sched: EAS: kick nohz idle CPU within the same sched domain
At present, nohz idle balancer finds and kicks the lowest idle CPU
in numerical order. Because of that when a misfit task on a big CPU
is in need of load balancing, currently scheduler can kick a little
CPU that will cause not only load balancing failure but also
meaningless idle CPU wake up since the woken up little CPU will
fail to pull task as the task won't fit into.
Try to kick nohz idle CPU within the same sched domain at first and
then kick the other cluster idle CPU when it's necessary.
Change-Id: Idf0f9a494e4596cb254f15ea58b558815847896b
Signed-off-by: Joonwoo Park <joonwoop@codeaurora.org>
1 file changed