Hi Srinivas,
On Mon, Feb 14, 2022 at 7:17 AM srinivas pandruvada srinivas.pandruvada@linux.intel.com wrote:
Hi Doug,
I think you use CONFIG_NO_HZ_FULL.
No. Here is the relevant excerpt from the kernel config file:
# # Timers subsystem # CONFIG_TICK_ONESHOT=y CONFIG_NO_HZ_COMMON=y # CONFIG_HZ_PERIODIC is not set CONFIG_NO_HZ_IDLE=y # CONFIG_NO_HZ_FULL is not set CONFIG_NO_HZ=y CONFIG_HIGH_RES_TIMERS=y # end of Timers subsystem
Here we are getting callback from scheduler. Can we check that if scheduler woke up on those CPUs?
As far as I can determine, yes. But note that I am unfamiliar with this area.
We can run "trace-cmd -e sched" and check in kernel shark if there is similar gaps in activity.
I do not use trace-cmd and had never heard of kernel shark. Nor do I actually run any desktop GUI on linux, only servers. I attempted to acquire what you wanted with primitive trace commands.
Workflow: All as before (the 347 Hertz work/sleep frequency test), with a 20 minute trace in the middle. Powers were monitored again to confirm differences, and just in case trace itself modified the system response (it didn't).
Power averages (excluding the sample where the trace file was being written to disk): Stock: 4.1 +37% Revert: 3.0
I only looked at a few of the CPUs data, the largest, smallest and a mid-range file sizes, excluding the main working CPU. Maximum times between "sched_wakeup", seconds:
Stock: CPU 2: 4.0 CPU 4: 4.0 CPU 9: 1.0
Revert: CPU 1: 2.0 CPU 2: 4.0 CPU 7: 1.54
I do not know if other stuff in the files might be odd or not.
... Doug