Re: [RFC PATCH 6/6] mm/memcontrol: Make memory.high tier-aware

[Donet Tom <donettom@linux.ibm.com>]

On 3/24/26 8:53 PM, Gregory Price wrote:
> On Tue, Mar 24, 2026 at 04:21:06PM +0530, Donet Tom wrote:
>> IIUC The intent of this patch is to partition cgroup memory such that
>> 0 → toptier_high is backed by higher-tier memory, and
>> toptier_high → max is backed by lower-tier memory.
>>
>> Based on this:
>>
>> 1.If top-tier usage exceeds toptier_high, pages should be
>>    demoted to the lower tier.
>>
>> 2. If lower-tier usage exceeds (max - toptier_high), pages
>>    should be swapped out.
>>
> This is not accurate and an incorrect heuristic.
>
> Transiently, lower-tier usage may exceed (max - toptier_high) for any
> number of reasons which should not be used as signal for pushing swap.
>
> driving swap usage is a function of (usage > memory.high) without regard
> for toptier / lowtier.
>
>> 3. If total memory usage exceeds max, demotion should be
>>    avoided and reclaim should directly swap out pages.
>>
> This is also incorrect, as it would drive agingin inversions.
> Demotion is a natural extension of the LRU infrastructure:
>
> toptier active -> toptier inactive -> lowtier inactive -> swap
>
> if you do (toptier inactive -> swap) you have inverted the LRU.


Thanks, Gregory, for the clarification.

One remaining concern is that under cgroup memory pressure,
demotion to the lower tier can still happen. Since demotion
does not uncharge the memcg, this could still trigger OOM.

Is this an issue we should address?


>
> As far as I know, from testing, we retain all the existing behavior - we
> are just managing a limited resource (top tier memory) to manage the
> noisy-neighbor issue.  So...
>
>
>> Should we also handle cases (2) and (3) in this patch?
> No, I don't think we should
>
> ~Gregory
>