On Tue, May 24, 2022 at 02:01:34PM +0200, Christoph Hellwig wrote: > On Tue, May 24, 2022 at 07:47:03PM +0800, Ming Lei wrote: > > Not look into details yet, but as one blk-mq debugfs user, I don't like > > the idea, since I often see io hang issue when calling > > blk_mq_freeze_queue_wait(), and blk-mq debugfs is very helpful for > > investigating this kind of issue. > > > > But now blk-mq debugfs is gone with this patch __before__ draining IO in > > del_gendisk, and it becomes not useful as before. > > This is the way hot it is set up, so in doubt I want it to be torn down > synchronously. There might be a way to move the teardown later, but > that will require a very careful audit. Then please move it to disk release at least, when blk-mq debug becomes less useful. In theory, blk_cleanup_queue() may have similar issue, but most of drivers release disk after blk_cleanup_queue() except for scsi disk. So from user viewpoint, the best place to tear down blk-mq debugfs is still in queue release handler. Thanks, Ming
