In order to save power, it would be useful to schedule light weight work on cpus that aren't IDLE instead of waking up an IDLE one.
By idle cpu (from scheduler's perspective) we mean: - Current task is idle task - nr_running == 0 - wake_list is empty
This is already implemented for timers as get_nohz_timer_target(). We can figure out few more users of this feature, like workqueues.
This patchset converts get_nohz_timer_target() into a generic API sched_select_cpu() so that other frameworks (like workqueue) can also use it.
This routine returns the cpu which is non-idle. It accepts a bitwise OR of SD_* flags present in linux/sched.h. If the local CPU isn't idle OR all cpus are idle, local cpu is returned back. If local cpu is idle, then we must look for another CPU which have all the flags passed as argument as set and isn't idle.
This patchset in first two patches creates generic API sched_select_cpu(). In the third patch we create a new set of APIs for workqueues to queue work on any cpu. All other patches migrate some of the users of workqueues which showed up significantly on my setup. Others can be migrated later.
Earlier discussions over v1 and v2 can be found here: http://www.mail-archive.com/linaro-dev@lists.linaro.org/msg13342.html http://lists.linaro.org/pipermail/linaro-dev/2012-November/014344.html
Earlier discussions over this concept were done at last LPC: http://summit.linuxplumbersconf.org/lpc-2012/meeting/90/lpc2012-sched-timer-...
Setup: ----- - ARM Vexpress TC2 - big.LITTLE CPU - Core 0-1: A15, 2-4: A7 - rootfs: linaro-ubuntu-devel
This patchset has been tested on a big LITTLE system (heterogeneous) but is useful for all other homogeneous systems as well. During these tests audio was played in background using aplay.
Results: -------
Cluster A15 Energy Cluster A7 Energy Total ------------------ ----------------- -----
Without this patchset (Energy in Joules): ---------------------
0.151162 2.183545 2.334707 0.223730 2.687067 2.910797 0.289687 2.732702 3.022389 0.454198 2.745908 3.200106 0.495552 2.746465 3.242017
Average: 0.322866 2.619137 2.942003
With this patchset (Energy in Joules): ---------------------
0.133361 2.267822 2.401183 0.260626 2.833389 3.094015 0.142365 2.277929 2.420294 0.246793 2.582550 2.829343 0.130462 2.257033 2.387495
Average: 0.182721 2.443745 2.626466
Above tests are repeated multiple times and events are tracked using trace-cmd and analysed using kernelshark. And it was easily noticeable that idle time for many cpus has increased considerably, which eventually saved some power.
These patches are applied here for others to test: http://git.linaro.org/gitweb?p=people/vireshk/linux.git%3Ba=shortlog%3Bh=ref...
Viresh Kumar (7): sched: Create sched_select_cpu() to give preferred CPU for power saving timer: hrtimer: Don't check idle_cpu() before calling get_nohz_timer_target() workqueue: Add helpers to schedule work on any cpu PHYLIB: queue work on any cpu mmc: queue work on any cpu block: queue work on any cpu fbcon: queue work on any cpu
block/blk-core.c | 6 +- block/blk-ioc.c | 2 +- block/genhd.c | 9 ++- drivers/mmc/core/core.c | 2 +- drivers/net/phy/phy.c | 9 +-- drivers/video/console/fbcon.c | 2 +- include/linux/sched.h | 21 +++++- include/linux/workqueue.h | 5 ++ kernel/hrtimer.c | 2 +- kernel/sched/core.c | 63 +++++++++------- kernel/timer.c | 2 +- kernel/workqueue.c | 163 +++++++++++++++++++++++++++++------------- 12 files changed, 192 insertions(+), 94 deletions(-)