Re: [PATCH v2 0/5] mm/hwpoison: Fix regressions in memory failure handling

[Borislav Petkov <bp@alien8.de>]

On Thu, Feb 20, 2025 at 05:50:14PM +0000, Luck, Tony wrote:
> Agreed. Shaui needs to harvest this thread to fill out the details in the commit
> messages.

Yap.

> There are probably other races. Two CPUs both take local #MC on the same page
> (maybe not all that rare in threaded processes ... or even with some hot code in 
> a shared library).

Yap, exactly. And I think there's nothing we can do - the hw is out there so
the sw needs to handle them cases correctly.

> Hmmm indeed. Needs some thought. Though failing to kill a process likely means
> it retries the access and comes right back to try again (without the race this time).

What happens if it fails to kill the process? It'll return to it, it'll try to
touch the faulty memory and raise another #MC? Right, I think so.


> > > On Intel that would mean not registering the notifier at all. What about AMD?
> > > Do you have similar races for MCE_DEFERRED_SEVERITY errors?
> >
> > Probably. Lemme ask around.

After talking to folks internally, yeah, I think we'll probably have a similar
thing. Haven't seen it happen yet.

> Linux tries to enable if LMCE is supported, but BIOS has veto power.
> See the bit in lmce_supported() that checks MSR_IA32_FEAT_CTL

I'm trying to educate our hw folks to not rely on OEM BIOS if possible. For
every chance I get. Otherwise you get crap like that and this is never getting
better.

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette