On Tue, May 30, 2023 at 4:54 PM Johannes Weiner <hannes@xxxxxxxxxxx> wrote:
>
> On Tue, May 30, 2023 at 09:02:51PM +0000, Yosry Ahmed wrote:
> > @@ -216,8 +216,13 @@ int __frontswap_load(struct page *page)
> >
> > /* Try loading from each implementation, until one succeeds. */
> > ret = frontswap_ops->load(type, offset, page);
> > - if (ret == 0)
> > + if (ret == 0) {
> > inc_frontswap_loads();
> > + if (frontswap_ops->exclusive_loads) {
> > + SetPageDirty(page);
> > + __frontswap_clear(sis, offset);
> > + }
>
> Somewhat tangential, but is there still a point to the frontswap
> layer? It seems usecases other than zswap have never materialized, at
> least not in tree. Life would be a lot easier if we were to just
> hardcode the zswap callbacks in the swap functions.
>
> It's not the patch's fault, but it highlights the boiler plate the
> indirection causes. ->load() already has the page and could just dirty
> it directly. Instead now both layers have to handle invalidation,
> which is a vector for bugs.
>
> Can somebody think of reasons to keep it? If not, I'd take a stab at
> removing it.
I was intending to eventually remove it as part of the swap
abstraction proposal I am working on, but this will take some time. If
you want to remove it now, even better :)
