf2576 | 2012-08-20 14:51:23 -0700 | [diff] [blame] | 1353 | |
Tejun Heo | c917808 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1354 | /* |
| 1355 | * If @work was previously on a different pool, it might still be |
| 1356 | * running there, in which case the work needs to be queued on that |
| 1357 | * pool to guarantee non-reentrancy. |
| 1358 | */ |
| 1359 | last_pool = get_work_pool(work); |
| 1360 | if (last_pool && last_pool != pwq->pool) { |
| 1361 | struct worker *worker; |
Tejun Heo | 18aa9ef | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 1362 | |
Tejun Heo | c917808 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1363 | spin_lock(&last_pool->lock); |
Tejun Heo | 18aa9ef | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 1364 | |
Tejun Heo | c917808 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1365 | worker = find_worker_executing_work(last_pool, work); |
Tejun Heo | 18aa9ef | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 1366 | |
Tejun Heo | c917808 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1367 | if (worker && worker->current_pwq->wq == wq) { |
| 1368 | pwq = worker->current_pwq; |
Tejun Heo | 8930cab | 2012-08-03 10:30:45 -0700 | [diff] [blame] | 1369 | } else { |
Tejun Heo | c917808 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1370 | /* meh... not running there, queue here */ |
| 1371 | spin_unlock(&last_pool->lock); |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1372 | spin_lock(&pwq->pool->lock); |
Tejun Heo | 8930cab | 2012-08-03 10:30:45 -0700 | [diff] [blame] | 1373 | } |
Tejun Heo | f342179 | 2010-07-02 10:03:51 +0200 | [diff] [blame] | 1374 | } else { |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1375 | spin_lock(&pwq->pool->lock); |
Tejun Heo | 502ca9d | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 1376 | } |
| 1377 | |
Tejun Heo | 9e8cd2f | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1378 | /* |
| 1379 | * pwq is determined and locked. For unbound pools, we could have |
| 1380 | * raced with pwq release and it could already be dead. If its |
| 1381 | * refcnt is zero, repeat pwq selection. Note that pwqs never die |
Tejun Heo | df2d5ae | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 1382 | * without another pwq replacing it in the numa_pwq_tbl or while |
| 1383 | * work items are executing on it, so the retrying is guaranteed to |
Tejun Heo | 9e8cd2f | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 1384 | * make forward-progress. |
| 1385 | */ |
| 1386 | if (unlikely(!pwq->refcnt)) { |
| 1387 | if (wq->flags & WQ_UNBOUND) { |
| 1388 | spin_unlock(&pwq->pool->lock); |
| 1389 | cpu_relax(); |
| 1390 | goto retry; |
| 1391 | } |
| 1392 | /* oops */ |
| 1393 | WARN_ONCE(true, "workqueue: per-cpu pwq for %s on cpu%d has 0 refcnt", |
| 1394 | wq->name, cpu); |
| 1395 | } |
| 1396 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1397 | /* pwq determined, queue */ |
| 1398 | trace_workqueue_queue_work(req_cpu, pwq, work); |
Tejun Heo | 502ca9d | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 1399 | |
Dan Carpenter | f5b2552 | 2012-04-13 22:06:58 +0300 | [diff] [blame] | 1400 | if (WARN_ON(!list_empty(&work->entry))) { |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1401 | spin_unlock(&pwq->pool->lock); |
Dan Carpenter | f5b2552 | 2012-04-13 22:06:58 +0300 | [diff] [blame] | 1402 | return; |
| 1403 | } |
Tejun Heo | 1e19ffc | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1404 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1405 | pwq->nr_in_flight[pwq->work_color]++; |
| 1406 | work_flags = work_color_to_flags(pwq->work_color); |
Tejun Heo | 1e19ffc | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1407 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1408 | if (likely(pwq->nr_active < pwq->max_active)) { |
Tejun Heo | cdadf00 | 2010-10-05 10:49:55 +0200 | [diff] [blame] | 1409 | trace_workqueue_activate_work(work); |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1410 | pwq->nr_active++; |
| 1411 | worklist = &pwq->pool->worklist; |
Tejun Heo | 8a2e8e5d | 2010-08-25 10:33:56 +0200 | [diff] [blame] | 1412 | } else { |
| 1413 | work_flags |= WORK_STRUCT_DELAYED; |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1414 | worklist = &pwq->delayed_works; |
Tejun Heo | 8a2e8e5d | 2010-08-25 10:33:56 +0200 | [diff] [blame] | 1415 | } |
Tejun Heo | 1e19ffc | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1416 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1417 | insert_work(pwq, work, worklist, work_flags); |
Tejun Heo | 1e19ffc | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1418 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1419 | spin_unlock(&pwq->pool->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1420 | } |
| 1421 | |
Rolf Eike Beer | 0fcb78c | 2006-07-30 03:03:42 -0700 | [diff] [blame] | 1422 | /** |
Zhang Rui | c1a220e | 2008-07-23 21:28:39 -0700 | [diff] [blame] | 1423 | * queue_work_on - queue work on specific cpu |
| 1424 | * @cpu: CPU number to execute work on |
| 1425 | * @wq: workqueue to use |
| 1426 | * @work: work to queue |
| 1427 | * |
Zhang Rui | c1a220e | 2008-07-23 21:28:39 -0700 | [diff] [blame] | 1428 | * We queue the work to a specific CPU, the caller must ensure it |
| 1429 | * can't go away. |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 1430 | * |
| 1431 | * Return: %false if @work was already on a queue, %true otherwise. |
Zhang Rui | c1a220e | 2008-07-23 21:28:39 -0700 | [diff] [blame] | 1432 | */ |
Tejun Heo | d4283e9 | 2012-08-03 10:30:44 -0700 | [diff] [blame] | 1433 | bool queue_work_on(int cpu, struct workqueue_struct *wq, |
| 1434 | struct work_struct *work) |
Zhang Rui | c1a220e | 2008-07-23 21:28:39 -0700 | [diff] [blame] | 1435 | { |
Tejun Heo | d4283e9 | 2012-08-03 10:30:44 -0700 | [diff] [blame] | 1436 | bool ret = false; |
Tejun Heo | 8930cab | 2012-08-03 10:30:45 -0700 | [diff] [blame] | 1437 | unsigned long flags; |
| 1438 | |
| 1439 | local_irq_save(flags); |
Zhang Rui | c1a220e | 2008-07-23 21:28:39 -0700 | [diff] [blame] | 1440 | |
Tejun Heo | 22df02b | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 1441 | 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] | 1442 | __queue_work(cpu, wq, work); |
Tejun Heo | d4283e9 | 2012-08-03 10:30:44 -0700 | [diff] [blame] | 1443 | ret = true; |
Zhang Rui | c1a220e | 2008-07-23 21:28:39 -0700 | [diff] [blame] | 1444 | } |
Tejun Heo | 8930cab | 2012-08-03 10:30:45 -0700 | [diff] [blame] | 1445 | |
| 1446 | local_irq_restore(flags); |
Zhang Rui | c1a220e | 2008-07-23 21:28:39 -0700 | [diff] [blame] | 1447 | return ret; |
| 1448 | } |
Marc Dionne | ad7b1f8 | 2013-05-06 17:44:55 -0400 | [diff] [blame] | 1449 | EXPORT_SYMBOL(queue_work_on); |
Zhang Rui | c1a220e | 2008-07-23 21:28:39 -0700 | [diff] [blame] | 1450 | |
Tejun Heo | d8e794d | 2012-08-03 10:30:45 -0700 | [diff] [blame] | 1451 | void delayed_work_timer_fn(unsigned long __data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1452 | { |
David Howells | 52bad64 | 2006-11-22 14:54:01 +0000 | [diff] [blame] | 1453 | struct delayed_work *dwork = (struct delayed_work *)__data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1454 | |
Tejun Heo | e0aecdd | 2012-08-21 13:18:24 -0700 | [diff] [blame] | 1455 | /* should have been called from irqsafe timer with irq already off */ |
Lai Jiangshan | 60c057b | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 1456 | __queue_work(dwork->cpu, dwork->wq, &dwork->work); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1457 | } |
Konstantin Khlebnikov | 1438ade5 | 2013-01-24 16:36:31 +0400 | [diff] [blame] | 1458 | EXPORT_SYMBOL(delayed_work_timer_fn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1459 | |
Tejun Heo | 7beb2ed | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1460 | static void __queue_delayed_work(int cpu, struct workqueue_struct *wq, |
| 1461 | struct delayed_work *dwork, unsigned long delay) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1462 | { |
Tejun Heo | 7beb2ed | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1463 | struct timer_list *timer = &dwork->timer; |
| 1464 | struct work_struct *work = &dwork->work; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1465 | |
Tejun Heo | 7beb2ed | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1466 | WARN_ON_ONCE(timer->function != delayed_work_timer_fn || |
| 1467 | timer->data != (unsigned long)dwork); |
Tejun Heo | fc4b514 | 2012-12-04 07:40:39 -0800 | [diff] [blame] | 1468 | WARN_ON_ONCE(timer_pending(timer)); |
| 1469 | WARN_ON_ONCE(!list_empty(&work->entry)); |
Tejun Heo | 7beb2ed | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1470 | |
Tejun Heo | 8852aac | 2012-12-01 16:23:42 -0800 | [diff] [blame] | 1471 | /* |
| 1472 | * If @delay is 0, queue @dwork->work immediately. This is for |
| 1473 | * both optimization and correctness. The earliest @timer can |
| 1474 | * expire is on the closest next tick and delayed_work users depend |
| 1475 | * on that there's no such delay when @delay is 0. |
| 1476 | */ |
| 1477 | if (!delay) { |
| 1478 | __queue_work(cpu, wq, &dwork->work); |
| 1479 | return; |
| 1480 | } |
| 1481 | |
Tejun Heo | 7beb2ed | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1482 | timer_stats_timer_set_start_info(&dwork->timer); |
| 1483 | |
Lai Jiangshan | 60c057b | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 1484 | dwork->wq = wq; |
Tejun Heo | 1265057 | 2012-08-08 09:38:42 -0700 | [diff] [blame] | 1485 | dwork->cpu = cpu; |
Tejun Heo | 7beb2ed | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1486 | timer->expires = jiffies + delay; |
| 1487 | |
| 1488 | if (unlikely(cpu != WORK_CPU_UNBOUND)) |
| 1489 | add_timer_on(timer, cpu); |
| 1490 | else |
| 1491 | add_timer(timer); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1492 | } |
| 1493 | |
Rolf Eike Beer | 0fcb78c | 2006-07-30 03:03:42 -0700 | [diff] [blame] | 1494 | /** |
| 1495 | * queue_delayed_work_on - queue work on specific CPU after delay |
| 1496 | * @cpu: CPU number to execute work on |
| 1497 | * @wq: workqueue to use |
Randy Dunlap | af9997e | 2006-12-22 01:06:52 -0800 | [diff] [blame] | 1498 | * @dwork: work to queue |
Rolf Eike Beer | 0fcb78c | 2006-07-30 03:03:42 -0700 | [diff] [blame] | 1499 | * @delay: number of jiffies to wait before queueing |
| 1500 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 1501 | * Return: %false if @work was already on a queue, %true otherwise. If |
Tejun Heo | 715f130 | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 1502 | * @delay is zero and @dwork is idle, it will be scheduled for immediate |
| 1503 | * execution. |
Rolf Eike Beer | 0fcb78c | 2006-07-30 03:03:42 -0700 | [diff] [blame] | 1504 | */ |
Tejun Heo | d4283e9 | 2012-08-03 10:30:44 -0700 | [diff] [blame] | 1505 | bool queue_delayed_work_on(int cpu, struct workqueue_struct *wq, |
| 1506 | struct delayed_work *dwork, unsigned long delay) |
Venkatesh Pallipadi | 7a6bc1c | 2006-06-28 13:50:33 -0700 | [diff] [blame] | 1507 | { |
David Howells | 52bad64 | 2006-11-22 14:54:01 +0000 | [diff] [blame] | 1508 | struct work_struct *work = &dwork->work; |
Tejun Heo | d4283e9 | 2012-08-03 10:30:44 -0700 | [diff] [blame] | 1509 | bool ret = false; |
Tejun Heo | 8930cab | 2012-08-03 10:30:45 -0700 | [diff] [blame] | 1510 | unsigned long flags; |
| 1511 | |
| 1512 | /* read the comment in __queue_work() */ |
| 1513 | local_irq_save(flags); |
Venkatesh Pallipadi | 7a6bc1c | 2006-06-28 13:50:33 -0700 | [diff] [blame] | 1514 | |
Tejun Heo | 22df02b | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 1515 | 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] | 1516 | __queue_delayed_work(cpu, wq, dwork, delay); |
Tejun Heo | d4283e9 | 2012-08-03 10:30:44 -0700 | [diff] [blame] | 1517 | ret = true; |
Venkatesh Pallipadi | 7a6bc1c | 2006-06-28 13:50:33 -0700 | [diff] [blame] | 1518 | } |
Tejun Heo | 8930cab | 2012-08-03 10:30:45 -0700 | [diff] [blame] | 1519 | |
| 1520 | local_irq_restore(flags); |
Venkatesh Pallipadi | 7a6bc1c | 2006-06-28 13:50:33 -0700 | [diff] [blame] | 1521 | return ret; |
| 1522 | } |
Marc Dionne | ad7b1f8 | 2013-05-06 17:44:55 -0400 | [diff] [blame] | 1523 | EXPORT_SYMBOL(queue_delayed_work_on); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1524 | |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1525 | /** |
Tejun Heo | 8376fe2 | 2012-08-03 10:30:47 -0700 | [diff] [blame] | 1526 | * mod_delayed_work_on - modify delay of or queue a delayed work on specific CPU |
| 1527 | * @cpu: CPU number to execute work on |
| 1528 | * @wq: workqueue to use |
| 1529 | * @dwork: work to queue |
| 1530 | * @delay: number of jiffies to wait before queueing |
| 1531 | * |
| 1532 | * If @dwork is idle, equivalent to queue_delayed_work_on(); otherwise, |
| 1533 | * modify @dwork's timer so that it expires after @delay. If @delay is |
| 1534 | * zero, @work is guaranteed to be scheduled immediately regardless of its |
| 1535 | * current state. |
| 1536 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 1537 | * Return: %false if @dwork was idle and queued, %true if @dwork was |
Tejun Heo | 8376fe2 | 2012-08-03 10:30:47 -0700 | [diff] [blame] | 1538 | * pending and its timer was modified. |
| 1539 | * |
Tejun Heo | e0aecdd | 2012-08-21 13:18:24 -0700 | [diff] [blame] | 1540 | * This function is safe to call from any context including IRQ handler. |
Tejun Heo | 8376fe2 | 2012-08-03 10:30:47 -0700 | [diff] [blame] | 1541 | * See try_to_grab_pending() for details. |
| 1542 | */ |
| 1543 | bool mod_delayed_work_on(int cpu, struct workqueue_struct *wq, |
| 1544 | struct delayed_work *dwork, unsigned long delay) |
| 1545 | { |
| 1546 | unsigned long flags; |
| 1547 | int ret; |
| 1548 | |
| 1549 | do { |
| 1550 | ret = try_to_grab_pending(&dwork->work, true, &flags); |
| 1551 | } while (unlikely(ret == -EAGAIN)); |
| 1552 | |
| 1553 | if (likely(ret >= 0)) { |
| 1554 | __queue_delayed_work(cpu, wq, dwork, delay); |
| 1555 | local_irq_restore(flags); |
| 1556 | } |
| 1557 | |
| 1558 | /* -ENOENT from try_to_grab_pending() becomes %true */ |
| 1559 | return ret; |
| 1560 | } |
| 1561 | EXPORT_SYMBOL_GPL(mod_delayed_work_on); |
| 1562 | |
| 1563 | /** |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1564 | * worker_enter_idle - enter idle state |
| 1565 | * @worker: worker which is entering idle state |
| 1566 | * |
| 1567 | * @worker is entering idle state. Update stats and idle timer if |
| 1568 | * necessary. |
| 1569 | * |
| 1570 | * LOCKING: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1571 | * spin_lock_irq(pool->lock). |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1572 | */ |
| 1573 | static void worker_enter_idle(struct worker *worker) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1574 | { |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1575 | struct worker_pool *pool = worker->pool; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1576 | |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 1577 | if (WARN_ON_ONCE(worker->flags & WORKER_IDLE) || |
| 1578 | WARN_ON_ONCE(!list_empty(&worker->entry) && |
| 1579 | (worker->hentry.next || worker->hentry.pprev))) |
| 1580 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1581 | |
Tejun Heo | cb44476 | 2010-07-02 10:03:50 +0200 | [diff] [blame] | 1582 | /* can't use worker_set_flags(), also called from start_worker() */ |
| 1583 | worker->flags |= WORKER_IDLE; |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1584 | pool->nr_idle++; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1585 | worker->last_active = jiffies; |
Peter Zijlstra | d5abe66 | 2006-12-06 20:37:26 -0800 | [diff] [blame] | 1586 | |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1587 | /* idle_list is LIFO */ |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1588 | list_add(&worker->entry, &pool->idle_list); |
Tejun Heo | db7bccf | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1589 | |
Tejun Heo | 628c78e | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 1590 | if (too_many_workers(pool) && !timer_pending(&pool->idle_timer)) |
| 1591 | mod_timer(&pool->idle_timer, jiffies + IDLE_WORKER_TIMEOUT); |
Tejun Heo | cb44476 | 2010-07-02 10:03:50 +0200 | [diff] [blame] | 1592 | |
Tejun Heo | 544ecf3 | 2012-05-14 15:04:50 -0700 | [diff] [blame] | 1593 | /* |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 1594 | * Sanity check nr_running. Because wq_unbind_fn() releases |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1595 | * pool->lock between setting %WORKER_UNBOUND and zapping |
Tejun Heo | 628c78e | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 1596 | * nr_running, the warning may trigger spuriously. Check iff |
| 1597 | * unbind is not in progress. |
Tejun Heo | 544ecf3 | 2012-05-14 15:04:50 -0700 | [diff] [blame] | 1598 | */ |
Tejun Heo | 2464757 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1599 | WARN_ON_ONCE(!(pool->flags & POOL_DISASSOCIATED) && |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1600 | pool->nr_workers == pool->nr_idle && |
Tejun Heo | e19e397 | 2013-01-24 11:39:44 -0800 | [diff] [blame] | 1601 | atomic_read(&pool->nr_running)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1602 | } |
| 1603 | |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1604 | /** |
| 1605 | * worker_leave_idle - leave idle state |
| 1606 | * @worker: worker which is leaving idle state |
| 1607 | * |
| 1608 | * @worker is leaving idle state. Update stats. |
| 1609 | * |
| 1610 | * LOCKING: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1611 | * spin_lock_irq(pool->lock). |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1612 | */ |
| 1613 | static void worker_leave_idle(struct worker *worker) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1614 | { |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1615 | struct worker_pool *pool = worker->pool; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1616 | |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 1617 | if (WARN_ON_ONCE(!(worker->flags & WORKER_IDLE))) |
| 1618 | return; |
Tejun Heo | d302f01 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 1619 | worker_clr_flags(worker, WORKER_IDLE); |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1620 | pool->nr_idle--; |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1621 | list_del_init(&worker->entry); |
| 1622 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1623 | |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1624 | /** |
Lai Jiangshan | f36dc67 | 2013-02-19 12:17:02 -0800 | [diff] [blame] | 1625 | * worker_maybe_bind_and_lock - try to bind %current to worker_pool and lock it |
| 1626 | * @pool: target worker_pool |
| 1627 | * |
| 1628 | * Bind %current to the cpu of @pool if it is associated and lock @pool. |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1629 | * |
| 1630 | * Works which are scheduled while the cpu is online must at least be |
| 1631 | * scheduled to a worker which is bound to the cpu so that if they are |
| 1632 | * flushed from cpu callbacks while cpu is going down, they are |
| 1633 | * guaranteed to execute on the cpu. |
| 1634 | * |
Lai Jiangshan | f5faa07 | 2013-02-19 12:17:02 -0800 | [diff] [blame] | 1635 | * This function is to be used by unbound workers and rescuers to bind |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1636 | * themselves to the target cpu and may race with cpu going down or |
| 1637 | * coming online. kthread_bind() can't be used because it may put the |
| 1638 | * worker to already dead cpu and set_cpus_allowed_ptr() can't be used |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 1639 | * verbatim as it's best effort and blocking and pool may be |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1640 | * [dis]associated in the meantime. |
| 1641 | * |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 1642 | * This function tries set_cpus_allowed() and locks pool and verifies the |
Tejun Heo | 2464757 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1643 | * binding against %POOL_DISASSOCIATED which is set during |
Tejun Heo | f2d5a0e | 2012-07-17 12:39:26 -0700 | [diff] [blame] | 1644 | * %CPU_DOWN_PREPARE and cleared during %CPU_ONLINE, so if the worker |
| 1645 | * enters idle state or fetches works without dropping lock, it can |
| 1646 | * guarantee the scheduling requirement described in the first paragraph. |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1647 | * |
| 1648 | * CONTEXT: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1649 | * Might sleep. Called without any lock but returns with pool->lock |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1650 | * held. |
| 1651 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 1652 | * Return: |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 1653 | * %true if the associated pool is online (@worker is successfully |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1654 | * bound), %false if offline. |
| 1655 | */ |
Lai Jiangshan | f36dc67 | 2013-02-19 12:17:02 -0800 | [diff] [blame] | 1656 | static bool worker_maybe_bind_and_lock(struct worker_pool *pool) |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1657 | __acquires(&pool->lock) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1658 | { |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1659 | while (true) { |
| 1660 | /* |
| 1661 | * The following call may fail, succeed or succeed |
| 1662 | * without actually migrating the task to the cpu if |
| 1663 | * it races with cpu hotunplug operation. Verify |
Tejun Heo | 2464757 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1664 | * against POOL_DISASSOCIATED. |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1665 | */ |
Tejun Heo | 2464757 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1666 | if (!(pool->flags & POOL_DISASSOCIATED)) |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 1667 | set_cpus_allowed_ptr(current, pool->attrs->cpumask); |
Oleg Nesterov | 85f4186 | 2007-05-09 02:34:20 -0700 | [diff] [blame] | 1668 | |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1669 | spin_lock_irq(&pool->lock); |
Tejun Heo | 2464757 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1670 | if (pool->flags & POOL_DISASSOCIATED) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1671 | return false; |
Lai Jiangshan | f5faa07 | 2013-02-19 12:17:02 -0800 | [diff] [blame] | 1672 | if (task_cpu(current) == pool->cpu && |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 1673 | cpumask_equal(¤t->cpus_allowed, pool->attrs->cpumask)) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1674 | return true; |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1675 | spin_unlock_irq(&pool->lock); |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 1676 | |
Tejun Heo | 5035b20 | 2011-04-29 18:08:37 +0200 | [diff] [blame] | 1677 | /* |
| 1678 | * We've raced with CPU hot[un]plug. Give it a breather |
| 1679 | * and retry migration. cond_resched() is required here; |
| 1680 | * otherwise, we might deadlock against cpu_stop trying to |
| 1681 | * bring down the CPU on non-preemptive kernel. |
| 1682 | */ |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1683 | cpu_relax(); |
Tejun Heo | 5035b20 | 2011-04-29 18:08:37 +0200 | [diff] [blame] | 1684 | cond_resched(); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1685 | } |
| 1686 | } |
| 1687 | |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1688 | static struct worker *alloc_worker(void) |
| 1689 | { |
| 1690 | struct worker *worker; |
| 1691 | |
| 1692 | worker = kzalloc(sizeof(*worker), GFP_KERNEL); |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1693 | if (worker) { |
| 1694 | INIT_LIST_HEAD(&worker->entry); |
Tejun Heo | affee4b | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1695 | INIT_LIST_HEAD(&worker->scheduled); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1696 | /* on creation a worker is in !idle && prep state */ |
| 1697 | worker->flags = WORKER_PREP; |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1698 | } |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1699 | return worker; |
| 1700 | } |
| 1701 | |
| 1702 | /** |
| 1703 | * create_worker - create a new workqueue worker |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1704 | * @pool: pool the new worker will belong to |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1705 | * |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1706 | * Create a new worker which is bound to @pool. The returned worker |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1707 | * can be started by calling start_worker() or destroyed using |
| 1708 | * destroy_worker(). |
| 1709 | * |
| 1710 | * CONTEXT: |
| 1711 | * Might sleep. Does GFP_KERNEL allocations. |
| 1712 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 1713 | * Return: |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1714 | * Pointer to the newly created worker. |
| 1715 | */ |
Tejun Heo | bc2ae0f | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 1716 | static struct worker *create_worker(struct worker_pool *pool) |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1717 | { |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1718 | struct worker *worker = NULL; |
Tejun Heo | f342179 | 2010-07-02 10:03:51 +0200 | [diff] [blame] | 1719 | int id = -1; |
Tejun Heo | e3c916a | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 1720 | char id_buf[16]; |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1721 | |
Tejun Heo | cd54968 | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 1722 | lockdep_assert_held(&pool->manager_mutex); |
| 1723 | |
Tejun Heo | 822d840 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 1724 | /* |
| 1725 | * ID is needed to determine kthread name. Allocate ID first |
| 1726 | * without installing the pointer. |
| 1727 | */ |
| 1728 | idr_preload(GFP_KERNEL); |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1729 | spin_lock_irq(&pool->lock); |
Tejun Heo | 822d840 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 1730 | |
| 1731 | id = idr_alloc(&pool->worker_idr, NULL, 0, 0, GFP_NOWAIT); |
| 1732 | |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1733 | spin_unlock_irq(&pool->lock); |
Tejun Heo | 822d840 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 1734 | idr_preload_end(); |
| 1735 | if (id < 0) |
| 1736 | goto fail; |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1737 | |
| 1738 | worker = alloc_worker(); |
| 1739 | if (!worker) |
| 1740 | goto fail; |
| 1741 | |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1742 | worker->pool = pool; |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1743 | worker->id = id; |
| 1744 | |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 1745 | if (pool->cpu >= 0) |
Tejun Heo | e3c916a | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 1746 | snprintf(id_buf, sizeof(id_buf), "%d:%d%s", pool->cpu, id, |
| 1747 | pool->attrs->nice < 0 ? "H" : ""); |
Tejun Heo | f342179 | 2010-07-02 10:03:51 +0200 | [diff] [blame] | 1748 | else |
Tejun Heo | e3c916a | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 1749 | snprintf(id_buf, sizeof(id_buf), "u%d:%d", pool->id, id); |
| 1750 | |
Tejun Heo | f3f90ad | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 1751 | worker->task = kthread_create_on_node(worker_thread, worker, pool->node, |
Tejun Heo | e3c916a | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 1752 | "kworker/%s", id_buf); |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1753 | if (IS_ERR(worker->task)) |
| 1754 | goto fail; |
| 1755 | |
Oleg Nesterov | 9115122 | 2013-11-14 12:56:18 +0100 | [diff] [blame] | 1756 | set_user_nice(worker->task, pool->attrs->nice); |
