We have come to discover some interesting things that certain
platforms will do that are legal with respect to the PCI spec.
Specifically, it's possible to be doing a burst write where all of the
byte-enabled are de-asserted for any of the transfers in the burst.
In other words, the cycle on the bus does not transfer data and serves
only to increment the address.  We have actually encountered this
happening.

For OGA's access to graphics memory, that is not a problem, because we
pay attention to the byte enables.  However, this IS a problem for
writing to control registers.  We completely ignore the byte enabled.
This is justified because a byte write to a config register is simply
(software) programmer error.  However, it's completely legit to have
ALL of the bytes deasserted (and we have to have no effects from it).
I'm not sure we can toss out the write in the XP10, because that could
throw off counters, so what I think we need to do is have the address
decode logic in the S3 figure it out and discard the dummy write.

-- 
Timothy Normand Miller
http://www.cse.ohio-state.edu/~millerti
Open Graphics Project
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)

Reply via email to