Closed oshmoun closed 6 years ago
please try with min_cpus="3"
We've been down this road before. It is a bug in the rqbalance driver and not the xml file
so you also tested increasing min_cpus to 3 and it didn't work? or did you only modify thresholds?
The problem is that the rqbalance governor will not recheck if the cores are balanced after an offlineing of a cpu until a cpufreq change happens which is not the case if all remaining core are at about 100% load.
This could also happen with 3cores at 100% load and the 4th core stuck offline. But it is much more unlikely to happen. You can see the PR i opened at the kernel repo for an explanation. You could also debug it by adding some pr_debug messages to the right methods in the rqbalance.c .
Ah understood, thanks for the explanation!
The previous thresholds declared were too high, and the device was lagging on video playback. Lower the thresholds to give the device more room to scale.