On Thu, Feb 11, 2016 at 9:58 AM, Ganapatrao Kulkarni [off-list ref] wrote:

Hi Rob,

On Thu, Feb 11, 2016 at 9:02 PM, Robert Richter
[off-list ref] wrote:

quoted

On 11.02.16 08:50:41, Rob Herring wrote:

quoted

On Tue, Feb 2, 2016 at 4:09 AM, Ganapatrao Kulkarni
[off-list ref] wrote:

quoted

dt node parsing for numa topology is done using device property
numa-node-id and device node distance-map.

How is it that powerpc doesn't need flat DT parsing for NUMA? Both
arches are memblock based and the binding is similar IIRC, so you
should be able to find a way to do this with the unflattened tree.
Ideally, there should be some common code shared as well.

powerpc has implemented numa binding using property named
"ibm,associativity", we too started adding using the same binding[
refer version V6]
however had discussion/review with Mark R about mapping and we could not map to
some of the topology like mesh and ring topology.

So we have defined and implemented this new binding which is simple
and scalable.
This binding is very much aligned to ACPI NUMA specification.

You didn't answer my main question. Presumably the NUMA setup needs to be done at the same point in time whether the binding is the same or not. Powerpc manages to do the setup 

Rob