The flush_work() annotation as introduced by commit: e159489baa71 ("workqueue: relax lockdep annotation on flush_work()") hits on the lockdep problem with recursive read locks. The situation as described is: Work W1: Work W2: Task: ARR(Q) ARR(Q) flush_workqueue(Q) A(W1) A(W2) A(Q) flush_work(W2) R(Q) A(W2) R(W2) if (special) A(Q) else ARR(Q) R(Q) where: A - acquire, ARR - acquire-read-recursive, R - release. Where under 'special' conditions we want to trigger a lock recursion deadlock, but otherwise allow the flush_work(). The allowing is done by using recursive read locks (ARR), but lockdep is broken for recursive stuff. However, there appears to be no need to acquire the lock if we're not 'special', so if we remove the 'else' clause things become much simpler and no longer need the recursion thing at all.
Showing the changes to kernel/workqueue.c (1 of 1 file). Show all
Unified · Split
| @@ -2091,7 +2091,7 @@ __acquires(&pool->lock) · 2090 unmodified lines | ||
| 2091 | 2091 | |
| 2092 | 2092 | spin_unlock_irq(&pool->lock); |
| 2093 | 2093 | |
| 2094 | − lock_map_acquire_read(&pwq->wq->lockdep_map); | |
| 2094 | + lock_map_acquire(&pwq->wq->lockdep_map); | |
| 2095 | 2095 | lock_map_acquire(&lockdep_map); |
| 2096 | 2096 | crossrelease_hist_start(XHLOCK_PROC); |
| 2097 | 2097 | trace_workqueue_execute_start(work); |
| @@ -2826,16 +2826,18 @@ static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr) · 728 unmodified lines | ||
| 2826 | 2826 | spin_unlock_irq(&pool->lock); |
| 2827 | 2827 | |
| 2828 | 2828 | /* |
| 2829 | − * If @max_active is 1 or rescuer is in use, flushing another work | |
| 2830 | − * item on the same workqueue may lead to deadlock. Make sure the | |
| 2831 | − * flusher is not running on the same workqueue by verifying write | |
| 2832 | − * access. | |
| 2829 | + * Force a lock recursion deadlock when using flush_work() inside a | |
| 2830 | + * single-threaded or rescuer equipped workqueue. | |
| 2831 | + * | |
| 2832 | + * For single threaded workqueues the deadlock happens when the work | |
| 2833 | + * is after the work issuing the flush_work(). For rescuer equipped | |
| 2834 | + * workqueues the deadlock happens when the rescuer stalls, blocking | |
| 2835 | + * forward progress. | |
| 2833 | 2836 | */ |
| 2834 | − if (pwq->wq->saved_max_active == 1 || pwq->wq->rescuer) | |
| 2837 | + if (pwq->wq->saved_max_active == 1 || pwq->wq->rescuer) { | |
| 2835 | 2838 | lock_map_acquire(&pwq->wq->lockdep_map); |
| 2836 | − else | |
| 2837 | − lock_map_acquire_read(&pwq->wq->lockdep_map); | |
| 2838 | − lock_map_release(&pwq->wq->lockdep_map); | |
| 2839 | + lock_map_release(&pwq->wq->lockdep_map); | |
| 2840 | + } | |
| 2839 | 2841 | |
| 2840 | 2842 | return true; |
| 2841 | 2843 | already_gone: |