Re: [PATCH 3/3] mm: Simplify anon_vma refcounts

[Peter Zijlstra <a.p.zijlstra@chello.nl>]

On Thu, 2011-02-17 at 10:30 -0800, Linus Torvalds wrote:
> On Thu, Feb 17, 2011 at 8:19 AM, Peter Zijlstra <a.p.zijlstra@chello.nl> wrote:
> >
> > +void __put_anon_vma(struct anon_vma *anon_vma)
> > +{
> > +       if (anon_vma->root != anon_vma)
> > +               put_anon_vma(anon_vma->root);
> > +       anon_vma_free(anon_vma);
> >  }
> 
> So this makes me nervous. It looks like recursion.
> 
> Now, I don't think we can ever get a chain of these things (because
> the root should be the root of everything),

Exactly.

>  but I still preferred the
> older code that made that "one-level root" case explicit, and didn't
> have recursion.
> 
> IOW, even though it should be entirely equivalent, I think I'd really
> prefer something like
> 
>   void __put_anon_vma(struct anon_vma *anon_vma)
>   {
>     struct anon_vma *root = anon_vma->root;
> 
>     if (root != anon_vma && atomic_dec_and_test(&root->refcount))
>       anon_vma_free(root);
>     anon_vma_free(anon_vma);
>   }
> 
> instead. Exactly because it makes it very clear that the "root" is a
> root, and we're not doing some possibly arbitrarily deep list like the
> dentry tree (which avoids recursion by open-coding its freeing as a
> loop).
> 
> Hmm? (The above is obviously untested, maybe it has some stupid bug)

Looks about right, I'll give it a spin.