Messages in this thread | | | Date | Wed, 12 May 2010 18:18:11 -0400 | From | Rik van Riel <> | Subject | Re: [PATCH 4/5] always lock the root (oldest) anon_vma |
| |
On 05/12/2010 05:55 PM, Linus Torvalds wrote:
> Wouldn't it be sufficient to do > > if (atomic_dec_and_test(&anon_vma->ksm_refcount)) { > anon_vma_lock(anon_vma); > > instead? The "atomic_dec_and_lock()" semantics are _much_ stricter than a > regular "decrement and test and then lock", and that strictness means that > it's way more complicated and expensive. So if you don't need the > semantics, you shouldn't use them.
I suspect the atomic_dec_and_lock in the KVM code is being used to prevent the following race:
1) KSM code reduces the refcount to 0
2) munmap on other CPU frees the anon_vma
3) KSM code takes the anon_vma lock, which now lives in freed memory
Am I totally confused by this and can we use a nicer approach?
-- All rights reversed
| |