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

[Sasha Levin <sashal@kernel.org>]

On Fri, Nov 29, 2024 at 09:34:53PM +0100, Pavel Machek wrote:
>On Fri 2024-11-29 08:38:48, Sasha Levin wrote:
>> On Fri, Nov 29, 2024 at 10:45:48AM +0100, Pavel Machek wrote:
>> > Hi!
>> >
>> > > > You've missed the 5.10 mail :)
>> > >
>> > > You mean in the flood? ;-P
>> > >
>> > > > Pavel objected to it so I've dropped it: https://lore.kernel.org/all/Zbli7QIGVFT8EtO4@sashalap/
>> > >
>> > > So we're not backporting those anymore? But everything else? :-P
>> > >
>> > > And 5.15 has it already...
>> > >
>> > > Frankly, with the amount of stuff going into stable, I see no problem with
>> > > backporting such patches. Especially if the people using stable kernels will
>> > > end up backporting it themselves and thus multiply work. I.e., Erwan's case.
>> >
>> > Well, some people would prefer -stable to only contain fixes for
>> > critical things, as documented.
>> >
>> > stable-kernel-rules.rst:
>> >
>> > - It must fix a problem that causes a build error (but not for things
>> >   marked CONFIG_BROKEN), an oops, a hang, data corruption, a real
>> >   security issue, or some "oh, that's not good" issue.  In short, something
>> >   critical.
>> >
>> > Now, you are right that reality and documentation are not exactly
>> > "aligned". I don't care much about which one is fixed, but I'd really
>> > like them to match (because that's what our users expect).
>>
>> You should consider reading past the first bullet in that section :)
>>
>>   - Serious issues as reported by a user of a distribution kernel may also
>>     be considered if they fix a notable performance or interactivity issue.
>>
>> It sounds like what's going on here, no?
>
>Is it? I'd not expect this to be visible in anything but
>microbenchmarks. Do you have user reports hitting this?
>
>It is not like this makes kernel build 10% slower, is it?

On Fri, Nov 29, 2024 at 10:30:11AM +0100, Erwan Velu wrote:
>This patch greatly impacts servers on production with AMD systems that
>have lasted since 5.11, having it backported really improves systems
>performance.
>Since this patch, I can share that our database team is no longer
>paged during the night, that's a real noticeable impact.

-- 
Thanks,
Sasha