Date:   Wed, 12 Jan 2022 04:32:34 +0000
From:   Matthew Wilcox <willy@infradead.org>
To:     Hugh Dickins <hughd@google.com>
Cc:     Peng Liang <liangpeng10@huawei.com>,
        David Hildenbrand <david@redhat.com>, linux-mm@kvack.org,
        linux-kernel@vger.kernel.org, akpm@linux-foundation.org,
        xiexiangyou@huawei.com, zhengchuan@huawei.com,
        wanghao232@huawei.com
Subject: Re: [RFC 0/1] memfd: Support mapping to zero page on reading
Message-ID: <Yd5Z4i+67+ae68HM@casper.infradead.org>
References: <20211222123400.1659635-1-liangpeng10@huawei.com>
 <4b1885b8-eb95-c50-2965-11e7c8efbf36@google.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <4b1885b8-eb95-c50-2965-11e7c8efbf36@google.com>
Precedence: bulk

On Tue, Jan 11, 2022 at 06:30:31PM -0800, Hugh Dickins wrote:
> But I have to say that use of ZERO_PAGE for shmem/memfd/tmpfs read-fault
> might (potentially) be very welcome.  Not as some MFD_ZEROPAGE special
> case, but as how it would always work.  Deleting the shmem_recalc_inode()
> cruft, which is there to correct accounting for the unmodified read-only
> pages, after page reclaim has got around to freeing them later.
> 
> It does require more work than you gave it in 1/1: mainly, as you call
> out above, there's a need to note in the mapping's XArray when ZERO_PAGE
> has been used at an offset, and do an rmap walk to unmap those ptes when
> a writable page is substituted - see __xip_unmap() in Linux 3.19's
> mm/filemap_xip.c for such an rmap walk.

I think putting a pointer to the zero page in the XArray would introduce
some unwelcome complexity, but the XArray has a special XA_ZERO_ENTRY
which might be usable for such a thing.  It would need some careful
analysis and testing, of course, but it might also let us remove
the special cases in the DAX code for DAX_ZERO_PAGE.

I agree with you that temporarily allocating pages has worked "well
enough", but maybe some workloads would benefit; even for files on block
device filesystems, reading a hole and never writing to it may be common
enough that this is an optimisation we've been missing for many years.