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: <afa1ac2c-a023-a91e-e596-60931b38247e@I-love.SAKURA.ne.jp>
Date:   Fri, 29 Jul 2022 11:28:40 +0900
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (Windows NT 6.3; Win64; x64; rv:91.0) Gecko/20100101
 Thunderbird/91.11.0
Subject: Re: [PATCH] workqueue: don't skip lockdep wq dependency in
 cancel_work_sync()
Content-Language: en-US
From:   Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
To:     Johannes Berg <johannes.berg@intel.com>
Cc:     Lai Jiangshan <jiangshanlai@gmail.com>,
        Hillf Danton <hdanton@sina.com>,
        LKML <linux-kernel@vger.kernel.org>, Tejun Heo <tj@kernel.org>
References: <21b9c1ac-64b7-7f4b-1e62-bf2f021fffcd@I-love.SAKURA.ne.jp>
 <YuK78Jiy12BJG/Tp@slm.duckdns.org>
 <0ad532b2-df5f-331a-ae7f-21460fc62fe2@I-love.SAKURA.ne.jp>
 <97cbf8a9-d5e1-376f-6a49-3474871ea6b4@I-love.SAKURA.ne.jp>
In-Reply-To: <97cbf8a9-d5e1-376f-6a49-3474871ea6b4@I-love.SAKURA.ne.jp>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit
Precedence: bulk

Thinking more, I might be confused by difference between

  the lockdep "struct work_struct" dependency handling

and

  the lockdep "struct work" dependency handling

. In an example that

      work3_function() { /* nothing */ }
      work4_function() { mutex_lock(&mutex); ... }

      other_function() {
        queue_work(ordered_wq, &work3);
        queue_work(ordered_wq, &work4);
        mutex_lock(&mutex);
        cancel_work_sync(&work4);
      }

, possibility of deadlock must be reported by lockdep via
the lockdep "struct work" dependency handling.

Then, prior to commit d6e89786bed977f3, lockdep flagging

  the lockdep "struct work_struct" dependency handling

as if cancel_work_sync() was flush_work() is a problem, and
lockdep not flagging

  the lockdep "struct work" dependency handling

 from cancel_work_sync() as if flush_work() is also a problem.

Then, commit d6e89786bed977f3 might be OK, but commit 87915adc3f0acdf0
was wrong in that it preserved lockdep not flagging

  the lockdep "struct work" dependency handling

 from cancel_work_sync() as if flush_work().