Re: [PATCH v1 1/1] mm: Fix OOM killer and proc stats inaccuracy on large many-core systems

[Andrew Morton]

On Tue, 13 Jan 2026 20:22:16 -0500 Mathieu Desnoyers <mathieu.desnoyers@xxxxxxxxxxxx> wrote:

> On 2026-01-13 18:55, Andrew Morton wrote:
> > On Tue, 13 Jan 2026 17:16:16 -0500 Mathieu Desnoyers <mathieu.desnoyers@xxxxxxxxxxxx> wrote:
> > 
> >> The hpcc series introduces an approximation which provides accuracy
> >> limits on the approximation that make the result is still somewhat
> >> meaninful on large many core systems.
> > 
> > Can we leave the non-oom related parts of procfs as-is for now, then
> > migrate them over to hpcc when that is available?  Safer that way.
> 
> Of course.
> 
> So AFAIU the plan is:
> 
> 1) update the oom accuracy fix to only use the precise sum for
>     the oom killer, no changes to procfs ABIs. This targets mm-new.
> 
> 2) update the hpcc series to base them on top of the new fix from (1).
>     Update their commit messages to indicate that they bring accuracy
>     improvements to the procfs ABI on large many-core systems, as well as
>     latency improvements to the oom killer. This will target upstreaming
>     after the next merge window, but I will still post it soon to gather
>     feedback.
> 
> Does that plan look OK ?

Perfect, thanks.  Except there is no "(1)".  We shall survive ;)