Date: Thu, 28 Apr 2016 17:08:31 +0200
From: Michal Hocko <mhocko@kernel.org>
To: Russell King - ARM Linux <linux@arm.linux.org.uk>
Cc: Andrew Morton <akpm@linux-foundation.org>, linux-mm@kvack.org,
        LKML <linux-kernel@vger.kernel.org>, linux-arch@vger.kernel.org
Subject: Re: [PATCH 04/20] arm: get rid of superfluous __GFP_REPEAT
Message-ID: <20160428150831.GK31489@dhcp22.suse.cz>
References: <1461849846-27209-1-git-send-email-mhocko@kernel.org>
 <1461849846-27209-5-git-send-email-mhocko@kernel.org>
 <20160428145545.GN19428@n2100.arm.linux.org.uk>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20160428145545.GN19428@n2100.arm.linux.org.uk>
User-Agent: Mutt/1.5.24 (2015-08-30)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1806
Lines: 43

On Thu 28-04-16 15:55:45, Russell King - ARM Linux wrote:
> On Thu, Apr 28, 2016 at 03:23:50PM +0200, Michal Hocko wrote:
> > From: Michal Hocko <mhocko@suse.com>
> > 
> > __GFP_REPEAT has a rather weak semantic but since it has been introduced
> > around 2.6.12 it has been ignored for low order allocations.
> > 
> > PGALLOC_GFP uses __GFP_REPEAT but none of the allocation which uses
> > this flag is for more than order-2. This means that this flag has never
> > been actually useful here because it has always been used only for
> > PAGE_ALLOC_COSTLY requests.
> 
> I'm unconvinced.  Back in 2013, I was seeing a lot of failures, so:
> 
> commit 8c65da6dc89ccb605d73773b1dd617e72982d971
> Author: Russell King <rmk+kernel@arm.linux.org.uk>
> Date:   Sat Nov 30 12:52:31 2013 +0000
> 
>     ARM: pgd allocation: retry on failure
> 
>     Make pgd allocation retry on failure; we really need this to succeed
>     otherwise fork() can trigger OOMs.
> 
>     Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
> 
> Maybe something has changed again in the MM layer which makes this flag
> unnecessary again, and it was a temporary blip around that time, I don't
> know.

PAGE_ALLOC_COSTLY_ORDER is defined to order 3 since 2007 and even before
the code was doing
-               if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
+               if ((order <= PAGE_ALLOC_COSTLY_ORDER) ||
+                                               (gfp_mask & __GFP_REPEAT))
                        do_retry = 1;

So an order-2 allocation which is the case for this particular code now
will trigger the OOM killer and fail only when the current task is
killed by the OOM killer. Other than that order-2 is basically
GFP_NOFAIL. Have a look at __alloc_pages_slowpath() for more details.
-- 
Michal Hocko
SUSE Labs