[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAM_iQpX7-BF=C+CAV3o=VeCZX7=CgdscZaazTD6QT-Tw1=XY9Q@mail.gmail.com>
Date: Mon, 6 Jan 2020 21:36:20 -0800
From: Cong Wang <xiyou.wangcong@...il.com>
To: syzbot <syzbot+4ec99438ed7450da6272@...kaller.appspotmail.com>
Cc: Taehee Yoo <ap420073@...il.com>,
Linux Kernel Network Developers <netdev@...r.kernel.org>
Subject: Re: WARNING: bad unlock balance in sch_direct_xmit
Hi, Taehee
On Sun, Jan 5, 2020 at 2:59 PM syzbot
<syzbot+4ec99438ed7450da6272@...kaller.appspotmail.com> wrote:
>
> syzbot has bisected this bug to:
>
> commit ab92d68fc22f9afab480153bd82a20f6e2533769
> Author: Taehee Yoo <ap420073@...il.com>
> Date: Mon Oct 21 18:47:51 2019 +0000
>
> net: core: add generic lockdep keys
Why netdev_update_lockdep_key() is needed?
It causes the bug here because the unregister and register are not
atomic although under RTNL, fast path could still lock with one key
and unlock with another key after the previous got unregistered.
>From my understand of lockdep here, as long as the device itself
is not changed, it doesn't need to update those keys.
Thanks.
Powered by blists - more mailing lists