Pull workqueue fix from Tejun Heo: "Lai's patch to fix highly unlikely but still possible workqueue stall during CPU hotunplug." * 'for-3.9-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq: workqueue: fix possible pool stall bug in wq_unbind_fn()
Showing the changes to kernel/workqueue.c (1 of 1 file). Show all
Unified · Split
| @@ -3447,28 +3447,34 @@ static void wq_unbind_fn(struct work_struct *work) · 3446 unmodified lines | ||
| 3447 | 3447 | |
| 3448 | 3448 | spin_unlock_irq(&pool->lock); |
| 3449 | 3449 | mutex_unlock(&pool->assoc_mutex); |
| 3450 | − } | |
| 3451 | 3450 | |
| 3452 | − /* | |
| 3453 | − * Call schedule() so that we cross rq->lock and thus can guarantee | |
| 3454 | − * sched callbacks see the %WORKER_UNBOUND flag. This is necessary | |
| 3455 | − * as scheduler callbacks may be invoked from other cpus. | |
| 3456 | − */ | |
| 3457 | − schedule(); | |
| 3451 | + /* | |
| 3452 | + * Call schedule() so that we cross rq->lock and thus can | |
| 3453 | + * guarantee sched callbacks see the %WORKER_UNBOUND flag. | |
| 3454 | + * This is necessary as scheduler callbacks may be invoked | |
| 3455 | + * from other cpus. | |
| 3456 | + */ | |
| 3457 | + schedule(); | |
| 3458 | 3458 | |
| 3459 | − /* | |
| 3460 | − * Sched callbacks are disabled now. Zap nr_running. After this, | |
| 3461 | − * nr_running stays zero and need_more_worker() and keep_working() | |
| 3462 | − * are always true as long as the worklist is not empty. Pools on | |
| 3463 | − * @cpu now behave as unbound (in terms of concurrency management) | |
| 3464 | − * pools which are served by workers tied to the CPU. | |
| 3465 | − * | |
| 3466 | − * On return from this function, the current worker would trigger | |
| 3467 | − * unbound chain execution of pending work items if other workers | |
| 3468 | − * didn't already. | |
| 3469 | − */ | |
| 3470 | − for_each_std_worker_pool(pool, cpu) | |
| 3459 | + /* | |
| 3460 | + * Sched callbacks are disabled now. Zap nr_running. | |
| 3461 | + * After this, nr_running stays zero and need_more_worker() | |
| 3462 | + * and keep_working() are always true as long as the | |
| 3463 | + * worklist is not empty. This pool now behaves as an | |
| 3464 | + * unbound (in terms of concurrency management) pool which | |
| 3465 | + * are served by workers tied to the pool. | |
| 3466 | + */ | |
| 3471 | 3467 | atomic_set(&pool->nr_running, 0); |
| 3468 | + | |
| 3469 | + /* | |
| 3470 | + * With concurrency management just turned off, a busy | |
| 3471 | + * worker blocking could lead to lengthy stalls. Kick off | |
| 3472 | + * unbound chain execution of currently pending work items. | |
| 3473 | + */ | |
| 3474 | + spin_lock_irq(&pool->lock); | |
| 3475 | + wake_up_worker(pool); | |
| 3476 | + spin_unlock_irq(&pool->lock); | |
| 3477 | + } | |
| 3472 | 3478 | } |
| 3473 | 3479 | |
| 3474 | 3480 | /* |