Received-SPF: pass (google.com: domain of linux-kernel+bounces-208823-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.48.161 as permitted sender) client-ip=147.75.48.161;
Sender: Tejun Heo <htejun@gmail.com>
Date: Mon, 10 Jun 2024 10:00:49 -1000
From: Tejun Heo <tj@kernel.org>
To: Oleg Nesterov <oleg@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>,
	"Eric W. Biederman" <ebiederm@xmission.com>,
	linux-kernel@vger.kernel.org, Zefan Li <lizefan.x@bytedance.com>,
	Johannes Weiner <hannes@cmpxchg.org>
Subject: Re: Q: css_task_iter_advance() && dying_tasks
Message-ID: <ZmdbcSFmkmh-Cl7R@slm.duckdns.org>
References: <20240609142342.GA11165@redhat.com>
 <20240610105028.GA21586@redhat.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20240610105028.GA21586@redhat.com>

Hello, Oleg.

On Mon, Jun 10, 2024 at 12:50:28PM +0200, Oleg Nesterov wrote:
> I never understood the code in kernel/cgroup/ even remotely, most probably
> I missed something, but let me ask a couple of stupid questions anyway.
> 
> cgroup_exit() does
> 
> 	css_set_move_task(tsk, cset, NULL, false);
> 	list_add_tail(&tsk->cg_list, &cset->dying_tasks);
> 
> but unless I am totally confused css_task_iter_advance() always ignores
> the "dying" sub-threads, so perhaps it should do, say,
> 
> 	css_set_move_task(tsk, cset, NULL, false);
> 	if (delay_group_leader(tsk))
> 		list_add_tail(&tsk->cg_list, &cset->dying_tasks);
> 
> and then cgroup_release() can check list_empty(cg_list) before it takes
> css_set_lock.
> 
> No ?

Yeah, I think so. The current code hasn't broken for quite a while but it
also hasn't received much attention after the iterator updates which were
very much in the spirit of just getting it to work. I don't think you're
missing anything.

> Or, perhaps we can do even better? Can't cgroup_exit() do something like
> 
> 	// group_dead should be passed from do_exit()
> 
> 	css_set_move_task(tsk, cset, NULL, false);
> 
> 	if (thread_group_leader(tsk) && !group_dead)
> 		list_add_tail(&tsk->cg_list, &cset->dying_tasks);
> 
> 	else if (!thread_group_leader(tsk) && group_dead) {
> 		leader = tsk->group_leader;
> 		if (!list_empty(leader->cg_list) {
> 			css_set_skip_task_iters(task_css_set(leader), leader);
> 			list_del_init(&leader->cg_list);
> 		}
> 	}
> 
> and then
> 
> 	- kill the atomic_read(&task->signal->live)) check in
> 	  css_task_iter_advance()
> 
> 	- kill the code under css_set_lock in cgroup_release()

That does sound a lot better than the current code. Care to send the patch?

Thanks.

-- 
tejun