Re: [RFC PATCH v2] mm: use per-numa-node atomics instead of percpu_counters

[Mathieu Desnoyers]

On 2025-04-03 13:59, Mathieu Desnoyers wrote:
> On 2025-04-02 20:00, Shakeel Butt wrote:
> > On Mon, Mar 31, 2025 at 06:35:14PM -0400, Sweet Tea Dorminy wrote:
> [...]
> > I am still not buying the 'good performance' point. To me we might need
> > to go with reduced batch size of existing approach or multi level
> > approach from Mathieu (I still have to see Mateusz and Kairui's
> > proposals).
>
> Here is an initial userspace prototype of my hierarchical split counters:
>
> https://github.com/compudj/librseq/blob/percpu-counter/include/rseq/ 
> percpu-counter.h
> https://github.com/compudj/librseq/blob/percpu-counter/src/percpu-counter.c

I've updated my branches to use unsigned long counters rather than
signed. This eliminates tricky 32-bit/64-bit overflow corner-cases and
should not have significant impact in practice.

Thanks,

Mathieu

--
Mathieu Desnoyers
EfficiOS Inc.
https://www.efficios.com