Skip to contentSkip to conversation

Linux 2.6.2, AMD kernel: MCE: The hardware reports a non fatal, correctable incident

DISCUSSION Andi Kleen 2 participants 2 messages Mar 4, 2004 2 messages not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): [email protected]

Andi Kleen

author
#2 lore raw
Quoted 2 lines · Dave Jones

I'm toying with the idea of marking it CONFIG_BROKEN for 2.6, and fixing it up later.

I would actually suggest to switch over to the rewritten MCE handler from x86-64 for i386 too. IMHO it is much better. It is race free, does not panic the machine if not needed, CPU independent, follows the Intel and AMD recommendations, run time sysfs configurable, logs to a separate device and does lots of other things much better [of course I'm biased on that a bit]. Disadvantage is that it isn't as well tested.

I haven't tried it on i386, but i wrote it to be easily portable to 32bit too. It does periodic MCEs too, but with a much lower frequency and they could be turned off. I'm considering to turn them off for x86-64 too, because they seem to only log one bit ECC errors all the time. But with the new separate log device it's much less of a problem.

The only thing you would lose is the support for P5 MCEs, but these could be relatively easily readded if that should be a problem. Extended register logging for P4 is also not implemented right now, but that hardly seems like a needed feature.

-Andi

#3 lore raw
Quoted 5 lines · Andi Kleen
Quoted 2 lines · Dave Jones

I'm toying with the idea of marking it CONFIG_BROKEN for 2.6, and fixing it up later.

Quoted 7 lines

I would actually suggest to switch over to the rewritten MCE handler from x86-64 for i386 too. IMHO it is much better. It is race free, does not panic the machine if not needed, CPU independent, follows the Intel and AMD recommendations, run time sysfs configurable, logs to a separate device and does lots of other things much better [of course I'm biased on that a bit]. Disadvantage is that it isn't as well tested.

Well, the only way to solve that problem is to test it, right? And what better way to test it than to switch i386 over to it too :-)

Quoted 9 lines

I haven't tried it on i386, but i wrote it to be easily portable to 32bit too. It does periodic MCEs too, but with a much lower frequency and they could be turned off. I'm considering to turn them off for x86-64 too, because they seem to only log one bit ECC errors all the time. But with the new separate log device it's much less of a problem.

The only thing you would lose is the support for P5 MCEs, but these could be relatively easily readded if that should be a problem.

Well, losing functionality would be bad.

Quoted 2 lines

Extended register logging for P4 is also not implemented right now, but that hardly seems like a needed feature.

No opinion here.

Regards: David Weinehall

Signature · 3 lines
 /) David Weinehall <[email protected]> /) Northern lights wander      (\
//  Maintainer of the v2.0 kernel   //  Dance across the winter sky //
\)  http://www.acc.umu.se/~tao/    (/   Full colour fire           (/