Received-SPF: pass (google.com: domain of linux-kernel+bounces-206856-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.48.161 as permitted sender) client-ip=147.75.48.161;
Precedence: bulk
MIME-Version: 1.0
References: <20240606184818.1566920-1-yosryahmed@google.com>
 <84d78362-e75c-40c8-b6c2-56d5d5292aa7@redhat.com> <CAJD7tkZH9C21nx75W9Erun=oUvmad5ujmDyGYWRRHEwPCCizUw@mail.gmail.com>
 <7507d075-9f4d-4a9b-836c-1fbb2fbd2257@redhat.com> <CAGsJ_4w_dEyLsPhdJCtRMReXDD116p+U979Yk-8EBu=zx5FWgQ@mail.gmail.com>
 <CAJD7tkYnZCvwUSTAx-N-08b+mVF4yHEdDTbp9oqYrVQXX8M1Cw@mail.gmail.com>
 <CAGsJ_4wAYbofjg99iEr7cEAptH9Ve9WcrSXho8A-Sw54vb8Msw@mail.gmail.com>
 <9374758d-9f81-4e4f-8405-1f972234173e@redhat.com> <CAJD7tkZBzSB_6pAZP0n0nq+W=J1XKQGFzZZLzPmSH0apwaqTNg@mail.gmail.com>
 <424c6430-e40d-4a60-8297-438fc33056c9@redhat.com> <CAJD7tkbaX-8OQm0JispxWWbuzX+PJrH=HN6yqEjqL2yxRcu=Aw@mail.gmail.com>
 <CAGsJ_4zu1_oMsskrndfpk0gaaHGDrTRnMKkQxF05jwCabxtfDQ@mail.gmail.com>
In-Reply-To: <CAGsJ_4zu1_oMsskrndfpk0gaaHGDrTRnMKkQxF05jwCabxtfDQ@mail.gmail.com>
From: Yosry Ahmed <yosryahmed@google.com>
Date: Fri, 7 Jun 2024 17:28:30 -0700
Message-ID: <CAJD7tkakM+1EcgoZyJFcH4Tk2NU4C-knkh43kWUN0f2oKHxyzw@mail.gmail.com>
Subject: Re: [PATCH] mm: zswap: add VM_BUG_ON() if large folio swapin is attempted
To: Barry Song <21cnbao@gmail.com>
Cc: David Hildenbrand <david@redhat.com>, Andrew Morton <akpm@linux-foundation.org>, 
	Johannes Weiner <hannes@cmpxchg.org>, Nhat Pham <nphamcs@gmail.com>, 
	Chengming Zhou <chengming.zhou@linux.dev>, Baolin Wang <baolin.wang@linux.alibaba.com>, 
	Chris Li <chrisl@kernel.org>, Ryan Roberts <ryan.roberts@arm.com>, 
	Matthew Wilcox <willy@infradead.org>, linux-mm@kvack.org, linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Fri, Jun 7, 2024 at 3:09=E2=80=AFPM Barry Song <21cnbao@gmail.com> wrote=
:
>
> On Sat, Jun 8, 2024 at 6:58=E2=80=AFAM Yosry Ahmed <yosryahmed@google.com=
> wrote:
> >
> > On Fri, Jun 7, 2024 at 11:52=E2=80=AFAM David Hildenbrand <david@redhat=
.com> wrote:
> > >
> > > >> I have no strong opinion on this one, but likely a VM_WARN_ON woul=
d also
> > > >> be sufficient to find such issues early during testing. No need to=
 crash
> > > >> the machine.
> > > >
> > > > I thought VM_BUG_ON() was less frowned-upon than BUG_ON(), but afte=
r
> > > > some digging I found your patches to checkpatch and Linus clearly
> > > > stating that it isn't.
> > >
> > > :) yes.
> > >
> > > VM_BUG_ON is not particularly helpful IMHO. If you want something to =
be
> > > found early during testing, VM_WARN_ON is good enough.
> > >
> > > Ever since Fedora stopped enabling CONFIG_DEBUG_VM, VM_* friends are
> > > primarily reported during early/development testing only. But maybe s=
ome
> > > distro out there still sets it.
> > >
> > > >
> > > > How about something like the following (untested), it is the minima=
l
> > > > recovery we can do but should work for a lot of cases, and does
> > > > nothing beyond a warning if we can swapin the large folio from disk=
:
> > > >
> > > > diff --git a/mm/page_io.c b/mm/page_io.c
> > > > index f1a9cfab6e748..8f441dd8e109f 100644
> > > > --- a/mm/page_io.c
> > > > +++ b/mm/page_io.c
> > > > @@ -517,7 +517,6 @@ void swap_read_folio(struct folio *folio, struc=
t
> > > > swap_iocb **plug)
> > > >          delayacct_swapin_start();
> > > >
> > > >          if (zswap_load(folio)) {
> > > > -               folio_mark_uptodate(folio);
> > > >                  folio_unlock(folio);
> > > >          } else if (data_race(sis->flags & SWP_FS_OPS)) {
> > > >                  swap_read_folio_fs(folio, plug);
> > > > diff --git a/mm/zswap.c b/mm/zswap.c
> > > > index 6007252429bb2..cc04db6bb217e 100644
> > > > --- a/mm/zswap.c
> > > > +++ b/mm/zswap.c
