Re: [PATCH v1 1/2] mm: zswap: Modified zswap_store_page() to process multiple pages in a folio.

[Yosry Ahmed]

On Mon, Dec 2, 2024 at 5:13 PM Sridhar, Kanchana P
<kanchana.p.sridhar@xxxxxxxxx> wrote:
>
>
> > -----Original Message-----
> > From: Yosry Ahmed <yosryahmed@xxxxxxxxxx>
> > Sent: Monday, December 2, 2024 11:34 AM
> > To: Sridhar, Kanchana P <kanchana.p.sridhar@xxxxxxxxx>
> > Cc: linux-kernel@xxxxxxxxxxxxxxx; linux-mm@xxxxxxxxx;
> > hannes@xxxxxxxxxxx; nphamcs@xxxxxxxxx; chengming.zhou@xxxxxxxxx;
> > usamaarif642@xxxxxxxxx; ryan.roberts@xxxxxxx; 21cnbao@xxxxxxxxx;
> > akpm@xxxxxxxxxxxxxxxxxxxx; Feghali, Wajdi K <wajdi.k.feghali@xxxxxxxxx>;
> > Gopal, Vinodh <vinodh.gopal@xxxxxxxxx>
> > Subject: Re: [PATCH v1 1/2] mm: zswap: Modified zswap_store_page() to
> > process multiple pages in a folio.
> >
> > On Wed, Nov 27, 2024 at 2:53 PM Kanchana P Sridhar
> > <kanchana.p.sridhar@xxxxxxxxx> wrote:
> > >
> > > Modified zswap_store() to store the folio in batches of
> > > SWAP_CRYPTO_BATCH_SIZE pages. Accordingly, refactored
> > zswap_store_page()
> > > into zswap_store_pages() that processes a range of pages in the folio.
> > > zswap_store_pages() is a vectorized version of zswap_store_page().
> > >
> > > For now, zswap_store_pages() will sequentially compress these pages with
> > > zswap_compress().
> > >
> > > These changes are follow-up to code review comments received for [1], and
> > > are intended to set up zswap_store() for batching with Intel IAA.
> > >
> > > [1]: https://patchwork.kernel.org/project/linux-
> > mm/patch/20241123070127.332773-11-kanchana.p.sridhar@xxxxxxxxx/
> > >
> > > Signed-off-by: Kanchana P Sridhar <kanchana.p.sridhar@xxxxxxxxx>
> > > ---
> > >  include/linux/zswap.h |   1 +
> > >  mm/zswap.c            | 154 ++++++++++++++++++++++++------------------
> > >  2 files changed, 88 insertions(+), 67 deletions(-)
> > >
> > > diff --git a/include/linux/zswap.h b/include/linux/zswap.h
> > > index d961ead91bf1..05a81e750744 100644
> > > --- a/include/linux/zswap.h
> > > +++ b/include/linux/zswap.h
> > > @@ -7,6 +7,7 @@
> > >
> > >  struct lruvec;
> > >
> > > +#define SWAP_CRYPTO_BATCH_SIZE 8UL
> > >  extern atomic_long_t zswap_stored_pages;
> > >
> > >  #ifdef CONFIG_ZSWAP
> > > diff --git a/mm/zswap.c b/mm/zswap.c
> > > index f6316b66fb23..b09d1023e775 100644
> > > --- a/mm/zswap.c
> > > +++ b/mm/zswap.c
> > > @@ -1409,78 +1409,96 @@ static void shrink_worker(struct work_struct
> > *w)
> > >  * main API
> > >  **********************************/
> > >
> > > -static ssize_t zswap_store_page(struct page *page,
> > > -                               struct obj_cgroup *objcg,
> > > -                               struct zswap_pool *pool)
> > > +/*
> > > + * Store multiple pages in @folio, starting from the page at index @si up to
> > > + * and including the page at index @ei.
> > > + */
> > > +static ssize_t zswap_store_pages(struct folio *folio,
> > > +                                long si,
> > > +                                long ei,
> > > +                                struct obj_cgroup *objcg,
> > > +                                struct zswap_pool *pool)
> > >  {
> > > -       swp_entry_t page_swpentry = page_swap_entry(page);
> > > +       struct page *page;
> > > +       swp_entry_t page_swpentry;
> > >         struct zswap_entry *entry, *old;
> > > +       size_t compressed_bytes = 0;
> > > +       u8 nr_pages = ei - si + 1;
> > > +       u8 i;
> > > +
> > > +       for (i = 0; i < nr_pages; ++i) {
> > > +               page = folio_page(folio, si + i);
> > > +               page_swpentry = page_swap_entry(page);
> > > +
> > > +               /* allocate entry */
> > > +               entry = zswap_entry_cache_alloc(GFP_KERNEL,
> > page_to_nid(page));
> > > +               if (!entry) {
> > > +                       zswap_reject_kmemcache_fail++;
> > > +                       return -EINVAL;
> > > +               }
> >
> > I think this patch is wrong on its own, for example if an allocation
> > fails in the above loop we exit without cleaning up previous
> > allocations. I think it's fixed in patch 2 but we cannot introduce
>
> I think there might be a misunderstanding. zswap_store_pages() will
> clean up local resources allocated during an iteration of the for loop,
> upon an error in that iteration. If you see the "goto store_failed" and
> "goto compress_failed" this would explain what I mean. If an allocation
> fails for an iteration, we simply return -EINVAL, and zswap_store()
> will goto the "put_pool" label with "ret" still false, which will delete
> all zswap entries for the folio (allocated up until the error iteration in
> zswap_store_pages(); or potentially already in the xarray).
>
> Hence, there is no bug and each of the two patches are correct by
> themselves AFAICT, but please let me know if I am missing anything. Thanks!

Uh yes, the cleanup is in zswap_store().

>
> > bugs in-between patches. I think the helpers in patch 2 don't really
> > help as I mentioned. Please combine the changes and keep them in the
> > main series (unless you have a reason not to).
>
> Sure. As noted in my earlier response to comments received for patch 2,
> I can either inline all iterations or create helpers for all iterations over
> the pages in a batch. Appreciate your suggestions on which would be a
> better approach.

I think leaving them open-coded will be clearer for now. We can
revisit the code organization later if it gets out of hand.