On Thu, Aug 10, 2023 at 11:42:50AM +0800, Chengming Zhou wrote: > Christoph, it would be very helpful if you share some thoughts > and directions. I am not quite sure where the problems could even be. I think the interesting aspects are: - we are not directly processing through the normal submission path instead of adding the command to the flush_data_in_flight list and then going through blk_kick_flush. So a command now gets executed earlier than it did before. - given that it only happens with SATA, it must be in some way related to the fact that ATA flush commands are not queud, that is when a flush is outstanding no other command can't.
