Re: [RFC] x86, NMI, Treat unknown NMI as hardware error

[Huang Ying <ying.huang@intel.com>]

On 05/15/2011 02:34 PM, Cyrill Gorcunov wrote:
> On 05/15/2011 04:06 AM, huang ying wrote:
> ...
>>>
>>> yes, is not good. But at least we *must* provide a way to turn this new feature off
>>> via command line I think. One of a reason for me is perf unknown nmis (at moment we seems
>>> to have captured and cured all parasite NMIs sources but there is no guarantee we wont
>>> meet them in future due to some code change or whatever). And bloating trap.c with
>>> new if()'s is not that good I guess, that is why I asked if there a way to do all the
>>> work via notifiers ;)
>>
>> Yes.  We should consider about perf unknown NMI issues.  But compared
>> with pushing all magic to user, I think the better way is to have a
>> better default behavior in kernel.  For example, we can turn off
>> unknown NMI as hwerr logic temporarily if there are more than 1 perf
>> NMI events in action.  Is that reasonable?
> 
>   I'm personally fine even if it's enabled by default, only worried to have
> an option to disable hwerr from boot line.

The white list mechanism is not sufficient?  Spurious unknown NMI can
occur on white list machines?  People don't want to protect their data?

>> And, I am not a big fan of notifiers, that makes code hard to be
>> understood.  If you have concerns about the size of traps.c, we can
>> move all NMI logic to a new file.
> 
>   Ying, the concern is rather related to the code scheme in general. Since
> we have notifiers I think the better way to be consistent here and use
> hwerr notifier too. But it's IMHO ;)

As for go notifiers or not.  IMHO, a rule can be:

- If it is something like a driver, than it should go notifier
- If it is architectural/PC defacto standard, it can sit outside of
notifier.

I think that seeing unknown NMI as hardware error should be part of PC
defacto standard.  Do you think so?

Best Regards,
Huang Ying