On Fri, May 17, 2002 at 10:08:31PM -0400, Nathan J. Williams wrote:
> Manuel Bouyer <bouyer@antioche.eu.org> writes:
> 
> > OK, the len is wrong. The command comes from cdrecord.
> > Can you try to remplace patch-aj by the one attached and rebuild cdrecord ?
> 
> This fixed the problem, and I was able to re-enable ultra-DMA without
> any apparant problems (though it's not as fast as I think it should
> be).

OK, so I commited this change to pkgsrc

> 
> Is it a bug that cdrecord can wedge the system this way, given write
> access to the cd device?

Well, as the data bus is 16bit wide it's not surprising if IDE DMA can't deal
with 8bit data transfers. Now, the controller wedging the system instead of
reporting an error is probably a silicon bug (I suspect the IDE controller
has wedged the PCI bus completely). I commited a change to atapi_wdc.c
to protect against this (force PIO if the data len is not a multiple of
2 bytes).

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