On Tue, May 23, 2023 at 01:15:55PM -0700, Bart Van Assche wrote: > On 5/23/23 01:17, Ming Lei wrote: > > On Tue, May 23, 2023 at 2:39 AM Bart Van Assche <bvanassche@xxxxxxx> wrote: > > > > > > Requeue requests instead of sending these to the dispatch list if a CPU > > > is unplugged. This gives the I/O scheduler the chance to preserve the > > > order of zoned write requests. > > > > But the affected code path is only for queue with none scheduler, do you > > think none can maintain the order for write requests? > > Hi Ming, > > Doesn't blk_mq_insert_requests() insert requests in ctx->rq_lists[type] > whether or not an I/O scheduler is active? blk_mq_insert_requests() is only called for inserting passthrough request in case of !q->elevator. > > I haven't found any code in blk_mq_hctx_notify_dead() that makes this > function behave differently based on whether or not an I/O scheduler has > been associated with the request queue. Did I perhaps overlook something? blk_mq_hctx_notify_dead() is only for handling request from sw queue which is used for !q->elevator. Thanks, Ming
