On Fri, Jan 12, 2024 at 06:22:59AM +0800, Ming Lei wrote: > On Thu, Jan 11, 2024 at 02:57:05PM +0100, Christoph Hellwig wrote: > > q_usage_counter is the only thing preventing us from the limits changing > > under us in __bio_split_to_limits, but blk_mq_submit_bio doesn't hold it. > > Can you share why the limits change matters wrt. split? If queue is > live, both new and old one are supposed to work, such as, we don't > freeze queue when changing `max_sectors_kb` via sysfs. Hardware reconfigurations for example, which can happen in nvme for just about every limit, in SCSI for limits controller by the ULD, and in quite a few places for disabling write zeroes or discard.
