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;
Sender: Tejun Heo <htejun@gmail.com>
Date:   Tue, 22 Feb 2022 08:26:55 -1000
From:   Tejun Heo <tj@kernel.org>
To:     Bart Van Assche <bvanassche@acm.org>
Cc:     Tetsuo Handa <penguin-kernel@i-love.sakura.ne.jp>,
        Haakon Bugge <haakon.bugge@oracle.com>,
        syzbot <syzbot+831661966588c802aae9@syzkaller.appspotmail.com>,
        Jason Gunthorpe <jgg@ziepe.ca>,
        LKML <linux-kernel@vger.kernel.org>,
        OFED mailing list <linux-rdma@vger.kernel.org>,
        "syzkaller-bugs@googlegroups.com" <syzkaller-bugs@googlegroups.com>,
        Lai Jiangshan <jiangshanlai@gmail.com>
Subject: Re: [syzbot] possible deadlock in worker_thread
Message-ID: <YhUq70Y/tKGGNbR0@slm.duckdns.org>
References: <309c86b7-2a4c-1332-585f-7bcd59cfd762@I-love.SAKURA.ne.jp>
 <aa2bf24e-981a-a811-c5d8-a75f0b8f693a@acm.org>
 <2959649d-cfbc-bdf2-02ac-053b8e7af030@I-love.SAKURA.ne.jp>
 <YgnQGZWT/n3VAITX@slm.duckdns.org>
 <8ebd003c-f748-69b4-3a4f-fb80a3f39d36@I-love.SAKURA.ne.jp>
 <YgqSsuSN5C7StvKx@slm.duckdns.org>
 <a07e464c-69c6-6165-e88c-5a2eded79138@I-love.SAKURA.ne.jp>
 <76616D2F-14F2-4D83-9DB4-576FB2ACB72C@oracle.com>
 <fb854eea-a7e7-b526-a989-95784c1c593c@I-love.SAKURA.ne.jp>
 <1b70929f-1f73-c549-64c1-94cea2c1a36c@acm.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <1b70929f-1f73-c549-64c1-94cea2c1a36c@acm.org>
Precedence: bulk

On Tue, Feb 15, 2022 at 09:05:38AM -0800, Bart Van Assche wrote:
> On 2/15/22 04:48, Tetsuo Handa wrote:
> > I do not want to do like
> > 
> > -	system_wq = alloc_workqueue("events", 0, 0);
> > +	system_wq = alloc_workqueue("events", __WQ_SYSTEM_WIDE, 0);
> > 
> > because the intent of this change is to ask developers to create their own WQs.
> 
> I want more developers to use the system-wide workqueues since that reduces
> memory usage. That matters for embedded devices running Linux.

Each wq is just a frontend interface to backend shard pool and doesn't
consume a lot of memory. The only consumption which would matter is when
WQ_MEM_RECLAIM is specified which creates its dedicated rescuer thread to
guarantee forward progress under memory contention, but we aren't talking
about those here.

Thanks.

-- 
tejun