powertop
powertop copied to clipboard
fenrus75
running powertop drops all scaling_max_frequency to lowest on arm64 yoga c630 laptop
Hello, On ubuntu 20.04 with powertop (from repo or newest git) when I run the program It sets my scaling_max_frequency to lowest possible frequency. My laptop has 4 slow cores at 1.7ghz and 4 fast cores at 2.9ghz. It seems mostly random on which cores it will make slow, usually (%65? of time) the slow cores get slowed, slightly less often the higher cores get slowed and sometimes both. I edited src/cpu/abstract_cpu.cpp and commented out lines 479 to 492 which are supposed to reapply correct max scaling freq after "wiggling" the cpu .Now the issue seems to have stopped. I can fix the cpu's by simply running; echo 2956800 > /sys/devices/system/cpu/cpufreq/policy4/scaling_max_freq echo 1766400 > /sys/devices/system/cpu/cpufreq/policy0/scaling_max_freq
If this is caused by the HMP multicore setup or something else Im not sure. If a fix could be mainlined even if a command line or compile time option would be great. Thank you
I've seen similar behavior on a Lenovo Thinkpad X1 gen 9. Powertop sets the scaling_max_frequency to the lowest possible frequency on a single core.
It seems like this is possible if:
- You exit powertop at just the right time, before it resets the max frequency in
abstract_cpu::wiggle - You have two instances of powertop running at the same time, both trying to "wiggle" the same CPU core
I believe a partial fix here would be use a destructor of a some class to reset the max frequency when the abstract_cpu::wiggle method exits.
I totally forgot all about this, i apparently even posted my fork on my github https://github.com/bm16ton/yoga-c630-powertop
But i have no idea what im.doing i just play around till it all seems to work. I just received a second yoga not long ago. Hopefully soon ill be able to look at this again and any changes made upstream since my last commit.
I'm seeing what appears to be the same problem on the Framework laptop with 13th gen Intel CPU (Raptor Lake, i5-1340p). Most frequently CPU 0 gets "stuck" at the lowest frequency and needs to be adjusted manually. I came to believe that it's powertop that causes it. I'm running version 2.15. This is a serious problem!
