AMD Zen Thermal/Power Reporting Improvements Could Hit Linux 5.6 But More Testing Needed

[KoenDG]

So, chances of this hitting Ubuntu 20.04 are really small? Or is the kernel version already locked in?

[groeck]

So, chances of this hitting Ubuntu 20.04 are really small? Or is the kernel version already locked in?

The kernel crystal ball predicts that v5.6 will be released on April 12, 2020. The kernel freeze data for Ubuntu 20.04 is April 9, 2020. Based on this, I would guess that the chances are pretty much zero.

[scorpio810]

In case someone is interested in testing additional temperature sensors on Zen2 based Threadripper CPUs:
https://patchwork.kernel.org/patch/11348239/

For now in your opinion, is better to use your patch or use latest Initial threadripper/epyc support on Zenpower?

https://github.com/ocerman/zenpower/issues/12#issuecomment-579178365

[groeck]

For now in your opinion, is better to use your patch or use latest Initial threadripper/epyc support on Zenpower?
https://github.com/ocerman/zenpower/...ment-579178365

This is really your call to make. If you prefer zenpower's attempt to scale current values, go ahead and use it. Similar, if you like that it provides power values, go ahead and use it. The kernel driver won't attempt to provide scaling (meaning any scaling has to be done in the sensors configuration file), and it won't provide calculated power values (because the numbers are not from the chip but calculated in the kernel from current and voltage, and that calculation can as well be done in userspace). In respect to the new Epyc/Threadripper support for voltage/current reporting, sure, by all means go ahead and use zenpower.
Regarding k10temp, I will likely spend some more time on it at some point in the future, but I'd like to avoid hacks. For example, I am quite sure that there are registers showing for each Current/Voltage register if it is active or not, just like the CCD temperature registers. Similar, we do know that the chips report Tjmax (per HwINFO forum). On top of that, Tjmax is _not_ identical on all Zen chips, so reporting 95 degrees C is not much better than reporting 70 degrees C. Unfortunately, all that information is only available under NDA (that story again), and trying to figure it all out from sprinkled posts here or there and by trial and error is extremely time consuming, so I'd expect k10temp to always be a bit behind zenpower.
Of course, I am always happy to accept patches into k10temp, so if someone figures out how to determine based on telemetry registers (current/voltage readings) if they are enabled, or how to read Tjmax from the chip, by all means please go ahead and submit patches.