| 1757 | |
| 1758 | /* prevent userland from meddling with cpumask of workqueue workers */ |
| 1759 | worker->task->flags |= PF_NO_SETAFFINITY; |
| 1760 | |
Tejun Heo | c5aa87b | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 1761 | /* |
| 1762 | * set_cpus_allowed_ptr() will fail if the cpumask doesn't have any |
| 1763 | * online CPUs. It'll be re-applied when any of the CPUs come up. |
| 1764 | */ |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 1765 | set_cpus_allowed_ptr(worker->task, pool->attrs->cpumask); |
Tejun Heo | 3270476 | 2012-07-13 22:16:45 -0700 | [diff] [blame] | 1766 | |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 1767 | /* |
| 1768 | * The caller is responsible for ensuring %POOL_DISASSOCIATED |
| 1769 | * remains stable across this function. See the comments above the |
| 1770 | * flag definition for details. |
| 1771 | */ |
| 1772 | if (pool->flags & POOL_DISASSOCIATED) |
Tejun Heo | bc2ae0f | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 1773 | worker->flags |= WORKER_UNBOUND; |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 1774 | |
Tejun Heo | 822d840 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 1775 | /* successful, commit the pointer to idr */ |
| 1776 | spin_lock_irq(&pool->lock); |
| 1777 | idr_replace(&pool->worker_idr, worker, worker->id); |
| 1778 | spin_unlock_irq(&pool->lock); |
| 1779 | |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1780 | return worker; |
Tejun Heo | 822d840 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 1781 | |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1782 | fail: |
| 1783 | if (id >= 0) { |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1784 | spin_lock_irq(&pool->lock); |
Tejun Heo | 822d840 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 1785 | idr_remove(&pool->worker_idr, id); |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1786 | spin_unlock_irq(&pool->lock); |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1787 | } |
| 1788 | kfree(worker); |
| 1789 | return NULL; |
| 1790 | } |
| 1791 | |
| 1792 | /** |
| 1793 | * start_worker - start a newly created worker |
| 1794 | * @worker: worker to start |
| 1795 | * |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 1796 | * Make the pool aware of @worker and start it. |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1797 | * |
| 1798 | * CONTEXT: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1799 | * spin_lock_irq(pool->lock). |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1800 | */ |
| 1801 | static void start_worker(struct worker *worker) |
| 1802 | { |
Tejun Heo | cb44476 | 2010-07-02 10:03:50 +0200 | [diff] [blame] | 1803 | worker->flags |= WORKER_STARTED; |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1804 | worker->pool->nr_workers++; |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1805 | worker_enter_idle(worker); |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1806 | wake_up_process(worker->task); |
| 1807 | } |
| 1808 | |
| 1809 | /** |
Tejun Heo | ebf44d1 | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 1810 | * create_and_start_worker - create and start a worker for a pool |
| 1811 | * @pool: the target pool |
| 1812 | * |
Tejun Heo | cd54968 | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 1813 | * Grab the managership of @pool and create and start a new worker for it. |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 1814 | * |
| 1815 | * Return: 0 on success. A negative error code otherwise. |
Tejun Heo | ebf44d1 | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 1816 | */ |
| 1817 | static int create_and_start_worker(struct worker_pool *pool) |
| 1818 | { |
| 1819 | struct worker *worker; |
| 1820 | |
Tejun Heo | cd54968 | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 1821 | mutex_lock(&pool->manager_mutex); |
| 1822 | |
Tejun Heo | ebf44d1 | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 1823 | worker = create_worker(pool); |
| 1824 | if (worker) { |
| 1825 | spin_lock_irq(&pool->lock); |
| 1826 | start_worker(worker); |
| 1827 | spin_unlock_irq(&pool->lock); |
| 1828 | } |
| 1829 | |
Tejun Heo | cd54968 | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 1830 | mutex_unlock(&pool->manager_mutex); |
| 1831 | |
Tejun Heo | ebf44d1 | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 1832 | return worker ? 0 : -ENOMEM; |
| 1833 | } |
| 1834 | |
| 1835 | /** |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1836 | * destroy_worker - destroy a workqueue worker |
| 1837 | * @worker: worker to be destroyed |
| 1838 | * |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 1839 | * Destroy @worker and adjust @pool stats accordingly. |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1840 | * |
| 1841 | * CONTEXT: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1842 | * spin_lock_irq(pool->lock) which is released and regrabbed. |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1843 | */ |
| 1844 | static void destroy_worker(struct worker *worker) |
| 1845 | { |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1846 | struct worker_pool *pool = worker->pool; |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1847 | |
Tejun Heo | cd54968 | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 1848 | lockdep_assert_held(&pool->manager_mutex); |
| 1849 | lockdep_assert_held(&pool->lock); |
| 1850 | |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1851 | /* sanity check frenzy */ |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 1852 | if (WARN_ON(worker->current_work) || |
| 1853 | WARN_ON(!list_empty(&worker->scheduled))) |
| 1854 | return; |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1855 | |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1856 | if (worker->flags & WORKER_STARTED) |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1857 | pool->nr_workers--; |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1858 | if (worker->flags & WORKER_IDLE) |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1859 | pool->nr_idle--; |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1860 | |
Lai Jiangshan | 5bdfff9 | 2014-02-15 22:02:28 +0800 | [diff] [blame] | 1861 | /* |
| 1862 | * Once WORKER_DIE is set, the kworker may destroy itself at any |
| 1863 | * point. Pin to ensure the task stays until we're done with it. |
| 1864 | */ |
| 1865 | get_task_struct(worker->task); |
| 1866 | |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1867 | list_del_init(&worker->entry); |
Tejun Heo | cb44476 | 2010-07-02 10:03:50 +0200 | [diff] [blame] | 1868 | worker->flags |= WORKER_DIE; |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1869 | |
Tejun Heo | 822d840 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 1870 | idr_remove(&pool->worker_idr, worker->id); |
| 1871 | |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1872 | spin_unlock_irq(&pool->lock); |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 1873 | |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1874 | kthread_stop(worker->task); |
Lai Jiangshan | 5bdfff9 | 2014-02-15 22:02:28 +0800 | [diff] [blame] | 1875 | put_task_struct(worker->task); |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1876 | kfree(worker); |
| 1877 | |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1878 | spin_lock_irq(&pool->lock); |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 1879 | } |
| 1880 | |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1881 | static void idle_worker_timeout(unsigned long __pool) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1882 | { |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1883 | struct worker_pool *pool = (void *)__pool; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1884 | |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1885 | spin_lock_irq(&pool->lock); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1886 | |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1887 | if (too_many_workers(pool)) { |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1888 | struct worker *worker; |
| 1889 | unsigned long expires; |
| 1890 | |
| 1891 | /* idle_list is kept in LIFO order, check the last one */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1892 | worker = list_entry(pool->idle_list.prev, struct worker, entry); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1893 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; |
| 1894 | |
| 1895 | if (time_before(jiffies, expires)) |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1896 | mod_timer(&pool->idle_timer, expires); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1897 | else { |
| 1898 | /* it's been idle for too long, wake up manager */ |
Tejun Heo | 11ebea5 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1899 | pool->flags |= POOL_MANAGE_WORKERS; |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1900 | wake_up_worker(pool); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1901 | } |
| 1902 | } |
| 1903 | |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1904 | spin_unlock_irq(&pool->lock); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1905 | } |
| 1906 | |
Tejun Heo | 493a172 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 1907 | static void send_mayday(struct work_struct *work) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1908 | { |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 1909 | struct pool_workqueue *pwq = get_work_pwq(work); |
| 1910 | struct workqueue_struct *wq = pwq->wq; |
Tejun Heo | 493a172 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 1911 | |
Tejun Heo | 2e109a2 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 1912 | lockdep_assert_held(&wq_mayday_lock); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1913 | |
Tejun Heo | 493008a | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 1914 | if (!wq->rescuer) |
Tejun Heo | 493a172 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 1915 | return; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1916 | |
| 1917 | /* mayday mayday mayday */ |
Tejun Heo | 493a172 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 1918 | if (list_empty(&pwq->mayday_node)) { |
| 1919 | list_add_tail(&pwq->mayday_node, &wq->maydays); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1920 | wake_up_process(wq->rescuer->task); |
Tejun Heo | 493a172 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 1921 | } |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1922 | } |
| 1923 | |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 1924 | static void pool_mayday_timeout(unsigned long __pool) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1925 | { |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1926 | struct worker_pool *pool = (void *)__pool; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1927 | struct work_struct *work; |
| 1928 | |
Tejun Heo | 2e109a2 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 1929 | spin_lock_irq(&wq_mayday_lock); /* for wq->maydays */ |
Tejun Heo | 493a172 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 1930 | spin_lock(&pool->lock); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1931 | |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1932 | if (need_to_create_worker(pool)) { |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1933 | /* |
| 1934 | * We've been trying to create a new worker but |
| 1935 | * haven't been successful. We might be hitting an |
| 1936 | * allocation deadlock. Send distress signals to |
| 1937 | * rescuers. |
| 1938 | */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1939 | list_for_each_entry(work, &pool->worklist, entry) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1940 | send_mayday(work); |
| 1941 | } |
| 1942 | |
Tejun Heo | 493a172 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 1943 | spin_unlock(&pool->lock); |
Tejun Heo | 2e109a2 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 1944 | spin_unlock_irq(&wq_mayday_lock); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1945 | |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1946 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INTERVAL); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1947 | } |
| 1948 | |
| 1949 | /** |
| 1950 | * maybe_create_worker - create a new worker if necessary |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1951 | * @pool: pool to create a new worker for |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1952 | * |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1953 | * Create a new worker for @pool if necessary. @pool is guaranteed to |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1954 | * have at least one idle worker on return from this function. If |
| 1955 | * creating a new worker takes longer than MAYDAY_INTERVAL, mayday is |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1956 | * sent to all rescuers with works scheduled on @pool to resolve |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1957 | * possible allocation deadlock. |
| 1958 | * |
Tejun Heo | c5aa87b | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 1959 | * On return, need_to_create_worker() is guaranteed to be %false and |
| 1960 | * may_start_working() %true. |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1961 | * |
| 1962 | * LOCKING: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1963 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1964 | * multiple times. Does GFP_KERNEL allocations. Called only from |
| 1965 | * manager. |
| 1966 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 1967 | * Return: |
Tejun Heo | c5aa87b | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 1968 | * %false if no action was taken and pool->lock stayed locked, %true |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1969 | * otherwise. |
| 1970 | */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1971 | static bool maybe_create_worker(struct worker_pool *pool) |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1972 | __releases(&pool->lock) |
| 1973 | __acquires(&pool->lock) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1974 | { |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1975 | if (!need_to_create_worker(pool)) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1976 | return false; |
| 1977 | restart: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1978 | spin_unlock_irq(&pool->lock); |
Tejun Heo | 9f9c236 | 2010-07-14 11:31:20 +0200 | [diff] [blame] | 1979 | |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1980 | /* 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] | 1981 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INITIAL_TIMEOUT); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1982 | |
| 1983 | while (true) { |
| 1984 | struct worker *worker; |
| 1985 | |
Tejun Heo | bc2ae0f | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 1986 | worker = create_worker(pool); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1987 | if (worker) { |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1988 | del_timer_sync(&pool->mayday_timer); |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 1989 | spin_lock_irq(&pool->lock); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1990 | start_worker(worker); |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 1991 | if (WARN_ON_ONCE(need_to_create_worker(pool))) |
| 1992 | goto restart; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1993 | return true; |
| 1994 | } |
| 1995 | |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 1996 | if (!need_to_create_worker(pool)) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1997 | break; |
| 1998 | |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 1999 | __set_current_state(TASK_INTERRUPTIBLE); |
| 2000 | schedule_timeout(CREATE_COOLDOWN); |
Tejun Heo | 9f9c236 | 2010-07-14 11:31:20 +0200 | [diff] [blame] | 2001 | |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2002 | if (!need_to_create_worker(pool)) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2003 | break; |
| 2004 | } |
| 2005 | |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2006 | del_timer_sync(&pool->mayday_timer); |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2007 | spin_lock_irq(&pool->lock); |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2008 | if (need_to_create_worker(pool)) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2009 | goto restart; |
| 2010 | return true; |
| 2011 | } |
| 2012 | |
| 2013 | /** |
| 2014 | * maybe_destroy_worker - destroy workers which have been idle for a while |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2015 | * @pool: pool to destroy workers for |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2016 | * |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2017 | * Destroy @pool workers which have been idle for longer than |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2018 | * IDLE_WORKER_TIMEOUT. |
| 2019 | * |
| 2020 | * LOCKING: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2021 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2022 | * multiple times. Called only from manager. |
| 2023 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 2024 | * Return: |
Tejun Heo | c5aa87b | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 2025 | * %false if no action was taken and pool->lock stayed locked, %true |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2026 | * otherwise. |
| 2027 | */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2028 | static bool maybe_destroy_workers(struct worker_pool *pool) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2029 | { |
| 2030 | bool ret = false; |
| 2031 | |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2032 | while (too_many_workers(pool)) { |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2033 | struct worker *worker; |
| 2034 | unsigned long expires; |
| 2035 | |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2036 | worker = list_entry(pool->idle_list.prev, struct worker, entry); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2037 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; |
| 2038 | |
| 2039 | if (time_before(jiffies, expires)) { |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2040 | mod_timer(&pool->idle_timer, expires); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2041 | break; |
| 2042 | } |
| 2043 | |
| 2044 | destroy_worker(worker); |
| 2045 | ret = true; |
| 2046 | } |
| 2047 | |
| 2048 | return ret; |
| 2049 | } |
| 2050 | |
| 2051 | /** |
| 2052 | * manage_workers - manage worker pool |
| 2053 | * @worker: self |
| 2054 | * |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 2055 | * Assume the manager role and manage the worker pool @worker belongs |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2056 | * 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] | 2057 | * pool. The exclusion is handled automatically by this function. |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2058 | * |
| 2059 | * The caller can safely start processing works on false return. On |
| 2060 | * true return, it's guaranteed that need_to_create_worker() is false |
| 2061 | * and may_start_working() is true. |
| 2062 | * |
| 2063 | * CONTEXT: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2064 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2065 | * multiple times. Does GFP_KERNEL allocations. |
| 2066 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 2067 | * Return: |
Libin | 2d498db | 2013-08-21 08:50:40 +0800 | [diff] [blame] | 2068 | * %false if the pool don't need management and the caller can safely start |
| 2069 | * processing works, %true indicates that the function released pool->lock |
| 2070 | * and reacquired it to perform some management function and that the |
| 2071 | * conditions that the caller verified while holding the lock before |
| 2072 | * calling the function might no longer be true. |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2073 | */ |
| 2074 | static bool manage_workers(struct worker *worker) |
| 2075 | { |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2076 | struct worker_pool *pool = worker->pool; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2077 | bool ret = false; |
| 2078 | |
Tejun Heo | bc3a1af | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 2079 | /* |
| 2080 | * Managership is governed by two mutexes - manager_arb and |
| 2081 | * manager_mutex. manager_arb handles arbitration of manager role. |
| 2082 | * Anyone who successfully grabs manager_arb wins the arbitration |
| 2083 | * and becomes the manager. mutex_trylock() on pool->manager_arb |
| 2084 | * failure while holding pool->lock reliably indicates that someone |
| 2085 | * else is managing the pool and the worker which failed trylock |
| 2086 | * can proceed to executing work items. This means that anyone |
| 2087 | * grabbing manager_arb is responsible for actually performing |
| 2088 | * manager duties. If manager_arb is grabbed and released without |
| 2089 | * actual management, the pool may stall indefinitely. |
| 2090 | * |
| 2091 | * manager_mutex is used for exclusion of actual management |
| 2092 | * operations. The holder of manager_mutex can be sure that none |
| 2093 | * of management operations, including creation and destruction of |
| 2094 | * workers, won't take place until the mutex is released. Because |
| 2095 | * manager_mutex doesn't interfere with manager role arbitration, |
| 2096 | * it is guaranteed that the pool's management, while may be |
| 2097 | * delayed, won't be disturbed by someone else grabbing |
| 2098 | * manager_mutex. |
| 2099 | */ |
Tejun Heo | 34a06bd | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 2100 | if (!mutex_trylock(&pool->manager_arb)) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2101 | return ret; |
| 2102 | |
Lai Jiangshan | ee378aa | 2012-09-10 10:03:44 -0700 | [diff] [blame] | 2103 | /* |
Tejun Heo | bc3a1af | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 2104 | * With manager arbitration won, manager_mutex would be free in |
| 2105 | * most cases. trylock first without dropping @pool->lock. |
Lai Jiangshan | ee378aa | 2012-09-10 10:03:44 -0700 | [diff] [blame] | 2106 | */ |
Tejun Heo | bc3a1af | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 2107 | if (unlikely(!mutex_trylock(&pool->manager_mutex))) { |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2108 | spin_unlock_irq(&pool->lock); |
Tejun Heo | bc3a1af | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 2109 | mutex_lock(&pool->manager_mutex); |
Joonsoo Kim | 8f174b1 | 2013-05-01 00:07:00 +0900 | [diff] [blame] | 2110 | spin_lock_irq(&pool->lock); |
Lai Jiangshan | ee378aa | 2012-09-10 10:03:44 -0700 | [diff] [blame] | 2111 | ret = true; |
| 2112 | } |
| 2113 | |
Tejun Heo | 11ebea5 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2114 | pool->flags &= ~POOL_MANAGE_WORKERS; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2115 | |
| 2116 | /* |
| 2117 | * Destroy and then create so that may_start_working() is true |
| 2118 | * on return. |
| 2119 | */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2120 | ret |= maybe_destroy_workers(pool); |
| 2121 | ret |= maybe_create_worker(pool); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2122 | |
Tejun Heo | bc3a1af | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 2123 | mutex_unlock(&pool->manager_mutex); |
Tejun Heo | 34a06bd | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 2124 | mutex_unlock(&pool->manager_arb); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2125 | return ret; |
| 2126 | } |
| 2127 | |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2128 | /** |
| 2129 | * process_one_work - process single work |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2130 | * @worker: self |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2131 | * @work: work to process |
| 2132 | * |
| 2133 | * Process @work. This function contains all the logics necessary to |
| 2134 | * process a single work including synchronization against and |
| 2135 | * interaction with other workers on the same cpu, queueing and |
| 2136 | * flushing. As long as context requirement is met, any worker can |
| 2137 | * call this function to process a work. |
| 2138 | * |
| 2139 | * CONTEXT: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2140 | * spin_lock_irq(pool->lock) which is released and regrabbed. |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2141 | */ |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2142 | static void process_one_work(struct worker *worker, struct work_struct *work) |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2143 | __releases(&pool->lock) |
| 2144 | __acquires(&pool->lock) |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2145 | { |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2146 | struct pool_workqueue *pwq = get_work_pwq(work); |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2147 | struct worker_pool *pool = worker->pool; |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2148 | bool cpu_intensive = pwq->wq->flags & WQ_CPU_INTENSIVE; |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2149 | int work_color; |
Tejun Heo | 7e11629 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 2150 | struct worker *collision; |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2151 | #ifdef CONFIG_LOCKDEP |
| 2152 | /* |
| 2153 | * It is permissible to free the struct work_struct from |
| 2154 | * inside the function that is called from it, this we need to |
| 2155 | * take into account for lockdep too. To avoid bogus "held |
| 2156 | * lock freed" warnings as well as problems when looking into |
| 2157 | * work->lockdep_map, make a copy and use that here. |
| 2158 | */ |
Peter Zijlstra | 4d82a1d | 2012-05-15 08:06:19 -0700 | [diff] [blame] | 2159 | struct lockdep_map lockdep_map; |
| 2160 | |
| 2161 | lockdep_copy_map(&lockdep_map, &work->lockdep_map); |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2162 | #endif |
Tejun Heo | 6fec10a | 2012-07-22 10:16:34 -0700 | [diff] [blame] | 2163 | /* |
| 2164 | * Ensure we're on the correct CPU. DISASSOCIATED test is |
| 2165 | * necessary to avoid spurious warnings from rescuers servicing the |
Tejun Heo | 2464757 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2166 | * unbound or a disassociated pool. |
Tejun Heo | 6fec10a | 2012-07-22 10:16:34 -0700 | [diff] [blame] | 2167 | */ |
Lai Jiangshan | 5f7dabf | 2012-09-18 09:59:23 -0700 | [diff] [blame] | 2168 | WARN_ON_ONCE(!(worker->flags & WORKER_UNBOUND) && |
Tejun Heo | 2464757 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2169 | !(pool->flags & POOL_DISASSOCIATED) && |
Tejun Heo | ec22ca5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2170 | raw_smp_processor_id() != pool->cpu); |
Tejun Heo | 25511a4 | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 2171 | |
Tejun Heo | 7e11629 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 2172 | /* |
| 2173 | * A single work shouldn't be executed concurrently by |
| 2174 | * multiple workers on a single cpu. Check whether anyone is |
| 2175 | * already processing the work. If so, defer the work to the |
| 2176 | * currently executing one. |
| 2177 | */ |
Tejun Heo | c9e7cf2 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2178 | collision = find_worker_executing_work(pool, work); |
Tejun Heo | 7e11629 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 2179 | if (unlikely(collision)) { |
| 2180 | move_linked_works(work, &collision->scheduled, NULL); |
