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: <20200214225849.108108-1-bgeffon@google.com> <20200214231954.GA29849@redhat.com>
In-Reply-To: <20200214231954.GA29849@redhat.com>
From:   Brian Geffon <bgeffon@google.com>
Date:   Sat, 15 Feb 2020 09:29:46 -0500
Message-ID: <CADyq12w3tBO5NfZ33R__B3jvF=ed7ys+o4horGwyUO3bNevObg@mail.gmail.com>
Subject: Re: [RFC PATCH] userfaultfd: Address race after fault.
To:     Andrea Arcangeli <aarcange@redhat.com>
Cc:     Andrew Morton <akpm@linux-foundation.org>,
        linux-mm <linux-mm@kvack.org>,
        LKML <linux-kernel@vger.kernel.org>,
        Mike Rapoport <rppt@linux.vnet.ibm.com>,
        Sonny Rao <sonnyrao@google.com>,
        "Kirill A . Shutemov" <kirill@shutemov.name>,
        Peter Xu <peterx@redhat.com>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Hi Andrea,
Thanks for the quick reply. That's great to hear that Peter has been
working on those improvements. I didn't try the entire patchset but I
did confirm that patch 13, not surprisingly, also resolves that issue
on at least on x86:
  https://lkml.org/lkml/2019/9/26/179

Given that seems pretty low risk and it definitely resolves a pretty
big issue for the non-cooperative userfaultfd case, any chance it
could be landed ahead of the rest of the series?

Thanks,
Brian

On Fri, Feb 14, 2020 at 6:20 PM Andrea Arcangeli <aarcange@redhat.com> wrote:
>
> Hello,
>
> this and other enhancements have already implemented by Peter (CC'ed)
> and in the right way, by altering the retry logic in the page fault
> code. This is a requirement for other kind of usages too, notably the
> UFFD_WRITEPROTECT ioctl after which multiple consecutive faults can
> happen and must be handled.
>
> IIRC Kirill asked at last LSF-MM uffd-wp talk if there's any
> particular reason the fault couldn't be retried currently. I had no
> sure answer other than there's apparently no strong reason why
> VM_FAULT_RETRY is only allowed 1 time currently, so there should be no
> issue in lifting that artificial restriction.
>
> I'm running with this patchset applied in my systems since Nov with no
> regression at all. I got sidetracked by various other issues, so
> unfortunately I didn' post a proper reviewed-by on the last submit yet
> (pending), but I did at least test it and it was rock solid so far.
>
> https://lore.kernel.org/lkml/20190926093904.5090-1-peterx@redhat.com/
>
> Can you test and verify it too if it solves your use case?
>
> Also note the complete uffd-WP support submit also from Peter:
>
> https://lore.kernel.org/lkml/20190620022008.19172-1-peterx@redhat.com/
>
> https://github.com/xzpeter/linux/tree/uffd-wp-merged
>
> Thanks,
> Andrea
>