On Fri, Apr 27, 2001 at 04:02:52PM -0700, Jason R Thorpe wrote:
> So, for HBA hardware that automatically does the request-sense (i.e.
> BusLogic, etc.), the mid-layer should be able to handle that.  That
> is, if there is an error, and XS_SENSE is in the scsipi_xfer error word,
> then the mid-layer should *NOT* request sense again :-)
> 
> For HBA hardware that requires software to do it (i.e. the 53c9x, etc.),
> the the mid-layer should do it and HBA drivers should not.

Well, it's not that clear: for tagged queuing drivers, when a devices has
sense pending you shouldn't sent any new commands until the request sense
is complete. Some tagged commands may already have been given to the adapter,
but not yet accepted by the device. Depending on the case, it may be easier to
send the request sense from the driver itself, which can place the request
sense in front of the adapter's queue immediatly.

For non-tagged adapter, it's clear however: just report the check condition
to scsipi.

--
Manuel Bouyer <bouyer@antioche.eu.org>
--