Re: [PATCH v1 7/8] mm: zswap: Optimally place members of struct zswap_pool, struct zswap_entry.

[Kanchana P. Sridhar]

On Thu, Oct 15, 2026 at 3:06 PM Yosry Ahmed <yosry@xxxxxxxxxx> wrote:
>
> On Thu, Oct 15, 2026 at 2:49 PM Kanchana P. Sridhar
> <kanchanapsridhar2026@xxxxxxxxx> wrote:
> >
> > On Wed, Oct 14, 2026 at 8:43 AM Yosry Ahmed <yosry@xxxxxxxxxx> wrote:
> > >
> > > On Thu, Oct 08, 2026 at 11:28:13AM -0700, Kanchana P. Sridhar wrote:
> > > > Place and align the members of struct zswap_pool and struct zswap_entry
> > > > such that members used together in compress/decompress hot paths are
> > > > able to share the same cacheline. Minimize holes within these structs
> > > > while doing so.
> > > >
> > > > Signed-off-by: Kanchana P. Sridhar <kanchanapsridhar2026@xxxxxxxxx>
> > > > ---
> > > >  mm/zswap.c | 10 +++++-----
> > > >  1 file changed, 5 insertions(+), 5 deletions(-)
> > > >
> > > > diff --git a/mm/zswap.c b/mm/zswap.c
> > > > index 8ae67fd10522..ade84ac4076e 100644
> > > > --- a/mm/zswap.c
> > > > +++ b/mm/zswap.c
> > > > @@ -160,11 +160,11 @@ struct crypto_acomp_ctx {
> > > >  struct zswap_pool {
> > > >       struct zs_pool *zs_pool;
> > > >       struct crypto_acomp_ctx __percpu *acomp_ctx;
> > > > -     u8 compr_batch_size;
> > > >       struct percpu_ref ref;
> > > > -     struct rcu_work release_rwork;
> > > > -     struct hlist_node node;
> > > >       u8 idx;
> > > > +     u8 compr_batch_size;
> > > > +     struct hlist_node node;
> > > > +     struct rcu_work release_rwork;
> > >
> > > compr_batch_size is added by this series, why not just add it before idx
> > > to begin with?
> >
> > In patch 0006, where compr_batch_size is introduced, I was attempting
> > to place it in the same cacheline as the acomp_ctx. This increased the
> > size of struct zswap_pool to 248 bytes. I later optimized this in
> > patch 0007, bringing back the size to 240 with the addition of
> > compr_batch_size.
>
> This doesn't make much sense, and the size of zswap_pool is not that
> important (usually only one globally). I would drop the patch and add
> compr_batch_size directly where it needs to end up.

Sounds good. Thanks!