sched: Fix SD_OVERLAP
authorPeter Zijlstra <peterz@infradead.org>
Wed, 23 May 2012 16:00:43 +0000 (18:00 +0200)
committerIngo Molnar <mingo@kernel.org>
Wed, 30 May 2012 12:02:24 +0000 (14:02 +0200)
commit74a5ce20e6eeeb3751340b390e7ac1d1d07bbf55
treeebbef56666aa11303eafbfb6adbfce56e7a4c605
parent2ea45800d8e1c3c51c45a233d6bd6289a297a386
sched: Fix SD_OVERLAP

SD_OVERLAP exists to allow overlapping groups, overlapping groups
appear in NUMA topologies that aren't fully connected.

The typical result of not fully connected NUMA is that each cpu (or
rather node) will have different spans for a particular distance.
However due to how sched domains are traversed -- only the first cpu
in the mask goes one level up -- the next level only cares about the
spans of the cpus that went up.

Due to this two things were observed to be broken:

 - build_overlap_sched_groups() -- since its possible the cpu we're
   building the groups for exists in multiple (or all) groups, the
   selection criteria of the first group didn't ensure there was a cpu
   for which is was true that cpumask_first(span) == cpu. Thus load-
   balancing would terminate.

 - update_group_power() -- assumed that the cpu span of the first
   group of the domain was covered by all groups of the child domain.
   The above explains why this isn't true, so deal with it.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: David Rientjes <rientjes@google.com>
Link: http://lkml.kernel.org/r/1337788843.9783.14.camel@laptop
Signed-off-by: Ingo Molnar <mingo@kernel.org>
kernel/sched/core.c
kernel/sched/fair.c