Re: [PATCH 13/15] sched: Detect child domain of NUMA (aka NODE) domain

[Andreas Herrmann <andreas.herrmann3@amd.com>]

On Mon, Aug 24, 2009 at 05:34:18PM +0200, Peter Zijlstra wrote:
> On Thu, 2009-08-20 at 15:43 +0200, Andreas Herrmann wrote:
> > On multi-node processors a NUMA node might not span a socket.
> > Instead a socket might span several NUMA nodes.
> > 
> > This patch introduces a check whether NODE domain is parent
> > of MC domain and sets s_data.numa_child_level accordingly.
> > (See previous patch for further details.)
> 
> right, except that the previous patch
> was rather cryptic :/

Sorry for that.

> So you're proposing to have the NODE level depend on multi-node and then
> flip NODE and CPU around?

Conditioned.

Only if a NUMA node does not span an entire socket, e.g.
  node 0: 0-3
  node 1: 4-7
  socket 0: 0-3, 4-7

You may have a SRAT that describes one NUMA node containing all
sockets, e.g.
  node 0: 0-7
  socket 0 : 0-7
If we have something like that on a multi-node processor system then
we don't need to flip NODE and CPU around.

Same is true if there is no SRAT or SRAT is bogus or CONFIG_ACPI_NUMA=n.

In theory, I could also think of node interleaving where a NUMA node
spans internal nodes of a socket on a multi-node processor -- no flip
in domain hierarchy needed.



On balance, as soon as a socket spans more than one NUMA node we have
to flip NODE and CPU.


Regards,
Andreas

-- 
Operating | Advanced Micro Devices GmbH
  System  | Karl-Hammerschmidt-Str. 34, 85609 Dornach b. München, Germany
 Research | Geschäftsführer: Thomas M. McCoy, Giuliano Meroni
  Center  | Sitz: Dornach, Gemeinde Aschheim, Landkreis München
  (OSRC)  | Registergericht München, HRB Nr. 43632