On Fri, 1 Jul 2022 at 10:54, Christophe Leroy
[off-list ref] wrote:

Hi Marco,

Le 28/06/2022 à 11:58, Marco Elver a écrit :

quoted

Internal data structures (cpu_bps, task_bps) of powerpc's hw_breakpoint
implementation have relied on nr_bp_mutex serializing access to them.

Before overhauling synchronization of kernel/events/hw_breakpoint.c,
introduce 2 spinlocks to synchronize cpu_bps and task_bps respectively,
thus avoiding reliance on callers synchronizing powerpc's hw_breakpoint.

We have an still opened old issue in our database related to
hw_breakpoint, I was wondering if it could have any link with the
changes you are doing and whether you could handle it at the same time.

https://github.com/linuxppc/issues/issues/38

Maybe it is completely unrelated, but as your series modifies only
powerpc and as the issue says that powerpc is the only one to do that, I
thought it might be worth a hand up.

I see the powerpc issue unrelated to the optimizations in this series;
perhaps by fixing the powerpc issue, it would also become more
optimal. But all I saw is that it just so happens that powerpc relied
on the nr_bp_mutex which is going away.

This series will become even more complex if I decided to add a
powerpc rework on top (notwithstanding the fact I don't have any ppc
hardware at my disposal either). A separate series/patch seems much
more appropriate.

Thanks,
-- Marco