| 2181 | return; |
| 2182 | } |
| 2183 | |
Tejun Heo | 8930cab | 2012-08-03 10:30:45 -0700 | [diff] [blame] | 2184 | /* claim and dequeue */ |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2185 | debug_work_deactivate(work); |
Tejun Heo | c9e7cf2 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2186 | hash_add(pool->busy_hash, &worker->hentry, (unsigned long)work); |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2187 | worker->current_work = work; |
Tejun Heo | a2c1c57 | 2012-12-18 10:35:02 -0800 | [diff] [blame] | 2188 | worker->current_func = work->func; |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2189 | worker->current_pwq = pwq; |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2190 | work_color = get_work_color(work); |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 2191 | |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2192 | list_del_init(&work->entry); |
| 2193 | |
Tejun Heo | 649027d | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2194 | /* |
Tejun Heo | fb0e7be | 2010-06-29 10:07:15 +0200 | [diff] [blame] | 2195 | * CPU intensive works don't participate in concurrency |
| 2196 | * management. They're the scheduler's responsibility. |
| 2197 | */ |
| 2198 | if (unlikely(cpu_intensive)) |
| 2199 | worker_set_flags(worker, WORKER_CPU_INTENSIVE, true); |
| 2200 | |
Tejun Heo | 974271c | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2201 | /* |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2202 | * Unbound pool isn't concurrency managed and work items should be |
Tejun Heo | 974271c | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2203 | * executed ASAP. Wake up another worker if necessary. |
| 2204 | */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2205 | if ((worker->flags & WORKER_UNBOUND) && need_more_worker(pool)) |
| 2206 | wake_up_worker(pool); |
Tejun Heo | 974271c | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2207 | |
Tejun Heo | 8930cab | 2012-08-03 10:30:45 -0700 | [diff] [blame] | 2208 | /* |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2209 | * Record the last pool and clear PENDING which should be the last |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2210 | * update to @work. Also, do this inside @pool->lock so that |
Tejun Heo | 23657bb | 2012-08-13 17:08:19 -0700 | [diff] [blame] | 2211 | * PENDING and queued state changes happen together while IRQ is |
| 2212 | * disabled. |
Tejun Heo | 8930cab | 2012-08-03 10:30:45 -0700 | [diff] [blame] | 2213 | */ |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2214 | set_work_pool_and_clear_pending(work, pool->id); |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2215 | |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2216 | spin_unlock_irq(&pool->lock); |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2217 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2218 | lock_map_acquire_read(&pwq->wq->lockdep_map); |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2219 | lock_map_acquire(&lockdep_map); |
Arjan van de Ven | e36c886 | 2010-08-21 13:07:26 -0700 | [diff] [blame] | 2220 | trace_workqueue_execute_start(work); |
Tejun Heo | a2c1c57 | 2012-12-18 10:35:02 -0800 | [diff] [blame] | 2221 | worker->current_func(work); |
Arjan van de Ven | e36c886 | 2010-08-21 13:07:26 -0700 | [diff] [blame] | 2222 | /* |
| 2223 | * While we must be careful to not use "work" after this, the trace |
| 2224 | * point will only record its address. |
| 2225 | */ |
| 2226 | trace_workqueue_execute_end(work); |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2227 | lock_map_release(&lockdep_map); |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2228 | lock_map_release(&pwq->wq->lockdep_map); |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2229 | |
| 2230 | if (unlikely(in_atomic() || lockdep_depth(current) > 0)) { |
Valentin Ilie | 044c782 | 2012-08-19 00:52:42 +0300 | [diff] [blame] | 2231 | pr_err("BUG: workqueue leaked lock or atomic: %s/0x%08x/%d\n" |
| 2232 | " last function: %pf\n", |
Tejun Heo | a2c1c57 | 2012-12-18 10:35:02 -0800 | [diff] [blame] | 2233 | current->comm, preempt_count(), task_pid_nr(current), |
| 2234 | worker->current_func); |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2235 | debug_show_held_locks(current); |
| 2236 | dump_stack(); |
| 2237 | } |
| 2238 | |
Tejun Heo | b22ce27 | 2013-08-28 17:33:37 -0400 | [diff] [blame] | 2239 | /* |
| 2240 | * The following prevents a kworker from hogging CPU on !PREEMPT |
| 2241 | * kernels, where a requeueing work item waiting for something to |
| 2242 | * happen could deadlock with stop_machine as such work item could |
| 2243 | * indefinitely requeue itself while all other CPUs are trapped in |
| 2244 | * stop_machine. |
| 2245 | */ |
| 2246 | cond_resched(); |
| 2247 | |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2248 | spin_lock_irq(&pool->lock); |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2249 | |
Tejun Heo | fb0e7be | 2010-06-29 10:07:15 +0200 | [diff] [blame] | 2250 | /* clear cpu intensive status */ |
| 2251 | if (unlikely(cpu_intensive)) |
| 2252 | worker_clr_flags(worker, WORKER_CPU_INTENSIVE); |
| 2253 | |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2254 | /* we're done with it, release */ |
Sasha Levin | 42f8570 | 2012-12-17 10:01:23 -0500 | [diff] [blame] | 2255 | hash_del(&worker->hentry); |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2256 | worker->current_work = NULL; |
Tejun Heo | a2c1c57 | 2012-12-18 10:35:02 -0800 | [diff] [blame] | 2257 | worker->current_func = NULL; |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2258 | worker->current_pwq = NULL; |
Tejun Heo | 3d1cb20 | 2013-04-30 15:27:22 -0700 | [diff] [blame] | 2259 | worker->desc_valid = false; |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2260 | pwq_dec_nr_in_flight(pwq, work_color); |
Tejun Heo | a62428c | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2261 | } |
| 2262 | |
Tejun Heo | affee4b | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2263 | /** |
| 2264 | * process_scheduled_works - process scheduled works |
| 2265 | * @worker: self |
| 2266 | * |
| 2267 | * Process all scheduled works. Please note that the scheduled list |
| 2268 | * may change while processing a work, so this function repeatedly |
| 2269 | * fetches a work from the top and executes it. |
| 2270 | * |
| 2271 | * CONTEXT: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2272 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
Tejun Heo | affee4b | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2273 | * multiple times. |
| 2274 | */ |
| 2275 | static void process_scheduled_works(struct worker *worker) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2276 | { |
Tejun Heo | affee4b | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2277 | while (!list_empty(&worker->scheduled)) { |
| 2278 | struct work_struct *work = list_first_entry(&worker->scheduled, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2279 | struct work_struct, entry); |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2280 | process_one_work(worker, work); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2281 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2282 | } |
| 2283 | |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2284 | /** |
| 2285 | * worker_thread - the worker thread function |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2286 | * @__worker: self |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2287 | * |
Tejun Heo | c5aa87b | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 2288 | * The worker thread function. All workers belong to a worker_pool - |
| 2289 | * either a per-cpu one or dynamic unbound one. These workers process all |
| 2290 | * work items regardless of their specific target workqueue. The only |
| 2291 | * exception is work items which belong to workqueues with a rescuer which |
| 2292 | * will be explained in rescuer_thread(). |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 2293 | * |
| 2294 | * Return: 0 |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2295 | */ |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2296 | static int worker_thread(void *__worker) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2297 | { |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2298 | struct worker *worker = __worker; |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2299 | struct worker_pool *pool = worker->pool; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2300 | |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2301 | /* tell the scheduler that this is a workqueue worker */ |
| 2302 | worker->task->flags |= PF_WQ_WORKER; |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2303 | woke_up: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2304 | spin_lock_irq(&pool->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2305 | |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 2306 | /* am I supposed to die? */ |
| 2307 | if (unlikely(worker->flags & WORKER_DIE)) { |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2308 | spin_unlock_irq(&pool->lock); |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 2309 | WARN_ON_ONCE(!list_empty(&worker->entry)); |
| 2310 | worker->task->flags &= ~PF_WQ_WORKER; |
| 2311 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2312 | } |
| 2313 | |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2314 | worker_leave_idle(worker); |
Tejun Heo | db7bccf | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2315 | recheck: |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2316 | /* no more worker necessary? */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2317 | if (!need_more_worker(pool)) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2318 | goto sleep; |
| 2319 | |
| 2320 | /* do we need to manage? */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2321 | if (unlikely(!may_start_working(pool)) && manage_workers(worker)) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2322 | goto recheck; |
| 2323 | |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2324 | /* |
| 2325 | * ->scheduled list can only be filled while a worker is |
| 2326 | * preparing to process a work or actually processing it. |
| 2327 | * Make sure nobody diddled with it while I was sleeping. |
| 2328 | */ |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 2329 | WARN_ON_ONCE(!list_empty(&worker->scheduled)); |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2330 | |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2331 | /* |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 2332 | * Finish PREP stage. We're guaranteed to have at least one idle |
| 2333 | * worker or that someone else has already assumed the manager |
| 2334 | * role. This is where @worker starts participating in concurrency |
| 2335 | * management if applicable and concurrency management is restored |
| 2336 | * after being rebound. See rebind_workers() for details. |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2337 | */ |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 2338 | worker_clr_flags(worker, WORKER_PREP | WORKER_REBOUND); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2339 | |
| 2340 | do { |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2341 | struct work_struct *work = |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2342 | list_first_entry(&pool->worklist, |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2343 | struct work_struct, entry); |
| 2344 | |
| 2345 | if (likely(!(*work_data_bits(work) & WORK_STRUCT_LINKED))) { |
| 2346 | /* optimization path, not strictly necessary */ |
| 2347 | process_one_work(worker, work); |
| 2348 | if (unlikely(!list_empty(&worker->scheduled))) |
| 2349 | process_scheduled_works(worker); |
| 2350 | } else { |
| 2351 | move_linked_works(work, &worker->scheduled, NULL); |
| 2352 | process_scheduled_works(worker); |
| 2353 | } |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2354 | } while (keep_working(pool)); |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2355 | |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2356 | worker_set_flags(worker, WORKER_PREP, false); |
Tejun Heo | d313dd8 | 2010-07-02 10:03:51 +0200 | [diff] [blame] | 2357 | sleep: |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2358 | if (unlikely(need_to_manage_workers(pool)) && manage_workers(worker)) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2359 | goto recheck; |
Tejun Heo | d313dd8 | 2010-07-02 10:03:51 +0200 | [diff] [blame] | 2360 | |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2361 | /* |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2362 | * pool->lock is held and there's no work to process and no need to |
| 2363 | * manage, sleep. Workers are woken up only while holding |
| 2364 | * pool->lock or from local cpu, so setting the current state |
| 2365 | * before releasing pool->lock is enough to prevent losing any |
| 2366 | * event. |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2367 | */ |
| 2368 | worker_enter_idle(worker); |
| 2369 | __set_current_state(TASK_INTERRUPTIBLE); |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2370 | spin_unlock_irq(&pool->lock); |
Tejun Heo | c8e55f3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2371 | schedule(); |
| 2372 | goto woke_up; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2373 | } |
| 2374 | |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2375 | /** |
| 2376 | * rescuer_thread - the rescuer thread function |
Tejun Heo | 111c225 | 2013-01-17 17:16:24 -0800 | [diff] [blame] | 2377 | * @__rescuer: self |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2378 | * |
| 2379 | * Workqueue rescuer thread function. There's one rescuer for each |
Tejun Heo | 493008a | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 2380 | * workqueue which has WQ_MEM_RECLAIM set. |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2381 | * |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 2382 | * 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] | 2383 | * worker which uses GFP_KERNEL allocation which has slight chance of |
| 2384 | * developing into deadlock if some works currently on the same queue |
| 2385 | * need to be processed to satisfy the GFP_KERNEL allocation. This is |
| 2386 | * the problem rescuer solves. |
| 2387 | * |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 2388 | * When such condition is possible, the pool summons rescuers of all |
| 2389 | * workqueues which have works queued on the pool and let them process |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2390 | * those works so that forward progress can be guaranteed. |
| 2391 | * |
| 2392 | * This should happen rarely. |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 2393 | * |
| 2394 | * Return: 0 |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2395 | */ |
Tejun Heo | 111c225 | 2013-01-17 17:16:24 -0800 | [diff] [blame] | 2396 | static int rescuer_thread(void *__rescuer) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2397 | { |
Tejun Heo | 111c225 | 2013-01-17 17:16:24 -0800 | [diff] [blame] | 2398 | struct worker *rescuer = __rescuer; |
| 2399 | struct workqueue_struct *wq = rescuer->rescue_wq; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2400 | struct list_head *scheduled = &rescuer->scheduled; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2401 | |
| 2402 | set_user_nice(current, RESCUER_NICE_LEVEL); |
Tejun Heo | 111c225 | 2013-01-17 17:16:24 -0800 | [diff] [blame] | 2403 | |
| 2404 | /* |
| 2405 | * Mark rescuer as worker too. As WORKER_PREP is never cleared, it |
| 2406 | * doesn't participate in concurrency management. |
| 2407 | */ |
| 2408 | rescuer->task->flags |= PF_WQ_WORKER; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2409 | repeat: |
| 2410 | set_current_state(TASK_INTERRUPTIBLE); |
| 2411 | |
Mike Galbraith | 412d32e | 2012-11-28 07:17:18 +0100 | [diff] [blame] | 2412 | if (kthread_should_stop()) { |
| 2413 | __set_current_state(TASK_RUNNING); |
Tejun Heo | 111c225 | 2013-01-17 17:16:24 -0800 | [diff] [blame] | 2414 | rescuer->task->flags &= ~PF_WQ_WORKER; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2415 | return 0; |
Mike Galbraith | 412d32e | 2012-11-28 07:17:18 +0100 | [diff] [blame] | 2416 | } |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2417 | |
Tejun Heo | 493a172 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 2418 | /* see whether any pwq is asking for help */ |
Tejun Heo | 2e109a2 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 2419 | spin_lock_irq(&wq_mayday_lock); |
Tejun Heo | 493a172 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 2420 | |
| 2421 | while (!list_empty(&wq->maydays)) { |
| 2422 | struct pool_workqueue *pwq = list_first_entry(&wq->maydays, |
| 2423 | struct pool_workqueue, mayday_node); |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2424 | struct worker_pool *pool = pwq->pool; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2425 | struct work_struct *work, *n; |
| 2426 | |
| 2427 | __set_current_state(TASK_RUNNING); |
Tejun Heo | 493a172 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 2428 | list_del_init(&pwq->mayday_node); |
| 2429 | |
Tejun Heo | 2e109a2 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 2430 | spin_unlock_irq(&wq_mayday_lock); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2431 | |
| 2432 | /* migrate to the target cpu if possible */ |
Lai Jiangshan | f36dc67 | 2013-02-19 12:17:02 -0800 | [diff] [blame] | 2433 | worker_maybe_bind_and_lock(pool); |
Lai Jiangshan | b310410 | 2013-02-19 12:17:02 -0800 | [diff] [blame] | 2434 | rescuer->pool = pool; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2435 | |
| 2436 | /* |
| 2437 | * Slurp in all works issued via this workqueue and |
| 2438 | * process'em. |
| 2439 | */ |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 2440 | WARN_ON_ONCE(!list_empty(&rescuer->scheduled)); |
Tejun Heo | bd7bdd4 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2441 | list_for_each_entry_safe(work, n, &pool->worklist, entry) |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2442 | if (get_work_pwq(work) == pwq) |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2443 | move_linked_works(work, scheduled, &n); |
| 2444 | |
| 2445 | process_scheduled_works(rescuer); |
Tejun Heo | 7576958 | 2011-02-14 14:04:46 +0100 | [diff] [blame] | 2446 | |
| 2447 | /* |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2448 | * Leave this pool. If keep_working() is %true, notify a |
Tejun Heo | 7576958 | 2011-02-14 14:04:46 +0100 | [diff] [blame] | 2449 | * regular worker; otherwise, we end up with 0 concurrency |
| 2450 | * and stalling the execution. |
| 2451 | */ |
Tejun Heo | 63d95a9 | 2012-07-12 14:46:37 -0700 | [diff] [blame] | 2452 | if (keep_working(pool)) |
| 2453 | wake_up_worker(pool); |
Tejun Heo | 7576958 | 2011-02-14 14:04:46 +0100 | [diff] [blame] | 2454 | |
Lai Jiangshan | b310410 | 2013-02-19 12:17:02 -0800 | [diff] [blame] | 2455 | rescuer->pool = NULL; |
Tejun Heo | 493a172 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 2456 | spin_unlock(&pool->lock); |
Tejun Heo | 2e109a2 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 2457 | spin_lock(&wq_mayday_lock); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2458 | } |
| 2459 | |
Tejun Heo | 2e109a2 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 2460 | spin_unlock_irq(&wq_mayday_lock); |
Tejun Heo | 493a172 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 2461 | |
Tejun Heo | 111c225 | 2013-01-17 17:16:24 -0800 | [diff] [blame] | 2462 | /* rescuers should never participate in concurrency management */ |
| 2463 | WARN_ON_ONCE(!(rescuer->flags & WORKER_NOT_RUNNING)); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 2464 | schedule(); |
| 2465 | goto repeat; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2466 | } |
| 2467 | |
Oleg Nesterov | fc2e4d7 | 2007-05-09 02:33:51 -0700 | [diff] [blame] | 2468 | struct wq_barrier { |
| 2469 | struct work_struct work; |
| 2470 | struct completion done; |
| 2471 | }; |
| 2472 | |
| 2473 | static void wq_barrier_func(struct work_struct *work) |
| 2474 | { |
| 2475 | struct wq_barrier *barr = container_of(work, struct wq_barrier, work); |
| 2476 | complete(&barr->done); |
| 2477 | } |
| 2478 | |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2479 | /** |
| 2480 | * insert_wq_barrier - insert a barrier work |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2481 | * @pwq: pwq to insert barrier into |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2482 | * @barr: wq_barrier to insert |
Tejun Heo | affee4b | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2483 | * @target: target work to attach @barr to |
| 2484 | * @worker: worker currently executing @target, NULL if @target is not executing |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2485 | * |
Tejun Heo | affee4b | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2486 | * @barr is linked to @target such that @barr is completed only after |
| 2487 | * @target finishes execution. Please note that the ordering |
| 2488 | * guarantee is observed only with respect to @target and on the local |
| 2489 | * cpu. |
| 2490 | * |
| 2491 | * Currently, a queued barrier can't be canceled. This is because |
| 2492 | * try_to_grab_pending() can't determine whether the work to be |
| 2493 | * grabbed is at the head of the queue and thus can't clear LINKED |
| 2494 | * flag of the previous work while there must be a valid next work |
| 2495 | * after a work with LINKED flag set. |
| 2496 | * |
| 2497 | * Note that when @worker is non-NULL, @target may be modified |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2498 | * underneath us, so we can't reliably determine pwq from @target. |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2499 | * |
| 2500 | * CONTEXT: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2501 | * spin_lock_irq(pool->lock). |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2502 | */ |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2503 | static void insert_wq_barrier(struct pool_workqueue *pwq, |
Tejun Heo | affee4b | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2504 | struct wq_barrier *barr, |
| 2505 | struct work_struct *target, struct worker *worker) |
Oleg Nesterov | fc2e4d7 | 2007-05-09 02:33:51 -0700 | [diff] [blame] | 2506 | { |
Tejun Heo | affee4b | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2507 | struct list_head *head; |
| 2508 | unsigned int linked = 0; |
| 2509 | |
Thomas Gleixner | dc186ad | 2009-11-16 01:09:48 +0900 | [diff] [blame] | 2510 | /* |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2511 | * debugobject calls are safe here even with pool->lock locked |
Thomas Gleixner | dc186ad | 2009-11-16 01:09:48 +0900 | [diff] [blame] | 2512 | * as we know for sure that this will not trigger any of the |
| 2513 | * checks and call back into the fixup functions where we |
| 2514 | * might deadlock. |
| 2515 | */ |
Andrew Morton | ca1cab3 | 2010-10-26 14:22:34 -0700 | [diff] [blame] | 2516 | INIT_WORK_ONSTACK(&barr->work, wq_barrier_func); |
Tejun Heo | 22df02b | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 2517 | __set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(&barr->work)); |
Oleg Nesterov | fc2e4d7 | 2007-05-09 02:33:51 -0700 | [diff] [blame] | 2518 | init_completion(&barr->done); |
Oleg Nesterov | 83c2252 | 2007-05-09 02:33:54 -0700 | [diff] [blame] | 2519 | |
Tejun Heo | affee4b | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2520 | /* |
| 2521 | * If @target is currently being executed, schedule the |
| 2522 | * barrier to the worker; otherwise, put it after @target. |
| 2523 | */ |
| 2524 | if (worker) |
| 2525 | head = worker->scheduled.next; |
| 2526 | else { |
| 2527 | unsigned long *bits = work_data_bits(target); |
| 2528 | |
| 2529 | head = target->entry.next; |
| 2530 | /* there can already be other linked works, inherit and set */ |
| 2531 | linked = *bits & WORK_STRUCT_LINKED; |
| 2532 | __set_bit(WORK_STRUCT_LINKED_BIT, bits); |
| 2533 | } |
| 2534 | |
Thomas Gleixner | dc186ad | 2009-11-16 01:09:48 +0900 | [diff] [blame] | 2535 | debug_work_activate(&barr->work); |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2536 | insert_work(pwq, &barr->work, head, |
Tejun Heo | affee4b | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 2537 | work_color_to_flags(WORK_NO_COLOR) | linked); |
Oleg Nesterov | fc2e4d7 | 2007-05-09 02:33:51 -0700 | [diff] [blame] | 2538 | } |
| 2539 | |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2540 | /** |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2541 | * flush_workqueue_prep_pwqs - prepare pwqs for workqueue flushing |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2542 | * @wq: workqueue being flushed |
| 2543 | * @flush_color: new flush color, < 0 for no-op |
| 2544 | * @work_color: new work color, < 0 for no-op |
| 2545 | * |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2546 | * Prepare pwqs for workqueue flushing. |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2547 | * |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2548 | * If @flush_color is non-negative, flush_color on all pwqs should be |
| 2549 | * -1. If no pwq has in-flight commands at the specified color, all |
| 2550 | * pwq->flush_color's stay at -1 and %false is returned. If any pwq |
| 2551 | * has in flight commands, its pwq->flush_color is set to |
| 2552 | * @flush_color, @wq->nr_pwqs_to_flush is updated accordingly, pwq |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2553 | * wakeup logic is armed and %true is returned. |
| 2554 | * |
| 2555 | * The caller should have initialized @wq->first_flusher prior to |
| 2556 | * calling this function with non-negative @flush_color. If |
| 2557 | * @flush_color is negative, no flush color update is done and %false |
| 2558 | * is returned. |
| 2559 | * |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2560 | * If @work_color is non-negative, all pwqs should have the same |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2561 | * work_color which is previous to @work_color and all will be |
| 2562 | * advanced to @work_color. |
| 2563 | * |
| 2564 | * CONTEXT: |
Lai Jiangshan | 3c25a55 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 2565 | * mutex_lock(wq->mutex). |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2566 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 2567 | * Return: |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2568 | * %true if @flush_color >= 0 and there's something to flush. %false |
| 2569 | * otherwise. |
| 2570 | */ |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2571 | static bool flush_workqueue_prep_pwqs(struct workqueue_struct *wq, |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2572 | int flush_color, int work_color) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2573 | { |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2574 | bool wait = false; |
Tejun Heo | 49e3cf4 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 2575 | struct pool_workqueue *pwq; |
Oleg Nesterov | 1444196 | 2007-05-23 13:57:57 -0700 | [diff] [blame] | 2576 | |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2577 | if (flush_color >= 0) { |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 2578 | WARN_ON_ONCE(atomic_read(&wq->nr_pwqs_to_flush)); |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2579 | atomic_set(&wq->nr_pwqs_to_flush, 1); |
Thomas Gleixner | dc186ad | 2009-11-16 01:09:48 +0900 | [diff] [blame] | 2580 | } |
