Re: [PATCH AUTOSEL 5.15 11/12] x86/barrier: Do not serialize MSR accesses on AMD

From: Borislav Petkov
Date: Fri Nov 29 2024 - 08:31:50 EST


On Thu, Nov 28, 2024 at 07:21:50PM -0500, Sasha Levin wrote:
> Suggestions welcome... I don't really insist on a massive flood and was
> mostly following what was the convention when I started doing this work.
>
> I thought about cutting it down to one mail per commit, but OTOH I had
> folks complain often enough that they missed a mail or that it wasn't
> obvious enough.

I don't know. It is a lot of mail and I doubt people look at most of them but
it's not like cutting down the mails from stable will alleviate the general
firehose drinking situation so...

> It might be all those thanksgiving drinks, but I can't find it in
> 5.15... I see it was part of 6.7.6 and 6.6.18, but nothing older.

Sorry, I meant 6.6.

> The stable kernel rules allow for "notable" performance fixes, and we
> already added it to 6.6.

Frankly, I don't have anything smarter than "well, it needs to be decided on
a case-by-case basis." As Erwan confirms, it really brings perf improvements
for their use case. So I guess that's a reason good enough to backport it
everywhere.

But cutting a general rule about such patches...? Nope, I can't think of any
good one.

> No objection to adding it to older kernels...
>
> Happy to do it after the current round of releases goes out.

Thanks.

--
Regards/Gruss,
Boris.

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