On Wed, Apr 11, 2018 at 10:11:05AM +0800, Ming Lei wrote: > On Tue, Apr 10, 2018 at 03:01:57PM -0600, Bart Van Assche wrote: > > The blk-mq timeout handling code ignores completions that occur after > > blk_mq_check_expired() has been called and before blk_mq_rq_timed_out() > > has reset rq->aborted_gstate. If a block driver timeout handler always > > returns BLK_EH_RESET_TIMER then the result will be that the request > > never terminates. > > Under this situation: > > IMO, if this request has been handled by driver's irq handler, and if > driver's .timeout still returns BLK_EH_RESET_TIMER, it is driver's bug, > and the correct return value should be BLK_EH_HANDLED. We have plenty drivers that do that, so we'll need to audit all the drivers first. I guess a start would be to find a way that disables timeouts entirely.