On Tue, May 24, 2022 at 10:44:44PM +0800, Ming Lei wrote: > The above line code may cause kernel panic in the following code paths: > > 1) blk_mq_debugfs_unregister_hctxs() called from __blk_mq_update_nr_hw_queues() > > 2) blk_mq_debugfs_unregister_sched_hctx() called from blk_mq_exit_sched() Yes. Then again we already have the same sort of race window in reverse at startup time, where nothing synchronizes between blk_mq_debugfs_register_hctxs from __blk_mq_update_nr_hw_queues and add_disk adding debugfs files. All this probably needs serializing using debugfs_mutex and a check of QUEUE_FLAG_REGISTERED.
