Re: [PATCH 2/2] x86, MCE: Drain mcelog buffer

[Borislav Petkov <bp@amd64.org>]

On Fri, Dec 09, 2011 at 02:08:59PM -0800, Luck, Tony wrote:
> > Well, off the top of my head, we could probably _not_ delete the already
> > logged MCEs (bool keep arg, or similar) and when the last one registers,
> > it passes keep=false and cleans them up. And since we probably know
> > which one is the last - order is enforced by the initcalls order -
> > we're done.
> 
> Knowing when the last one has come is hard if there are modules as well
> as built-in registrants.  We could save the whole list, and deliver a
> copy of the history of the world so far to each as they register (never
> deleting anything). But that sounds odd - why should a module loaded
> 6 months after system boot expect to see everything.

Well, currently the buffer overflows at the 32th entry so something that
registers 6 months after system boot will probably see unrelated errors
from 6 months ago depending on the error rate.

> I think if we ensure that the cpu decoder gets first chance to register,
> we should be OK with giving it all the pended stuff. We can defer doing
> something more complicated until the point that someone has a real use
> case.

Agreed, and this is currently the case on AMD and Intel, AFAICT, where
the i7 or sb edac module registers. We can always change it later if
required.

-- 
Regards/Gruss,
Boris.

Advanced Micro Devices GmbH
Einsteinring 24, 85609 Dornach
GM: Alberto Bozzo
Reg: Dornach, Landkreis Muenchen
HRB Nr. 43632 WEEE Registernr: 129 19551