Re: [PATCH v1] perf arm: Workaround ARM PMUs cpu maps having offline cpus

[Leo Yan]

Hi Ian,

On 6/10/24 18:33, Ian Rogers wrote:

[...]

> > Just a nitpick and I think it is not an issue caused by this patch.
> > After hotplug off one CPU and then if specify the CPU with option '-C',
> > the 'perf stat' command still continues to run. This is inconsistent
> > with the 'perf record' which exits with failures immediately.
> >
> > Maybe consider to send an extra patch to address this issue?
>
> As you say, this doesn't relate to the problem fixed here. I don't
> have a problem with the command line behavior change but I think my
> getting shouted at budget is fairly well exhausted.

I understand you put much efforts on fixing issues, on Arm platforms and
other platforms. This is also why I want to contribute a bit for testing
the patches.

> How about we trade issues, if the following get fixed:
>
> Renaming of the cycles event in arm_dsu_pmu.c - I'd say this is a top
> priority issue right now.

I cannot promise this. The main reason is that I still believe the
'cycles' event (or, generally speaking, all events) should be managed by
the tool rather than by the uncore PMU drivers. Additionally, the perf
tool currently has handled these symbolic events effectively.

> Fixing ARM hybrid default perf stat (it is crazy we can accept this
> being broken), opening events on both BIG.little PMUs as done here.

Yeah, James is working on this.

> I'll address this.

Anyway, I appreciate your work.

Thanks,
Leo