[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <CAFgQCTvu7vcp0DqG43XxFQmoOOqXWbCfRDNcWUDm7vro5GmdtA@mail.gmail.com>
Date: Tue, 4 Jun 2019 16:50:39 +0800
From: Pingfan Liu <kernelfans@...il.com>
To: "Aneesh Kumar K.V" <aneesh.kumar@...ux.ibm.com>
Cc: linux-mm@...ck.org, Andrew Morton <akpm@...ux-foundation.org>,
Ira Weiny <ira.weiny@...el.com>,
Dan Williams <dan.j.williams@...el.com>,
Thomas Gleixner <tglx@...utronix.de>,
Mike Rapoport <rppt@...ux.ibm.com>,
John Hubbard <jhubbard@...dia.com>,
Keith Busch <keith.busch@...el.com>,
LKML <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] mm/gup: remove unnecessary check against CMA in __gup_longterm_locked()
On Tue, Jun 4, 2019 at 4:30 PM Aneesh Kumar K.V
<aneesh.kumar@...ux.ibm.com> wrote:
>
> On 6/4/19 12:56 PM, Pingfan Liu wrote:
> > The PF_MEMALLOC_NOCMA is set by memalloc_nocma_save(), which is finally
> > cast to ~_GFP_MOVABLE. So __get_user_pages_locked() will get pages from
> > non CMA area and pin them. There is no need to
> > check_and_migrate_cma_pages().
>
>
> That is not completely correct. We can fault in that pages outside
> get_user_pages_longterm at which point those pages can get allocated
> from CMA region. memalloc_nocma_save() as added as an optimization to
> avoid unnecessary page migration.
Yes, you are right.
Thanks,
Pingfan
Powered by blists - more mailing lists