On Wed, Oct 19, 2022 at 09:30:35AM +0200, Christoph Hellwig wrote: > Based on the per-tagset srcu for quiesce discusion I've been wondering > why we need the queue quiesce around elevator_exit and rq_qos_exit in > del_gendisk. At the point where we call it, we've stopped new fs > I/O submissions, and the queue is frozen. What does the quiesce still > protect against, and if anything can we come up with a cheaper way to > do that? There might be in-progress run queue not finished yet, only rcu can guarantee that. Thanks, Ming