Oleg Nesterov | 1444196 | 2007-05-23 13:57:57 -0700 | [diff] [blame] | 2581 | |
Tejun Heo | 49e3cf4 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 2582 | for_each_pwq(pwq, wq) { |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2583 | struct worker_pool *pool = pwq->pool; |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2584 | |
Lai Jiangshan | b09f4fd | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 2585 | spin_lock_irq(&pool->lock); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2586 | |
| 2587 | if (flush_color >= 0) { |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 2588 | WARN_ON_ONCE(pwq->flush_color != -1); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2589 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2590 | if (pwq->nr_in_flight[flush_color]) { |
| 2591 | pwq->flush_color = flush_color; |
| 2592 | atomic_inc(&wq->nr_pwqs_to_flush); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2593 | wait = true; |
| 2594 | } |
| 2595 | } |
| 2596 | |
| 2597 | if (work_color >= 0) { |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 2598 | WARN_ON_ONCE(work_color != work_next_color(pwq->work_color)); |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2599 | pwq->work_color = work_color; |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2600 | } |
| 2601 | |
Lai Jiangshan | b09f4fd | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 2602 | spin_unlock_irq(&pool->lock); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2603 | } |
| 2604 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2605 | 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] | 2606 | complete(&wq->first_flusher->done); |
| 2607 | |
| 2608 | return wait; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2609 | } |
| 2610 | |
Rolf Eike Beer | 0fcb78c | 2006-07-30 03:03:42 -0700 | [diff] [blame] | 2611 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2612 | * flush_workqueue - ensure that any scheduled work has run to completion. |
Rolf Eike Beer | 0fcb78c | 2006-07-30 03:03:42 -0700 | [diff] [blame] | 2613 | * @wq: workqueue to flush |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2614 | * |
Tejun Heo | c5aa87b | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 2615 | * This function sleeps until all work items which were queued on entry |
| 2616 | * have finished execution, but it is not livelocked by new incoming ones. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2617 | */ |
Harvey Harrison | 7ad5b3a | 2008-02-08 04:19:53 -0800 | [diff] [blame] | 2618 | void flush_workqueue(struct workqueue_struct *wq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2619 | { |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2620 | struct wq_flusher this_flusher = { |
| 2621 | .list = LIST_HEAD_INIT(this_flusher.list), |
| 2622 | .flush_color = -1, |
| 2623 | .done = COMPLETION_INITIALIZER_ONSTACK(this_flusher.done), |
| 2624 | }; |
| 2625 | int next_color; |
Oleg Nesterov | b1f4ec1 | 2007-05-09 02:34:12 -0700 | [diff] [blame] | 2626 | |
Ingo Molnar | 3295f0e | 2008-08-11 10:30:30 +0200 | [diff] [blame] | 2627 | lock_map_acquire(&wq->lockdep_map); |
| 2628 | lock_map_release(&wq->lockdep_map); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2629 | |
Lai Jiangshan | 3c25a55 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 2630 | mutex_lock(&wq->mutex); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2631 | |
| 2632 | /* |
| 2633 | * Start-to-wait phase |
| 2634 | */ |
| 2635 | next_color = work_next_color(wq->work_color); |
| 2636 | |
| 2637 | if (next_color != wq->flush_color) { |
| 2638 | /* |
| 2639 | * Color space is not full. The current work_color |
| 2640 | * becomes our flush_color and work_color is advanced |
| 2641 | * by one. |
| 2642 | */ |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 2643 | WARN_ON_ONCE(!list_empty(&wq->flusher_overflow)); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2644 | this_flusher.flush_color = wq->work_color; |
| 2645 | wq->work_color = next_color; |
| 2646 | |
| 2647 | if (!wq->first_flusher) { |
| 2648 | /* no flush in progress, become the first flusher */ |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 2649 | WARN_ON_ONCE(wq->flush_color != this_flusher.flush_color); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2650 | |
| 2651 | wq->first_flusher = &this_flusher; |
| 2652 | |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2653 | if (!flush_workqueue_prep_pwqs(wq, wq->flush_color, |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2654 | wq->work_color)) { |
| 2655 | /* nothing to flush, done */ |
| 2656 | wq->flush_color = next_color; |
| 2657 | wq->first_flusher = NULL; |
| 2658 | goto out_unlock; |
| 2659 | } |
| 2660 | } else { |
| 2661 | /* wait in queue */ |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 2662 | WARN_ON_ONCE(wq->flush_color == this_flusher.flush_color); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2663 | list_add_tail(&this_flusher.list, &wq->flusher_queue); |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2664 | flush_workqueue_prep_pwqs(wq, -1, wq->work_color); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 2665 | } |
| 2666 | } else { |
| 2667 | /* |
| 2668 | * Oops, color space is full, wait on overflow queue. |
| 2669 | * The next flush completion will assign us |
| 2670 | * flush_color and transfer to flusher_queue. |
| 2671 | */ |
| 2672 | list_add_tail(&this_flusher.list, &wq->flusher_overflow); |
| 2673 | } |
| 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 | } |
Dave Jones | ae90dd5 | 2006-06-30 01:40:45 -0400 | [diff] [blame] | 2762 | EXPORT_SYMBOL_GPL(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 |
| 2771 | * repeatedly until it becomes empty. The number of flushing is detemined |
| 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 | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2847 | insert_wq_barrier(pwq, barr, work, worker); |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2848 | spin_unlock_irq(&pool->lock); |
Tejun Heo | baf5902 | 2010-09-16 10:42:16 +0200 | [diff] [blame] | 2849 | |
Tejun Heo | e159489 | 2011-01-09 23:32:15 +0100 | [diff] [blame] | 2850 | /* |
| 2851 | * If @max_active is 1 or rescuer is in use, flushing another work |
| 2852 | * item on the same workqueue may lead to deadlock. Make sure the |
| 2853 | * flusher is not running on the same workqueue by verifying write |
| 2854 | * access. |
| 2855 | */ |
Tejun Heo | 493008a | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 2856 | if (pwq->wq->saved_max_active == 1 || pwq->wq->rescuer) |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2857 | lock_map_acquire(&pwq->wq->lockdep_map); |
Tejun Heo | e159489 | 2011-01-09 23:32:15 +0100 | [diff] [blame] | 2858 | else |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 2859 | lock_map_acquire_read(&pwq->wq->lockdep_map); |
| 2860 | lock_map_release(&pwq->wq->lockdep_map); |
Tejun Heo | e159489 | 2011-01-09 23:32:15 +0100 | [diff] [blame] | 2861 | |
Tejun Heo | baf5902 | 2010-09-16 10:42:16 +0200 | [diff] [blame] | 2862 | return true; |
| 2863 | already_gone: |
Tejun Heo | d565ed6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2864 | spin_unlock_irq(&pool->lock); |
Tejun Heo | baf5902 | 2010-09-16 10:42:16 +0200 | [diff] [blame] | 2865 | return false; |
| 2866 | } |
| 2867 | |
Oleg Nesterov | db70089 | 2008-07-25 01:47:49 -0700 | [diff] [blame] | 2868 | /** |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2869 | * flush_work - wait for a work to finish executing the last queueing instance |
| 2870 | * @work: the work to flush |
Oleg Nesterov | db70089 | 2008-07-25 01:47:49 -0700 | [diff] [blame] | 2871 | * |
Tejun Heo | 606a502 | 2012-08-20 14:51:23 -0700 | [diff] [blame] | 2872 | * Wait until @work has finished execution. @work is guaranteed to be idle |
| 2873 | * on return if it hasn't been requeued since flush started. |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2874 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 2875 | * Return: |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2876 | * %true if flush_work() waited for the work to finish execution, |
| 2877 | * %false if it was already idle. |
Oleg Nesterov | db70089 | 2008-07-25 01:47:49 -0700 | [diff] [blame] | 2878 | */ |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2879 | bool flush_work(struct work_struct *work) |
Oleg Nesterov | db70089 | 2008-07-25 01:47:49 -0700 | [diff] [blame] | 2880 | { |
Bjorn Helgaas | 12997d1 | 2013-11-18 11:00:29 -0700 | [diff] [blame] | 2881 | struct wq_barrier barr; |
| 2882 | |
Stephen Boyd | 0976dfc | 2012-04-20 17:28:50 -0700 | [diff] [blame] | 2883 | lock_map_acquire(&work->lockdep_map); |
| 2884 | lock_map_release(&work->lockdep_map); |
| 2885 | |
Bjorn Helgaas | 12997d1 | 2013-11-18 11:00:29 -0700 | [diff] [blame] | 2886 | if (start_flush_work(work, &barr)) { |
| 2887 | wait_for_completion(&barr.done); |
| 2888 | destroy_work_on_stack(&barr.work); |
| 2889 | return true; |
| 2890 | } else { |
| 2891 | return false; |
| 2892 | } |
Oleg Nesterov | db70089 | 2008-07-25 01:47:49 -0700 | [diff] [blame] | 2893 | } |
| 2894 | EXPORT_SYMBOL_GPL(flush_work); |
| 2895 | |
Tejun Heo | 36e227d | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 2896 | static bool __cancel_work_timer(struct work_struct *work, bool is_dwork) |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2897 | { |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 2898 | unsigned long flags; |
Oleg Nesterov | 1f1f642 | 2007-07-15 23:41:44 -0700 | [diff] [blame] | 2899 | int ret; |
| 2900 | |
| 2901 | do { |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 2902 | ret = try_to_grab_pending(work, is_dwork, &flags); |
| 2903 | /* |
| 2904 | * If someone else is canceling, wait for the same event it |
| 2905 | * would be waiting for before retrying. |
| 2906 | */ |
| 2907 | if (unlikely(ret == -ENOENT)) |
Tejun Heo | 606a502 | 2012-08-20 14:51:23 -0700 | [diff] [blame] | 2908 | flush_work(work); |
Oleg Nesterov | 1f1f642 | 2007-07-15 23:41:44 -0700 | [diff] [blame] | 2909 | } while (unlikely(ret < 0)); |
| 2910 | |
Tejun Heo | bbb68df | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 2911 | /* tell other tasks trying to grab @work to back off */ |
| 2912 | mark_work_canceling(work); |
| 2913 | local_irq_restore(flags); |
| 2914 | |
Tejun Heo | 606a502 | 2012-08-20 14:51:23 -0700 | [diff] [blame] | 2915 | flush_work(work); |
Tejun Heo | 7a22ad7 | 2010-06-29 10:07:13 +0200 | [diff] [blame] | 2916 | clear_work_data(work); |
Oleg Nesterov | 1f1f642 | 2007-07-15 23:41:44 -0700 | [diff] [blame] | 2917 | return ret; |
| 2918 | } |
| 2919 | |
Oleg Nesterov | 6e84d64 | 2007-05-09 02:34:46 -0700 | [diff] [blame] | 2920 | /** |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2921 | * cancel_work_sync - cancel a work and wait for it to finish |
| 2922 | * @work: the work to cancel |
Oleg Nesterov | 6e84d64 | 2007-05-09 02:34:46 -0700 | [diff] [blame] | 2923 | * |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2924 | * Cancel @work and wait for its execution to finish. This function |
| 2925 | * can be used even if the work re-queues itself or migrates to |
| 2926 | * another workqueue. On return from this function, @work is |
| 2927 | * guaranteed to be not pending or executing on any CPU. |
Oleg Nesterov | 1f1f642 | 2007-07-15 23:41:44 -0700 | [diff] [blame] | 2928 | * |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2929 | * cancel_work_sync(&delayed_work->work) must not be used for |
| 2930 | * delayed_work's. Use cancel_delayed_work_sync() instead. |
Oleg Nesterov | 6e84d64 | 2007-05-09 02:34:46 -0700 | [diff] [blame] | 2931 | * |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2932 | * The caller must ensure that the workqueue on which @work was last |
Oleg Nesterov | 6e84d64 | 2007-05-09 02:34:46 -0700 | [diff] [blame] | 2933 | * queued can't be destroyed before this function returns. |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2934 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 2935 | * Return: |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2936 | * %true if @work was pending, %false otherwise. |
Oleg Nesterov | 6e84d64 | 2007-05-09 02:34:46 -0700 | [diff] [blame] | 2937 | */ |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2938 | bool cancel_work_sync(struct work_struct *work) |
Oleg Nesterov | 6e84d64 | 2007-05-09 02:34:46 -0700 | [diff] [blame] | 2939 | { |
Tejun Heo | 36e227d | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 2940 | return __cancel_work_timer(work, false); |
Oleg Nesterov | b89deed | 2007-05-09 02:33:52 -0700 | [diff] [blame] | 2941 | } |
Oleg Nesterov | 28e53bd | 2007-05-09 02:34:22 -0700 | [diff] [blame] | 2942 | EXPORT_SYMBOL_GPL(cancel_work_sync); |
Oleg Nesterov | b89deed | 2007-05-09 02:33:52 -0700 | [diff] [blame] | 2943 | |
Oleg Nesterov | 6e84d64 | 2007-05-09 02:34:46 -0700 | [diff] [blame] | 2944 | /** |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2945 | * flush_delayed_work - wait for a dwork to finish executing the last queueing |
| 2946 | * @dwork: the delayed work to flush |
Oleg Nesterov | 6e84d64 | 2007-05-09 02:34:46 -0700 | [diff] [blame] | 2947 | * |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2948 | * Delayed timer is cancelled and the pending work is queued for |
| 2949 | * immediate execution. Like flush_work(), this function only |
| 2950 | * considers the last queueing instance of @dwork. |
Oleg Nesterov | 1f1f642 | 2007-07-15 23:41:44 -0700 | [diff] [blame] | 2951 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 2952 | * Return: |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2953 | * %true if flush_work() waited for the work to finish execution, |
| 2954 | * %false if it was already idle. |
Oleg Nesterov | 6e84d64 | 2007-05-09 02:34:46 -0700 | [diff] [blame] | 2955 | */ |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2956 | bool flush_delayed_work(struct delayed_work *dwork) |
| 2957 | { |
Tejun Heo | 8930cab | 2012-08-03 10:30:45 -0700 | [diff] [blame] | 2958 | local_irq_disable(); |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2959 | if (del_timer_sync(&dwork->timer)) |
Lai Jiangshan | 60c057b | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 2960 | __queue_work(dwork->cpu, dwork->wq, &dwork->work); |
Tejun Heo | 8930cab | 2012-08-03 10:30:45 -0700 | [diff] [blame] | 2961 | local_irq_enable(); |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 2962 | return flush_work(&dwork->work); |
| 2963 | } |
| 2964 | EXPORT_SYMBOL(flush_delayed_work); |
| 2965 | |
| 2966 | /** |
Tejun Heo | 57b30ae | 2012-08-21 13:18:24 -0700 | [diff] [blame] | 2967 | * cancel_delayed_work - cancel a delayed work |
| 2968 | * @dwork: delayed_work to cancel |
Tejun Heo | 0938349 | 2010-09-16 10:48:29 +0200 | [diff] [blame] | 2969 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 2970 | * Kill off a pending delayed_work. |
| 2971 | * |
| 2972 | * Return: %true if @dwork was pending and canceled; %false if it wasn't |
| 2973 | * pending. |
| 2974 | * |
| 2975 | * Note: |
| 2976 | * The work callback function may still be running on return, unless |
| 2977 | * it returns %true and the work doesn't re-arm itself. Explicitly flush or |
| 2978 | * use cancel_delayed_work_sync() to wait on it. |
Tejun Heo | 0938349 | 2010-09-16 10:48:29 +0200 | [diff] [blame] | 2979 | * |
Tejun Heo | 57b30ae | 2012-08-21 13:18:24 -0700 | [diff] [blame] | 2980 | * This function is safe to call from any context including IRQ handler. |
Tejun Heo | 0938349 | 2010-09-16 10:48:29 +0200 | [diff] [blame] | 2981 | */ |
Tejun Heo | 57b30ae | 2012-08-21 13:18:24 -0700 | [diff] [blame] | 2982 | bool cancel_delayed_work(struct delayed_work *dwork) |
Tejun Heo | 0938349 | 2010-09-16 10:48:29 +0200 | [diff] [blame] | 2983 | { |
Tejun Heo | 57b30ae | 2012-08-21 13:18:24 -0700 | [diff] [blame] | 2984 | unsigned long flags; |
| 2985 | int ret; |
| 2986 | |
| 2987 | do { |
| 2988 | ret = try_to_grab_pending(&dwork->work, true, &flags); |
| 2989 | } while (unlikely(ret == -EAGAIN)); |
| 2990 | |
| 2991 | if (unlikely(ret < 0)) |
| 2992 | return false; |
| 2993 | |
Tejun Heo | 7c3eed5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 2994 | set_work_pool_and_clear_pending(&dwork->work, |
| 2995 | get_work_pool_id(&dwork->work)); |
Tejun Heo | 57b30ae | 2012-08-21 13:18:24 -0700 | [diff] [blame] | 2996 | local_irq_restore(flags); |
Dan Magenheimer | c0158ca | 2012-10-18 16:31:37 -0700 | [diff] [blame] | 2997 | return ret; |
Tejun Heo | 0938349 | 2010-09-16 10:48:29 +0200 | [diff] [blame] | 2998 | } |
Tejun Heo | 57b30ae | 2012-08-21 13:18:24 -0700 | [diff] [blame] | 2999 | EXPORT_SYMBOL(cancel_delayed_work); |
Tejun Heo | 0938349 | 2010-09-16 10:48:29 +0200 | [diff] [blame] | 3000 | |
| 3001 | /** |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 3002 | * cancel_delayed_work_sync - cancel a delayed work and wait for it to finish |
| 3003 | * @dwork: the delayed work cancel |
| 3004 | * |
| 3005 | * This is cancel_work_sync() for delayed works. |
| 3006 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 3007 | * Return: |
Tejun Heo | 401a8d0 | 2010-09-16 10:36:00 +0200 | [diff] [blame] | 3008 | * %true if @dwork was pending, %false otherwise. |
| 3009 | */ |
| 3010 | bool cancel_delayed_work_sync(struct delayed_work *dwork) |
Oleg Nesterov | 6e84d64 | 2007-05-09 02:34:46 -0700 | [diff] [blame] | 3011 | { |
Tejun Heo | 36e227d | 2012-08-03 10:30:46 -0700 | [diff] [blame] | 3012 | return __cancel_work_timer(&dwork->work, true); |
Oleg Nesterov | 6e84d64 | 2007-05-09 02:34:46 -0700 | [diff] [blame] | 3013 | } |
Oleg Nesterov | f5a421a | 2007-07-15 23:41:44 -0700 | [diff] [blame] | 3014 | EXPORT_SYMBOL(cancel_delayed_work_sync); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3015 | |
Rolf Eike Beer | 0fcb78c | 2006-07-30 03:03:42 -0700 | [diff] [blame] | 3016 | /** |
Tejun Heo | 31ddd87 | 2010-10-19 11:14:49 +0200 | [diff] [blame] | 3017 | * schedule_on_each_cpu - execute a function synchronously on each online CPU |
Andrew Morton | b613677 | 2006-06-25 05:47:49 -0700 | [diff] [blame] | 3018 | * @func: the function to call |
Andrew Morton | b613677 | 2006-06-25 05:47:49 -0700 | [diff] [blame] | 3019 | * |
Tejun Heo | 31ddd87 | 2010-10-19 11:14:49 +0200 | [diff] [blame] | 3020 | * schedule_on_each_cpu() executes @func on each online CPU using the |
| 3021 | * system workqueue and blocks until all CPUs have completed. |
Andrew Morton | b613677 | 2006-06-25 05:47:49 -0700 | [diff] [blame] | 3022 | * schedule_on_each_cpu() is very slow. |
Tejun Heo | 31ddd87 | 2010-10-19 11:14:49 +0200 | [diff] [blame] | 3023 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 3024 | * Return: |
Tejun Heo | 31ddd87 | 2010-10-19 11:14:49 +0200 | [diff] [blame] | 3025 | * 0 on success, -errno on failure. |
Andrew Morton | b613677 | 2006-06-25 05:47:49 -0700 | [diff] [blame] | 3026 | */ |
David Howells | 65f27f3 | 2006-11-22 14:55:48 +0000 | [diff] [blame] | 3027 | int schedule_on_each_cpu(work_func_t func) |
Christoph Lameter | 15316ba | 2006-01-08 01:00:43 -0800 | [diff] [blame] | 3028 | { |
| 3029 | int cpu; |
Namhyung Kim | 38f5156 | 2010-08-08 14:24:09 +0200 | [diff] [blame] | 3030 | struct work_struct __percpu *works; |
Christoph Lameter | 15316ba | 2006-01-08 01:00:43 -0800 | [diff] [blame] | 3031 | |
Andrew Morton | b613677 | 2006-06-25 05:47:49 -0700 | [diff] [blame] | 3032 | works = alloc_percpu(struct work_struct); |
| 3033 | if (!works) |
Christoph Lameter | 15316ba | 2006-01-08 01:00:43 -0800 | [diff] [blame] | 3034 | return -ENOMEM; |
Andrew Morton | b613677 | 2006-06-25 05:47:49 -0700 | [diff] [blame] | 3035 | |
Gautham R Shenoy | 95402b3 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 3036 | get_online_cpus(); |
Tejun Heo | 9398180 | 2009-11-17 14:06:20 -0800 | [diff] [blame] | 3037 | |
Christoph Lameter | 15316ba | 2006-01-08 01:00:43 -0800 | [diff] [blame] | 3038 | for_each_online_cpu(cpu) { |
Ingo Molnar | 9bfb183 | 2006-12-18 20:05:09 +0100 | [diff] [blame] | 3039 | struct work_struct *work = per_cpu_ptr(works, cpu); |
| 3040 | |
| 3041 | INIT_WORK(work, func); |
Tejun Heo | b71ab8c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 3042 | schedule_work_on(cpu, work); |
Andi Kleen | 65a6446 | 2009-10-14 06:22:47 +0200 | [diff] [blame] | 3043 | } |
Tejun Heo | 9398180 | 2009-11-17 14:06:20 -0800 | [diff] [blame] | 3044 | |
| 3045 | for_each_online_cpu(cpu) |
| 3046 | flush_work(per_cpu_ptr(works, cpu)); |
| 3047 | |
Gautham R Shenoy | 95402b3 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 3048 | put_online_cpus(); |
Andrew Morton | b613677 | 2006-06-25 05:47:49 -0700 | [diff] [blame] | 3049 | free_percpu(works); |
Christoph Lameter | 15316ba | 2006-01-08 01:00:43 -0800 | [diff] [blame] | 3050 | return 0; |
| 3051 | } |
| 3052 | |
Alan Stern | eef6a7d | 2010-02-12 17:39:21 +0900 | [diff] [blame] | 3053 | /** |
| 3054 | * flush_scheduled_work - ensure that any scheduled work has run to completion. |
| 3055 | * |
| 3056 | * Forces execution of the kernel-global workqueue and blocks until its |
| 3057 | * completion. |
| 3058 | * |
| 3059 | * Think twice before calling this function! It's very easy to get into |
| 3060 | * trouble if you don't take great care. Either of the following situations |
| 3061 | * will lead to deadlock: |
| 3062 | * |
| 3063 | * One of the work items currently on the workqueue needs to acquire |
| 3064 | * a lock held by your code or its caller. |
| 3065 | * |
| 3066 | * Your code is running in the context of a work routine. |
| 3067 | * |
| 3068 | * They will be detected by lockdep when they occur, but the first might not |
| 3069 | * occur very often. It depends on what work items are on the workqueue and |
| 3070 | * what locks they need, which you have no control over. |
| 3071 | * |
| 3072 | * In most situations flushing the entire workqueue is overkill; you merely |
| 3073 | * need to know that a particular work item isn't queued and isn't running. |
| 3074 | * In such cases you should use cancel_delayed_work_sync() or |
| 3075 | * cancel_work_sync() instead. |
| 3076 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3077 | void flush_scheduled_work(void) |
| 3078 | { |
Tejun Heo | d320c03 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 3079 | flush_workqueue(system_wq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3080 | } |
Dave Jones | ae90dd5 | 2006-06-30 01:40:45 -0400 | [diff] [blame] | 3081 | EXPORT_SYMBOL(flush_scheduled_work); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3082 | |
| 3083 | /** |
James Bottomley | 1fa44ec | 2006-02-23 12:43:43 -0600 | [diff] [blame] | 3084 | * execute_in_process_context - reliably execute the routine with user context |
| 3085 | * @fn: the function to execute |
James Bottomley | 1fa44ec | 2006-02-23 12:43:43 -0600 | [diff] [blame] | 3086 | * @ew: guaranteed storage for the execute work structure (must |
| 3087 | * be available when the work executes) |
| 3088 | * |
| 3089 | * Executes the function immediately if process context is available, |
| 3090 | * otherwise schedules the function for delayed execution. |
| 3091 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 3092 | * Return: 0 - function was executed |
James Bottomley | 1fa44ec | 2006-02-23 12:43:43 -0600 | [diff] [blame] | 3093 | * 1 - function was scheduled for execution |
| 3094 | */ |
David Howells | 65f27f3 | 2006-11-22 14:55:48 +0000 | [diff] [blame] | 3095 | 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] | 3096 | { |
| 3097 | if (!in_interrupt()) { |
David Howells | 65f27f3 | 2006-11-22 14:55:48 +0000 | [diff] [blame] | 3098 | fn(&ew->work); |
James Bottomley | 1fa44ec | 2006-02-23 12:43:43 -0600 | [diff] [blame] | 3099 | return 0; |
| 3100 | } |
| 3101 | |
David Howells | 65f27f3 | 2006-11-22 14:55:48 +0000 | [diff] [blame] | 3102 | INIT_WORK(&ew->work, fn); |
James Bottomley | 1fa44ec | 2006-02-23 12:43:43 -0600 | [diff] [blame] | 3103 | schedule_work(&ew->work); |
| 3104 | |
| 3105 | return 1; |
| 3106 | } |
| 3107 | EXPORT_SYMBOL_GPL(execute_in_process_context); |
| 3108 | |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 3109 | #ifdef CONFIG_SYSFS |
| 3110 | /* |
| 3111 | * Workqueues with WQ_SYSFS flag set is visible to userland via |
| 3112 | * /sys/bus/workqueue/devices/WQ_NAME. All visible workqueues have the |
| 3113 | * following attributes. |
| 3114 | * |
| 3115 | * per_cpu RO bool : whether the workqueue is per-cpu or unbound |
| 3116 | * max_active RW int : maximum number of in-flight work items |
| 3117 | * |
| 3118 | * Unbound workqueues have the following extra attributes. |
| 3119 | * |
| 3120 | * id RO int : the associated pool ID |
| 3121 | * nice RW int : nice value of the workers |
| 3122 | * cpumask RW mask : bitmask of allowed CPUs for the workers |
| 3123 | */ |
| 3124 | struct wq_device { |
| 3125 | struct workqueue_struct *wq; |
| 3126 | struct device dev; |
| 3127 | }; |
| 3128 | |
| 3129 | static struct workqueue_struct *dev_to_wq(struct device *dev) |
| 3130 | { |
| 3131 | struct wq_device *wq_dev = container_of(dev, struct wq_device, dev); |
| 3132 | |
| 3133 | return wq_dev->wq; |
| 3134 | } |
| 3135 | |
Greg Kroah-Hartman | 1a6661d | 2013-08-23 14:24:41 -0700 | [diff] [blame] | 3136 | static ssize_t per_cpu_show(struct device *dev, struct device_attribute *attr, |
| 3137 | char *buf) |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 3138 | { |
| 3139 | struct workqueue_struct *wq = dev_to_wq(dev); |
| 3140 | |
| 3141 | return scnprintf(buf, PAGE_SIZE, "%d\n", (bool)!(wq->flags & WQ_UNBOUND)); |
| 3142 | } |
Greg Kroah-Hartman | 1a6661d | 2013-08-23 14:24:41 -0700 | [diff] [blame] | 3143 | static DEVICE_ATTR_RO(per_cpu); |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 3144 | |
Greg Kroah-Hartman | 1a6661d | 2013-08-23 14:24:41 -0700 | [diff] [blame] | 3145 | static ssize_t max_active_show(struct device *dev, |
| 3146 | struct device_attribute *attr, char *buf) |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 3147 | { |
| 3148 | struct workqueue_struct *wq = dev_to_wq(dev); |
| 3149 | |
| 3150 | return scnprintf(buf, PAGE_SIZE, "%d\n", wq->saved_max_active); |
| 3151 | } |
| 3152 | |
Greg Kroah-Hartman | 1a6661d | 2013-08-23 14:24:41 -0700 | [diff] [blame] | 3153 | static ssize_t max_active_store(struct device *dev, |
| 3154 | struct device_attribute *attr, const char *buf, |
| 3155 | size_t count) |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 3156 | { |
| 3157 | struct workqueue_struct *wq = dev_to_wq(dev); |
| 3158 | int val; |
| 3159 | |
| 3160 | if (sscanf(buf, "%d", &val) != 1 || val <= 0) |
| 3161 | return -EINVAL; |
| 3162 | |
| 3163 | workqueue_set_max_active(wq, val); |
| 3164 | return count; |
| 3165 | } |
Greg Kroah-Hartman | 1a6661d | 2013-08-23 14:24:41 -0700 | [diff] [blame] | 3166 | static DEVICE_ATTR_RW(max_active); |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 3167 | |
Greg Kroah-Hartman | 1a6661d | 2013-08-23 14:24:41 -0700 | [diff] [blame] | 3168 | static struct attribute *wq_sysfs_attrs[] = { |
| 3169 | &dev_attr_per_cpu.attr, |
| 3170 | &dev_attr_max_active.attr, |
| 3171 | NULL, |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 3172 | }; |
Greg Kroah-Hartman | 1a6661d | 2013-08-23 14:24:41 -0700 | [diff] [blame] | 3173 | ATTRIBUTE_GROUPS(wq_sysfs); |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 3174 | |
Tejun Heo | d55262c | 2013-04-01 11:23:38 -0700 | [diff] [blame] | 3175 | static ssize_t wq_pool_ids_show(struct device *dev, |
| 3176 | struct device_attribute *attr, char *buf) |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 3177 | { |
