Re: [syzbot] WARNING in call_timer_fn

From: Thomas Gleixner <tglx@linutronix.de>
To: Hillf Danton <hdanton@sina.com>
Cc: syzbot <syzbot+6fb78d577e89e69602f9@syzkaller.appspotmail.com>,
	linux-kernel@vger.kernel.org, pbonzini@redhat.com,
	syzkaller-bugs@googlegroups.com,
	Steven Rostedt <rosted@goodmis.org>,
	Marcel Holtmann <marcel@holtmann.org>,
	Luiz Augusto von Dentz <luiz.dentz@gmail.com>
Subject: Re: [syzbot] WARNING in call_timer_fn
Date: Thu, 17 Nov 2022 12:54:28 +0100	[thread overview]
Message-ID: <87fsehn6dn.ffs@tglx> (raw)
In-Reply-To: <20221117024511.3606-1-hdanton@sina.com>

On Thu, Nov 17 2022 at 10:45, Hillf Danton wrote:
> On 16 Nov 2022 19:29:27 +0100 Thomas Gleixner <tglx@linutronix.de>
>> 
>
> I doubt your fix helps in the domain of workqueue in general, and in this
> report if it is delayed work that triggered the warning.

Indeed. I misread the backtrace and my brain connected the wrong dots
here. It's a different bluethooth hickup.

> --- x/net/bluetooth/hci_sync.c
> +++ y/net/bluetooth/hci_sync.c
> @@ -4806,6 +4806,7 @@ int hci_dev_close_sync(struct hci_dev *h
>  	/* Avoid potential lockdep warnings from the *_flush() calls by
>  	 * ensuring the workqueue is empty up front.
>  	 */
> +	cancel_delayed_work_sync(&hdev->cmd_timer);
>  	drain_workqueue(hdev->workqueue);

The work has been canceled already before in the same function and there
are some more delayed works which can trigger this.

So no, this whole close_sync() function is prone to teardown races and
just slapping a single cancel here without deeper analysis does not cut
it.

Thanks,

        tglx