Martin, On 3/1/17 11:52, Martin K. Petersen wrote: >>>>>> "Christoph" == Christoph Hellwig <hch@xxxxxx> writes: > > Christoph> I don't really like this too much - this is too many SCSI > Christoph> specifics for the block layer to care. Maybe using bios for > Christoph> the zone ops was a mistake after all, and we should just have > Christoph> operations in struct block_device instead.. > > Yeah, I'm afraid I don't like this either. > > Short term I still think we should adjust in sd. And then maybe longer > term revisit whether these hybrid commands should be something other > than bios. OK. I will think about it. The problem remains that the mpt3sas driver needs fixing. As you suggest, we can do that in sd, or directly in mpt3sas. I tried to do a clean fix in sd, but always end up consuming a lot of aspirin because of all the potential corner cases to deal with. So I would prefer really just a fix in mpt3sas for now and think of the sd changes (if any) together with a redesign of the zone commands execution (not as BIOs). The file sd.h has the inline function scsi_medium_access_command() defined. We could move that to include/scsi/scsi.h (or scsi_proto.h) and use it in place of blk_rq_accesses_medium() in the mpt3sas driver to not force unaligned resid corrections for zone commands. Would that be acceptable ? Best regards. -- Damien Le Moal, Ph.D. Sr. Manager, System Software Research Group, Western Digital Corporation Damien.LeMoal@xxxxxxx (+81) 0466-98-3593 (ext. 513593) 1 kirihara-cho, Fujisawa, Kanagawa, 252-0888 Japan www.wdc.com, www.hgst.com
