On Tue, Feb 04, 2025 at 08:02:17AM -0500, Martin K. Petersen wrote: > > Christoph, > > > 0xffff as the max and a weirdly uneven number even for partity RAID > > seems like a good value to ignore by default. > > Quite a few SCSI devices report 0xffff to indicate that the optimal > transfer length is the same as the maximum transfer length which for > low-byte commands is capped at 0xffff. That's where the odd value comes > from in some cases.
Hmm, optimal == max is odd, and I'm pretty sure it's going to confuse the hell out of whatever non-sophisticated users of the optimal I/O size that Mikulas is dealing with.