> > > > @@ -1557,6 +1557,22 @@ bool zswap_load(struct folio *folio)
> > > >
> > > >          VM_WARN_ON_ONCE(!folio_test_locked(folio));
> > > >
> > > > +       /*
> > > > +        * Large folios should not be swapped in while zswap is bei=
ng used, as
> > > > +        * they are not properly handled.
> > > > +        *
> > > > +        * If any of the subpages are in zswap, reading from disk w=
ould result
> > > > +        * in data corruption, so return true without marking the f=
olio uptodate
> > > > +        * so that an IO error is emitted (e.g. do_swap_page() will=
 sigfault).
> > > > +        *
> > > > +        * Otherwise, return false and read the folio from disk.
> > > > +        */
> > > > +       if (WARN_ON_ONCE(folio_test_large(folio))) {
> > > > +               if (xa_find(tree, &offset, offset +
> > > > folio_nr_pages(folio) - 1, 0))
> > > > +                       return true;
> > > > +               return false;
> > > > +       }
> > > > +
> > > >          /*
> > > >           * When reading into the swapcache, invalidate our entry. =
The
> > > >           * swapcache can be the authoritative owner of the page an=
d
> > > > @@ -1593,7 +1609,7 @@ bool zswap_load(struct folio *folio)
> > > >                  zswap_entry_free(entry);
> > > >                  folio_mark_dirty(folio);
> > > >          }
> > > > -
> > > > +       folio_mark_uptodate(folio);
> > > >          return true;
> > > >   }
> > > >
> > > > One problem is that even if zswap was never enabled, the warning wi=
ll
> > > > be emitted just if CONFIG_ZSWAP is on. Perhaps we need a variable o=
r
> > > > static key if zswap was "ever" enabled.
> > >
> > > We should use WARN_ON_ONCE() only for things that cannot happen. So i=
f
> > > there are cases where this could be triggered today, it would be
> > > problematic -- especially if it can be triggered from unprivileged us=
er
> > > space. But if we're concerned of other code messing up our invariant =
in
> > > the future (e.g., enabling large folios without taking proper care ab=
out
> > > zswap etc), we're good to add it.
> >
> > Right now I can't see any paths allocating large folios for swapin, so
> > I think it cannot happen. Once someone tries adding it, the warning
> > will fire if CONFIG_ZSWAP is used, even if zswap is disabled.
> >
> > At this point we will have several options:
> > - Make large folios swapin depend on !CONFIG_ZSWAP for now.
>
> It appears quite problematic. We lack control over whether the kernel bui=
ld
> will enable CONFIG_ZSWAP, particularly when aiming for a common
> defconfig across all platforms to streamline configurations. For instance=
,
> in the case of ARM, this was once a significant goal.
>
> Simply trigger a single WARN or BUG if an attempt is made to load
> large folios in zswap_load, while ensuring that zswap_is_enabled()
> remains unaffected. In the mainline code, large folio swap-in support
> is absent, so this warning is intended for debugging purposes and
> targets a very small audience=E2=80=94perhaps fewer than five individuals
> worldwide. Real users won=E2=80=99t encounter this warning, as it remains
> hidden from their view.

I can make the warning only fire if any part of the folio is in zswap
to avoid getting warnings from zswap_load() if we never actually use
zswap, that's reasonable. I wanted to warn if we reach zswap_load()
with any large folio at all for higher coverage only.

I will send something out in the next week or so.


>
> > - Keep track if zswap was ever enabled and make the warning
> > conditional on it. We should also always fallback to order-0 if zswap
> > was ever enabled.
> > - Properly handle large folio swapin with zswap.
> >
> > Does this sound reasonable to you?
>
> Thanks
> Barry