Re: [PATCH 1/4] loop: Use worker per cgroup instead of kworker

[Dan Schatzberg]

On Tue, Apr 21, 2020 at 10:48:45AM +0800, Hillf Danton wrote:
> 
> On Mon, 20 Apr 2020 18:39:29 -0400 Dan Schatzberg wrote:
> > 
> > @@ -1140,8 +1215,17 @@ static int __loop_clr_fd(struct loop_device *lo, bool release)
> >  	blk_mq_freeze_queue(lo->lo_queue);
> >  
> >  	spin_lock_irq(&lo->lo_lock);
> > +	destroy_workqueue(lo->workqueue);
> 
> Destruct it out of atomic context.

I may as well do this, but it doesn't matter, does it? The
blk_mq_freeze_queue above should drain all I/O so the workqueue will
be idle.