]> git.karo-electronics.de Git - linux-beck.git/commit
sched: Move the second half of ttwu() to the remote cpu
authorPeter Zijlstra <a.p.zijlstra@chello.nl>
Tue, 5 Apr 2011 15:23:58 +0000 (17:23 +0200)
committerIngo Molnar <mingo@elte.hu>
Thu, 14 Apr 2011 06:52:41 +0000 (08:52 +0200)
commit317f394160e9beb97d19a84c39b7e5eb3d7815a8
tree486f268e9ec010744c17333f0f543da9d3909ff4
parentc05fbafba1c5482bee399b360288fa405415e126
sched: Move the second half of ttwu() to the remote cpu

Now that we've removed the rq->lock requirement from the first part of
ttwu() and can compute placement without holding any rq->lock, ensure
we execute the second half of ttwu() on the actual cpu we want the
task to run on.

This avoids having to take rq->lock and doing the task enqueue
remotely, saving lots on cacheline transfers.

As measured using: http://oss.oracle.com/~mason/sembench.c

  $ for i in /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor ; do echo performance > $i; done
  $ echo 4096 32000 64 128 > /proc/sys/kernel/sem
  $ ./sembench -t 2048 -w 1900 -o 0

  unpatched: run time 30 seconds 647278 worker burns per second
  patched:   run time 30 seconds 816715 worker burns per second

Reviewed-by: Frank Rowand <frank.rowand@am.sony.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Nick Piggin <npiggin@kernel.dk>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/20110405152729.515897185@chello.nl
include/linux/sched.h
init/Kconfig
kernel/sched.c
kernel/sched_features.h