Re: [PATCH 3/6] bdi: Use refcount_t for reference counting instead atomic_t

[Sebastian Andrzej Siewior <bigeasy@linutronix.de>]

On 2018-07-16 15:57:16 [-0700], Andrew Morton wrote:
> > --- a/mm/backing-dev.c
> > +++ b/mm/backing-dev.c
> > @@ -438,10 +438,10 @@ wb_congested_get_create(struct backing_dev_info *bdi, int blkcg_id, gfp_t gfp)
> >  	if (new_congested) {
> >  		/* !found and storage for new one already allocated, insert */
> >  		congested = new_congested;
> > -		new_congested = NULL;
> >  		rb_link_node(&congested->rb_node, parent, node);
> >  		rb_insert_color(&congested->rb_node, &bdi->cgwb_congested_tree);
> > -		goto found;
> > +		spin_unlock_irqrestore(&cgwb_lock, flags);
> > +		return congested;
> >  	}
> >  
> >  	spin_unlock_irqrestore(&cgwb_lock, flags);
> > @@ -451,13 +451,13 @@ wb_congested_get_create(struct backing_dev_info *bdi, int blkcg_id, gfp_t gfp)
> >  	if (!new_congested)
> >  		return NULL;
> >  
> > -	atomic_set(&new_congested->refcnt, 0);
> > +	refcount_set(&new_congested->refcnt, 1);
> >  	new_congested->__bdi = bdi;
> >  	new_congested->blkcg_id = blkcg_id;
> >  	goto retry;
> >  
> >  found:
> > -	atomic_inc(&congested->refcnt);
> > +	refcount_inc(&congested->refcnt);
> >  	spin_unlock_irqrestore(&cgwb_lock, flags);
> >  	kfree(new_congested);
> >  	return congested;
> >
> > ...
> >
> 
> I'm not sure that the restructuring of wb_congested_get_create() was
> desirable and it does make the patch harder to review.  But it looks
> OK to me.

By `restructuring' you mean the addition of return statement instead
using the goto label in the first hunk? If so, then you would have

	refcount_set(&new_congested->refcnt, 0);
	refcount_inc(&congested->refcnt);

which is a 0 -> 1 transition and is forbidden by refcount_t. So I had to
avoid this one.

Thank you applying the patches!

You applied the bdi and userns switch from atomic_t to refcount_t.
There were also the patches
  [PATCH 4/6] bdi: Use irqsave variant of refcount_dec_and_lock()
  [PATCH 6/6] userns: Use irqsave variant of refcount_dec_and_lock()

in the series which make use the irqsave version of
refcount_dec_and_lock(). Did you miss them by chance or skipped them on
purpose?

Sebastian