Re: [PATCH v3] mm/zswap: invalidate old entry when store fail or !zswap_enabled

[Yosry Ahmed]

> > @@ -1608,14 +1598,12 @@ bool zswap_store(struct folio *folio)
> >         /* map */
> >         spin_lock(&tree->lock);
> >         /*
> > -        * A duplicate entry should have been removed at the beginning of this
> > -        * function. Since the swap entry should be pinned, if a duplicate is
> > -        * found again here it means that something went wrong in the swap
> > -        * cache.
> > +        * The folio may have been dirtied again, invalidate the
> > +        * possibly stale entry before inserting the new entry.
> >          */
> > -       while (zswap_rb_insert(&tree->rbroot, entry, &dupentry) == -EEXIST) {
> > -               WARN_ON(1);
> > +       if (zswap_rb_insert(&tree->rbroot, entry, &dupentry) == -EEXIST) {
> >                 zswap_invalidate_entry(tree, dupentry);
> > +               VM_WARN_ON(zswap_rb_insert(&tree->rbroot, entry, &dupentry));
>
> It seems there is only one path called zswap_rb_insert() and there is
> no loop to repeat the insert any more. Can we have the
> zswap_rb_insert() install the entry and return the dupentry? We can
> still just call zswap_invalidate_entry() on the duplicate. The mapping
> of the dupentry has been removed when  zswap_rb_insert() returns. That
> will save a repeat lookup on the duplicate case.
> After this change, the zswap_rb_insert() will map to the xarray
> xa_store() pretty nicely.

I brought this up in v1 [1]. We agreed to leave it as-is for now since
we expect the xarray conversion soon-ish. No need to update
zswap_rb_insert() only to replace it with xa_store() later anyway.

[1] https://lore.kernel.org/lkml/ZcFne336KJdbrvvS@xxxxxxxxxx/