Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <97667b25-11e6-d918-2468-2cf17532c3ad@gmail.com>
Date:   Sat, 5 Feb 2022 12:30:03 +0800
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:91.0) Gecko/20100101
 Thunderbird/91.3.0
Subject: Re: [BUG] tty: serial: possible deadlock in uart_remove_one_port()
 and uart_hangup()
To:     Hillf Danton <hdanton@sina.com>
Cc:     Greg KH <gregkh@linuxfoundation.org>, jirislaby@kernel.org,
        linux-serial@vger.kernel.org,
        linux-kernel <linux-kernel@vger.kernel.org>
References: <ab5d9322-6bea-9845-c61b-fb68e3bb3a87@gmail.com>
 <YfUPlYwoWpMjhvpR@kroah.com> <20220201090740.3816-1-hdanton@sina.com>
From:   Jia-Ju Bai <baijiaju1990@gmail.com>
In-Reply-To: <20220201090740.3816-1-hdanton@sina.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk


On 2022/2/1 17:07, Hillf Danton wrote:
> On Tue, 1 Feb 2022 14:51:09 +0800 Jia-Ju Bai wrote:
>> On 2022/1/29 17:57, Greg KH wrote:
>>> On Sat, Jan 29, 2022 at 05:34:05PM +0800, Jia-Ju Bai wrote:
>>>> Hello,
>>>>
>>>> My static analysis tool reports a possible deadlock in the tty driver in
>>>> Linux 5.10:
>>> 5.10 was released over a year ago and over 100 thousand changes ago.
>>> Please redo your check on 5.16 at the oldest.
>> My static analysis tool checks the tty driver in Linux 5.16, and also
>> finds this possible deadlock:
>>
>> uart_remove_one_port()
>>     mutex_lock(&port->mutex); --> Line 3032 (Lock A)
>>     wait_event(state->remove_wait, ...); --> Line 3034 (Wait X)
>>     mutex_unlock(&port->mutex); --> Line 3036 (Unlock A)
>>
>> uart_hangup()
>>     mutex_lock(&port->mutex); --> Line 1669 (Lock A)
>>     uart_flush_buffer()
>>       uart_port_unlock()
>>         uart_port_deref()
>>           wake_up(&uport->state->remove_wait); --> Line 68 (Wake X)
>>     mutex_unlock(&port->mutex); --> Line 1686 (Unlock A)
>>
>> When uart_remove_one_port() is executed, "Wait X" is performed by
>> holding "Lock A". If uart_hangup() is executed at this time, "Wake X"
>> cannot be performed to wake up "Wait X" in uart_remove_one_port(),
>> because "Lock A" has been already hold by uart_remove_one_port(),
>> causing a possible deadlock.
>>
>> I am not quite sure whether this possible problem is real and how to fix
>> it if it is real.
>> Maybe we can call wait_event() before mutex_lock() in
>> uart_remove_one_port().
>> Any feedback would be appreciated, thanks :)
>>
>>
>> Best wishes,
>> Jia-Ju Bai
> Hey Jia-Ju
>
> Thank you for reporting it.
>
> In uart_flush_buffer(), uart_port_unlock() pairs with uart_port_lock()
> which bumps refcount up. OTOH no wakep is needed without refcount
> incremented, so the wakeup above in the hangup path is not waited for
> in the remove path.

Hi Hillf,

Thanks for the explanation :)
So I wonder which wait_event() can be paired with 
wake_up(&uport->state->remove_wait) in uart_port_deref()?


Best wishes,
Jia-Ju Bai