Messages in this thread | | | Date | Thu, 17 Dec 2009 09:41:39 +0100 | From | Andi Kleen <> | Subject | Re: [mm][RFC][PATCH 0/11] mm accessor updates. |
| |
On Thu, Dec 17, 2009 at 12:01:55AM +0100, Peter Zijlstra wrote: > On Wed, 2009-12-16 at 10:27 -0600, Christoph Lameter wrote: > > On Wed, 16 Dec 2009, Andi Kleen wrote: > > > > > > Do you have alternative recommendation rather than wrapping all accesses by > > > > special functions ? > > > > > > Work out what changes need to be done for ranged mmap locks and do them all > > > in one pass. > > > > Locking ranges is already possible through the split ptlock and > > could be enhanced through placing locks in the vma structures. > > > > That does nothing solve the basic locking issues of mmap_sem. We need > > Kame-sans abstraction layer. A vma based lock or a ptlock still needs to > > ensure that the mm struct does not vanish while the lock is held. > > It should, you shouldn't be able to remove a mm while there's still > vma's around, and you shouldn't be able to remove a vma when there's > still pagetables around. And if you rcu-free all of them you're stable > enough for lots of speculative behaviour.
Yes, the existing reference counts are probably sufficient for all that.
Still need list stability.
> As for per-vma locks, those are pretty much useless too, there's plenty > applications doing lots of work on a few very large vmas.
True. -Andi
-- ak@linux.intel.com -- Speaking for myself only.
| |