Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <1594107889-32228-1-git-send-email-iamjoonsoo.kim@lge.com>
 <1594107889-32228-7-git-send-email-iamjoonsoo.kim@lge.com>
 <409b6e24-d143-a61c-95a3-1a55e1a6008e@suse.cz> <20200707190013.GZ5913@dhcp22.suse.cz>
In-Reply-To: <20200707190013.GZ5913@dhcp22.suse.cz>
From:   Joonsoo Kim <js1304@gmail.com>
Date:   Thu, 9 Jul 2020 16:15:07 +0900
Message-ID: <CAAmzW4PVp_PX-PySjy5axMzyy5n_4cHmzrhfT=tFH2VEzxVFRQ@mail.gmail.com>
Subject: Re: [PATCH v4 06/11] mm/migrate: make a standard migration target
 allocation function
To:     Michal Hocko <mhocko@kernel.org>
Cc:     Vlastimil Babka <vbabka@suse.cz>,
        Andrew Morton <akpm@linux-foundation.org>,
        Linux Memory Management List <linux-mm@kvack.org>,
        LKML <linux-kernel@vger.kernel.org>, kernel-team@lge.com,
        Christoph Hellwig <hch@infradead.org>,
        Roman Gushchin <guro@fb.com>,
        Mike Kravetz <mike.kravetz@oracle.com>,
        Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>,
        Joonsoo Kim <iamjoonsoo.kim@lge.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

2020=EB=85=84 7=EC=9B=94 8=EC=9D=BC (=EC=88=98) =EC=98=A4=EC=A0=84 4:00, Mi=
chal Hocko <mhocko@kernel.org>=EB=8B=98=EC=9D=B4 =EC=9E=91=EC=84=B1:
>
> On Tue 07-07-20 16:49:51, Vlastimil Babka wrote:
> > On 7/7/20 9:44 AM, js1304@gmail.com wrote:
> > > From: Joonsoo Kim <iamjoonsoo.kim@lge.com>
> > >
> > > There are some similar functions for migration target allocation.  Si=
nce
> > > there is no fundamental difference, it's better to keep just one rath=
er
> > > than keeping all variants.  This patch implements base migration targ=
et
> > > allocation function.  In the following patches, variants will be conv=
erted
> > > to use this function.
> > >
> > > Changes should be mechanical but there are some differences. First, S=
ome
> > > callers' nodemask is assgined to NULL since NULL nodemask will be
> > > considered as all available nodes, that is, &node_states[N_MEMORY].
> > > Second, for hugetlb page allocation, gfp_mask is ORed since a user co=
uld
> > > provide a gfp_mask from now on.
> >
> > I think that's wrong. See how htlb_alloc_mask() determines between
> > GFP_HIGHUSER_MOVABLE and GFP_HIGHUSER, but then you OR it with __GFP_MO=
VABLE so
> > it's always GFP_HIGHUSER_MOVABLE.

Indeed.

> Right you are! Not that it would make any real difference because only
> migrateable hugetlb pages will get __GFP_MOVABLE and so we shouldn't
> really end up here for !movable pages in the first place (not sure about
> soft offlining at this moment). But yeah it would be simply better to
> override gfp mask for hugetlb which we have been doing anyway.

Override gfp mask doesn't work since some users will call this function wit=
h
__GFP_THISNODE.  I will use hugepage_movable_supported() here and
clear __GFP_MOVABLE if needed.

Thanks.

Thanks.