Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Wed, 6 Nov 2019 10:59:21 -0800 (PST)
From:   Hugh Dickins <hughd@google.com>
To:     Yang Shi <yang.shi@linux.alibaba.com>
cc:     Michal Hocko <mhocko@kernel.org>, hughd@google.com,
        akpm@linux-foundation.org, linux-mm@kvack.org,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH] mm: shmem: use proper gfp flags for shmem_writepage()
In-Reply-To: <733100ea-97aa-db27-4b43-cf42317afaf8@linux.alibaba.com>
Message-ID: <alpine.LSU.2.11.1911061039540.1357@eggly.anvils>
References: <1572991351-86061-1-git-send-email-yang.shi@linux.alibaba.com> <20191106151820.GB8138@dhcp22.suse.cz> <733100ea-97aa-db27-4b43-cf42317afaf8@linux.alibaba.com>
User-Agent: Alpine 2.11 (LSU 23 2013-08-11)
MIME-Version: 1.0
Content-Type: MULTIPART/MIXED; BOUNDARY="0-1214476646-1573066773=:1357"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

  This message is in MIME format.  The first part should be readable text,
  while the remaining parts are likely unreadable without MIME-aware tools.

--0-1214476646-1573066773=:1357
Content-Type: TEXT/PLAIN; charset=UTF-8
Content-Transfer-Encoding: QUOTED-PRINTABLE

On Wed, 6 Nov 2019, Yang Shi wrote:
> On 11/6/19 7:18 AM, Michal Hocko wrote:
> > On Wed 06-11-19 06:02:31, Yang Shi wrote:
> > > The shmem_writepage() uses GFP_ATOMIC to allocate swap cache.
> > > GFP_ATOMIC used to mean __GFP_HIGH, but now it means __GFP_HIGH |
> > > __GFP_ATOMIC | __GFP_KSWAPD_RECLAIM.  However, shmem_writepage() shou=
ld
> > > write out to swap only in response to memory pressure, so
> > > __GFP_KSWAPD_RECLAIM looks useless since the caller may be kswapd its=
elf
> > > or in direct reclaim already.
> > What kind of problem are you trying to fix here?
>=20
> I didn't run into any visible problem. I just happened to find this
> inconsistency when I was looking into the other problem.

Yes, I don't think it fixes any actual problem: just a cleanup to
make the two calls look the same when they don't need to be different
(whereas the call from __read_swap_cache_async() rightly uses a=20
lower priority gfp).

If it does fix a problem, then you need to worry also about the
=09 * TODO: this could cause a theoretical memory reclaim
=09 * deadlock in the swap out path.
comment still against the call in add_to_swap(): but I think that
is equally theoretical, demanding no attention since 2.6.12.

>=20
> The add_to_swap() does:
>=20
> int add_to_swap(struct page *page)
> {
> ...
> err =3D add_to_swap_cache(page, entry,
> =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=
=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 __GFP_HI=
GH|__GFP_NOMEMALLOC|__GFP_NOWARN);
> ...
> }
>=20
> Actually, shmem_writepage() does almost the same thing and both of them a=
re
> called in reclaim context, so I didn't see why they should use different =
gfp
> flag. And, GFP_ATOMIC is also different from the old definition as I
> mentioned in the commit log.
>=20
> >=20
> > > In addition, XArray node allocations from PF_MEMALLOC contexts could
> > > completely exhaust the page allocator, __GFP_NOMEMALLOC stops emergen=
cy
> > > reserves from being allocated.
> > I am not really familiar with XArray much, could you be more specific
> > please?
>=20
> It comes from the comments of add_to_swap(), says:
>=20
> /*
> =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 * XArray node allocation=
s from PF_MEMALLOC contexts could
> =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 * completely exhaust the=
 page allocator. __GFP_NOMEMALLOC
> =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 * stops emergency reserv=
es from being allocated.
>=20
> And, it looks the original comment came from pre-git time, TBH I'm not qu=
ite
> sure about the specific problem which incurred this. I suspect it may be
> because PF_MEMALLOC context allows ALLOC_NO_WATERMARK.
>=20
> >=20
> > > Here just copy the gfp flags used by add_to_swap().
> > >=20
> > > Cc: Hugh Dickins <hughd@google.com>
> > > Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com>
> > > ---
> > >   mm/shmem.c | 3 ++-
> > >   1 file changed, 2 insertions(+), 1 deletion(-)
> > >=20
> > > diff --git a/mm/shmem.c b/mm/shmem.c
> > > index 220be9f..9691dec 100644
> > > --- a/mm/shmem.c
> > > +++ b/mm/shmem.c
> > > @@ -1369,7 +1369,8 @@ static int shmem_writepage(struct page *page,
> > > struct writeback_control *wbc)
> > >   =09if (list_empty(&info->swaplist))
> > >   =09=09list_add(&info->swaplist, &shmem_swaplist);
> > >   -=09if (add_to_swap_cache(page, swap, GFP_ATOMIC) =3D=3D 0) {
> > > +=09if (add_to_swap_cache(page, swap,
> > > +=09=09=09__GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN) =3D=3D 0) {
> > >   =09=09spin_lock_irq(&info->lock);
> > >   =09=09shmem_recalc_inode(inode);
> > >   =09=09info->swapped++;
> > > --=20
> > > 1.8.3.1
--0-1214476646-1573066773=:1357--