Re: [PATCH] zswap: initialize entry->pool on same filled entry

[Yosry Ahmed <yosryahmed@google.com>]

On Thu, Mar 21, 2024 at 4:53 PM Chris Li <chrisl@kernel.org> wrote:
>
> Current zswap will leave the entry->pool uninitialized if
> the page is same  filled. The entry->pool pointer can
> contain data written by previous usage.
>
> Initialize entry->pool to zero for the same filled zswap entry.
>
> Signed-off-by: Chris Li <chrisl@kernel.org>
> ---
> Per Yosry's suggestion to split out this clean up
> from the zxwap rb tree to xarray patch.
>
> https://lore.kernel.org/all/ZemDuW25YxjqAjm-@google.com/
> ---
>  mm/zswap.c | 1 +
>  1 file changed, 1 insertion(+)
>
> diff --git a/mm/zswap.c b/mm/zswap.c
> index b31c977f53e9..f04a75a36236 100644
> --- a/mm/zswap.c
> +++ b/mm/zswap.c
> @@ -1527,6 +1527,7 @@ bool zswap_store(struct folio *folio)
>                         kunmap_local(src);
>                         entry->length = 0;
>                         entry->value = value;
> +                       entry->pool = 0;

This should be NULL.

That being said, I am working on a series that should make non-filled
entries not use a zswap_entry at all. So I think this cleanup is
unnecessary, especially that it is documented in the definition of
struct zswap_entry that entry->pool is invalid for same-filled
entries.

>                         atomic_inc(&zswap_same_filled_pages);
>                         goto insert_entry;
>                 }
>
> ---
> base-commit: a824831a082f1d8f9b51a4c0598e633d38555fcf
> change-id: 20240315-zswap-fill-f65f44574760
>
> Best regards,
> --
> Chris Li <chrisl@kernel.org>
>