How does one do that?

(I'd also still love to see more info that can be used to determine where the CPU time is going.)

I'm not sure. Sonny do you know?

I'm also not sure how to find out where those cycles are spent. I saw 300ms of CPU busy time in powerd in a trace, but that doesn't tell you what powerd is doing :(

+dtor

The right way is to put it into a cpuset that is limited to little cores.  We have cpusets for chrome processes and for the android container but not one for general system processes.

This is probably something we want to do for other general processes as well -- perhaps update_engine should also go in there.

Shrug. I just spent some time trying to get useful data from perf, but I can't figure out how to get it to print symbols or even to consistently provide data about powerd instead of about other random processes. It also makes powerd hang every time I try to stop it.

Sonny any secrets to capturing perf you can share?  Do you have some existing kevin traces you could share?

Are there any guidelines for pushing lower priority tasks to little cores?

I thought in a perfect world the scheduler would be tuned to do this accordingly?  I worry forcing certain processes to run on little vs big will have undesirable consequences like higher power or jank elsewhere with a task that got cast to little cores alone. 

This may not be necessary anymore with the new scheduler scheme -- important tasks will get prioritized onto littles -- I'm going to mark WontFix unless someone points out a real trace where this is still a problem