On Mon, Feb 23, 2004 at 06:17:11PM -0600, James Bottomley wrote:
> On Tue, 2004-01-20 at 15:56, Martin Hicks wrote:
> > Updated patches against Linus' bk tree from earlier today.  The patch
> > that fixes up the scsi_sysfs.c error path, that was sent out by me on
> > Jan 16, is required for the core patch to apply cleanly.
> > 
> > I think I've addressed all the comments from the last round with the
> > exception of doing Lazy registration of the transport classes.  I'm
> > still not totally convinced that this is really required.
> 
> OK, I finally found time to look at all this and try it out.

Thanks!

> 
> My big objection is still that the scsi mid-layer needs to know about
> the transports.  In testing, I recoded this so that no mid-layer
> transport knowledge is necessary (patch attached).  I didn't opt for a
> transport registration API in the mid-layer, but I could see that such a
> thing might have been a better solution.

Agreed.  Thanks for getting the cruft out of the midlayer.

> I also see no necessity to allow hosts to add arbitrary transport
> attributes: they can simply add extra device attributes to the same
> effect.  This means that the attribute list is fixed, and we develop a
> private API between the transport and the device driver to update
> attributes (I didn't code this, but you can see where I'm going with
> it).

This is a future-needs thing, as far as I can see.  I'm going to leave
it as-is for now.  When the time comes that we want r/w sysfs attributes
to callback into the drivers it can be added then.

> There were a few other initialisation touches (Like making the
> attributes part of the host structure directly), and initialising the
> default transportt correctly in the host_alloc.

These are great.  Thanks.

> I did my implementation for the 53c700 which is the driver I can readily
> use.  As you can see, the other half of this is shifting the driver
> internally to use the transport attribute values (so that whatever we
> find there can be directly exported to the user).

I've expanded your changes to fix up the Fiber Channel transport
attributes using your new ideas.  I've also attached driver updates
for qla2xxx and qla1280.

thanks
mh

--
Martin Hicks                Wild Open Source Inc.
mort@wildopensource.com     613-266-2296