Re: [PATCH 2/3] btrfs: switch extent_buffer spinning_writers from atomic to int

[David Sterba <dsterba@suse.cz>]

On Fri, May 31, 2019 at 12:19:15PM +0300, Nikolay Borisov wrote:
> >  {
> > -	WARN_ON(atomic_read(&eb->spinning_writers));
> > -	atomic_inc(&eb->spinning_writers);
> > +	WARN_ON(eb->spinning_writers);
> > +	eb->spinning_writers++;
> >  }
> >  
> >  static void btrfs_assert_spinning_writers_put(struct extent_buffer *eb)
> >  {
> > -	WARN_ON(atomic_read(&eb->spinning_writers) != 1);
> > -	atomic_dec(&eb->spinning_writers);
> > +	WARN_ON(eb->spinning_writers != 1);
> > +	eb->spinning_writers--;
> >  }
> >  
> >  static void btrfs_assert_no_spinning_writers(struct extent_buffer *eb)
> >  {
> > -	WARN_ON(atomic_read(&eb->spinning_writers));
> > +	WARN_ON(eb->spinning_writers);
> >  }
> 
> IMO longterm  it will be good if those debug functions contained
> lockdep_assert_held_exclusive/read macros for posterity.

The functions are not public and used only inside implementation of
locks, so the chances of wrong use are low so I don't see much value
adding it.