Re: [PATCH] x86/mm: Rip out the TLB benchmarking knob

[Andy Lutomirski]

On Sat, Oct 14, 2017 at 5:34 AM, Borislav Petkov <bp@xxxxxxxxx> wrote:
> On Sat, Oct 14, 2017 at 03:49:12AM -0700, tip-bot for Andy Lutomirski wrote:
>> Commit-ID:  b956575bed91ecfb136a8300742ecbbf451471ab
>> Gitweb:     https://git.kernel.org/tip/b956575bed91ecfb136a8300742ecbbf451471ab
>> Author:     Andy Lutomirski <luto@xxxxxxxxxx>
>> AuthorDate: Mon, 9 Oct 2017 09:50:49 -0700
>> Committer:  Ingo Molnar <mingo@xxxxxxxxxx>
>> CommitDate: Sat, 14 Oct 2017 09:21:24 +0200
>>
>> x86/mm: Flush more aggressively in lazy TLB mode
>
> Andy,
>
> I know you're busy so lemme send out the piece removing the benchmarking
> knob. The good thing is, if you revert this patch, you get the knob back
> so it is actually there for those who wanna do benchmarking :-)
>
> It boots, but please check me nevertheless.

Ingo, please don't apply -- please use the one I just sent instead.

>
> Thx.
>
> ---
> From: Borislav Petkov <bp@xxxxxxx>
> Date: Sat, 14 Oct 2017 14:11:14 +0200
> Subject: [PATCH] x86/mm: Rip out the TLB benchmarking knob
>
> This is not really useful to have it upstream as it is only for
> benchmarking PCID vs !PCID and once the code is settled it will become
> one useless knob.

It's actually for benchmarking the two modes against each other on a
PCID CPU.  It's possible, though rather unlikely, that some workloads
are improved significantly when the knob changed.