On Mon, 24 Oct 2005, Greg A. Woods wrote:

> Hmmm.... but I've already long ago pulled that patch up into my local
> 1.6.x tree that my running kernel was built from.
>
> Also FYI, My old machine had 1.5GB of RAM and never saw this problem
> (with the patch installed of course), the new one currently has 2GB.

  Hmm, I didn't notice that this was a AS4100, but it looks like it's
the symptoms I saw on the ES40.  In my case the mlx change fixed the
problem, but it's certainly possible the problem could be triggered by
something else.  What was happening to me was a direction flag was
being left in a bus_dma structure, which caused an extra spill page to be
allocated at the end of a transfer block, but that extra page was not
being accounted for when unloading the bus_dma mapping - resulting in
a request that didn't match the allocated extent.

  You probably should add the change in revision 1.59 of
alpha/common/bus_dma.c and revision 1.31 of alpha/common/sgmap_typedep.c
if you haven't already.  These will check to be sure there aren't any
stale flags.

  I don't know if revisions 1.29 and 1.30 of alpha/common/sgmap_typedep.c
would affect this or not, but they shouldn't hurt anything.


--
Michael L. Hitch			mhitch@montana.edu
Computer Consultant
Information Technology Center
Montana State University	Bozeman, MT	USA