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: <1601283046-15329-1-git-send-email-iamjoonsoo.kim@lge.com>
 <20200928165215.f46924bfff9a109131048f81@linux-foundation.org>
 <CAAmzW4OY7z+bF=aFOsNuadc8M_f1Pb7jifuxzQo5AL6mCuO5Ng@mail.gmail.com> <20200928215055.02ff9b3ff5e0c392b2403411@linux-foundation.org>
In-Reply-To: <20200928215055.02ff9b3ff5e0c392b2403411@linux-foundation.org>
From:   Joonsoo Kim <js1304@gmail.com>
Date:   Tue, 29 Sep 2020 15:46:02 +0900
Message-ID: <CAAmzW4NPv+mu5P9=jHf0XgcrPjbCKJZR1hQkUkPvcTHb-0q50g@mail.gmail.com>
Subject: Re: [PATCH v2 for v5.9] mm/page_alloc: handle a missing case for
 memalloc_nocma_{save/restore} APIs
To:     Andrew Morton <akpm@linux-foundation.org>
Cc:     Linux Memory Management List <linux-mm@kvack.org>,
        LKML <linux-kernel@vger.kernel.org>,
        Michal Hocko <mhocko@kernel.org>,
        Vlastimil Babka <vbabka@suse.cz>,
        "Aneesh Kumar K . V" <aneesh.kumar@linux.ibm.com>,
        Mel Gorman <mgorman@techsingularity.net>, kernel-team@lge.com,
        Joonsoo Kim <iamjoonsoo.kim@lge.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

2020=EB=85=84 9=EC=9B=94 29=EC=9D=BC (=ED=99=94) =EC=98=A4=ED=9B=84 1:50, A=
ndrew Morton <akpm@linux-foundation.org>=EB=8B=98=EC=9D=B4 =EC=9E=91=EC=84=
=B1:
>
> On Tue, 29 Sep 2020 10:28:05 +0900 Joonsoo Kim <js1304@gmail.com> wrote:
>
> > > What about manually emptying the pcplists beforehand?
> >
> > It also increases the probability. schedule() or interrupt after emptyi=
ng but
> > before the allocation could invalidate the effect.
>
> Keep local interrupts disabled across the pcp drain and the allocation
> attempt.

As said before, it's an allocation context API and actual allocation
happens later.
Doing such things there is not an easy job.

> > > Or byassing the pcplists for this caller and calling __rmqueue() dire=
ctly?
> >
> > What this patch does is this one.
>
> I meant via a different function rather than by adding overhead to the
> existing commonly-used function.

Got it. One idea could be disabling/enabling pcp list on these APIs but
it's overhead would not be appropriate on these APIs. I don't have
another idea that doesn't touch the allocation path.

Thanks.