On 12/04/2012 08:18 PM, Minchan Kim wrote:
> On Tue, Dec 04, 2012 at 11:13:40AM -0800, John Stultz wrote:
>> I don't think the problem is when vmas being marked VM_VOLATILE are
>> being merged, its that when we mark the vma as *non-volatile*, and
>> remove the VM_VOLATILE flag we merge the non-volatile vmas with
>> neighboring vmas. So preserving the purged flag during that merge is
>> important. Again, the example I used to trigger this was an
>> alternating pattern of volatile and non volatile vmas, then marking
>> the entire range non-volatile (though sometimes in two overlapping
>> passes).
> Understood. Thanks.
> Below patch solves your problems? It's simple than yours.

Yea, this is nicer then my fix.
Although I still need the purged handling in the vma merge code for me 
to see the behavior I expect in my tests.

I've integrated your patch and repushed my queue here:
http://git.linaro.org/gitweb?p=people/jstultz/android-dev.git;a=shortlog;h=refs/heads/dev/minchan-anonvol

git://git.linaro.org/people/jstultz/android-dev.git dev/minchan-anonvol

> Anyway, both yours and mine are not right fix.
> As I mentioned, locking scheme is broken.
> We need anon_vma_lock to handle purged and we should consider fork
> case, too.
Hrm. I'm sure you're right, as I've not yet fully grasped all the 
locking rules here.  Could you clarify how it is broken? And why is the 
anon_vma_lock needed to manage the purged state that is part of the vma 
itself?

thanks
-john