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: <20200921211744.24758-1-peterx@redhat.com> <20200921212031.25233-1-peterx@redhat.com>
 <20200922120505.GH8409@ziepe.ca> <20200923152409.GC59978@xz-x1>
 <CAHbLzkr_C-4g+ArA3JXyvFHiAzU1ObHA9f4t1anj=N9b8oi7wQ@mail.gmail.com> <20200924154750.GG79898@xz-x1>
In-Reply-To: <20200924154750.GG79898@xz-x1>
From:   Yang Shi <shy828301@gmail.com>
Date:   Thu, 24 Sep 2020 10:29:27 -0700
Message-ID: <CAHbLzkp1wuKQ8PxeKVEs94+hjgtdX20Eo88i+9iQba6x0swnWA@mail.gmail.com>
Subject: Re: [PATCH 5/5] mm/thp: Split huge pmds/puds if they're pinned when fork()
To:     Peter Xu <peterx@redhat.com>
Cc:     Jason Gunthorpe <jgg@ziepe.ca>, Linux MM <linux-mm@kvack.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Linus Torvalds <torvalds@linux-foundation.org>,
        Michal Hocko <mhocko@suse.com>,
        Kirill Shutemov <kirill@shutemov.name>,
        Jann Horn <jannh@google.com>, Oleg Nesterov <oleg@redhat.com>,
        Kirill Tkhai <ktkhai@virtuozzo.com>,
        Hugh Dickins <hughd@google.com>,
        Leon Romanovsky <leonro@nvidia.com>, Jan Kara <jack@suse.cz>,
        John Hubbard <jhubbard@nvidia.com>,
        Christoph Hellwig <hch@lst.de>,
        Andrew Morton <akpm@linux-foundation.org>,
        Andrea Arcangeli <aarcange@redhat.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Thu, Sep 24, 2020 at 8:47 AM Peter Xu <peterx@redhat.com> wrote:
>
> On Wed, Sep 23, 2020 at 09:07:49AM -0700, Yang Shi wrote:
> > For tmpfs map, the pmd split just clears the pmd entry without
> > reinstalling ptes (oppositely anonymous map would reinstall ptes). It
> > looks this patch intends to copy at pte level by splitting pmd. But
> > I'm afraid this may not work for tmpfs mappings.
>
> IIUC that's exactly what we want.
>
> We only want to make sure the pinned tmpfs shared pages will be kept there in
> the parent.  It's not a must to copy the pages to the child, as long as they
> can be faulted in later correctly.

Aha, got your point. Yes, they can be refaulted in later. This is how
the file THP pmd split was designed.

>
> --
> Peter Xu
>