| 3178 | struct workqueue_struct *wq = dev_to_wq(dev); |
Tejun Heo | d55262c | 2013-04-01 11:23:38 -0700 | [diff] [blame] | 3179 | const char *delim = ""; |
| 3180 | int node, written = 0; |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 3181 | |
| 3182 | rcu_read_lock_sched(); |
Tejun Heo | d55262c | 2013-04-01 11:23:38 -0700 | [diff] [blame] | 3183 | for_each_node(node) { |
| 3184 | written += scnprintf(buf + written, PAGE_SIZE - written, |
| 3185 | "%s%d:%d", delim, node, |
| 3186 | unbound_pwq_by_node(wq, node)->pool->id); |
| 3187 | delim = " "; |
| 3188 | } |
| 3189 | written += scnprintf(buf + written, PAGE_SIZE - written, "\n"); |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 3190 | rcu_read_unlock_sched(); |
| 3191 | |
| 3192 | return written; |
| 3193 | } |
| 3194 | |
| 3195 | static ssize_t wq_nice_show(struct device *dev, struct device_attribute *attr, |
| 3196 | char *buf) |
| 3197 | { |
| 3198 | struct workqueue_struct *wq = dev_to_wq(dev); |
| 3199 | int written; |
| 3200 | |
Tejun Heo | 6029a91 | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 3201 | mutex_lock(&wq->mutex); |
| 3202 | written = scnprintf(buf, PAGE_SIZE, "%d\n", wq->unbound_attrs->nice); |
| 3203 | mutex_unlock(&wq->mutex); |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 3204 | |
| 3205 | return written; |
| 3206 | } |
| 3207 | |
| 3208 | /* prepare workqueue_attrs for sysfs store operations */ |
| 3209 | static struct workqueue_attrs *wq_sysfs_prep_attrs(struct workqueue_struct *wq) |
| 3210 | { |
| 3211 | struct workqueue_attrs *attrs; |
| 3212 | |
| 3213 | attrs = alloc_workqueue_attrs(GFP_KERNEL); |
| 3214 | if (!attrs) |
| 3215 | return NULL; |
| 3216 | |
Tejun Heo | 6029a91 | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 3217 | mutex_lock(&wq->mutex); |
| 3218 | copy_workqueue_attrs(attrs, wq->unbound_attrs); |
| 3219 | mutex_unlock(&wq->mutex); |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 3220 | return attrs; |
| 3221 | } |
| 3222 | |
| 3223 | static ssize_t wq_nice_store(struct device *dev, struct device_attribute *attr, |
| 3224 | const char *buf, size_t count) |
| 3225 | { |
| 3226 | struct workqueue_struct *wq = dev_to_wq(dev); |
| 3227 | struct workqueue_attrs *attrs; |
| 3228 | int ret; |
| 3229 | |
| 3230 | attrs = wq_sysfs_prep_attrs(wq); |
| 3231 | if (!attrs) |
| 3232 | return -ENOMEM; |
| 3233 | |
| 3234 | if (sscanf(buf, "%d", &attrs->nice) == 1 && |
Dongsheng Yang | 1448184 | 2014-02-11 15:34:52 +0800 | [diff] [blame] | 3235 | attrs->nice >= MIN_NICE && attrs->nice <= MAX_NICE) |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 3236 | ret = apply_workqueue_attrs(wq, attrs); |
| 3237 | else |
| 3238 | ret = -EINVAL; |
| 3239 | |
| 3240 | free_workqueue_attrs(attrs); |
| 3241 | return ret ?: count; |
| 3242 | } |
| 3243 | |
| 3244 | static ssize_t wq_cpumask_show(struct device *dev, |
| 3245 | struct device_attribute *attr, char *buf) |
| 3246 | { |
| 3247 | struct workqueue_struct *wq = dev_to_wq(dev); |
| 3248 | int written; |
| 3249 | |
Tejun Heo | 6029a91 | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 3250 | mutex_lock(&wq->mutex); |
| 3251 | written = cpumask_scnprintf(buf, PAGE_SIZE, wq->unbound_attrs->cpumask); |
| 3252 | mutex_unlock(&wq->mutex); |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 3253 | |
| 3254 | written += scnprintf(buf + written, PAGE_SIZE - written, "\n"); |
| 3255 | return written; |
| 3256 | } |
| 3257 | |
| 3258 | static ssize_t wq_cpumask_store(struct device *dev, |
| 3259 | struct device_attribute *attr, |
| 3260 | const char *buf, size_t count) |
| 3261 | { |
| 3262 | struct workqueue_struct *wq = dev_to_wq(dev); |
| 3263 | struct workqueue_attrs *attrs; |
| 3264 | int ret; |
| 3265 | |
| 3266 | attrs = wq_sysfs_prep_attrs(wq); |
| 3267 | if (!attrs) |
| 3268 | return -ENOMEM; |
| 3269 | |
| 3270 | ret = cpumask_parse(buf, attrs->cpumask); |
| 3271 | if (!ret) |
| 3272 | ret = apply_workqueue_attrs(wq, attrs); |
| 3273 | |
| 3274 | free_workqueue_attrs(attrs); |
| 3275 | return ret ?: count; |
| 3276 | } |
| 3277 | |
Tejun Heo | d55262c | 2013-04-01 11:23:38 -0700 | [diff] [blame] | 3278 | static ssize_t wq_numa_show(struct device *dev, struct device_attribute *attr, |
| 3279 | char *buf) |
| 3280 | { |
| 3281 | struct workqueue_struct *wq = dev_to_wq(dev); |
| 3282 | int written; |
| 3283 | |
| 3284 | mutex_lock(&wq->mutex); |
| 3285 | written = scnprintf(buf, PAGE_SIZE, "%d\n", |
| 3286 | !wq->unbound_attrs->no_numa); |
| 3287 | mutex_unlock(&wq->mutex); |
| 3288 | |
| 3289 | return written; |
| 3290 | } |
| 3291 | |
| 3292 | static ssize_t wq_numa_store(struct device *dev, struct device_attribute *attr, |
| 3293 | const char *buf, size_t count) |
| 3294 | { |
| 3295 | struct workqueue_struct *wq = dev_to_wq(dev); |
| 3296 | struct workqueue_attrs *attrs; |
| 3297 | int v, ret; |
| 3298 | |
| 3299 | attrs = wq_sysfs_prep_attrs(wq); |
| 3300 | if (!attrs) |
| 3301 | return -ENOMEM; |
| 3302 | |
| 3303 | ret = -EINVAL; |
| 3304 | if (sscanf(buf, "%d", &v) == 1) { |
| 3305 | attrs->no_numa = !v; |
| 3306 | ret = apply_workqueue_attrs(wq, attrs); |
| 3307 | } |
| 3308 | |
| 3309 | free_workqueue_attrs(attrs); |
| 3310 | return ret ?: count; |
| 3311 | } |
| 3312 | |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 3313 | static struct device_attribute wq_sysfs_unbound_attrs[] = { |
Tejun Heo | d55262c | 2013-04-01 11:23:38 -0700 | [diff] [blame] | 3314 | __ATTR(pool_ids, 0444, wq_pool_ids_show, NULL), |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 3315 | __ATTR(nice, 0644, wq_nice_show, wq_nice_store), |
| 3316 | __ATTR(cpumask, 0644, wq_cpumask_show, wq_cpumask_store), |
Tejun Heo | d55262c | 2013-04-01 11:23:38 -0700 | [diff] [blame] | 3317 | __ATTR(numa, 0644, wq_numa_show, wq_numa_store), |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 3318 | __ATTR_NULL, |
| 3319 | }; |
| 3320 | |
| 3321 | static struct bus_type wq_subsys = { |
| 3322 | .name = "workqueue", |
Greg Kroah-Hartman | 1a6661d | 2013-08-23 14:24:41 -0700 | [diff] [blame] | 3323 | .dev_groups = wq_sysfs_groups, |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 3324 | }; |
| 3325 | |
| 3326 | static int __init wq_sysfs_init(void) |
| 3327 | { |
| 3328 | return subsys_virtual_register(&wq_subsys, NULL); |
| 3329 | } |
| 3330 | core_initcall(wq_sysfs_init); |
| 3331 | |
| 3332 | static void wq_device_release(struct device *dev) |
| 3333 | { |
| 3334 | struct wq_device *wq_dev = container_of(dev, struct wq_device, dev); |
| 3335 | |
| 3336 | kfree(wq_dev); |
| 3337 | } |
| 3338 | |
| 3339 | /** |
| 3340 | * workqueue_sysfs_register - make a workqueue visible in sysfs |
| 3341 | * @wq: the workqueue to register |
| 3342 | * |
| 3343 | * Expose @wq in sysfs under /sys/bus/workqueue/devices. |
| 3344 | * alloc_workqueue*() automatically calls this function if WQ_SYSFS is set |
| 3345 | * which is the preferred method. |
| 3346 | * |
| 3347 | * Workqueue user should use this function directly iff it wants to apply |
| 3348 | * workqueue_attrs before making the workqueue visible in sysfs; otherwise, |
| 3349 | * apply_workqueue_attrs() may race against userland updating the |
| 3350 | * attributes. |
| 3351 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 3352 | * Return: 0 on success, -errno on failure. |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 3353 | */ |
| 3354 | int workqueue_sysfs_register(struct workqueue_struct *wq) |
| 3355 | { |
| 3356 | struct wq_device *wq_dev; |
| 3357 | int ret; |
| 3358 | |
| 3359 | /* |
| 3360 | * Adjusting max_active or creating new pwqs by applyting |
| 3361 | * attributes breaks ordering guarantee. Disallow exposing ordered |
| 3362 | * workqueues. |
| 3363 | */ |
| 3364 | if (WARN_ON(wq->flags & __WQ_ORDERED)) |
| 3365 | return -EINVAL; |
| 3366 | |
| 3367 | wq->wq_dev = wq_dev = kzalloc(sizeof(*wq_dev), GFP_KERNEL); |
| 3368 | if (!wq_dev) |
| 3369 | return -ENOMEM; |
| 3370 | |
| 3371 | wq_dev->wq = wq; |
| 3372 | wq_dev->dev.bus = &wq_subsys; |
| 3373 | wq_dev->dev.init_name = wq->name; |
| 3374 | wq_dev->dev.release = wq_device_release; |
| 3375 | |
| 3376 | /* |
| 3377 | * unbound_attrs are created separately. Suppress uevent until |
| 3378 | * everything is ready. |
| 3379 | */ |
| 3380 | dev_set_uevent_suppress(&wq_dev->dev, true); |
| 3381 | |
| 3382 | ret = device_register(&wq_dev->dev); |
| 3383 | if (ret) { |
| 3384 | kfree(wq_dev); |
| 3385 | wq->wq_dev = NULL; |
| 3386 | return ret; |
| 3387 | } |
| 3388 | |
| 3389 | if (wq->flags & WQ_UNBOUND) { |
| 3390 | struct device_attribute *attr; |
| 3391 | |
| 3392 | for (attr = wq_sysfs_unbound_attrs; attr->attr.name; attr++) { |
| 3393 | ret = device_create_file(&wq_dev->dev, attr); |
| 3394 | if (ret) { |
| 3395 | device_unregister(&wq_dev->dev); |
| 3396 | wq->wq_dev = NULL; |
| 3397 | return ret; |
| 3398 | } |
| 3399 | } |
| 3400 | } |
| 3401 | |
| 3402 | kobject_uevent(&wq_dev->dev.kobj, KOBJ_ADD); |
| 3403 | return 0; |
| 3404 | } |
| 3405 | |
| 3406 | /** |
| 3407 | * workqueue_sysfs_unregister - undo workqueue_sysfs_register() |
| 3408 | * @wq: the workqueue to unregister |
| 3409 | * |
| 3410 | * If @wq is registered to sysfs by workqueue_sysfs_register(), unregister. |
| 3411 | */ |
| 3412 | static void workqueue_sysfs_unregister(struct workqueue_struct *wq) |
| 3413 | { |
| 3414 | struct wq_device *wq_dev = wq->wq_dev; |
| 3415 | |
| 3416 | if (!wq->wq_dev) |
| 3417 | return; |
| 3418 | |
| 3419 | wq->wq_dev = NULL; |
| 3420 | device_unregister(&wq_dev->dev); |
| 3421 | } |
| 3422 | #else /* CONFIG_SYSFS */ |
| 3423 | static void workqueue_sysfs_unregister(struct workqueue_struct *wq) { } |
| 3424 | #endif /* CONFIG_SYSFS */ |
| 3425 | |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 3426 | /** |
| 3427 | * free_workqueue_attrs - free a workqueue_attrs |
| 3428 | * @attrs: workqueue_attrs to free |
| 3429 | * |
| 3430 | * Undo alloc_workqueue_attrs(). |
| 3431 | */ |
| 3432 | void free_workqueue_attrs(struct workqueue_attrs *attrs) |
| 3433 | { |
| 3434 | if (attrs) { |
| 3435 | free_cpumask_var(attrs->cpumask); |
| 3436 | kfree(attrs); |
| 3437 | } |
| 3438 | } |
| 3439 | |
| 3440 | /** |
| 3441 | * alloc_workqueue_attrs - allocate a workqueue_attrs |
| 3442 | * @gfp_mask: allocation mask to use |
| 3443 | * |
| 3444 | * Allocate a new workqueue_attrs, initialize with default settings and |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 3445 | * return it. |
| 3446 | * |
| 3447 | * Return: The allocated new workqueue_attr on success. %NULL on failure. |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 3448 | */ |
| 3449 | struct workqueue_attrs *alloc_workqueue_attrs(gfp_t gfp_mask) |
| 3450 | { |
| 3451 | struct workqueue_attrs *attrs; |
| 3452 | |
| 3453 | attrs = kzalloc(sizeof(*attrs), gfp_mask); |
| 3454 | if (!attrs) |
| 3455 | goto fail; |
| 3456 | if (!alloc_cpumask_var(&attrs->cpumask, gfp_mask)) |
| 3457 | goto fail; |
| 3458 | |
Tejun Heo | 13e2e55 | 2013-04-01 11:23:31 -0700 | [diff] [blame] | 3459 | cpumask_copy(attrs->cpumask, cpu_possible_mask); |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 3460 | return attrs; |
| 3461 | fail: |
| 3462 | free_workqueue_attrs(attrs); |
| 3463 | return NULL; |
| 3464 | } |
| 3465 | |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3466 | static void copy_workqueue_attrs(struct workqueue_attrs *to, |
| 3467 | const struct workqueue_attrs *from) |
| 3468 | { |
| 3469 | to->nice = from->nice; |
| 3470 | cpumask_copy(to->cpumask, from->cpumask); |
Shaohua Li | 2865a8f | 2013-08-01 09:56:36 +0800 | [diff] [blame] | 3471 | /* |
| 3472 | * Unlike hash and equality test, this function doesn't ignore |
| 3473 | * ->no_numa as it is used for both pool and wq attrs. Instead, |
| 3474 | * get_unbound_pool() explicitly clears ->no_numa after copying. |
| 3475 | */ |
| 3476 | to->no_numa = from->no_numa; |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3477 | } |
| 3478 | |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3479 | /* hash value of the content of @attr */ |
| 3480 | static u32 wqattrs_hash(const struct workqueue_attrs *attrs) |
| 3481 | { |
| 3482 | u32 hash = 0; |
| 3483 | |
| 3484 | hash = jhash_1word(attrs->nice, hash); |
Tejun Heo | 13e2e55 | 2013-04-01 11:23:31 -0700 | [diff] [blame] | 3485 | hash = jhash(cpumask_bits(attrs->cpumask), |
| 3486 | BITS_TO_LONGS(nr_cpumask_bits) * sizeof(long), hash); |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3487 | return hash; |
| 3488 | } |
| 3489 | |
| 3490 | /* content equality test */ |
| 3491 | static bool wqattrs_equal(const struct workqueue_attrs *a, |
| 3492 | const struct workqueue_attrs *b) |
| 3493 | { |
| 3494 | if (a->nice != b->nice) |
| 3495 | return false; |
| 3496 | if (!cpumask_equal(a->cpumask, b->cpumask)) |
| 3497 | return false; |
| 3498 | return true; |
| 3499 | } |
| 3500 | |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 3501 | /** |
| 3502 | * init_worker_pool - initialize a newly zalloc'd worker_pool |
| 3503 | * @pool: worker_pool to initialize |
| 3504 | * |
| 3505 | * Initiailize a newly zalloc'd @pool. It also allocates @pool->attrs. |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 3506 | * |
| 3507 | * Return: 0 on success, -errno on failure. Even on failure, all fields |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3508 | * inside @pool proper are initialized and put_unbound_pool() can be called |
| 3509 | * on @pool safely to release it. |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 3510 | */ |
| 3511 | static int init_worker_pool(struct worker_pool *pool) |
Tejun Heo | 4e1a1f9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 3512 | { |
| 3513 | spin_lock_init(&pool->lock); |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3514 | pool->id = -1; |
| 3515 | pool->cpu = -1; |
Tejun Heo | f3f90ad | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 3516 | pool->node = NUMA_NO_NODE; |
Tejun Heo | 4e1a1f9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 3517 | pool->flags |= POOL_DISASSOCIATED; |
| 3518 | INIT_LIST_HEAD(&pool->worklist); |
| 3519 | INIT_LIST_HEAD(&pool->idle_list); |
| 3520 | hash_init(pool->busy_hash); |
| 3521 | |
| 3522 | init_timer_deferrable(&pool->idle_timer); |
| 3523 | pool->idle_timer.function = idle_worker_timeout; |
| 3524 | pool->idle_timer.data = (unsigned long)pool; |
| 3525 | |
| 3526 | setup_timer(&pool->mayday_timer, pool_mayday_timeout, |
| 3527 | (unsigned long)pool); |
| 3528 | |
| 3529 | mutex_init(&pool->manager_arb); |
Tejun Heo | bc3a1af | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 3530 | mutex_init(&pool->manager_mutex); |
Tejun Heo | 822d840 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 3531 | idr_init(&pool->worker_idr); |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 3532 | |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3533 | INIT_HLIST_NODE(&pool->hash_node); |
| 3534 | pool->refcnt = 1; |
| 3535 | |
| 3536 | /* shouldn't fail above this point */ |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 3537 | pool->attrs = alloc_workqueue_attrs(GFP_KERNEL); |
| 3538 | if (!pool->attrs) |
| 3539 | return -ENOMEM; |
| 3540 | return 0; |
Tejun Heo | 4e1a1f9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 3541 | } |
| 3542 | |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3543 | static void rcu_free_pool(struct rcu_head *rcu) |
| 3544 | { |
| 3545 | struct worker_pool *pool = container_of(rcu, struct worker_pool, rcu); |
| 3546 | |
Tejun Heo | 822d840 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 3547 | idr_destroy(&pool->worker_idr); |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3548 | free_workqueue_attrs(pool->attrs); |
| 3549 | kfree(pool); |
| 3550 | } |
| 3551 | |
| 3552 | /** |
| 3553 | * put_unbound_pool - put a worker_pool |
| 3554 | * @pool: worker_pool to put |
| 3555 | * |
| 3556 | * 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] | 3557 | * safe manner. get_unbound_pool() calls this function on its failure path |
| 3558 | * and this function should be able to release pools which went through, |
| 3559 | * successfully or not, init_worker_pool(). |
Tejun Heo | a892cac | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 3560 | * |
| 3561 | * Should be called with wq_pool_mutex held. |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3562 | */ |
| 3563 | static void put_unbound_pool(struct worker_pool *pool) |
| 3564 | { |
| 3565 | struct worker *worker; |
| 3566 | |
Tejun Heo | a892cac | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 3567 | lockdep_assert_held(&wq_pool_mutex); |
| 3568 | |
| 3569 | if (--pool->refcnt) |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3570 | return; |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3571 | |
| 3572 | /* sanity checks */ |
| 3573 | if (WARN_ON(!(pool->flags & POOL_DISASSOCIATED)) || |
Tejun Heo | a892cac | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 3574 | WARN_ON(!list_empty(&pool->worklist))) |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3575 | return; |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3576 | |
| 3577 | /* release id and unhash */ |
| 3578 | if (pool->id >= 0) |
| 3579 | idr_remove(&worker_pool_idr, pool->id); |
| 3580 | hash_del(&pool->hash_node); |
| 3581 | |
Tejun Heo | c5aa87b | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 3582 | /* |
| 3583 | * Become the manager and destroy all workers. Grabbing |
| 3584 | * manager_arb prevents @pool's workers from blocking on |
| 3585 | * manager_mutex. |
| 3586 | */ |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3587 | mutex_lock(&pool->manager_arb); |
Tejun Heo | cd54968 | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 3588 | mutex_lock(&pool->manager_mutex); |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3589 | spin_lock_irq(&pool->lock); |
| 3590 | |
| 3591 | while ((worker = first_worker(pool))) |
| 3592 | destroy_worker(worker); |
| 3593 | WARN_ON(pool->nr_workers || pool->nr_idle); |
| 3594 | |
| 3595 | spin_unlock_irq(&pool->lock); |
Tejun Heo | cd54968 | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 3596 | mutex_unlock(&pool->manager_mutex); |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3597 | mutex_unlock(&pool->manager_arb); |
| 3598 | |
| 3599 | /* shut down the timers */ |
| 3600 | del_timer_sync(&pool->idle_timer); |
| 3601 | del_timer_sync(&pool->mayday_timer); |
| 3602 | |
| 3603 | /* sched-RCU protected to allow dereferences from get_work_pool() */ |
| 3604 | call_rcu_sched(&pool->rcu, rcu_free_pool); |
| 3605 | } |
| 3606 | |
| 3607 | /** |
| 3608 | * get_unbound_pool - get a worker_pool with the specified attributes |
| 3609 | * @attrs: the attributes of the worker_pool to get |
| 3610 | * |
| 3611 | * Obtain a worker_pool which has the same attributes as @attrs, bump the |
| 3612 | * reference count and return it. If there already is a matching |
| 3613 | * worker_pool, it will be used; otherwise, this function attempts to |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 3614 | * create a new one. |
Tejun Heo | a892cac | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 3615 | * |
| 3616 | * Should be called with wq_pool_mutex held. |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 3617 | * |
| 3618 | * Return: On success, a worker_pool with the same attributes as @attrs. |
| 3619 | * On failure, %NULL. |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3620 | */ |
| 3621 | static struct worker_pool *get_unbound_pool(const struct workqueue_attrs *attrs) |
| 3622 | { |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3623 | u32 hash = wqattrs_hash(attrs); |
| 3624 | struct worker_pool *pool; |
Tejun Heo | f3f90ad | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 3625 | int node; |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3626 | |
Tejun Heo | a892cac | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 3627 | lockdep_assert_held(&wq_pool_mutex); |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3628 | |
| 3629 | /* do we already have a matching pool? */ |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3630 | hash_for_each_possible(unbound_pool_hash, pool, hash_node, hash) { |
| 3631 | if (wqattrs_equal(pool->attrs, attrs)) { |
| 3632 | pool->refcnt++; |
| 3633 | goto out_unlock; |
| 3634 | } |
| 3635 | } |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3636 | |
| 3637 | /* nope, create a new one */ |
| 3638 | pool = kzalloc(sizeof(*pool), GFP_KERNEL); |
| 3639 | if (!pool || init_worker_pool(pool) < 0) |
| 3640 | goto fail; |
| 3641 | |
Lai Jiangshan | 12ee4fc | 2013-03-20 03:28:01 +0800 | [diff] [blame] | 3642 | if (workqueue_freezing) |
| 3643 | pool->flags |= POOL_FREEZING; |
| 3644 | |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3645 | lockdep_set_subclass(&pool->lock, 1); /* see put_pwq() */ |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3646 | copy_workqueue_attrs(pool->attrs, attrs); |
| 3647 | |
Shaohua Li | 2865a8f | 2013-08-01 09:56:36 +0800 | [diff] [blame] | 3648 | /* |
| 3649 | * no_numa isn't a worker_pool attribute, always clear it. See |
| 3650 | * 'struct workqueue_attrs' comments for detail. |
| 3651 | */ |
| 3652 | pool->attrs->no_numa = false; |
| 3653 | |
Tejun Heo | f3f90ad | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 3654 | /* if cpumask is contained inside a NUMA node, we belong to that node */ |
| 3655 | if (wq_numa_enabled) { |
| 3656 | for_each_node(node) { |
| 3657 | if (cpumask_subset(pool->attrs->cpumask, |
| 3658 | wq_numa_possible_cpumask[node])) { |
| 3659 | pool->node = node; |
| 3660 | break; |
| 3661 | } |
| 3662 | } |
| 3663 | } |
| 3664 | |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3665 | if (worker_pool_assign_id(pool) < 0) |
| 3666 | goto fail; |
| 3667 | |
| 3668 | /* create and start the initial worker */ |
Tejun Heo | ebf44d1 | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 3669 | if (create_and_start_worker(pool) < 0) |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3670 | goto fail; |
| 3671 | |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3672 | /* install */ |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3673 | hash_add(unbound_pool_hash, &pool->hash_node, hash); |
| 3674 | out_unlock: |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3675 | return pool; |
| 3676 | fail: |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 3677 | if (pool) |
| 3678 | put_unbound_pool(pool); |
| 3679 | return NULL; |
| 3680 | } |
| 3681 | |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3682 | static void rcu_free_pwq(struct rcu_head *rcu) |
| 3683 | { |
| 3684 | kmem_cache_free(pwq_cache, |
| 3685 | container_of(rcu, struct pool_workqueue, rcu)); |
| 3686 | } |
| 3687 | |
| 3688 | /* |
| 3689 | * Scheduled on system_wq by put_pwq() when an unbound pwq hits zero refcnt |
| 3690 | * and needs to be destroyed. |
| 3691 | */ |
| 3692 | static void pwq_unbound_release_workfn(struct work_struct *work) |
| 3693 | { |
| 3694 | struct pool_workqueue *pwq = container_of(work, struct pool_workqueue, |
| 3695 | unbound_release_work); |
| 3696 | struct workqueue_struct *wq = pwq->wq; |
| 3697 | struct worker_pool *pool = pwq->pool; |
Tejun Heo | bc0caf0 | 2013-04-01 11:23:31 -0700 | [diff] [blame] | 3698 | bool is_last; |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3699 | |
| 3700 | if (WARN_ON_ONCE(!(wq->flags & WQ_UNBOUND))) |
| 3701 | return; |
| 3702 | |
Tejun Heo | 75ccf59 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3703 | /* |
Lai Jiangshan | 3c25a55 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 3704 | * Unlink @pwq. Synchronization against wq->mutex isn't strictly |
Tejun Heo | 75ccf59 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3705 | * necessary on release but do it anyway. It's easier to verify |
| 3706 | * and consistent with the linking path. |
| 3707 | */ |
Lai Jiangshan | 3c25a55 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 3708 | mutex_lock(&wq->mutex); |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3709 | list_del_rcu(&pwq->pwqs_node); |
Tejun Heo | bc0caf0 | 2013-04-01 11:23:31 -0700 | [diff] [blame] | 3710 | is_last = list_empty(&wq->pwqs); |
Lai Jiangshan | 3c25a55 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 3711 | mutex_unlock(&wq->mutex); |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3712 | |
Tejun Heo | a892cac | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 3713 | mutex_lock(&wq_pool_mutex); |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3714 | put_unbound_pool(pool); |
Tejun Heo | a892cac | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 3715 | mutex_unlock(&wq_pool_mutex); |
| 3716 | |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3717 | call_rcu_sched(&pwq->rcu, rcu_free_pwq); |
| 3718 | |
| 3719 | /* |
| 3720 | * If we're the last pwq going away, @wq is already dead and no one |
| 3721 | * is gonna access it anymore. Free it. |
| 3722 | */ |
Tejun Heo | 6029a91 | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 3723 | if (is_last) { |
| 3724 | free_workqueue_attrs(wq->unbound_attrs); |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3725 | kfree(wq); |
Tejun Heo | 6029a91 | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 3726 | } |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3727 | } |
| 3728 | |
Tejun Heo | 0fbd95a | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3729 | /** |
Tejun Heo | 699ce09 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3730 | * 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] | 3731 | * @pwq: target pool_workqueue |
Tejun Heo | 0fbd95a | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3732 | * |
Tejun Heo | 699ce09 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3733 | * If @pwq isn't freezing, set @pwq->max_active to the associated |
| 3734 | * workqueue's saved_max_active and activate delayed work items |
| 3735 | * accordingly. If @pwq is freezing, clear @pwq->max_active to zero. |
Tejun Heo | 0fbd95a | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3736 | */ |
Tejun Heo | 699ce09 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3737 | static void pwq_adjust_max_active(struct pool_workqueue *pwq) |
Tejun Heo | 0fbd95a | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3738 | { |
Tejun Heo | 699ce09 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3739 | struct workqueue_struct *wq = pwq->wq; |
| 3740 | bool freezable = wq->flags & WQ_FREEZABLE; |
Tejun Heo | 0fbd95a | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3741 | |
Tejun Heo | 699ce09 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3742 | /* for @wq->saved_max_active */ |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 3743 | lockdep_assert_held(&wq->mutex); |
Tejun Heo | 699ce09 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3744 | |
| 3745 | /* fast exit for non-freezable wqs */ |
| 3746 | if (!freezable && pwq->max_active == wq->saved_max_active) |
| 3747 | return; |
| 3748 | |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 3749 | spin_lock_irq(&pwq->pool->lock); |
Tejun Heo | 699ce09 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3750 | |
| 3751 | if (!freezable || !(pwq->pool->flags & POOL_FREEZING)) { |
| 3752 | pwq->max_active = wq->saved_max_active; |
| 3753 | |
| 3754 | while (!list_empty(&pwq->delayed_works) && |
| 3755 | pwq->nr_active < pwq->max_active) |
| 3756 | pwq_activate_first_delayed(pwq); |
Lai Jiangshan | 951a078 | 2013-03-20 10:52:30 -0700 | [diff] [blame] | 3757 | |
| 3758 | /* |
| 3759 | * Need to kick a worker after thawed or an unbound wq's |
| 3760 | * max_active is bumped. It's a slow path. Do it always. |
