Re: [PATCH -mm v2 01/11] mm: track free size between VMAs in VMA rbtree

[Peter Zijlstra <peterz@infradead.org>]

On Tue, 2012-07-03 at 16:16 -0700, Michel Lespinasse wrote:
> On Tue, Jul 3, 2012 at 2:37 PM, Rik van Riel <riel@redhat.com> wrote:
> > On 06/29/2012 07:46 PM, Michel Lespinasse wrote:
> >> Basically, I think lib/rbtree.c should provide augmented rbtree support
> >> in the form of (versions of) rb_insert_color() and rb_erase() being able
> >> to
> >> callback to adjust the augmented node information around tree rotations,
> >> instead of using (conservative, overkill) loops to adjust the augmented
> >> node information after the fact
> >
> > That is what I originally worked on.
> >
> > I threw out that code after people told me (at LSF/MM) in
> > no uncertain terms that I should use the augmented rbtree
> > code :)
> 
> Well, bummer. Could you summarize what their argument was ? In other
> words, what are the constraints besides not adding overhead to the
> scheduler rbtree use case and keeping the code size reasonable ?

Mostly those.. but if you can do the thing with the __always_inline and
__flatten stuff from Daniel I'm ok with GCC generating more code, as
long we don't don't have to replicate the RB tree logic in C.