On Fri, Sep 22, 2023 at 05:57:40PM +0000, Yosry Ahmed wrote:
> memcg_rstat_updated() uses the value of the state update to keep track
> of the magnitude of pending updates, so that we only do a stats flush
> when it's worth the work. Most values passed into memcg_rstat_updated()
> are in pages, however, a few of them are actually in bytes or KBs.
>
> To put this into perspective, a 512 byte slab allocation today would
> look the same as allocating 512 pages. This may result in premature
> flushes, which means unnecessary work and latency.
Yikes.
I'm somewhat less concerned about the performance as I am about the
variance in flushing cost that could be quite difficult to pinpoint.
IMO this is a correctness fix and a code cleanup, not a performance
thing.
> Normalize all the state values passed into memcg_rstat_updated() to
> pages. Round up non-zero sub-page to 1 page, because
> memcg_rstat_updated() ignores 0 page updates.
>
> Fixes: 5b3be698a872 ("memcg: better bounds on the memcg stats updates")
> Signed-off-by: Yosry Ahmed <yosryahmed@xxxxxxxxxx>
Acked-by: Johannes Weiner <hannes@xxxxxxxxxxx>
