Re: Opteron ECC Syndrome Issues

Stefan Reinauer  writes:
> Eric,
>
> with the latest CVS I am seeing memory errors on all addresses of the
> second dimm module in our solo systems. This does not happen on the
> HDAMA. Memory size is correctly detected though. Any good starting point
> for debugging this? The first dimm module works perfectly fine.

Is this ECC SDRAM?  If so the current code in the public CVS tree
is incorrect.  To get the check bits correct the memory needs to be
explicitly written to.

There are also a couple of B3 stepping errata that can cause bugs like
that.

And I believe there is an errata that says something about the first
dimm slot needing to be populated for the Athlon64.

So I would start by ensuring the know errata are fixed.  And testing
to see what the problems follows if you swap the DIMMS.

I need to push the bug fixes from my internal tree back to the public
tree.  This is a please bug me until I get it done request.  I will
see what I can do later today.

Eric