On 05/29/2012 09:25 PM, Christoph Lameter wrote:

quoted

I will try to at least have the page accounting done in a consistent way. How

quoted

 about that?

Ok. What do you mean by "consistent"? Since objects and pages can be used
in a shared way and since accounting in many areas of the kernel is
intentional fuzzy to avoid performance issues there is not that much of a
requirement for accuracy. We have problems even just defining what the
exact set of pages belonging to a task/process is.

I mean consistent between allocators, done in a shared place.
Note that what we do here *is* still fuzzy to some degree. We can have a 
level of sharing, and always account to the first one to touch, we don't 
go worrying about this because that would be hell. We also don't
carry a process' objects around when we send it to a different cgroup. 
Those are all already simplications for performance and simplicity sake.

But we really need a page to be filled with objects from the same 
cgroup, and the non-shared objects to be accounted to the right place.

Otherwise, I don't think we can meet even the lighter of isolation 
guarantees.