Starbook ultra: cpu frequency scaling

i’ve been running qubes on laptops for over a decade, and i’ve observed various cpu frequency scaling issues over the past ~7 years. i just recently bought a star labs starbook ultra mk vii, and it seems to be experiencing a variation upon this years-long pattern i have observed of cpu frequency scaling problems with various laptops.

the specific problem i am experiencing currently is that the cpu base frequencies for both performance and efficient cores in this ultra 7 165h do not match specs, and the machine runs hot, even when it is idle. xenpm get-cpufreq-para shows performance and efficient cores with a base frequency of 3.1 ghz, whereas the intel specs show performance, efficient, and low power efficient core base frequencies as 1.4 ghz, 900 mhz, and 700 mhz respectively. the xenpm output does show the max frequencies of 5 ghz and 3.8 ghz for performance and efficient cores properly, and the scaling_driver shows as hwp for all cpus.

i recognize that qubes has limitations, and i have experienced various types of frequency scaling issues with thinkpad t480 and t490, purism librem 14, and now starbook ultra. in each case, the machine runs hot, makes cpu fan noise, and the cpu frequency scaling settings visible via xenpm are not what you would expect. i have also read most of the comments on several of the more active qubes issues, e.g. 4604, but i am not sure i understand specifically what is causing these issues.

my inference regarding what is going on here is that xen is a product mainly intended to run on bare metal servers used by large corporations to host vpses of various types. qubes aims to run on both desktop and laptop machines, but laptop machines often make aggressive use of frequency scaling for the reasons cited above (heat, battery life, noise, wear and tear). a side effect of xen being built to run on servers is that they don’t care much about these laptop-driven considerations, and xen correspondingly doesn’t work great on the laptops.

ofc, i could be entirely wrong about all this, but at this point i’m keen to understand why and how qubes so consistently has these issues with frequency scaling across multiple laptops. it would be great to address my specific current issue, and my sense is that it’s a broader problem that spans years and may be difficult for qubes devs to address, e.g. changes needed in xen codebase.

thanks for reading

Check this TLP evolving into xenpm topic:

i’ve been re-reviewing the tools available for documenting what is going on, and i have been running powertop. i am also giving 4604 comment a close re-read to get refreshed on the topic.

i have 5 sys vms + 10 appvms running, and when the appvms are doing almost nothing, i see ~20 W usage and if i scroll up and down on this forum in tor browser, it goes up to ~30 W. i was surprised to see that i could not easily get cpu temp and fan speed information, so any advice on how to do this would be appreciated.

from a leg heat and fan noise perspective, this is literally a downgrade going from a thinkpad t480 to this starbook ultra. the heat coming from the starbook when it’s doing almost nothing is on par with the t480 while i was watching yt.

i know some kernel devs that may be able to assist with actually fixing this problem, but without knowing more of the constraints of the problem, e.g. maintaining a substantive xen patch is too much of a burden for qubes devs, getting a xen patch merged upstream is a political headache, i’m reticent to get more involved.