On Friday, November 19, 2021 at 11:56:55 AM Christoph Hellwig wrote: > >Direct reply without a quote a the formatting of your mail is completely >messed up. Sorry about that - hopefully this is better. >We don't treat a FUA a a flush. If the device supports >FUA is is just passed on. Does this mean a FUA request for a device that natively supports FUA is eventually "passed on" to the scheduler? Or does it go straight to dispatch? >But if the device does not support FUA it >is sequenced into doing the write first and then a flush before returning >completion to the caller to guarantee the FUA semantics. Because of that >the command needs special treatment and be handed over to the flush >state machine, but it won't do anything interesting if FUA is actually >natively supported. > Yes, I can see why this flush logic is needed for devices that don't support FUA. Thanks.
