Re: [PATCH v1 2/2] mm/memory: add anonymous mTHP folios to deferred split list

Hi Barry,

On Tue, Aug 4, 2026 at 3:07 PM Barry Song <[email protected]> wrote:
>
> BTW, do you see any issue around 16? To me, it seems more
> sensible to use different thresholds for splitting and collapse,
> with the splitting threshold higher than the collapse threshold.
> That way, there is some hysteresis between the two, avoiding
> thrashing around the threshold.

I think PMD right now uses the same threshold for splitting and
collapse. It seems like if using the same splitting and collapse
threshold leads to thrashing, it would also manifest in the PMD case?
Just curious, have there been reports of this? For mTHP today though,
intermediary values for collapse can't be set, so I think that
thrashing scenario is automatically avoided.

If we wanted to separate the split and collapse threshold though,
maybe it makes more sense to set it internally (eg split threshold =
collapse threshold + some scale of nr_pages) instead of adding it as a
tunable. afaict, the only scenario where there's the thrashing
feedback loop on the same folio without any changes to the folio is if
someone sets the split threshold below the collapse threshold.

Thanks,
Joanne