On Mon, May 21, 2018 at 08:33:31AM -0700, Tejun Heo wrote:
> Hello,
> 
> On Mon, May 21, 2018 at 03:38:08PM +0800, Boqun Feng wrote:
> > > [   28.387571] WARNING: possible irq lock inversion dependency detected
> > > [   28.389198] 4.17.0-rc1-00027-gb37d049 #6 Not tainted
> > > [   28.390617] --------------------------------------------------------
> > > [   28.392205] systemd/1 just changed the state of lock:
> > > [   28.393627] 00000000fe57773b (css_set_lock){..-.}, at: cgroup_free+0xf2/0x12a
> > > [   28.395456] but this lock took another, SOFTIRQ-unsafe lock in the past:
> > > [   28.397152]  (tasklist_lock){.+.+}
> > > [   28.397155] 
> > > [   28.397155] 
> > > [   28.397155] and interrupts could create inverse lock ordering between them.
> > > [   28.397155] 
> > > [   28.402049] 
> > > [   28.402049] other info that might help us debug this:
> > > [   28.404203]  Possible interrupt unsafe locking scenario:
> > > [   28.404203] 
> > > [   28.406379]        CPU0                    CPU1
> > > [   28.407712]        ----                    ----
> > > [   28.409107]   lock(tasklist_lock);
> > > [   28.410207]                                local_irq_disable();
> > > [   28.411822]                                lock(css_set_lock);
> > > [   28.413423]                                lock(tasklist_lock);
> > > [   28.415094]   <Interrupt>
> > > [   28.416106]     lock(css_set_lock);
> > > [   28.417263] 
> > > [   28.417263]  *** DEADLOCK ***
> > > [   28.417263] 
> 
> Would something like the following work?
> 

Yep! Looks good to me.

Regards,
Boqun

> Thanks.
> 
> diff --git a/kernel/cgroup/cgroup.c b/kernel/cgroup/cgroup.c
> index a662bfc..3b68174 100644
> --- a/kernel/cgroup/cgroup.c
> +++ b/kernel/cgroup/cgroup.c
> @@ -1782,13 +1782,6 @@ static void cgroup_enable_task_cg_lists(void)
>  {
>  	struct task_struct *p, *g;
>  
> -	spin_lock_irq(&css_set_lock);
> -
> -	if (use_task_css_set_links)
> -		goto out_unlock;
> -
> -	use_task_css_set_links = true;
> -
>  	/*
>  	 * We need tasklist_lock because RCU is not safe against
>  	 * while_each_thread(). Besides, a forking task that has passed
> @@ -1797,6 +1790,13 @@ static void cgroup_enable_task_cg_lists(void)
>  	 * tasklist if we walk through it with RCU.
>  	 */
>  	read_lock(&tasklist_lock);
> +	spin_lock_irq(&css_set_lock);
> +
> +	if (use_task_css_set_links)
> +		goto out_unlock;
> +
> +	use_task_css_set_links = true;
> +
>  	do_each_thread(g, p) {
>  		WARN_ON_ONCE(!list_empty(&p->cg_list) ||
>  			     task_css_set(p) != &init_css_set);
> @@ -1824,9 +1824,9 @@ static void cgroup_enable_task_cg_lists(void)
>  		}
>  		spin_unlock(&p->sighand->siglock);
>  	} while_each_thread(g, p);
> -	read_unlock(&tasklist_lock);
>  out_unlock:
>  	spin_unlock_irq(&css_set_lock);
> +	read_unlock(&tasklist_lock);
>  }
>  
>  static void init_cgroup_housekeeping(struct cgroup *cgrp)