On Thu, Mar 16, 2000 at 07:44:21AM -0800, Jason R Thorpe wrote:
> This is a known problem.  Basically, the `ti' driver isn't properly
> bus_space'd; the access to the card shared memory is done via direct
> access to structs.
> 
> I was going to fix this, but I changed jobs and no longer have access to
> the cards.  One thing that's going to definitely need fixing is the implied
> bus space barriers that bus_space_{read,write}_*() currently do.  It's nice
> to have implied barriers in some cases, not in others.  A fix for this is
> fairly straight-forward.
> 
> I would like to work on it, and will be able to hack on it here and there.
> Would you be willing to test patches for me?

Certainly, but I'll have to put this box in production someday :)
I planned to look at converting it to bus_space this evening.
I'm not sure what you mean with "bus space barriers", I guess I need to read
more closely the code :)


> 
> FWIW, it would work on a DS20 if you kludge a few things, because the DS20's
> PCI memory space is not sparse (it's a BWX-capable space).

I removed BUS_SPACE_MAP_LINEAR and it worked ... but I agree this is not a good
solution :)

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