Hi,
If slow path frequency changes are conducted in a SCHED_OTHER context then they may be delayed for some amount of time, including indefinitely, when real time or deadline activity is taking place.
Move the slow path to a real time kernel thread using the kthread worker infrastructure. In the future the thread should be made SCHED_DEADLINE. The RT priority is arbitrarily set to 50 for now.
This was tested with Hackbench on ARM Exynos, dual core A15 platform and no regressions were seen. The third patch has more details on it.
This work was started by Steve Muckle, where he used a simple kthread instead of kthread-worker and that wasn't sufficient as some guarantees weren't met.
I was wondering if the same should be done for ondemand/conservative governors as well ?
V1->V2: - first patch is new based on Peter's suggestions. - fixed indented label - Moved kthread creation/destruction into separate routines - Used MACRO instead of magic number '50' - minor formatting, commenting and improved commit logs
-- viresh
Viresh Kumar (4): cpufreq: schedutil: Avoid indented labels cpufreq: schedutil: enable fast switch earlier cpufreq: schedutil: move slow path from workqueue to SCHED_FIFO task cpufreq: schedutil: irq-work and mutex are only used in slow path
kernel/sched/cpufreq_schedutil.c | 119 ++++++++++++++++++++++++++++++++------- 1 file changed, 99 insertions(+), 20 deletions(-)