On 09/21/2012 09:01 AM, Borislav Petkov wrote:
> Ah ok, my bad, I have 4 more patches ontop which should deal with the
> channel doubling, please repull the same branch which I've just updated.
>
> git://git.kernel.org/pub/scm/linux/kernel/git/bp/bp.git edac-queue
>
> and retest.
>
> And you're right about WIDTH_128 - it actually denotes the channel width
> and the code wrongly uses it to say that there are two channels on K8.
>
> However, this has been the case since the driver went upstream and I
> need to think a bit harder when correcting that so that I don't upset
> something else in it.
>
> Also, it would be very helpful if you send me full dmesg after you've
> applied the latest patches so that I can have another reference when
> changing the WIDTH_128 thing.
>
> Thanks for your help.

OK so progress, but still not there.

So this is correct now:
root@x.x.x.x:/sys/devices/system/edac/mc/mc0# grep . csrow*/size_mb
csrow0/size_mb:1024
csrow1/size_mb:1024
csrow2/size_mb:1024
csrow3/size_mb:1024

But this is not:
root@x.x.x.x:/sys/devices/system/edac/mc/mc0# grep . size_mb
8192

Attached patch seems to resolve reading size_mb from the mc in sysfs. 
I'm also attaching the new dmesg.

Josh