Re: [PATCH] mm,thp,shmem: limit shmem THP alloc gfp_mask

From: Rik van Riel <riel@surriel.com>
To: Michal Hocko <mhocko@suse.com>
Cc: Hugh Dickins <hughd@google.com>,
	Andrew Morton <akpm@linux-foundation.org>,
	 linux-mm@kvack.org, linux-kernel@vger.kernel.org,
	kernel-team@fb.com, Mel Gorman <mgorman@suse.de>,
	Andrea Arcangeli <aarcange@redhat.com>,
	Matthew Wilcox <willy@infradead.org>
Subject: Re: [PATCH] mm,thp,shmem: limit shmem THP alloc gfp_mask
Date: Thu, 22 Oct 2020 12:06:01 -0400	[thread overview]
Message-ID: <1b2d262c30bd839ac433587532a27ad800df4771.camel@surriel.com> (raw)
In-Reply-To: <20201022155022.GO23790@dhcp22.suse.cz>

[-- Attachment #1: Type: text/plain, Size: 2225 bytes --]

On Thu, 2020-10-22 at 17:50 +0200, Michal Hocko wrote:
> On Thu 22-10-20 09:25:21, Rik van Riel wrote:
> > On Thu, 2020-10-22 at 10:15 +0200, Michal Hocko wrote:
> > > On Wed 21-10-20 23:48:46, Rik van Riel wrote:
> > > > 
> > > > diff --git a/mm/shmem.c b/mm/shmem.c
> > > > index 537c137698f8..d1290eb508e5 100644
> > > > --- a/mm/shmem.c
> > > > +++ b/mm/shmem.c
> > > > @@ -1545,8 +1545,11 @@ static struct page
> > > > *shmem_alloc_hugepage(gfp_t gfp,
> > > >  		return NULL;
> > > >  
> > > >  	shmem_pseudo_vma_init(&pvma, info, hindex);
> > > > -	page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY
> > > > |
> > > > __GFP_NOWARN,
> > > > -			HPAGE_PMD_ORDER, &pvma, 0,
> > > > numa_node_id(),
> > > > true);
> > > > +	/* Limit the gfp mask according to THP configuration.
> > > > */
> > > > +	gfp |= __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN;
> > > 
> > > What is the reason for these when alloc_hugepage_direct_gfpmask
> > > provides
> > > the full mask?
> > 
> > The mapping_gfp_mask for the shmem file might have additional
> > restrictions above and beyond the gfp mask returned by
> > alloc_hugepage_direct_gfpmask, and I am not sure we should just
> > ignore the mapping_gfp_mask.
> 
> No, we shouldn't. But I do not see why you should be adding the above
> set of flags on top.

Because THP allocations are higher order and optimistic,
and we want them to:
1) be annotated as compound allocations
2) fail (and fall back to 4kB allocations) when they cannot
   be easily satisfied, and
3) not create a spew of allocation failure backtraces on
   the (serial) console when these THP allocations fail

> > That is why this patch takes the union of both gfp masks.
> > 
> > However, digging into things more, it looks like shmem inodes
> > always have the mapping gfp mask set to GFP_HIGHUSER_MOVABLE,
> > and it is never changed, so simply using the output from
> > alloc_hugepage_direct_gfpmask should be fine.
> > 
> > I can do the patch either way. Just let me know what you prefer.
> 
> I would just and the given gfp with alloc_hugepage_direct_gfpmask

That would miss the three things we definitely want
from above.

-- 
All Rights Reversed.

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 488 bytes --]