Johannes, > here's a early preview of my SCSI results rework so we can eventually > discuss things next week at LSF/MM (it still has compiler errors on > aic7xxx and scsi_debug). Here's what I'd like to see: - Forget about cramming everything into that dreaded u32 result. Bart's suggestion of making it a union is fine but it's perpetuating that these fields are one thing. And they really aren't. I'd rather not be confined to the crusty old model throughout SCSI. We can synthesize the u32 result the few places where it's actually required (i.e. returned to userland). - Create separate scsi_cmnd status fields for host, driver, target (SAM), and transport (msg). And maybe block layer errno as suggested. - Fix the naming so it makes sense. No more DID_SPOT_RUN. Let's have HOST_FOO / DRIVER_BAR / SAM_STAT_BAZ so it's clear who said what. - For the first iteration, maybe we can keep the message goo. But ideally I'd like to see the SPI-specifics moved to the SPI transport class. And then have an opaque transport error field in the scsi_cmnd that other transports can use as well. - If the interface naming is sensible, I don't think we need wrappers. But if it helps the driver conversion, I don't have a problem having them as long as they are self-documenting: scsi_set_command_status(HOST_FOO, DRIVER_SENSE, SAM_STAT_BAZ); is infinitely less error prone than: result = (NON_SENSE << 24) | (DID_RUN << 16) | (SPOT_FORGOT_TO_SHIFT); I'll set aside a slot next week... -- Martin K. Petersen Oracle Linux Engineering
