Re: [PATCH v2] x86/mce: Distirbute the clear operation of mces_seen to Per-CPU rather than only monarch CPU

[Chen Yucong <slaoub@gmail.com>]

On Tue, 2014-05-20 at 19:33 +0200, Borislav Petkov wrote:
> On Tue, May 20, 2014 at 10:11:25AM +0800, Chen Yucong wrote:
> > mces_seen is a Per-CPU variable which should only be accessed by
> > Per-CPU as possible. So the clear operation of mces_seen should also
> > be lcoal to Per-CPU rather than monarch CPU.
> >
> > Meanwhile, there is also a potential risk that mces_seen will not
> > be be cleared if a timeout occors in mce_end for monarch CPU. As a
> > reuslt, the stale value of mces_seen will reappear on the next mce.
> 
> I don't know how many times I have to tell you this already: if we reach
> the timeout, we have a much bigger friggin' problem!

Even if we do not take into account timeout, we should distribute the
clear operation of mces_seen to Per-CPU rather then monarch CPU.
mce_regin, which is only called by monarch CPU, can be used for system
panics as quickly as possible if there is a truly data corrupting error.
But Monarch CPU don't have to help all other CPU to clean mces_clean.
One advantage of Per-CPU is the isolation of errors propagation, being
so, why do not we clean mces_seen by Per-CPU?

You say, "you need to do the cleaning in mce_reign because the monarch
cpu has to run last after all other cpus have scanned their mce banks."
But this is not an adequate explanation.

thx!
cyc

> 
> What you could do instead is make the machine panic in the tolerant==1,
> i.e., the default case, in mce_timed_out().
> 
> Basically, in the case any core is stuck and we reach a timeout, we want
> to panic the whole box immediately. There's a very little chance we can
> recover so panic is the only sane thing left to do.
> 
> Ok?
>