Re: [PATCH 0/2] x86/intel_rdt and perf/x86: Fix lack of coordination with perf

From: Reinette Chatre
Date: Fri Aug 03 2018 - 11:20:34 EST


Hi Peter,

On 8/3/2018 3:49 AM, Peter Zijlstra wrote:
> On Thu, Aug 02, 2018 at 01:43:42PM -0700, Reinette Chatre wrote:
>
>> The goal of this work is to use the existing PMU hardware coordination
>> mechanism to ensure that perf and resctrl will not interfere with each
>> other.
>
> I understand what it does.. but if I'd seen you frobbing at the PMU
> earlier your patches would've never gotten merged.
>
> It's simply not going to happen.

Your two-fold guidance is clear to me: (1) do not use PMU registers
directly, (2) use perf API instead.

You state that you understand what we are trying to do and I hope that I
convinced you that we are not able to accomplish the same by following
your guidance.

Could you please guide us how to obtain the accurate results we obtain
at this time while satisfying your requirements?

Earlier your response to a similar question was "That's the best you're
going to get" - unfortunately I cannot respectfully say that to our
customers when they indeed have become used to working with accurate
data for more than a year. I really want to find a solution that is
acceptable to mainline instead of having to maintain a solution needed
by customers out of tree.

Reinette