On Mon, Nov 21, 2022 at 12:03:02AM -0800, Christoph Hellwig wrote: > I've been wading through this thread a bit, and while I can't follow > everything due some pretty bad reply trimming I'm really really > confused by this whole discussion. > > Every single storage system, be that hard drives, SSDs, file systems > or what else much preferres sequential write. Any code that turns > perfectly sequential writes into reverse sequential writes is a massive > performance bug, zoned or not. > > So totally independent of zone device support in ublk this needs to be > fixed ASAP. I will send one fix soon. BTW, blk-mq has such similar issue, which is reported for while, but not addressed yet. https://lore.kernel.org/linux-nvme/CAHex0coMNvFa1TPzK+5mZHsiie4d1Jd0Z8ejcZk1Vi1_4F7eRg@xxxxxxxxxxxxxx/T/#mbcb9b52ff61aed6fdd1b6630c6e62e55a4ed898f Thanks, Ming