On Thu, Jan 20, 2022 at 05:29:50PM +0100, Daniel Wagner wrote: > On Thu, Jan 20, 2022 at 11:37:00PM +0800, Ming Lei wrote: > > Seems more likely nvme recovery issue, tags is changed even though > > ->q_usage_counter is grabbed in blk_mq_queue_tag_busy_iter(), which is > > added in v4.19. > > > > You may check dmesg log and see if there is any nvme error recovery info. > > Thanks for the tip. It's likely that a recovery is running although > there is no entry in the logs. At least I have an idea what could be the > problem here. FWIW, a846a8e6c9a5 ("blk-mq: don't free tags if the tag_set is used by other device in queue initialztion") was missing our tree.