Messages in this thread | | | Date | Thu, 6 May 2010 15:25:31 +0100 | From | Mel Gorman <> | Subject | Re: [PATCH 1/2] mm,migration: Prevent rmap_walk_[anon|ksm] seeing the wrong VMA information |
| |
On Thu, May 06, 2010 at 07:15:31AM -0700, Linus Torvalds wrote: > > > On Thu, 6 May 2010, Mel Gorman wrote: > > > > > > What makes this ok is the fact that it must be running under the RCU read > > > lock, and anon_vma's thus cannot be released. > > > > This is very subtle in itself. RCU guarantees that the anon_vma exists > > but does it guarantee that it's the same one we expect and that it > > hasn't been freed and reused? > > Nothing. And we shouldn't care. > > If it's been freed and re-used, then all the anon_vma's (and vma's) > associated with the original anon_vma (and page) have been free'd. > > And that, in turn, means that we don't really need to lock anything at > all. The fact that we end up locking an anon_vma that _used_ to be the > root anon_vma is immaterial - the lock won't _help_, but it shouldn't hurt > either, since it's still a valid spinlock. >
I can't see any problem with the logic.
> Now, the above is only true as far as the anon_vma itself is concerned. > It's entirely possible that any _other_ data structures would need to be > double-checked after getting the lock. For example, is the _page_ still > associated with that anon_vma? But that's an external issue as far as the > anon_vma locking is concerned - presumably the 'rmap_walk()' caller will > have made sure that the page itself is stable somehow. >
It does, by having the page locked as it performs the walk.
-- Mel Gorman Part-time Phd Student Linux Technology Center University of Limerick IBM Dublin Software Lab
| |