| 3761 | */ |
| 3762 | wake_up_worker(pwq->pool); |
Tejun Heo | 699ce09 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3763 | } else { |
| 3764 | pwq->max_active = 0; |
| 3765 | } |
| 3766 | |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 3767 | spin_unlock_irq(&pwq->pool->lock); |
Tejun Heo | 0fbd95a | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3768 | } |
| 3769 | |
Tejun Heo | e50aba9 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3770 | /* initialize newly alloced @pwq which is associated with @wq and @pool */ |
Tejun Heo | f147f29 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3771 | static void init_pwq(struct pool_workqueue *pwq, struct workqueue_struct *wq, |
| 3772 | struct worker_pool *pool) |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3773 | { |
| 3774 | BUG_ON((unsigned long)pwq & WORK_STRUCT_FLAG_MASK); |
| 3775 | |
Tejun Heo | e50aba9 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3776 | memset(pwq, 0, sizeof(*pwq)); |
| 3777 | |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3778 | pwq->pool = pool; |
| 3779 | pwq->wq = wq; |
| 3780 | pwq->flush_color = -1; |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3781 | pwq->refcnt = 1; |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3782 | INIT_LIST_HEAD(&pwq->delayed_works); |
Tejun Heo | 1befcf3 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3783 | INIT_LIST_HEAD(&pwq->pwqs_node); |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3784 | INIT_LIST_HEAD(&pwq->mayday_node); |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3785 | INIT_WORK(&pwq->unbound_release_work, pwq_unbound_release_workfn); |
Tejun Heo | f147f29 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3786 | } |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3787 | |
Tejun Heo | f147f29 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3788 | /* 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] | 3789 | static void link_pwq(struct pool_workqueue *pwq) |
Tejun Heo | f147f29 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3790 | { |
| 3791 | struct workqueue_struct *wq = pwq->wq; |
| 3792 | |
| 3793 | lockdep_assert_held(&wq->mutex); |
Tejun Heo | 75ccf59 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3794 | |
Tejun Heo | 1befcf3 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3795 | /* may be called multiple times, ignore if already linked */ |
| 3796 | if (!list_empty(&pwq->pwqs_node)) |
| 3797 | return; |
| 3798 | |
Tejun Heo | 983ca25 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3799 | /* |
| 3800 | * Set the matching work_color. This is synchronized with |
Lai Jiangshan | 3c25a55 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 3801 | * wq->mutex to avoid confusing flush_workqueue(). |
Tejun Heo | 983ca25 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3802 | */ |
Tejun Heo | 75ccf59 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3803 | pwq->work_color = wq->work_color; |
Tejun Heo | 983ca25 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 3804 | |
| 3805 | /* sync max_active to the current setting */ |
| 3806 | pwq_adjust_max_active(pwq); |
| 3807 | |
| 3808 | /* link in @pwq */ |
Tejun Heo | 9e8cd2f | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3809 | list_add_rcu(&pwq->pwqs_node, &wq->pwqs); |
Tejun Heo | f147f29 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3810 | } |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 3811 | |
Tejun Heo | f147f29 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3812 | /* obtain a pool matching @attr and create a pwq associating the pool and @wq */ |
| 3813 | static struct pool_workqueue *alloc_unbound_pwq(struct workqueue_struct *wq, |
| 3814 | const struct workqueue_attrs *attrs) |
| 3815 | { |
| 3816 | struct worker_pool *pool; |
| 3817 | struct pool_workqueue *pwq; |
| 3818 | |
| 3819 | lockdep_assert_held(&wq_pool_mutex); |
| 3820 | |
| 3821 | pool = get_unbound_pool(attrs); |
| 3822 | if (!pool) |
| 3823 | return NULL; |
| 3824 | |
Tejun Heo | e50aba9 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3825 | pwq = kmem_cache_alloc_node(pwq_cache, GFP_KERNEL, pool->node); |
Tejun Heo | f147f29 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3826 | if (!pwq) { |
| 3827 | put_unbound_pool(pool); |
| 3828 | return NULL; |
Tejun Heo | df2d5ae | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3829 | } |
Tejun Heo | 6029a91 | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 3830 | |
Tejun Heo | f147f29 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3831 | init_pwq(pwq, wq, pool); |
| 3832 | return pwq; |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3833 | } |
| 3834 | |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3835 | /* undo alloc_unbound_pwq(), used only in the error path */ |
| 3836 | static void free_unbound_pwq(struct pool_workqueue *pwq) |
| 3837 | { |
| 3838 | lockdep_assert_held(&wq_pool_mutex); |
| 3839 | |
| 3840 | if (pwq) { |
| 3841 | put_unbound_pool(pwq->pool); |
Wei Yongjun | cece95d | 2013-04-09 14:29:11 +0800 | [diff] [blame] | 3842 | kmem_cache_free(pwq_cache, pwq); |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3843 | } |
| 3844 | } |
| 3845 | |
| 3846 | /** |
| 3847 | * wq_calc_node_mask - calculate a wq_attrs' cpumask for the specified node |
| 3848 | * @attrs: the wq_attrs of interest |
| 3849 | * @node: the target NUMA node |
| 3850 | * @cpu_going_down: if >= 0, the CPU to consider as offline |
| 3851 | * @cpumask: outarg, the resulting cpumask |
| 3852 | * |
| 3853 | * Calculate the cpumask a workqueue with @attrs should use on @node. If |
| 3854 | * @cpu_going_down is >= 0, that cpu is considered offline during |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 3855 | * calculation. The result is stored in @cpumask. |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3856 | * |
| 3857 | * If NUMA affinity is not enabled, @attrs->cpumask is always used. If |
| 3858 | * enabled and @node has online CPUs requested by @attrs, the returned |
| 3859 | * cpumask is the intersection of the possible CPUs of @node and |
| 3860 | * @attrs->cpumask. |
| 3861 | * |
| 3862 | * The caller is responsible for ensuring that the cpumask of @node stays |
| 3863 | * stable. |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 3864 | * |
| 3865 | * Return: %true if the resulting @cpumask is different from @attrs->cpumask, |
| 3866 | * %false if equal. |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3867 | */ |
| 3868 | static bool wq_calc_node_cpumask(const struct workqueue_attrs *attrs, int node, |
| 3869 | int cpu_going_down, cpumask_t *cpumask) |
| 3870 | { |
Tejun Heo | d55262c | 2013-04-01 11:23:38 -0700 | [diff] [blame] | 3871 | if (!wq_numa_enabled || attrs->no_numa) |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3872 | goto use_dfl; |
| 3873 | |
| 3874 | /* does @node have any online CPUs @attrs wants? */ |
| 3875 | cpumask_and(cpumask, cpumask_of_node(node), attrs->cpumask); |
| 3876 | if (cpu_going_down >= 0) |
| 3877 | cpumask_clear_cpu(cpu_going_down, cpumask); |
| 3878 | |
| 3879 | if (cpumask_empty(cpumask)) |
| 3880 | goto use_dfl; |
| 3881 | |
| 3882 | /* yeap, return possible CPUs in @node that @attrs wants */ |
| 3883 | cpumask_and(cpumask, attrs->cpumask, wq_numa_possible_cpumask[node]); |
| 3884 | return !cpumask_equal(cpumask, attrs->cpumask); |
| 3885 | |
| 3886 | use_dfl: |
| 3887 | cpumask_copy(cpumask, attrs->cpumask); |
| 3888 | return false; |
| 3889 | } |
| 3890 | |
Tejun Heo | 1befcf3 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3891 | /* install @pwq into @wq's numa_pwq_tbl[] for @node and return the old pwq */ |
| 3892 | static struct pool_workqueue *numa_pwq_tbl_install(struct workqueue_struct *wq, |
| 3893 | int node, |
| 3894 | struct pool_workqueue *pwq) |
| 3895 | { |
| 3896 | struct pool_workqueue *old_pwq; |
| 3897 | |
| 3898 | lockdep_assert_held(&wq->mutex); |
| 3899 | |
| 3900 | /* link_pwq() can handle duplicate calls */ |
| 3901 | link_pwq(pwq); |
| 3902 | |
| 3903 | old_pwq = rcu_access_pointer(wq->numa_pwq_tbl[node]); |
| 3904 | rcu_assign_pointer(wq->numa_pwq_tbl[node], pwq); |
| 3905 | return old_pwq; |
| 3906 | } |
| 3907 | |
Tejun Heo | 9e8cd2f | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3908 | /** |
| 3909 | * apply_workqueue_attrs - apply new workqueue_attrs to an unbound workqueue |
| 3910 | * @wq: the target workqueue |
| 3911 | * @attrs: the workqueue_attrs to apply, allocated with alloc_workqueue_attrs() |
| 3912 | * |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3913 | * Apply @attrs to an unbound workqueue @wq. Unless disabled, on NUMA |
| 3914 | * machines, this function maps a separate pwq to each NUMA node with |
| 3915 | * possibles CPUs in @attrs->cpumask so that work items are affine to the |
| 3916 | * NUMA node it was issued on. Older pwqs are released as in-flight work |
| 3917 | * items finish. Note that a work item which repeatedly requeues itself |
| 3918 | * back-to-back will stay on its current pwq. |
Tejun Heo | 9e8cd2f | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3919 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 3920 | * Performs GFP_KERNEL allocations. |
| 3921 | * |
| 3922 | * Return: 0 on success and -errno on failure. |
Tejun Heo | 9e8cd2f | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3923 | */ |
| 3924 | int apply_workqueue_attrs(struct workqueue_struct *wq, |
| 3925 | const struct workqueue_attrs *attrs) |
| 3926 | { |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3927 | struct workqueue_attrs *new_attrs, *tmp_attrs; |
| 3928 | struct pool_workqueue **pwq_tbl, *dfl_pwq; |
Tejun Heo | f147f29 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3929 | int node, ret; |
Tejun Heo | 9e8cd2f | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3930 | |
Tejun Heo | 8719dce | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3931 | /* only unbound workqueues can change attributes */ |
Tejun Heo | 9e8cd2f | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3932 | if (WARN_ON(!(wq->flags & WQ_UNBOUND))) |
| 3933 | return -EINVAL; |
| 3934 | |
Tejun Heo | 8719dce | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3935 | /* creating multiple pwqs breaks ordering guarantee */ |
| 3936 | if (WARN_ON((wq->flags & __WQ_ORDERED) && !list_empty(&wq->pwqs))) |
| 3937 | return -EINVAL; |
| 3938 | |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3939 | pwq_tbl = kzalloc(wq_numa_tbl_len * sizeof(pwq_tbl[0]), GFP_KERNEL); |
Tejun Heo | 13e2e55 | 2013-04-01 11:23:31 -0700 | [diff] [blame] | 3940 | new_attrs = alloc_workqueue_attrs(GFP_KERNEL); |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3941 | tmp_attrs = alloc_workqueue_attrs(GFP_KERNEL); |
| 3942 | if (!pwq_tbl || !new_attrs || !tmp_attrs) |
Tejun Heo | 13e2e55 | 2013-04-01 11:23:31 -0700 | [diff] [blame] | 3943 | goto enomem; |
| 3944 | |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3945 | /* make a copy of @attrs and sanitize it */ |
Tejun Heo | 13e2e55 | 2013-04-01 11:23:31 -0700 | [diff] [blame] | 3946 | copy_workqueue_attrs(new_attrs, attrs); |
| 3947 | cpumask_and(new_attrs->cpumask, new_attrs->cpumask, cpu_possible_mask); |
| 3948 | |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3949 | /* |
| 3950 | * We may create multiple pwqs with differing cpumasks. Make a |
| 3951 | * copy of @new_attrs which will be modified and used to obtain |
| 3952 | * pools. |
| 3953 | */ |
| 3954 | copy_workqueue_attrs(tmp_attrs, new_attrs); |
Tejun Heo | 9e8cd2f | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 3955 | |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3956 | /* |
| 3957 | * CPUs should stay stable across pwq creations and installations. |
| 3958 | * Pin CPUs, determine the target cpumask for each node and create |
| 3959 | * pwqs accordingly. |
| 3960 | */ |
| 3961 | get_online_cpus(); |
| 3962 | |
| 3963 | mutex_lock(&wq_pool_mutex); |
| 3964 | |
| 3965 | /* |
| 3966 | * If something goes wrong during CPU up/down, we'll fall back to |
| 3967 | * the default pwq covering whole @attrs->cpumask. Always create |
| 3968 | * it even if we don't use it immediately. |
| 3969 | */ |
| 3970 | dfl_pwq = alloc_unbound_pwq(wq, new_attrs); |
| 3971 | if (!dfl_pwq) |
| 3972 | goto enomem_pwq; |
| 3973 | |
| 3974 | for_each_node(node) { |
| 3975 | if (wq_calc_node_cpumask(attrs, node, -1, tmp_attrs->cpumask)) { |
| 3976 | pwq_tbl[node] = alloc_unbound_pwq(wq, tmp_attrs); |
| 3977 | if (!pwq_tbl[node]) |
| 3978 | goto enomem_pwq; |
| 3979 | } else { |
| 3980 | dfl_pwq->refcnt++; |
| 3981 | pwq_tbl[node] = dfl_pwq; |
| 3982 | } |
| 3983 | } |
| 3984 | |
| 3985 | mutex_unlock(&wq_pool_mutex); |
| 3986 | |
| 3987 | /* all pwqs have been created successfully, let's install'em */ |
Tejun Heo | f147f29 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3988 | mutex_lock(&wq->mutex); |
| 3989 | |
Tejun Heo | f147f29 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3990 | copy_workqueue_attrs(wq->unbound_attrs, new_attrs); |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3991 | |
| 3992 | /* save the previous pwq and install the new one */ |
Tejun Heo | f147f29 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3993 | for_each_node(node) |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 3994 | pwq_tbl[node] = numa_pwq_tbl_install(wq, node, pwq_tbl[node]); |
| 3995 | |
| 3996 | /* @dfl_pwq might not have been used, ensure it's linked */ |
| 3997 | link_pwq(dfl_pwq); |
| 3998 | swap(wq->dfl_pwq, dfl_pwq); |
Tejun Heo | f147f29 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 3999 | |
| 4000 | mutex_unlock(&wq->mutex); |
| 4001 | |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 4002 | /* put the old pwqs */ |
| 4003 | for_each_node(node) |
| 4004 | put_pwq_unlocked(pwq_tbl[node]); |
| 4005 | put_pwq_unlocked(dfl_pwq); |
| 4006 | |
| 4007 | put_online_cpus(); |
Tejun Heo | 4862125 | 2013-04-01 11:23:31 -0700 | [diff] [blame] | 4008 | ret = 0; |
| 4009 | /* fall through */ |
| 4010 | out_free: |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 4011 | free_workqueue_attrs(tmp_attrs); |
Tejun Heo | 4862125 | 2013-04-01 11:23:31 -0700 | [diff] [blame] | 4012 | free_workqueue_attrs(new_attrs); |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 4013 | kfree(pwq_tbl); |
Tejun Heo | 4862125 | 2013-04-01 11:23:31 -0700 | [diff] [blame] | 4014 | return ret; |
Tejun Heo | 13e2e55 | 2013-04-01 11:23:31 -0700 | [diff] [blame] | 4015 | |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 4016 | enomem_pwq: |
| 4017 | free_unbound_pwq(dfl_pwq); |
| 4018 | for_each_node(node) |
| 4019 | if (pwq_tbl && pwq_tbl[node] != dfl_pwq) |
| 4020 | free_unbound_pwq(pwq_tbl[node]); |
| 4021 | mutex_unlock(&wq_pool_mutex); |
| 4022 | put_online_cpus(); |
Tejun Heo | 13e2e55 | 2013-04-01 11:23:31 -0700 | [diff] [blame] | 4023 | enomem: |
Tejun Heo | 4862125 | 2013-04-01 11:23:31 -0700 | [diff] [blame] | 4024 | ret = -ENOMEM; |
| 4025 | goto out_free; |
Tejun Heo | 9e8cd2f | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4026 | } |
| 4027 | |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 4028 | /** |
| 4029 | * wq_update_unbound_numa - update NUMA affinity of a wq for CPU hot[un]plug |
| 4030 | * @wq: the target workqueue |
| 4031 | * @cpu: the CPU coming up or going down |
| 4032 | * @online: whether @cpu is coming up or going down |
| 4033 | * |
| 4034 | * This function is to be called from %CPU_DOWN_PREPARE, %CPU_ONLINE and |
| 4035 | * %CPU_DOWN_FAILED. @cpu is being hot[un]plugged, update NUMA affinity of |
| 4036 | * @wq accordingly. |
| 4037 | * |
| 4038 | * If NUMA affinity can't be adjusted due to memory allocation failure, it |
| 4039 | * falls back to @wq->dfl_pwq which may not be optimal but is always |
| 4040 | * correct. |
| 4041 | * |
| 4042 | * Note that when the last allowed CPU of a NUMA node goes offline for a |
| 4043 | * workqueue with a cpumask spanning multiple nodes, the workers which were |
| 4044 | * already executing the work items for the workqueue will lose their CPU |
| 4045 | * affinity and may execute on any CPU. This is similar to how per-cpu |
| 4046 | * workqueues behave on CPU_DOWN. If a workqueue user wants strict |
| 4047 | * affinity, it's the user's responsibility to flush the work item from |
| 4048 | * CPU_DOWN_PREPARE. |
| 4049 | */ |
| 4050 | static void wq_update_unbound_numa(struct workqueue_struct *wq, int cpu, |
| 4051 | bool online) |
| 4052 | { |
| 4053 | int node = cpu_to_node(cpu); |
| 4054 | int cpu_off = online ? -1 : cpu; |
| 4055 | struct pool_workqueue *old_pwq = NULL, *pwq; |
| 4056 | struct workqueue_attrs *target_attrs; |
| 4057 | cpumask_t *cpumask; |
| 4058 | |
| 4059 | lockdep_assert_held(&wq_pool_mutex); |
| 4060 | |
| 4061 | if (!wq_numa_enabled || !(wq->flags & WQ_UNBOUND)) |
| 4062 | return; |
| 4063 | |
| 4064 | /* |
| 4065 | * We don't wanna alloc/free wq_attrs for each wq for each CPU. |
| 4066 | * Let's use a preallocated one. The following buf is protected by |
| 4067 | * CPU hotplug exclusion. |
| 4068 | */ |
| 4069 | target_attrs = wq_update_unbound_numa_attrs_buf; |
| 4070 | cpumask = target_attrs->cpumask; |
| 4071 | |
| 4072 | mutex_lock(&wq->mutex); |
Tejun Heo | d55262c | 2013-04-01 11:23:38 -0700 | [diff] [blame] | 4073 | if (wq->unbound_attrs->no_numa) |
| 4074 | goto out_unlock; |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 4075 | |
| 4076 | copy_workqueue_attrs(target_attrs, wq->unbound_attrs); |
| 4077 | pwq = unbound_pwq_by_node(wq, node); |
| 4078 | |
| 4079 | /* |
| 4080 | * Let's determine what needs to be done. If the target cpumask is |
| 4081 | * different from wq's, we need to compare it to @pwq's and create |
| 4082 | * a new one if they don't match. If the target cpumask equals |
| 4083 | * wq's, the default pwq should be used. If @pwq is already the |
| 4084 | * default one, nothing to do; otherwise, install the default one. |
| 4085 | */ |
| 4086 | if (wq_calc_node_cpumask(wq->unbound_attrs, node, cpu_off, cpumask)) { |
| 4087 | if (cpumask_equal(cpumask, pwq->pool->attrs->cpumask)) |
| 4088 | goto out_unlock; |
| 4089 | } else { |
| 4090 | if (pwq == wq->dfl_pwq) |
| 4091 | goto out_unlock; |
| 4092 | else |
| 4093 | goto use_dfl_pwq; |
| 4094 | } |
| 4095 | |
| 4096 | mutex_unlock(&wq->mutex); |
| 4097 | |
| 4098 | /* create a new pwq */ |
| 4099 | pwq = alloc_unbound_pwq(wq, target_attrs); |
| 4100 | if (!pwq) { |
| 4101 | pr_warning("workqueue: allocation failed while updating NUMA affinity of \"%s\"\n", |
| 4102 | wq->name); |
| 4103 | goto out_unlock; |
| 4104 | } |
| 4105 | |
| 4106 | /* |
| 4107 | * Install the new pwq. As this function is called only from CPU |
| 4108 | * hotplug callbacks and applying a new attrs is wrapped with |
| 4109 | * get/put_online_cpus(), @wq->unbound_attrs couldn't have changed |
| 4110 | * inbetween. |
| 4111 | */ |
| 4112 | mutex_lock(&wq->mutex); |
| 4113 | old_pwq = numa_pwq_tbl_install(wq, node, pwq); |
| 4114 | goto out_unlock; |
| 4115 | |
| 4116 | use_dfl_pwq: |
| 4117 | spin_lock_irq(&wq->dfl_pwq->pool->lock); |
| 4118 | get_pwq(wq->dfl_pwq); |
| 4119 | spin_unlock_irq(&wq->dfl_pwq->pool->lock); |
| 4120 | old_pwq = numa_pwq_tbl_install(wq, node, wq->dfl_pwq); |
| 4121 | out_unlock: |
| 4122 | mutex_unlock(&wq->mutex); |
| 4123 | put_pwq_unlocked(old_pwq); |
| 4124 | } |
| 4125 | |
Tejun Heo | 30cdf24 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4126 | static int alloc_and_link_pwqs(struct workqueue_struct *wq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4127 | { |
Tejun Heo | 49e3cf4 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4128 | bool highpri = wq->flags & WQ_HIGHPRI; |
Tejun Heo | 8a2b753 | 2013-09-05 12:30:04 -0400 | [diff] [blame] | 4129 | int cpu, ret; |
Frederic Weisbecker | e1d8aa9 | 2009-01-12 23:15:46 +0100 | [diff] [blame] | 4130 | |
Tejun Heo | 30cdf24 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4131 | if (!(wq->flags & WQ_UNBOUND)) { |
Tejun Heo | 420c0dd | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 4132 | wq->cpu_pwqs = alloc_percpu(struct pool_workqueue); |
| 4133 | if (!wq->cpu_pwqs) |
Tejun Heo | 30cdf24 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4134 | return -ENOMEM; |
| 4135 | |
| 4136 | for_each_possible_cpu(cpu) { |
Tejun Heo | 7fb98ea | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4137 | struct pool_workqueue *pwq = |
| 4138 | per_cpu_ptr(wq->cpu_pwqs, cpu); |
Tejun Heo | 7a62c2c | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 4139 | struct worker_pool *cpu_pools = |
Tejun Heo | f02ae73 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 4140 | per_cpu(cpu_worker_pools, cpu); |
Tejun Heo | 30cdf24 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4141 | |
Tejun Heo | f147f29 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 4142 | init_pwq(pwq, wq, &cpu_pools[highpri]); |
| 4143 | |
| 4144 | mutex_lock(&wq->mutex); |
Tejun Heo | 1befcf3 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 4145 | link_pwq(pwq); |
Tejun Heo | f147f29 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 4146 | mutex_unlock(&wq->mutex); |
Tejun Heo | 30cdf24 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4147 | } |
Tejun Heo | 9e8cd2f | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4148 | return 0; |
Tejun Heo | 8a2b753 | 2013-09-05 12:30:04 -0400 | [diff] [blame] | 4149 | } else if (wq->flags & __WQ_ORDERED) { |
| 4150 | ret = apply_workqueue_attrs(wq, ordered_wq_attrs[highpri]); |
| 4151 | /* there should only be single pwq for ordering guarantee */ |
| 4152 | WARN(!ret && (wq->pwqs.next != &wq->dfl_pwq->pwqs_node || |
| 4153 | wq->pwqs.prev != &wq->dfl_pwq->pwqs_node), |
| 4154 | "ordering guarantee broken for workqueue %s\n", wq->name); |
| 4155 | return ret; |
Tejun Heo | 30cdf24 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4156 | } else { |
Tejun Heo | 9e8cd2f | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4157 | return apply_workqueue_attrs(wq, unbound_std_wq_attrs[highpri]); |
Tejun Heo | 30cdf24 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4158 | } |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4159 | } |
| 4160 | |
Tejun Heo | f342179 | 2010-07-02 10:03:51 +0200 | [diff] [blame] | 4161 | static int wq_clamp_max_active(int max_active, unsigned int flags, |
| 4162 | const char *name) |
Tejun Heo | b71ab8c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4163 | { |
Tejun Heo | f342179 | 2010-07-02 10:03:51 +0200 | [diff] [blame] | 4164 | int lim = flags & WQ_UNBOUND ? WQ_UNBOUND_MAX_ACTIVE : WQ_MAX_ACTIVE; |
| 4165 | |
| 4166 | if (max_active < 1 || max_active > lim) |
Valentin Ilie | 044c782 | 2012-08-19 00:52:42 +0300 | [diff] [blame] | 4167 | pr_warn("workqueue: max_active %d requested for %s is out of range, clamping between %d and %d\n", |
| 4168 | max_active, name, 1, lim); |
Tejun Heo | b71ab8c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4169 | |
Tejun Heo | f342179 | 2010-07-02 10:03:51 +0200 | [diff] [blame] | 4170 | return clamp_val(max_active, 1, lim); |
Tejun Heo | b71ab8c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4171 | } |
| 4172 | |
Tejun Heo | b196be8 | 2012-01-10 15:11:35 -0800 | [diff] [blame] | 4173 | struct workqueue_struct *__alloc_workqueue_key(const char *fmt, |
Tejun Heo | d320c03 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4174 | unsigned int flags, |
| 4175 | int max_active, |
| 4176 | struct lock_class_key *key, |
Tejun Heo | b196be8 | 2012-01-10 15:11:35 -0800 | [diff] [blame] | 4177 | const char *lock_name, ...) |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4178 | { |
Tejun Heo | df2d5ae | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 4179 | size_t tbl_size = 0; |
Tejun Heo | ecf6881 | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 4180 | va_list args; |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4181 | struct workqueue_struct *wq; |
Tejun Heo | 49e3cf4 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4182 | struct pool_workqueue *pwq; |
Tejun Heo | b196be8 | 2012-01-10 15:11:35 -0800 | [diff] [blame] | 4183 | |
Viresh Kumar | cee22a1 | 2013-04-08 16:45:40 +0530 | [diff] [blame] | 4184 | /* see the comment above the definition of WQ_POWER_EFFICIENT */ |
| 4185 | if ((flags & WQ_POWER_EFFICIENT) && wq_power_efficient) |
| 4186 | flags |= WQ_UNBOUND; |
| 4187 | |
Tejun Heo | ecf6881 | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 4188 | /* allocate wq and format name */ |
Tejun Heo | df2d5ae | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 4189 | if (flags & WQ_UNBOUND) |
| 4190 | tbl_size = wq_numa_tbl_len * sizeof(wq->numa_pwq_tbl[0]); |
| 4191 | |
| 4192 | wq = kzalloc(sizeof(*wq) + tbl_size, GFP_KERNEL); |
Tejun Heo | b196be8 | 2012-01-10 15:11:35 -0800 | [diff] [blame] | 4193 | if (!wq) |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4194 | return NULL; |
Tejun Heo | b196be8 | 2012-01-10 15:11:35 -0800 | [diff] [blame] | 4195 | |
Tejun Heo | 6029a91 | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 4196 | if (flags & WQ_UNBOUND) { |
| 4197 | wq->unbound_attrs = alloc_workqueue_attrs(GFP_KERNEL); |
| 4198 | if (!wq->unbound_attrs) |
| 4199 | goto err_free_wq; |
| 4200 | } |
| 4201 | |
Tejun Heo | ecf6881 | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 4202 | va_start(args, lock_name); |
| 4203 | vsnprintf(wq->name, sizeof(wq->name), fmt, args); |
Tejun Heo | b196be8 | 2012-01-10 15:11:35 -0800 | [diff] [blame] | 4204 | va_end(args); |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4205 | |
Tejun Heo | d320c03 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4206 | max_active = max_active ?: WQ_DFL_ACTIVE; |
Tejun Heo | b196be8 | 2012-01-10 15:11:35 -0800 | [diff] [blame] | 4207 | max_active = wq_clamp_max_active(max_active, flags, wq->name); |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4208 | |
Tejun Heo | b196be8 | 2012-01-10 15:11:35 -0800 | [diff] [blame] | 4209 | /* init wq */ |
Tejun Heo | 97e37d7 | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 4210 | wq->flags = flags; |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4211 | wq->saved_max_active = max_active; |
Lai Jiangshan | 3c25a55 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4212 | mutex_init(&wq->mutex); |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 4213 | atomic_set(&wq->nr_pwqs_to_flush, 0); |
Tejun Heo | 30cdf24 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4214 | INIT_LIST_HEAD(&wq->pwqs); |
Tejun Heo | 73f53c4 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 4215 | INIT_LIST_HEAD(&wq->flusher_queue); |
| 4216 | INIT_LIST_HEAD(&wq->flusher_overflow); |
Tejun Heo | 493a172 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 4217 | INIT_LIST_HEAD(&wq->maydays); |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4218 | |
Johannes Berg | eb13ba8 | 2008-01-16 09:51:58 +0100 | [diff] [blame] | 4219 | lockdep_init_map(&wq->lockdep_map, lock_name, key, 0); |
Oleg Nesterov | cce1a16 | 2007-05-09 02:34:13 -0700 | [diff] [blame] | 4220 | INIT_LIST_HEAD(&wq->list); |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4221 | |
Tejun Heo | 30cdf24 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4222 | if (alloc_and_link_pwqs(wq) < 0) |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4223 | goto err_free_wq; |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4224 | |
Tejun Heo | 493008a | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 4225 | /* |
| 4226 | * Workqueues which may be used during memory reclaim should |
| 4227 | * have a rescuer to guarantee forward progress. |
| 4228 | */ |
