Series improving plugging for fast devices, but some fixes in here too. Patch 1 changes the logic for when we sort the plug list, switching it to only doing it if we have requests for multiple queues in the plug (and enough requests to warrant an actual sort). 2-6 add a ->commit_rqs() hook and implement it in drivers that use (or will use) bd->last to optimize IO submission. If a driver currently uses bd->last to know if it's needed to kick the hardware into action, there are cases where we flag bd->last == false, but then fail to submit any further IO due to other resource constraints. We probably get saved by the fact that this happens for the case where we have pending IO and that will eventually guarantee forward progress, but we really should kick IO into gear at that point. 7 improves plugging for blk-mq. In terms of improvements, for virtualized nvme, I've seen a 2-x IOPS improvement with proper handling of bd->last with this series. Changes since v1: - Fold the plug case for blk-mq, do it if we have 1 hw queue, or if we have ->commit_rqs(). - Add disable/enable IRQ to ataflop - Drop nvme ring index helper - Rework when to sort patch -- Jens Axboe
