Re: [PATCH][RT] x86: Fix an RT MCE crash

[Corey Minyard <minyard@acm.org>]

On 06/30/2016 12:26 PM, Borislav Petkov wrote:
> On Thu, Jun 30, 2016 at 12:18:01PM -0500, Corey Minyard wrote:
>> This is on 3.10-rt with PREEMPT_RT enabled.  It appears that from 3.18-rt
>> and later it has code like the change I have proposed, so it does not crash.
>>
>> I could add a something to see if the interrupt is coming in early to
>> 4.6-rt,
>> is that what you are looking for?
> Actually, I'd like to know first whether the unpatched upstream kernel -
> not -rt - is crashing.

It won't crash.  If you disable PREEMPT_RT on the 3.10-rt kernel it won't
crash (which I have tested).  With PREEMPT_RT, the kernel creates a
separate thread that is woken on mce notifications.  The trouble is
that the interrupts are initialized before the thread is created.

> And then 4.6-rt.
>
> Because from looking at your splat, you're getting a thresholding
> interrupt the moment you enable the local APIC and from staring at the
> MCE code upstream, I think we should be prepared for that scenario.
>
> AFAICT, both -rt and upstream should handle that case just fine and I'm
> guessing upstream was fixed at some point and -rt grew another fix which
> is probably not needed and it should take the upstream one instead...

This is not a bug in mainline.  This is only an RT bug, and only
with PREEMPT_RT enabled.

I can try these things if you really want, but it doesn't seem like
a useful activity to me.

It looks like in 3.18-rt someone noticed this issue and fixed it,
but the fix wasn't backported to earlier kernels.  I'm really just
trying to get that fix backported.

-corey