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: <309c86b7-2a4c-1332-585f-7bcd59cfd762@I-love.SAKURA.ne.jp>
Date:   Sun, 13 Feb 2022 02:14:09 +0900
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (Windows NT 6.3; Win64; x64; rv:91.0) Gecko/20100101
 Thunderbird/91.6.0
Subject: Re: [syzbot] possible deadlock in worker_thread
Content-Language: en-US
To:     Bart Van Assche <bvanassche@acm.org>,
        syzbot <syzbot+831661966588c802aae9@syzkaller.appspotmail.com>,
        jgg@ziepe.ca, linux-kernel@vger.kernel.org,
        linux-rdma@vger.kernel.org, syzkaller-bugs@googlegroups.com,
        Tejun Heo <tj@kernel.org>,
        Lai Jiangshan <jiangshanlai@gmail.com>
References: <0000000000005975a605d7aef05e@google.com>
 <8ea57ddf-a09c-43f2-4285-4dfb908ad967@acm.org>
 <ccd04d8a-154b-543e-e1c3-84bc655508d1@I-love.SAKURA.ne.jp>
 <71d6f14e-46af-cc5a-bc70-af1cdc6de8d5@acm.org>
From:   Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
In-Reply-To: <71d6f14e-46af-cc5a-bc70-af1cdc6de8d5@acm.org>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 2022/02/13 1:37, Bart Van Assche wrote:
> On 2/11/22 21:31, Tetsuo Handa wrote:
>> But this report might be suggesting us that we should consider
>> deprecating (and eventually getting rid of) system-wide workqueues
>> (declared in include/linux/workqueue.h), for since flush_workqueue()
>> synchronously waits for completion, sharing system-wide workqueues
>> among multiple modules can generate unexpected locking dependency
>> chain (like this report).
> 
> I do not agree with deprecating system-wide workqueues. I think that
> all flush_workqueue(system_long_wq) calls should be reviewed since
> these are deadlock-prone.
> 
> Thanks,
> 
> Bart.

The loop module is not using flush_workqueue(system_long_wq) call; it only
scheduled a work via system_long_wq which will call flush_workqueue() (of
a local workqueue) from drain_workqueue() from destroy_workqueue() from
work function.

How can reviewing all flush_workqueue(system_long_wq) calls help?