Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: 7bit
MIME-Version: 1.0
From:   "NeilBrown" <neilb@suse.de>
To:     "Mel Gorman" <mgorman@techsingularity.net>
Cc:     "Michal Hocko" <mhocko@suse.com>,
        "Andrew Morton" <akpm@linux-foundation.org>,
        "Thierry Reding" <thierry.reding@gmail.com>,
        "Matthew Wilcox" <willy@infradead.org>, linux-mm@kvack.org,
        linux-kernel@vger.kernel.org, "Mel Gorman" <mgorman@suse.de>
Subject: Re: [PATCH] MM: discard __GFP_ATOMIC
In-reply-to: <20211118132753.GB3366@techsingularity.net>
References: <163712397076.13692.4727608274002939094@noble.neil.brown.name>,
 <YZYbXC/ycmif8WIE@dhcp22.suse.cz>,
 <20211118132753.GB3366@techsingularity.net>
Date:   Fri, 19 Nov 2021 10:02:55 +1100
Message-id: <163727657594.13692.10357464624495712361@noble.neil.brown.name>
Precedence: bulk

On Fri, 19 Nov 2021, Mel Gorman wrote:
> On Thu, Nov 18, 2021 at 10:22:36AM +0100, Michal Hocko wrote:
> > [Cc Mel]
> > 
> 
> I think this patch should be ok. There are few direct users of __GFP_HIGH
> and some of them are borderline silly (e.g. mm/shmem.c specifying
> __GFP_HIGH|__GFP_NOMEMALLOC) while others just look questionable (
> drivers/md/raid10.c seems to assume __GFP_HIGH guarantees allocation
> success). Xen appears to be the worst abuser of __GFP_HIGH.

That __GFP_HIGH in raid10.c is passed to mempool_alloc(), so there is no
assumption that __GFP_HIGH will provide guarantees - the mempool does
that.
The comment - which I wrote 4 years ago and don't recall at all -
suggest it was purely about performance - get error handling out of the
way quickly.  I doubt I could justify it if challenged...

Thanks,
NeilBrown


> 
> -- 
> Mel Gorman
> SUSE Labs
> 
>