Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <eabca57aa14f4df723173b24891f4a2d9c501f21.1543526537.git.jstancek@redhat.com>
 <c440d69879e34209feba21e12d236d06bc0a25db.1543577156.git.jstancek@redhat.com>
 <CADnJP=vsum7_YYWBpknpahTQFAzm7G40_E2dLMB_poFEhPKEfw@mail.gmail.com> <997509746.100933786.1549350874925.JavaMail.zimbra@redhat.com>
In-Reply-To: <997509746.100933786.1549350874925.JavaMail.zimbra@redhat.com>
From:   Lars Persson <lists@bofh.nu>
Date:   Mon, 18 Feb 2019 14:43:58 +0100
Message-ID: <CADnJP=t25=AcVq7z3w8iG1+ywnSNN4Vbow3-7tOai+qnyD5ACQ@mail.gmail.com>
Subject: Re: [PATCH v2] mm: page_mapped: don't assume compound page is huge or THP
To:     Jan Stancek <jstancek@redhat.com>
Cc:     linux-mm@kvack.org, lersek@redhat.com,
        alex williamson <alex.williamson@redhat.com>,
        aarcange@redhat.com, rientjes@google.com, kirill@shutemov.name,
        mgorman@techsingularity.net, mhocko@suse.com,
        linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Tue, Feb 5, 2019 at 8:14 AM Jan Stancek <jstancek@redhat.com> wrote:
> Hi,
>
> are you using THP (CONFIG_TRANSPARENT_HUGEPAGE)?
>
> The changed line should affect only THP and normal compound pages,
> so a test with THP disabled might be interesting.
>
> >
> > The breakage consists of random processes dying with SIGILL or SIGSEGV
> > when we stress test the system with high memory pressure and explicit
> > memory compaction requested through /proc/sys/vm/compact_memory.
> > Reverting this patch fixes the crashes.
> >
> > We can put some effort on debugging if there are no obvious
> > explanations for this. Keep in mind that this is 32-bit system with
> > HIGHMEM.
>
> Nothing obvious that I can see. I've been trying to reproduce on
> 32-bit x86 Fedora with no luck so far.
>

Hi

Thanks for looking in to it. After some deep dive in MM code, I think
it is safe to say this patch was innocent.

All traces studied so far points to a missing cache coherency call in
mm/migrate.c:migrate_page that is needed only for those evil MIPSes
that lack I/D cache coherency. I will send a write-up to linux-mips
about this. Basically for a non-mapped page it does only a copy of
page data and metadata but no flush_dcache_page() call will be done.
This races with subsequent use of the page.

BR,
 Lars