Message-ID: <CAD-N9QWsu8d_ubihGD0B1xf5YWv=WTw6iy4uNhV-73jA9xYbjw@mail.gmail.com>
Date:   Fri, 22 Jan 2021 13:47:11 +0800
From:   慕冬亮 <mudongliangabcd@...il.com>
To:     Lukas Bulwahn <lukas.bulwahn@...il.com>
Cc:     Greg KH <gregkh@...uxfoundation.org>, jirislaby@...nel.org,
        linux-kernel <linux-kernel@...r.kernel.org>,
        syzkaller-bugs <syzkaller-bugs@...glegroups.com>,
        syzkaller <syzkaller@...glegroups.com>
Subject: Re: "possible deadlock in console_lock_spinning_enable" and "possible
 deadlock in console_unlock" should be duplicate crash behaviors

On Thu, Jan 21, 2021 at 8:49 PM Lukas Bulwahn <lukas.bulwahn@...il.com> wrote:
>
> On Thu, Jan 21, 2021 at 6:37 AM 慕冬亮 <mudongliangabcd@...il.com> wrote:
> >
> > Dear kernel developers,
> >
> > I found that on the syzbot dashboard, “possible deadlock in
> > console_lock_spinning_enable”[1] and "possible deadlock in
> > console_unlock"[2] should share the same root cause.
> >
> > The reasons for the above statement:
> > 1) the stack trace is the same, and this title difference is due to
> > the inline property of "console_lock_spinning_enable";
> > 2) their PoCs are the same as each other;
> >
> > If you can have any issues with this statement or our information is
> > useful to you, please let us know. Thanks very much.
> >
> > [1] “possible deadlock in console_lock_spinning_enable” -
> > https://syzkaller.appspot.com/bug?id=2820deb61d92a8d7ab17a56ced58e963e65d76d0
> > [2] “possible deadlock in console_unlock” -
> > https://syzkaller.appspot.com/bug?id=39ea6caa479af471183997376dc7e90bc7d64a6a
> >
> >
>
> Dongliang, what is the purpose of this activity?

Lukas,

We are conducting some research on the crash deduplication (or
identifying unique bugs) of kernel crash reports. We would like to
share some results from our research to facilitate the bugfix in the
syzbot dashboard.

>
> Why do inform the kernel maintainers that two issues share the root cause?
>
> How does this activity contribute to fixing the bugs? Why does it
> become easier to fix the issue/create a patch with the information you
> provide?

I do this for three reasons:

(1) I think the reports sharing the same root cause may expedite the
patching processing and help generate more complete patches. After
patching bugs in one case, we can close the other cases quicker.
Without these reports, one developer might be misled to develop an
incomplete patch due to a lack of understanding of the underlying bug
[1].
(2) I think it might help maintainers to better assess the severity of
the bug and thus could prioritize their effort.
(3) Multiple reports might better help maintainers diagnose the bug's
root cause.

[1]  https://groups.google.com/g/syzkaller-bugs/c/9u_hEFvNbLw/m/CO9bfF8zCQAJ

> (Honestly, I do not see how it does. I believe if anyone becomes
> active and fixes the issue due to either one of the two reports, the
> one report would be closed by the reported-by tag and the other report
> would simply disappear after time because it could never be reproduced
> and hence, syzbot would close it.)
>
> Would it not be more reasonable to fix issues rather than identifying
> duplicates in the automatically filled and managed database?

Yes, fixing issues or bugs is the ultimate goal. However, crash
deduplication does benefit the bugfix process, and can reduce the
heavy burden on the kernel developers. To make our analysis more
useful, we will try our best to add some root cause analysis and how
to fix the underlying bug.

>
> Best regards,
>
> Lukas

Powered by blists - more mailing lists