| 4229 | if (flags & WQ_MEM_RECLAIM) { |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4230 | struct worker *rescuer; |
| 4231 | |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4232 | rescuer = alloc_worker(); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4233 | if (!rescuer) |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4234 | goto err_destroy; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4235 | |
Tejun Heo | 111c225 | 2013-01-17 17:16:24 -0800 | [diff] [blame] | 4236 | rescuer->rescue_wq = wq; |
| 4237 | rescuer->task = kthread_create(rescuer_thread, rescuer, "%s", |
Tejun Heo | b196be8 | 2012-01-10 15:11:35 -0800 | [diff] [blame] | 4238 | wq->name); |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4239 | if (IS_ERR(rescuer->task)) { |
| 4240 | kfree(rescuer); |
| 4241 | goto err_destroy; |
| 4242 | } |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4243 | |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4244 | wq->rescuer = rescuer; |
Tejun Heo | 14a40ff | 2013-03-19 13:45:20 -0700 | [diff] [blame] | 4245 | rescuer->task->flags |= PF_NO_SETAFFINITY; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4246 | wake_up_process(rescuer->task); |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4247 | } |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4248 | |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 4249 | if ((wq->flags & WQ_SYSFS) && workqueue_sysfs_register(wq)) |
| 4250 | goto err_destroy; |
| 4251 | |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4252 | /* |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4253 | * wq_pool_mutex protects global freeze state and workqueues list. |
| 4254 | * Grab it, adjust max_active and add the new @wq to workqueues |
| 4255 | * list. |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4256 | */ |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4257 | mutex_lock(&wq_pool_mutex); |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4258 | |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 4259 | mutex_lock(&wq->mutex); |
Tejun Heo | 699ce09 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 4260 | for_each_pwq(pwq, wq) |
| 4261 | pwq_adjust_max_active(pwq); |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 4262 | mutex_unlock(&wq->mutex); |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4263 | |
Tejun Heo | 1537663 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 4264 | list_add(&wq->list, &workqueues); |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4265 | |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4266 | mutex_unlock(&wq_pool_mutex); |
Tejun Heo | 1537663 | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 4267 | |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4268 | return wq; |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4269 | |
| 4270 | err_free_wq: |
Tejun Heo | 6029a91 | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 4271 | free_workqueue_attrs(wq->unbound_attrs); |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4272 | kfree(wq); |
| 4273 | return NULL; |
| 4274 | err_destroy: |
| 4275 | destroy_workqueue(wq); |
Tejun Heo | 4690c4a | 2010-06-29 10:07:10 +0200 | [diff] [blame] | 4276 | return NULL; |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4277 | } |
Tejun Heo | d320c03 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4278 | EXPORT_SYMBOL_GPL(__alloc_workqueue_key); |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4279 | |
| 4280 | /** |
| 4281 | * destroy_workqueue - safely terminate a workqueue |
| 4282 | * @wq: target workqueue |
| 4283 | * |
| 4284 | * Safely destroy a workqueue. All work currently pending will be done first. |
| 4285 | */ |
| 4286 | void destroy_workqueue(struct workqueue_struct *wq) |
| 4287 | { |
Tejun Heo | 49e3cf4 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4288 | struct pool_workqueue *pwq; |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 4289 | int node; |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4290 | |
Tejun Heo | 9c5a2ba | 2011-04-05 18:01:44 +0200 | [diff] [blame] | 4291 | /* drain it before proceeding with destruction */ |
| 4292 | drain_workqueue(wq); |
Tejun Heo | c8efcc2 | 2010-12-20 19:32:04 +0100 | [diff] [blame] | 4293 | |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4294 | /* sanity checks */ |
Lai Jiangshan | b09f4fd | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 4295 | mutex_lock(&wq->mutex); |
Tejun Heo | 49e3cf4 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4296 | for_each_pwq(pwq, wq) { |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4297 | int i; |
| 4298 | |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4299 | for (i = 0; i < WORK_NR_COLORS; i++) { |
| 4300 | if (WARN_ON(pwq->nr_in_flight[i])) { |
Lai Jiangshan | b09f4fd | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 4301 | mutex_unlock(&wq->mutex); |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4302 | return; |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4303 | } |
| 4304 | } |
| 4305 | |
Lai Jiangshan | 5c52959 | 2013-04-04 10:05:38 +0800 | [diff] [blame] | 4306 | if (WARN_ON((pwq != wq->dfl_pwq) && (pwq->refcnt > 1)) || |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4307 | WARN_ON(pwq->nr_active) || |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4308 | WARN_ON(!list_empty(&pwq->delayed_works))) { |
Lai Jiangshan | b09f4fd | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 4309 | mutex_unlock(&wq->mutex); |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4310 | return; |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4311 | } |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4312 | } |
Lai Jiangshan | b09f4fd | 2013-03-25 16:57:18 -0700 | [diff] [blame] | 4313 | mutex_unlock(&wq->mutex); |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4314 | |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4315 | /* |
| 4316 | * wq list is used to freeze wq, remove from list after |
| 4317 | * flushing is complete in case freeze races us. |
| 4318 | */ |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4319 | mutex_lock(&wq_pool_mutex); |
Tejun Heo | d2c1d40 | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4320 | list_del_init(&wq->list); |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4321 | mutex_unlock(&wq_pool_mutex); |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4322 | |
Tejun Heo | 226223a | 2013-03-12 11:30:05 -0700 | [diff] [blame] | 4323 | workqueue_sysfs_unregister(wq); |
| 4324 | |
Tejun Heo | 493008a | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 4325 | if (wq->rescuer) { |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4326 | kthread_stop(wq->rescuer->task); |
Xiaotian Feng | 8d9df9f | 2010-08-16 09:54:28 +0200 | [diff] [blame] | 4327 | kfree(wq->rescuer); |
Tejun Heo | 493008a | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 4328 | wq->rescuer = NULL; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4329 | } |
| 4330 | |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4331 | if (!(wq->flags & WQ_UNBOUND)) { |
| 4332 | /* |
| 4333 | * The base ref is never dropped on per-cpu pwqs. Directly |
| 4334 | * free the pwqs and wq. |
| 4335 | */ |
| 4336 | free_percpu(wq->cpu_pwqs); |
| 4337 | kfree(wq); |
| 4338 | } else { |
| 4339 | /* |
| 4340 | * We're the sole accessor of @wq at this point. Directly |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 4341 | * access numa_pwq_tbl[] and dfl_pwq to put the base refs. |
| 4342 | * @wq will be freed when the last pwq is released. |
Tejun Heo | 8864b4e | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4343 | */ |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 4344 | for_each_node(node) { |
| 4345 | pwq = rcu_access_pointer(wq->numa_pwq_tbl[node]); |
| 4346 | RCU_INIT_POINTER(wq->numa_pwq_tbl[node], NULL); |
| 4347 | put_pwq_unlocked(pwq); |
| 4348 | } |
| 4349 | |
| 4350 | /* |
| 4351 | * Put dfl_pwq. @wq may be freed any time after dfl_pwq is |
| 4352 | * put. Don't access it afterwards. |
| 4353 | */ |
| 4354 | pwq = wq->dfl_pwq; |
| 4355 | wq->dfl_pwq = NULL; |
Tejun Heo | dce90d4 | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 4356 | put_pwq_unlocked(pwq); |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 4357 | } |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4358 | } |
| 4359 | EXPORT_SYMBOL_GPL(destroy_workqueue); |
| 4360 | |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4361 | /** |
| 4362 | * workqueue_set_max_active - adjust max_active of a workqueue |
| 4363 | * @wq: target workqueue |
| 4364 | * @max_active: new max_active value. |
| 4365 | * |
| 4366 | * Set max_active of @wq to @max_active. |
| 4367 | * |
| 4368 | * CONTEXT: |
| 4369 | * Don't call from IRQ context. |
| 4370 | */ |
| 4371 | void workqueue_set_max_active(struct workqueue_struct *wq, int max_active) |
| 4372 | { |
Tejun Heo | 49e3cf4 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4373 | struct pool_workqueue *pwq; |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4374 | |
Tejun Heo | 8719dce | 2013-03-12 11:30:04 -0700 | [diff] [blame] | 4375 | /* disallow meddling with max_active for ordered workqueues */ |
| 4376 | if (WARN_ON(wq->flags & __WQ_ORDERED)) |
| 4377 | return; |
| 4378 | |
Tejun Heo | f342179 | 2010-07-02 10:03:51 +0200 | [diff] [blame] | 4379 | max_active = wq_clamp_max_active(max_active, wq->flags, wq->name); |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4380 | |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 4381 | mutex_lock(&wq->mutex); |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4382 | |
| 4383 | wq->saved_max_active = max_active; |
| 4384 | |
Tejun Heo | 699ce09 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 4385 | for_each_pwq(pwq, wq) |
| 4386 | pwq_adjust_max_active(pwq); |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4387 | |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 4388 | mutex_unlock(&wq->mutex); |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4389 | } |
| 4390 | EXPORT_SYMBOL_GPL(workqueue_set_max_active); |
| 4391 | |
| 4392 | /** |
Tejun Heo | e6267616 | 2013-03-12 17:41:37 -0700 | [diff] [blame] | 4393 | * current_is_workqueue_rescuer - is %current workqueue rescuer? |
| 4394 | * |
| 4395 | * Determine whether %current is a workqueue rescuer. Can be used from |
| 4396 | * 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] | 4397 | * |
| 4398 | * Return: %true if %current is a workqueue rescuer. %false otherwise. |
Tejun Heo | e6267616 | 2013-03-12 17:41:37 -0700 | [diff] [blame] | 4399 | */ |
| 4400 | bool current_is_workqueue_rescuer(void) |
| 4401 | { |
| 4402 | struct worker *worker = current_wq_worker(); |
| 4403 | |
Lai Jiangshan | 6a092df | 2013-03-20 03:28:03 +0800 | [diff] [blame] | 4404 | return worker && worker->rescue_wq; |
Tejun Heo | e6267616 | 2013-03-12 17:41:37 -0700 | [diff] [blame] | 4405 | } |
| 4406 | |
| 4407 | /** |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4408 | * workqueue_congested - test whether a workqueue is congested |
| 4409 | * @cpu: CPU in question |
| 4410 | * @wq: target workqueue |
| 4411 | * |
| 4412 | * Test whether @wq's cpu workqueue for @cpu is congested. There is |
| 4413 | * no synchronization around this function and the test result is |
| 4414 | * unreliable and only useful as advisory hints or for debugging. |
| 4415 | * |
Tejun Heo | d325185 | 2013-05-10 11:10:17 -0700 | [diff] [blame] | 4416 | * If @cpu is WORK_CPU_UNBOUND, the test is performed on the local CPU. |
| 4417 | * Note that both per-cpu and unbound workqueues may be associated with |
| 4418 | * multiple pool_workqueues which have separate congested states. A |
| 4419 | * workqueue being congested on one CPU doesn't mean the workqueue is also |
| 4420 | * contested on other CPUs / NUMA nodes. |
| 4421 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 4422 | * Return: |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4423 | * %true if congested, %false otherwise. |
| 4424 | */ |
Tejun Heo | d84ff05 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 4425 | bool workqueue_congested(int cpu, struct workqueue_struct *wq) |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4426 | { |
Tejun Heo | 7fb98ea | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4427 | struct pool_workqueue *pwq; |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4428 | bool ret; |
| 4429 | |
Lai Jiangshan | 8810945 | 2013-03-20 03:28:10 +0800 | [diff] [blame] | 4430 | rcu_read_lock_sched(); |
Tejun Heo | 7fb98ea | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4431 | |
Tejun Heo | d325185 | 2013-05-10 11:10:17 -0700 | [diff] [blame] | 4432 | if (cpu == WORK_CPU_UNBOUND) |
| 4433 | cpu = smp_processor_id(); |
| 4434 | |
Tejun Heo | 7fb98ea | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4435 | if (!(wq->flags & WQ_UNBOUND)) |
| 4436 | pwq = per_cpu_ptr(wq->cpu_pwqs, cpu); |
| 4437 | else |
Tejun Heo | df2d5ae | 2013-04-01 11:23:35 -0700 | [diff] [blame] | 4438 | pwq = unbound_pwq_by_node(wq, cpu_to_node(cpu)); |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4439 | |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4440 | ret = !list_empty(&pwq->delayed_works); |
Lai Jiangshan | 8810945 | 2013-03-20 03:28:10 +0800 | [diff] [blame] | 4441 | rcu_read_unlock_sched(); |
Tejun Heo | 76af4d9 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4442 | |
| 4443 | return ret; |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4444 | } |
| 4445 | EXPORT_SYMBOL_GPL(workqueue_congested); |
| 4446 | |
| 4447 | /** |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4448 | * work_busy - test whether a work is currently pending or running |
| 4449 | * @work: the work to be tested |
| 4450 | * |
| 4451 | * Test whether @work is currently pending or running. There is no |
| 4452 | * synchronization around this function and the test result is |
| 4453 | * unreliable and only useful as advisory hints or for debugging. |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4454 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 4455 | * Return: |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4456 | * OR'd bitmask of WORK_BUSY_* bits. |
| 4457 | */ |
| 4458 | unsigned int work_busy(struct work_struct *work) |
| 4459 | { |
Tejun Heo | fa1b54e | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4460 | struct worker_pool *pool; |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4461 | unsigned long flags; |
| 4462 | unsigned int ret = 0; |
| 4463 | |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4464 | if (work_pending(work)) |
| 4465 | ret |= WORK_BUSY_PENDING; |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4466 | |
Tejun Heo | fa1b54e | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4467 | local_irq_save(flags); |
| 4468 | pool = get_work_pool(work); |
Lai Jiangshan | 038366c | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 4469 | if (pool) { |
Tejun Heo | fa1b54e | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4470 | spin_lock(&pool->lock); |
Lai Jiangshan | 038366c | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 4471 | if (find_worker_executing_work(pool, work)) |
| 4472 | ret |= WORK_BUSY_RUNNING; |
Tejun Heo | fa1b54e | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4473 | spin_unlock(&pool->lock); |
Lai Jiangshan | 038366c | 2013-02-06 18:04:53 -0800 | [diff] [blame] | 4474 | } |
Tejun Heo | fa1b54e | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 4475 | local_irq_restore(flags); |
Tejun Heo | dcd989c | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4476 | |
| 4477 | return ret; |
| 4478 | } |
| 4479 | EXPORT_SYMBOL_GPL(work_busy); |
| 4480 | |
Tejun Heo | 3d1cb20 | 2013-04-30 15:27:22 -0700 | [diff] [blame] | 4481 | /** |
| 4482 | * set_worker_desc - set description for the current work item |
| 4483 | * @fmt: printf-style format string |
| 4484 | * @...: arguments for the format string |
| 4485 | * |
| 4486 | * This function can be called by a running work function to describe what |
| 4487 | * the work item is about. If the worker task gets dumped, this |
| 4488 | * information will be printed out together to help debugging. The |
| 4489 | * description can be at most WORKER_DESC_LEN including the trailing '\0'. |
| 4490 | */ |
| 4491 | void set_worker_desc(const char *fmt, ...) |
| 4492 | { |
| 4493 | struct worker *worker = current_wq_worker(); |
| 4494 | va_list args; |
| 4495 | |
| 4496 | if (worker) { |
| 4497 | va_start(args, fmt); |
| 4498 | vsnprintf(worker->desc, sizeof(worker->desc), fmt, args); |
| 4499 | va_end(args); |
| 4500 | worker->desc_valid = true; |
| 4501 | } |
| 4502 | } |
| 4503 | |
| 4504 | /** |
| 4505 | * print_worker_info - print out worker information and description |
| 4506 | * @log_lvl: the log level to use when printing |
| 4507 | * @task: target task |
| 4508 | * |
| 4509 | * If @task is a worker and currently executing a work item, print out the |
| 4510 | * name of the workqueue being serviced and worker description set with |
| 4511 | * set_worker_desc() by the currently executing work item. |
| 4512 | * |
| 4513 | * This function can be safely called on any task as long as the |
| 4514 | * task_struct itself is accessible. While safe, this function isn't |
| 4515 | * synchronized and may print out mixups or garbages of limited length. |
| 4516 | */ |
| 4517 | void print_worker_info(const char *log_lvl, struct task_struct *task) |
| 4518 | { |
| 4519 | work_func_t *fn = NULL; |
| 4520 | char name[WQ_NAME_LEN] = { }; |
| 4521 | char desc[WORKER_DESC_LEN] = { }; |
| 4522 | struct pool_workqueue *pwq = NULL; |
| 4523 | struct workqueue_struct *wq = NULL; |
| 4524 | bool desc_valid = false; |
| 4525 | struct worker *worker; |
| 4526 | |
| 4527 | if (!(task->flags & PF_WQ_WORKER)) |
| 4528 | return; |
| 4529 | |
| 4530 | /* |
| 4531 | * This function is called without any synchronization and @task |
| 4532 | * could be in any state. Be careful with dereferences. |
| 4533 | */ |
| 4534 | worker = probe_kthread_data(task); |
| 4535 | |
| 4536 | /* |
| 4537 | * Carefully copy the associated workqueue's workfn and name. Keep |
| 4538 | * the original last '\0' in case the original contains garbage. |
| 4539 | */ |
| 4540 | probe_kernel_read(&fn, &worker->current_func, sizeof(fn)); |
| 4541 | probe_kernel_read(&pwq, &worker->current_pwq, sizeof(pwq)); |
| 4542 | probe_kernel_read(&wq, &pwq->wq, sizeof(wq)); |
| 4543 | probe_kernel_read(name, wq->name, sizeof(name) - 1); |
| 4544 | |
| 4545 | /* copy worker description */ |
| 4546 | probe_kernel_read(&desc_valid, &worker->desc_valid, sizeof(desc_valid)); |
| 4547 | if (desc_valid) |
| 4548 | probe_kernel_read(desc, worker->desc, sizeof(desc) - 1); |
| 4549 | |
| 4550 | if (fn || name[0] || desc[0]) { |
| 4551 | printk("%sWorkqueue: %s %pf", log_lvl, name, fn); |
| 4552 | if (desc[0]) |
| 4553 | pr_cont(" (%s)", desc); |
| 4554 | pr_cont("\n"); |
| 4555 | } |
| 4556 | } |
| 4557 | |
Tejun Heo | db7bccf | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4558 | /* |
| 4559 | * CPU hotplug. |
| 4560 | * |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4561 | * There are two challenges in supporting CPU hotplug. Firstly, there |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 4562 | * are a lot of assumptions on strong associations among work, pwq and |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 4563 | * pool which make migrating pending and scheduled works very |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4564 | * difficult to implement without impacting hot paths. Secondly, |
Tejun Heo | 94cf58b | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 4565 | * 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] | 4566 | * blocked draining impractical. |
| 4567 | * |
Tejun Heo | 2464757 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 4568 | * 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] | 4569 | * running as an unbound one and allowing it to be reattached later if the |
| 4570 | * cpu comes back online. |
Tejun Heo | db7bccf | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4571 | */ |
| 4572 | |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 4573 | static void wq_unbind_fn(struct work_struct *work) |
Tejun Heo | db7bccf | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4574 | { |
Tejun Heo | 38db41d | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 4575 | int cpu = smp_processor_id(); |
Tejun Heo | 4ce62e9 | 2012-07-13 22:16:44 -0700 | [diff] [blame] | 4576 | struct worker_pool *pool; |
Tejun Heo | db7bccf | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4577 | struct worker *worker; |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4578 | int wi; |
Tejun Heo | db7bccf | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4579 | |
Tejun Heo | f02ae73 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 4580 | for_each_cpu_worker_pool(pool, cpu) { |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4581 | WARN_ON_ONCE(cpu != smp_processor_id()); |
Tejun Heo | 94cf58b | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 4582 | |
Tejun Heo | bc3a1af | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 4583 | mutex_lock(&pool->manager_mutex); |
Tejun Heo | 94cf58b | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 4584 | spin_lock_irq(&pool->lock); |
| 4585 | |
| 4586 | /* |
Tejun Heo | bc3a1af | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 4587 | * We've blocked all manager operations. Make all workers |
Tejun Heo | 94cf58b | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 4588 | * unbound and set DISASSOCIATED. Before this, all workers |
| 4589 | * except for the ones which are still executing works from |
| 4590 | * before the last CPU down must be on the cpu. After |
| 4591 | * this, they may become diasporas. |
| 4592 | */ |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4593 | for_each_pool_worker(worker, wi, pool) |
Tejun Heo | c9e7cf2 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 4594 | worker->flags |= WORKER_UNBOUND; |
Tejun Heo | db7bccf | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4595 | |
Tejun Heo | 2464757 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 4596 | pool->flags |= POOL_DISASSOCIATED; |
Tejun Heo | f2d5a0e | 2012-07-17 12:39:26 -0700 | [diff] [blame] | 4597 | |
Tejun Heo | 94cf58b | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 4598 | spin_unlock_irq(&pool->lock); |
Tejun Heo | bc3a1af | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 4599 | mutex_unlock(&pool->manager_mutex); |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 4600 | |
Lai Jiangshan | eb28342 | 2013-03-08 15:18:28 -0800 | [diff] [blame] | 4601 | /* |
| 4602 | * Call schedule() so that we cross rq->lock and thus can |
| 4603 | * guarantee sched callbacks see the %WORKER_UNBOUND flag. |
| 4604 | * This is necessary as scheduler callbacks may be invoked |
| 4605 | * from other cpus. |
| 4606 | */ |
| 4607 | schedule(); |
Tejun Heo | 628c78e | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 4608 | |
Lai Jiangshan | eb28342 | 2013-03-08 15:18:28 -0800 | [diff] [blame] | 4609 | /* |
| 4610 | * Sched callbacks are disabled now. Zap nr_running. |
| 4611 | * After this, nr_running stays zero and need_more_worker() |
| 4612 | * and keep_working() are always true as long as the |
| 4613 | * worklist is not empty. This pool now behaves as an |
| 4614 | * unbound (in terms of concurrency management) pool which |
| 4615 | * are served by workers tied to the pool. |
| 4616 | */ |
Tejun Heo | e19e397 | 2013-01-24 11:39:44 -0800 | [diff] [blame] | 4617 | atomic_set(&pool->nr_running, 0); |
Lai Jiangshan | eb28342 | 2013-03-08 15:18:28 -0800 | [diff] [blame] | 4618 | |
| 4619 | /* |
| 4620 | * With concurrency management just turned off, a busy |
| 4621 | * worker blocking could lead to lengthy stalls. Kick off |
| 4622 | * unbound chain execution of currently pending work items. |
| 4623 | */ |
| 4624 | spin_lock_irq(&pool->lock); |
| 4625 | wake_up_worker(pool); |
| 4626 | spin_unlock_irq(&pool->lock); |
| 4627 | } |
Tejun Heo | db7bccf | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4628 | } |
| 4629 | |
Tejun Heo | bd7c089 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4630 | /** |
| 4631 | * rebind_workers - rebind all workers of a pool to the associated CPU |
| 4632 | * @pool: pool of interest |
| 4633 | * |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4634 | * @pool->cpu is coming online. Rebind all workers to the CPU. |
Tejun Heo | bd7c089 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4635 | */ |
| 4636 | static void rebind_workers(struct worker_pool *pool) |
| 4637 | { |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4638 | struct worker *worker; |
| 4639 | int wi; |
Tejun Heo | bd7c089 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4640 | |
| 4641 | lockdep_assert_held(&pool->manager_mutex); |
Tejun Heo | bd7c089 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4642 | |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4643 | /* |
| 4644 | * Restore CPU affinity of all workers. As all idle workers should |
| 4645 | * be on the run-queue of the associated CPU before any local |
| 4646 | * wake-ups for concurrency management happen, restore CPU affinty |
| 4647 | * of all workers first and then clear UNBOUND. As we're called |
| 4648 | * from CPU_ONLINE, the following shouldn't fail. |
| 4649 | */ |
| 4650 | for_each_pool_worker(worker, wi, pool) |
| 4651 | WARN_ON_ONCE(set_cpus_allowed_ptr(worker->task, |
| 4652 | pool->attrs->cpumask) < 0); |
| 4653 | |
| 4654 | spin_lock_irq(&pool->lock); |
| 4655 | |
| 4656 | for_each_pool_worker(worker, wi, pool) { |
| 4657 | unsigned int worker_flags = worker->flags; |
Tejun Heo | bd7c089 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4658 | |
| 4659 | /* |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4660 | * A bound idle worker should actually be on the runqueue |
| 4661 | * of the associated CPU for local wake-ups targeting it to |
| 4662 | * work. Kick all idle workers so that they migrate to the |
| 4663 | * associated CPU. Doing this in the same loop as |
| 4664 | * replacing UNBOUND with REBOUND is safe as no worker will |
| 4665 | * be bound before @pool->lock is released. |
Tejun Heo | bd7c089 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4666 | */ |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4667 | if (worker_flags & WORKER_IDLE) |
| 4668 | wake_up_process(worker->task); |
| 4669 | |
| 4670 | /* |
| 4671 | * We want to clear UNBOUND but can't directly call |
| 4672 | * worker_clr_flags() or adjust nr_running. Atomically |
| 4673 | * replace UNBOUND with another NOT_RUNNING flag REBOUND. |
| 4674 | * @worker will clear REBOUND using worker_clr_flags() when |
| 4675 | * it initiates the next execution cycle thus restoring |
| 4676 | * concurrency management. Note that when or whether |
| 4677 | * @worker clears REBOUND doesn't affect correctness. |
| 4678 | * |
| 4679 | * ACCESS_ONCE() is necessary because @worker->flags may be |
| 4680 | * tested without holding any lock in |
| 4681 | * wq_worker_waking_up(). Without it, NOT_RUNNING test may |
| 4682 | * fail incorrectly leading to premature concurrency |
| 4683 | * management operations. |
| 4684 | */ |
| 4685 | WARN_ON_ONCE(!(worker_flags & WORKER_UNBOUND)); |
| 4686 | worker_flags |= WORKER_REBOUND; |
| 4687 | worker_flags &= ~WORKER_UNBOUND; |
| 4688 | ACCESS_ONCE(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 | spin_unlock_irq(&pool->lock); |
Tejun Heo | bd7c089 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4692 | } |
