Re: [PATCH] thp+memcg-numa: fix BUG at include/linux/mm.h:370!

[Johannes Weiner <jweiner@redhat.com>]

On Mon, Mar 14, 2011 at 06:17:31PM +0100, Andrea Arcangeli wrote:
> On Mon, Mar 14, 2011 at 09:56:10AM -0700, Linus Torvalds wrote:
> > Does mem_cgroup_newpage_charge() even _need_ the mmap_sem at all? And
> > if not, why not release the read-lock early? And even if it _does_
> > need it, why not do

[...]

> About mem_cgroup_newpage_charge I think you're right it won't need the
> mmap_sem. Running it under it is sure safe. But if it's not needed we
> can move the up_read before the mem_cgroup_newpage_charge like you
> suggested. Johannes/Minchan could you confirm the mmap_sem isn't
> needed around mem_cgroup_newpage_charge? The mm and new_page are
> stable without the mmap_sem, only the vma goes away but the memcg
> shouldn't care.

We don't care about the vma.  It's all about assigning the physical
page to the memcg that mm->owner belongs to.

It would be the first callsite not holding the mmap_sem, but that is
only because all existing sites are fault handlers that don't drop the
lock for other reasons.

I am not aware of anything that would rely on the lock in there, or
would not deserve to break if it did.