perf: fix reproducibility issues

perf: fix reproducibility issues

[Denys Zagorui -X (dzagorui - GLOBALLOGIC INC@Cisco)]

Hi Bruce

I've just found that were working on the same reproducibility issues for perf package. You fixed that in yocto 
http://cgit.openembedded.org/openembedded-core/commit/meta/recipes-kernel/perf/perf.bb?id=dea57aa1cd879df4feae1b9595b38c5b6ff74574
i prepared a patch to kernel that i want to send to community, maybe you have concerns about this.

Regards,
Denys

Re: perf: fix reproducibility issues

[Bruce Ashfield]

On Wed, Mar 10, 2021 at 2:21 PM Denys Zagorui -X (dzagorui -
GLOBALLOGIC INC at Cisco) <dzagorui@cisco.com> wrote:
>
> Hi Bruce
>
> I've just found that were working on the same reproducibility issues for perf package. You fixed that in yocto
> http://cgit.openembedded.org/openembedded-core/commit/meta/recipes-kernel/perf/perf.bb?id=dea57aa1cd879df4feae1b9595b38c5b6ff74574
> i prepared a patch to kernel that i want to send to community, maybe you have concerns about this.

The fixes in the yocto tree aren't enough for full reproducibility. If
you've also fixed the pmu-events.c issues, then that is of interest,
since they still remain.

Due to the way we build perf, we can't actually patch it for fixes, it
must be source modifications like I've done in my initial pass.
Eventually once upstream is fully reproducible, and our oldest support
kernel is newer than those versions, we'll drop the perf recipe
changes.

The right thing is to fix it in the kernel, which is what I was going
to submit after we had proven our changes, but it sounds like I won't
have to do that part :D If you have a pointer to any submitted
patches, that would be great .. since I took a look at my lkml
subscription and didn't see anything obvious.

Cheers,

Bruce

>
> Regards,
> Denys




--
- Thou shalt not follow the NULL pointer, for chaos and madness await
thee at its end
- "Use the force Harry" - Gandalf, Star Trek II