| 4693 | |
Tejun Heo | 7dbc725 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4694 | /** |
| 4695 | * restore_unbound_workers_cpumask - restore cpumask of unbound workers |
| 4696 | * @pool: unbound pool of interest |
| 4697 | * @cpu: the CPU which is coming up |
| 4698 | * |
| 4699 | * An unbound pool may end up with a cpumask which doesn't have any online |
| 4700 | * CPUs. When a worker of such pool get scheduled, the scheduler resets |
| 4701 | * its cpus_allowed. If @cpu is in @pool's cpumask which didn't have any |
| 4702 | * online CPU before, cpus_allowed of all its workers should be restored. |
| 4703 | */ |
| 4704 | static void restore_unbound_workers_cpumask(struct worker_pool *pool, int cpu) |
| 4705 | { |
| 4706 | static cpumask_t cpumask; |
| 4707 | struct worker *worker; |
| 4708 | int wi; |
| 4709 | |
| 4710 | lockdep_assert_held(&pool->manager_mutex); |
| 4711 | |
| 4712 | /* is @cpu allowed for @pool? */ |
| 4713 | if (!cpumask_test_cpu(cpu, pool->attrs->cpumask)) |
| 4714 | return; |
| 4715 | |
| 4716 | /* is @cpu the only online CPU? */ |
| 4717 | cpumask_and(&cpumask, pool->attrs->cpumask, cpu_online_mask); |
| 4718 | if (cpumask_weight(&cpumask) != 1) |
| 4719 | return; |
| 4720 | |
| 4721 | /* as we're called from CPU_ONLINE, the following shouldn't fail */ |
| 4722 | for_each_pool_worker(worker, wi, pool) |
| 4723 | WARN_ON_ONCE(set_cpus_allowed_ptr(worker->task, |
| 4724 | pool->attrs->cpumask) < 0); |
| 4725 | } |
| 4726 | |
Tejun Heo | 8db25e7 | 2012-07-17 12:39:28 -0700 | [diff] [blame] | 4727 | /* |
| 4728 | * Workqueues should be brought up before normal priority CPU notifiers. |
| 4729 | * This will be registered high priority CPU notifier. |
| 4730 | */ |
Paul Gortmaker | 0db0628 | 2013-06-19 14:53:51 -0400 | [diff] [blame] | 4731 | static int workqueue_cpu_up_callback(struct notifier_block *nfb, |
Tejun Heo | 8db25e7 | 2012-07-17 12:39:28 -0700 | [diff] [blame] | 4732 | unsigned long action, |
| 4733 | void *hcpu) |
Oleg Nesterov | 3af24433 | 2007-05-09 02:34:09 -0700 | [diff] [blame] | 4734 | { |
Tejun Heo | d84ff05 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 4735 | int cpu = (unsigned long)hcpu; |
Tejun Heo | 4ce62e9 | 2012-07-13 22:16:44 -0700 | [diff] [blame] | 4736 | struct worker_pool *pool; |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 4737 | struct workqueue_struct *wq; |
Tejun Heo | 7dbc725 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4738 | int pi; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4739 | |
Tejun Heo | 8db25e7 | 2012-07-17 12:39:28 -0700 | [diff] [blame] | 4740 | switch (action & ~CPU_TASKS_FROZEN) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4741 | case CPU_UP_PREPARE: |
Tejun Heo | f02ae73 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 4742 | for_each_cpu_worker_pool(pool, cpu) { |
Tejun Heo | 3ce6337 | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 4743 | if (pool->nr_workers) |
| 4744 | continue; |
Tejun Heo | ebf44d1 | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 4745 | if (create_and_start_worker(pool) < 0) |
Tejun Heo | 3ce6337 | 2012-07-17 12:39:27 -0700 | [diff] [blame] | 4746 | return NOTIFY_BAD; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4747 | } |
Tejun Heo | db7bccf | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4748 | break; |
Oleg Nesterov | 00dfcaf | 2008-04-29 01:00:27 -0700 | [diff] [blame] | 4749 | |
Tejun Heo | 6575820 | 2012-07-17 12:39:26 -0700 | [diff] [blame] | 4750 | case CPU_DOWN_FAILED: |
| 4751 | case CPU_ONLINE: |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4752 | mutex_lock(&wq_pool_mutex); |
Tejun Heo | 7dbc725 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4753 | |
| 4754 | for_each_pool(pool, pi) { |
Tejun Heo | bc3a1af | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 4755 | mutex_lock(&pool->manager_mutex); |
Tejun Heo | 94cf58b | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 4756 | |
Tejun Heo | 7dbc725 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4757 | if (pool->cpu == cpu) { |
| 4758 | spin_lock_irq(&pool->lock); |
| 4759 | pool->flags &= ~POOL_DISASSOCIATED; |
| 4760 | spin_unlock_irq(&pool->lock); |
Tejun Heo | a9ab775 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4761 | |
Tejun Heo | 7dbc725 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4762 | rebind_workers(pool); |
| 4763 | } else if (pool->cpu < 0) { |
| 4764 | restore_unbound_workers_cpumask(pool, cpu); |
| 4765 | } |
Tejun Heo | 94cf58b | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 4766 | |
Tejun Heo | bc3a1af | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 4767 | mutex_unlock(&pool->manager_mutex); |
Tejun Heo | 94cf58b | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 4768 | } |
Tejun Heo | 7dbc725 | 2013-03-19 13:45:21 -0700 | [diff] [blame] | 4769 | |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 4770 | /* update NUMA affinity of unbound workqueues */ |
| 4771 | list_for_each_entry(wq, &workqueues, list) |
| 4772 | wq_update_unbound_numa(wq, cpu, true); |
| 4773 | |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4774 | mutex_unlock(&wq_pool_mutex); |
Tejun Heo | 8db25e7 | 2012-07-17 12:39:28 -0700 | [diff] [blame] | 4775 | break; |
Tejun Heo | 6575820 | 2012-07-17 12:39:26 -0700 | [diff] [blame] | 4776 | } |
| 4777 | return NOTIFY_OK; |
| 4778 | } |
| 4779 | |
| 4780 | /* |
| 4781 | * Workqueues should be brought down after normal priority CPU notifiers. |
| 4782 | * This will be registered as low priority CPU notifier. |
| 4783 | */ |
Paul Gortmaker | 0db0628 | 2013-06-19 14:53:51 -0400 | [diff] [blame] | 4784 | static int workqueue_cpu_down_callback(struct notifier_block *nfb, |
Tejun Heo | 6575820 | 2012-07-17 12:39:26 -0700 | [diff] [blame] | 4785 | unsigned long action, |
| 4786 | void *hcpu) |
| 4787 | { |
Tejun Heo | d84ff05 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 4788 | int cpu = (unsigned long)hcpu; |
Tejun Heo | 8db25e7 | 2012-07-17 12:39:28 -0700 | [diff] [blame] | 4789 | struct work_struct unbind_work; |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 4790 | struct workqueue_struct *wq; |
Tejun Heo | 8db25e7 | 2012-07-17 12:39:28 -0700 | [diff] [blame] | 4791 | |
Tejun Heo | 6575820 | 2012-07-17 12:39:26 -0700 | [diff] [blame] | 4792 | switch (action & ~CPU_TASKS_FROZEN) { |
| 4793 | case CPU_DOWN_PREPARE: |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 4794 | /* unbinding per-cpu workers should happen on the local CPU */ |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 4795 | INIT_WORK_ONSTACK(&unbind_work, wq_unbind_fn); |
Joonsoo Kim | 7635d2f | 2012-08-15 23:25:41 +0900 | [diff] [blame] | 4796 | queue_work_on(cpu, system_highpri_wq, &unbind_work); |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 4797 | |
| 4798 | /* update NUMA affinity of unbound workqueues */ |
| 4799 | mutex_lock(&wq_pool_mutex); |
| 4800 | list_for_each_entry(wq, &workqueues, list) |
| 4801 | wq_update_unbound_numa(wq, cpu, false); |
| 4802 | mutex_unlock(&wq_pool_mutex); |
| 4803 | |
| 4804 | /* wait for per-cpu unbinding to finish */ |
Tejun Heo | 8db25e7 | 2012-07-17 12:39:28 -0700 | [diff] [blame] | 4805 | flush_work(&unbind_work); |
Chuansheng Liu | 440a113 | 2014-01-11 22:26:33 -0500 | [diff] [blame] | 4806 | destroy_work_on_stack(&unbind_work); |
Tejun Heo | 8db25e7 | 2012-07-17 12:39:28 -0700 | [diff] [blame] | 4807 | break; |
Tejun Heo | 6575820 | 2012-07-17 12:39:26 -0700 | [diff] [blame] | 4808 | } |
| 4809 | return NOTIFY_OK; |
| 4810 | } |
| 4811 | |
Rusty Russell | 2d3854a | 2008-11-05 13:39:10 +1100 | [diff] [blame] | 4812 | #ifdef CONFIG_SMP |
Rusty Russell | 8ccad40 | 2009-01-16 15:31:15 -0800 | [diff] [blame] | 4813 | |
Rusty Russell | 2d3854a | 2008-11-05 13:39:10 +1100 | [diff] [blame] | 4814 | struct work_for_cpu { |
Tejun Heo | ed48ece | 2012-09-18 12:48:43 -0700 | [diff] [blame] | 4815 | struct work_struct work; |
Rusty Russell | 2d3854a | 2008-11-05 13:39:10 +1100 | [diff] [blame] | 4816 | long (*fn)(void *); |
| 4817 | void *arg; |
| 4818 | long ret; |
| 4819 | }; |
| 4820 | |
Tejun Heo | ed48ece | 2012-09-18 12:48:43 -0700 | [diff] [blame] | 4821 | static void work_for_cpu_fn(struct work_struct *work) |
Rusty Russell | 2d3854a | 2008-11-05 13:39:10 +1100 | [diff] [blame] | 4822 | { |
Tejun Heo | ed48ece | 2012-09-18 12:48:43 -0700 | [diff] [blame] | 4823 | struct work_for_cpu *wfc = container_of(work, struct work_for_cpu, work); |
| 4824 | |
Rusty Russell | 2d3854a | 2008-11-05 13:39:10 +1100 | [diff] [blame] | 4825 | wfc->ret = wfc->fn(wfc->arg); |
| 4826 | } |
| 4827 | |
| 4828 | /** |
| 4829 | * work_on_cpu - run a function in user context on a particular cpu |
| 4830 | * @cpu: the cpu to run on |
| 4831 | * @fn: the function to run |
| 4832 | * @arg: the function arg |
| 4833 | * |
Rusty Russell | 31ad908 | 2009-01-16 15:31:15 -0800 | [diff] [blame] | 4834 | * 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] | 4835 | * 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] | 4836 | * |
| 4837 | * Return: The value @fn returns. |
Rusty Russell | 2d3854a | 2008-11-05 13:39:10 +1100 | [diff] [blame] | 4838 | */ |
Tejun Heo | d84ff05 | 2013-03-12 11:29:59 -0700 | [diff] [blame] | 4839 | long work_on_cpu(int cpu, long (*fn)(void *), void *arg) |
Rusty Russell | 2d3854a | 2008-11-05 13:39:10 +1100 | [diff] [blame] | 4840 | { |
Tejun Heo | ed48ece | 2012-09-18 12:48:43 -0700 | [diff] [blame] | 4841 | struct work_for_cpu wfc = { .fn = fn, .arg = 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 | INIT_WORK_ONSTACK(&wfc.work, work_for_cpu_fn); |
| 4844 | schedule_work_on(cpu, &wfc.work); |
Bjorn Helgaas | 12997d1 | 2013-11-18 11:00:29 -0700 | [diff] [blame] | 4845 | flush_work(&wfc.work); |
Chuansheng Liu | 440a113 | 2014-01-11 22:26:33 -0500 | [diff] [blame] | 4846 | destroy_work_on_stack(&wfc.work); |
Rusty Russell | 2d3854a | 2008-11-05 13:39:10 +1100 | [diff] [blame] | 4847 | return wfc.ret; |
| 4848 | } |
| 4849 | EXPORT_SYMBOL_GPL(work_on_cpu); |
| 4850 | #endif /* CONFIG_SMP */ |
| 4851 | |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4852 | #ifdef CONFIG_FREEZER |
Rusty Russell | e7577c5 | 2009-01-01 10:12:25 +1030 | [diff] [blame] | 4853 | |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4854 | /** |
| 4855 | * freeze_workqueues_begin - begin freezing workqueues |
| 4856 | * |
Tejun Heo | 58a69cb | 2011-02-16 09:25:31 +0100 | [diff] [blame] | 4857 | * Start freezing workqueues. After this function returns, all freezable |
Tejun Heo | c5aa87b | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 4858 | * workqueues will queue new works to their delayed_works list instead of |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 4859 | * pool->worklist. |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4860 | * |
| 4861 | * CONTEXT: |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 4862 | * Grabs and releases wq_pool_mutex, wq->mutex and pool->lock's. |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4863 | */ |
| 4864 | void freeze_workqueues_begin(void) |
| 4865 | { |
Tejun Heo | 1711696 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4866 | struct worker_pool *pool; |
Tejun Heo | 24b8a84 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4867 | struct workqueue_struct *wq; |
| 4868 | struct pool_workqueue *pwq; |
Tejun Heo | 611c92a | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 4869 | int pi; |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4870 | |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4871 | mutex_lock(&wq_pool_mutex); |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4872 | |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4873 | WARN_ON_ONCE(workqueue_freezing); |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4874 | workqueue_freezing = true; |
| 4875 | |
Tejun Heo | 24b8a84 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4876 | /* set FREEZING */ |
Tejun Heo | 611c92a | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 4877 | for_each_pool(pool, pi) { |
Tejun Heo | 5bcab33 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 4878 | spin_lock_irq(&pool->lock); |
Tejun Heo | 1711696 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4879 | WARN_ON_ONCE(pool->flags & POOL_FREEZING); |
| 4880 | pool->flags |= POOL_FREEZING; |
Tejun Heo | 5bcab33 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 4881 | spin_unlock_irq(&pool->lock); |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4882 | } |
| 4883 | |
Tejun Heo | 24b8a84 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4884 | list_for_each_entry(wq, &workqueues, list) { |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 4885 | mutex_lock(&wq->mutex); |
Tejun Heo | 699ce09 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 4886 | for_each_pwq(pwq, wq) |
| 4887 | pwq_adjust_max_active(pwq); |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 4888 | mutex_unlock(&wq->mutex); |
Tejun Heo | 24b8a84 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4889 | } |
Tejun Heo | 5bcab33 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 4890 | |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4891 | mutex_unlock(&wq_pool_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4892 | } |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4893 | |
| 4894 | /** |
Tejun Heo | 58a69cb | 2011-02-16 09:25:31 +0100 | [diff] [blame] | 4895 | * freeze_workqueues_busy - are freezable workqueues still busy? |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4896 | * |
| 4897 | * Check whether freezing is complete. This function must be called |
| 4898 | * between freeze_workqueues_begin() and thaw_workqueues(). |
| 4899 | * |
| 4900 | * CONTEXT: |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4901 | * Grabs and releases wq_pool_mutex. |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4902 | * |
Yacine Belkadi | d185af3 | 2013-07-31 14:59:24 -0700 | [diff] [blame] | 4903 | * Return: |
Tejun Heo | 58a69cb | 2011-02-16 09:25:31 +0100 | [diff] [blame] | 4904 | * %true if some freezable workqueues are still busy. %false if freezing |
| 4905 | * is complete. |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4906 | */ |
| 4907 | bool freeze_workqueues_busy(void) |
| 4908 | { |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4909 | bool busy = false; |
Tejun Heo | 24b8a84 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4910 | struct workqueue_struct *wq; |
| 4911 | struct pool_workqueue *pwq; |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4912 | |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4913 | mutex_lock(&wq_pool_mutex); |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4914 | |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4915 | WARN_ON_ONCE(!workqueue_freezing); |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4916 | |
Tejun Heo | 24b8a84 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4917 | list_for_each_entry(wq, &workqueues, list) { |
| 4918 | if (!(wq->flags & WQ_FREEZABLE)) |
| 4919 | continue; |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4920 | /* |
| 4921 | * nr_active is monotonically decreasing. It's safe |
| 4922 | * to peek without lock. |
| 4923 | */ |
Lai Jiangshan | 8810945 | 2013-03-20 03:28:10 +0800 | [diff] [blame] | 4924 | rcu_read_lock_sched(); |
Tejun Heo | 24b8a84 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4925 | for_each_pwq(pwq, wq) { |
Tejun Heo | 6183c00 | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 4926 | WARN_ON_ONCE(pwq->nr_active < 0); |
Tejun Heo | 112202d | 2013-02-13 19:29:12 -0800 | [diff] [blame] | 4927 | if (pwq->nr_active) { |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4928 | busy = true; |
Lai Jiangshan | 8810945 | 2013-03-20 03:28:10 +0800 | [diff] [blame] | 4929 | rcu_read_unlock_sched(); |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4930 | goto out_unlock; |
| 4931 | } |
| 4932 | } |
Lai Jiangshan | 8810945 | 2013-03-20 03:28:10 +0800 | [diff] [blame] | 4933 | rcu_read_unlock_sched(); |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4934 | } |
| 4935 | out_unlock: |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4936 | mutex_unlock(&wq_pool_mutex); |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4937 | return busy; |
| 4938 | } |
| 4939 | |
| 4940 | /** |
| 4941 | * thaw_workqueues - thaw workqueues |
| 4942 | * |
| 4943 | * Thaw workqueues. Normal queueing is restored and all collected |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 4944 | * frozen works are transferred to their respective pool worklists. |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4945 | * |
| 4946 | * CONTEXT: |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 4947 | * Grabs and releases wq_pool_mutex, wq->mutex and pool->lock's. |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4948 | */ |
| 4949 | void thaw_workqueues(void) |
| 4950 | { |
Tejun Heo | 24b8a84 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4951 | struct workqueue_struct *wq; |
| 4952 | struct pool_workqueue *pwq; |
| 4953 | struct worker_pool *pool; |
Tejun Heo | 611c92a | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 4954 | int pi; |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4955 | |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4956 | mutex_lock(&wq_pool_mutex); |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4957 | |
| 4958 | if (!workqueue_freezing) |
| 4959 | goto out_unlock; |
| 4960 | |
Tejun Heo | 24b8a84 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4961 | /* clear FREEZING */ |
Tejun Heo | 611c92a | 2013-03-13 16:51:36 -0700 | [diff] [blame] | 4962 | for_each_pool(pool, pi) { |
Tejun Heo | 5bcab33 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 4963 | spin_lock_irq(&pool->lock); |
Tejun Heo | 24b8a84 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4964 | WARN_ON_ONCE(!(pool->flags & POOL_FREEZING)); |
| 4965 | pool->flags &= ~POOL_FREEZING; |
Tejun Heo | 5bcab33 | 2013-03-13 19:47:40 -0700 | [diff] [blame] | 4966 | spin_unlock_irq(&pool->lock); |
Tejun Heo | 24b8a84 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4967 | } |
Tejun Heo | 8b03ae3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4968 | |
Tejun Heo | 24b8a84 | 2013-03-12 11:29:58 -0700 | [diff] [blame] | 4969 | /* restore max_active and repopulate worklist */ |
| 4970 | list_for_each_entry(wq, &workqueues, list) { |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 4971 | mutex_lock(&wq->mutex); |
Tejun Heo | 699ce09 | 2013-03-13 16:51:35 -0700 | [diff] [blame] | 4972 | for_each_pwq(pwq, wq) |
| 4973 | pwq_adjust_max_active(pwq); |
Lai Jiangshan | a357fc0 | 2013-03-25 16:57:19 -0700 | [diff] [blame] | 4974 | mutex_unlock(&wq->mutex); |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4975 | } |
| 4976 | |
| 4977 | workqueue_freezing = false; |
| 4978 | out_unlock: |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 4979 | mutex_unlock(&wq_pool_mutex); |
Tejun Heo | a0a1a5f | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 4980 | } |
| 4981 | #endif /* CONFIG_FREEZER */ |
| 4982 | |
Tejun Heo | bce9038 | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 4983 | static void __init wq_numa_init(void) |
| 4984 | { |
| 4985 | cpumask_var_t *tbl; |
| 4986 | int node, cpu; |
| 4987 | |
| 4988 | /* determine NUMA pwq table len - highest node id + 1 */ |
| 4989 | for_each_node(node) |
| 4990 | wq_numa_tbl_len = max(wq_numa_tbl_len, node + 1); |
| 4991 | |
| 4992 | if (num_possible_nodes() <= 1) |
| 4993 | return; |
| 4994 | |
Tejun Heo | d55262c | 2013-04-01 11:23:38 -0700 | [diff] [blame] | 4995 | if (wq_disable_numa) { |
| 4996 | pr_info("workqueue: NUMA affinity support disabled\n"); |
| 4997 | return; |
| 4998 | } |
| 4999 | |
Tejun Heo | 4c16bd3 | 2013-04-01 11:23:36 -0700 | [diff] [blame] | 5000 | wq_update_unbound_numa_attrs_buf = alloc_workqueue_attrs(GFP_KERNEL); |
| 5001 | BUG_ON(!wq_update_unbound_numa_attrs_buf); |
| 5002 | |
Tejun Heo | bce9038 | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 5003 | /* |
| 5004 | * We want masks of possible CPUs of each node which isn't readily |
| 5005 | * available. Build one from cpu_to_node() which should have been |
| 5006 | * fully initialized by now. |
| 5007 | */ |
| 5008 | tbl = kzalloc(wq_numa_tbl_len * sizeof(tbl[0]), GFP_KERNEL); |
| 5009 | BUG_ON(!tbl); |
| 5010 | |
| 5011 | for_each_node(node) |
Tejun Heo | 1be0c25 | 2013-05-15 14:24:24 -0700 | [diff] [blame] | 5012 | BUG_ON(!alloc_cpumask_var_node(&tbl[node], GFP_KERNEL, |
| 5013 | node_online(node) ? node : NUMA_NO_NODE)); |
Tejun Heo | bce9038 | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 5014 | |
| 5015 | for_each_possible_cpu(cpu) { |
| 5016 | node = cpu_to_node(cpu); |
| 5017 | if (WARN_ON(node == NUMA_NO_NODE)) { |
| 5018 | pr_warn("workqueue: NUMA node mapping not available for cpu%d, disabling NUMA support\n", cpu); |
| 5019 | /* happens iff arch is bonkers, let's just proceed */ |
| 5020 | return; |
| 5021 | } |
| 5022 | cpumask_set_cpu(cpu, tbl[node]); |
| 5023 | } |
| 5024 | |
| 5025 | wq_numa_possible_cpumask = tbl; |
| 5026 | wq_numa_enabled = true; |
| 5027 | } |
| 5028 | |
Suresh Siddha | 6ee0578 | 2010-07-30 14:57:37 -0700 | [diff] [blame] | 5029 | static int __init init_workqueues(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5030 | { |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 5031 | int std_nice[NR_STD_WORKER_POOLS] = { 0, HIGHPRI_NICE_LEVEL }; |
| 5032 | int i, cpu; |
Tejun Heo | c34056a | 2010-06-29 10:07:11 +0200 | [diff] [blame] | 5033 | |
Tejun Heo | e904e6c | 2013-03-12 11:29:57 -0700 | [diff] [blame] | 5034 | WARN_ON(__alignof__(struct pool_workqueue) < __alignof__(long long)); |
| 5035 | |
| 5036 | pwq_cache = KMEM_CACHE(pool_workqueue, SLAB_PANIC); |
| 5037 | |
Tejun Heo | 6575820 | 2012-07-17 12:39:26 -0700 | [diff] [blame] | 5038 | cpu_notifier(workqueue_cpu_up_callback, CPU_PRI_WORKQUEUE_UP); |
Lai Jiangshan | a5b4e57 | 2012-09-18 09:59:23 -0700 | [diff] [blame] | 5039 | hotcpu_notifier(workqueue_cpu_down_callback, CPU_PRI_WORKQUEUE_DOWN); |
Tejun Heo | 8b03ae3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 5040 | |
Tejun Heo | bce9038 | 2013-04-01 11:23:32 -0700 | [diff] [blame] | 5041 | wq_numa_init(); |
| 5042 | |
Tejun Heo | 706026c | 2013-01-24 11:01:34 -0800 | [diff] [blame] | 5043 | /* initialize CPU pools */ |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 5044 | for_each_possible_cpu(cpu) { |
Tejun Heo | 4ce62e9 | 2012-07-13 22:16:44 -0700 | [diff] [blame] | 5045 | struct worker_pool *pool; |
Tejun Heo | 8b03ae3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 5046 | |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 5047 | i = 0; |
Tejun Heo | f02ae73 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 5048 | for_each_cpu_worker_pool(pool, cpu) { |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 5049 | BUG_ON(init_worker_pool(pool)); |
Tejun Heo | ec22ca5 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 5050 | pool->cpu = cpu; |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 5051 | cpumask_copy(pool->attrs->cpumask, cpumask_of(cpu)); |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 5052 | pool->attrs->nice = std_nice[i++]; |
Tejun Heo | f3f90ad | 2013-04-01 11:23:34 -0700 | [diff] [blame] | 5053 | pool->node = cpu_to_node(cpu); |
Tejun Heo | 7a4e344 | 2013-03-12 11:30:00 -0700 | [diff] [blame] | 5054 | |
Tejun Heo | 9daf9e6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 5055 | /* alloc pool ID */ |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 5056 | mutex_lock(&wq_pool_mutex); |
Tejun Heo | 9daf9e6 | 2013-01-24 11:01:33 -0800 | [diff] [blame] | 5057 | BUG_ON(worker_pool_assign_id(pool)); |
Lai Jiangshan | 68e13a6 | 2013-03-25 16:57:17 -0700 | [diff] [blame] | 5058 | mutex_unlock(&wq_pool_mutex); |
Tejun Heo | 4ce62e9 | 2012-07-13 22:16:44 -0700 | [diff] [blame] | 5059 | } |
Tejun Heo | 8b03ae3 | 2010-06-29 10:07:12 +0200 | [diff] [blame] | 5060 | } |
| 5061 | |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 5062 | /* create the initial worker */ |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 5063 | for_each_online_cpu(cpu) { |
Tejun Heo | 4ce62e9 | 2012-07-13 22:16:44 -0700 | [diff] [blame] | 5064 | struct worker_pool *pool; |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 5065 | |
Tejun Heo | f02ae73 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 5066 | for_each_cpu_worker_pool(pool, cpu) { |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 5067 | pool->flags &= ~POOL_DISASSOCIATED; |
Tejun Heo | ebf44d1 | 2013-03-13 19:47:39 -0700 | [diff] [blame] | 5068 | BUG_ON(create_and_start_worker(pool) < 0); |
Tejun Heo | 4ce62e9 | 2012-07-13 22:16:44 -0700 | [diff] [blame] | 5069 | } |
Tejun Heo | e22bee7 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 5070 | } |
| 5071 | |
Tejun Heo | 8a2b753 | 2013-09-05 12:30:04 -0400 | [diff] [blame] | 5072 | /* create default unbound and ordered wq attrs */ |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 5073 | for (i = 0; i < NR_STD_WORKER_POOLS; i++) { |
| 5074 | struct workqueue_attrs *attrs; |
| 5075 | |
| 5076 | BUG_ON(!(attrs = alloc_workqueue_attrs(GFP_KERNEL))); |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 5077 | attrs->nice = std_nice[i]; |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 5078 | unbound_std_wq_attrs[i] = attrs; |
Tejun Heo | 8a2b753 | 2013-09-05 12:30:04 -0400 | [diff] [blame] | 5079 | |
| 5080 | /* |
| 5081 | * An ordered wq should have only one pwq as ordering is |
| 5082 | * guaranteed by max_active which is enforced by pwqs. |
| 5083 | * Turn off NUMA so that dfl_pwq is used for all nodes. |
| 5084 | */ |
| 5085 | BUG_ON(!(attrs = alloc_workqueue_attrs(GFP_KERNEL))); |
| 5086 | attrs->nice = std_nice[i]; |
| 5087 | attrs->no_numa = true; |
| 5088 | ordered_wq_attrs[i] = attrs; |
Tejun Heo | 29c91e9 | 2013-03-12 11:30:03 -0700 | [diff] [blame] | 5089 | } |
| 5090 | |
Tejun Heo | d320c03 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 5091 | system_wq = alloc_workqueue("events", 0, 0); |
Joonsoo Kim | 1aabe90 | 2012-08-15 23:25:39 +0900 | [diff] [blame] | 5092 | system_highpri_wq = alloc_workqueue("events_highpri", WQ_HIGHPRI, 0); |
Tejun Heo | d320c03 | 2010-06-29 10:07:14 +0200 | [diff] [blame] | 5093 | system_long_wq = alloc_workqueue("events_long", 0, 0); |
Tejun Heo | f342179 | 2010-07-02 10:03:51 +0200 | [diff] [blame] | 5094 | system_unbound_wq = alloc_workqueue("events_unbound", WQ_UNBOUND, |
| 5095 | WQ_UNBOUND_MAX_ACTIVE); |
Tejun Heo | 24d51ad | 2011-02-21 09:52:50 +0100 | [diff] [blame] | 5096 | system_freezable_wq = alloc_workqueue("events_freezable", |
| 5097 | WQ_FREEZABLE, 0); |
Viresh Kumar | 0668106 | 2013-04-24 17:12:54 +0530 | [diff] [blame] | 5098 | system_power_efficient_wq = alloc_workqueue("events_power_efficient", |
| 5099 | WQ_POWER_EFFICIENT, 0); |
| 5100 | system_freezable_power_efficient_wq = alloc_workqueue("events_freezable_power_efficient", |
| 5101 | WQ_FREEZABLE | WQ_POWER_EFFICIENT, |
| 5102 | 0); |
Joonsoo Kim | 1aabe90 | 2012-08-15 23:25:39 +0900 | [diff] [blame] | 5103 | BUG_ON(!system_wq || !system_highpri_wq || !system_long_wq || |
Viresh Kumar | 0668106 | 2013-04-24 17:12:54 +0530 | [diff] [blame] | 5104 | !system_unbound_wq || !system_freezable_wq || |
| 5105 | !system_power_efficient_wq || |
| 5106 | !system_freezable_power_efficient_wq); |
Suresh Siddha | 6ee0578 | 2010-07-30 14:57:37 -0700 | [diff] [blame] | 5107 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5108 | } |
Suresh Siddha | 6ee0578 | 2010-07-30 14:57:37 -0700 | [diff] [blame] | 5109 | early_initcall(init_workqueues); |