Re: PREEMPT_RCU breaks anon_vma locking ?

[Paul E. McKenney]

On Sun, Feb 25, 2007 at 11:05:50PM +0300, Oleg Nesterov wrote:
> On 02/24, Hugh Dickins wrote:
> >
> > On Sat, 24 Feb 2007, Oleg Nesterov wrote:
> > 
> > > So page_lock_anon_vma() works correctly due to SLAB_DESTROY_BY_RCU even if
> > > anon_vma_unlink() has already freed anon_vma. In that case we should see
> > > list_empty(&anon_vma->head), we are safe.
> > 
> > (It doesn't affect your argument, but we won't necessarily see list_empty
> > there: the anon_vma slot may already have got reused for a different
> > bundle of vmas completely; but its lock remains a lock and its list
> > remains a list of vmas, and the worst that happens is that
> > page_referenced_anon or try_to_unmap_anon wanders through an irrelevant
> > bundle of vmas, looking for a page that cannot be found there.)
> 
> Yes, but in that case we are safe, right? We hold the lock, anon_vma can't be
> freed. But thanks for clarification! Somehow I missed that not only unlock()
> is unsafe (in theory). If anon_vma's memory was re-used for something else, we
> can't assume that we will see list_empty(&anon_vma->head).
> 
> > > 	static inline void page_lock_anon_vma(struct anon_vma *anon_vma)
> > 
> > It might be wiser to call that one page_unlock_anon_vma ;)
> 
> Congratulations, you passed the test! Paul didn't :)

What is in a name?  ;-)

						Thanx, Paul

> > (I'm slightly disgruntled that page_lock_anon_vma takes a struct page *,
> > but page_unlock_anon_vma no struct page *.  But it would be silly to do
> > it differently, or mess with the naming: besides, it's a static function
> > and the prototype guards against error anyway.)
> 
> OK. I thought about "unlock_anon_vma", but symmetry is good indeed.
> 
> Oleg.
> 
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/