The AMD Ryzen Thermal / Power Linux Reporting Improvements Working Well - V2 Up For Testing

[geearf]

Other than proper upstream kernel integration and not betrothing out-of-tree stuff?

Well I already use a bunch of non-upstream patches for my kernel anyway.
Maybe I should have phrased it more like: Nothing for me to really look forward to then :/

[rene]

Well I already use a bunch of non-upstream patches for my kernel anyway.
Maybe I should have phrased it more like: Nothing for me to really look forward to then :/

long term maintenance of the code. Out-of-tree drivers are the root of all evil. And bitrot.

[remenic]

A difference is, this k10temp driver only shows voltage and current readings while the zenpower driver also shows a reading for power usage.

Which is probably just voltage * current, so you're not missing anything.

[birdie]

Here's a patched file (v2) for the lazy ;-)

sha256sum *
0937189bd97e3ed6c90ad87172fa00980a3199769f81067a7e41e5440d2cbf40  k10temp.c

For my Ryzen 7 3700X:

sensors
k10temp-pci-00c3
Adapter: PCI adapter
Vcore:        +0.91 V
Vsoc:         +1.09 V
Tdie:         +34.5°C
Tctl:         +34.5°C
Tccd1:        +34.0°C
Icore:        +4.61 A
Isoc:         +7.65 A

Speaking of the zenpower driver - for my CPU it always showed absolutely incorrect power usage numbers. A simple multiplication of voltage and current doesn't work. Either there are hidden parameters or something else. My CPU eats up to 91W under load. 1.4 * 27 equals ~38W which is far from truth.

[geearf]

Speaking of the zenpower driver - for my CPU it always showed absolutely incorrect power usage numbers. A simple multiplication of voltage and current doesn't work. Either there are hidden parameters or something else. My CPU eats up to 91W under load. 1.4 * 27 equals ~38W which is far from truth.

Hmm, I just tried and on my system zenmonitor follows U=RI quite well for the SOC.

[ThoreauHD]

I think somrbody needs to do a cumulative post of the state of linux driver support on reddit's /r/AMD. Lisa Su reads that. She doesn't read this.

[birdie]

I think somrbody needs to do a cumulative post of the state of linux driver support on reddit's /r/AMD. Lisa Su reads that. She doesn't read this.

Lisa's PR managers could read that reddit. There's no way on Earth a CEO of a major corp would even touch reddit.

[muncrief]

It didn't compile on Manjaro linux-mainline 5.5-rc6. After applying the patches here's what I got when trying to compile:

AR drivers/gpu/built-in.a
make: *** [Makefile:1693: drivers] Error 2
==> ERROR: A failure occurred in build().
Aborting...

I don't know if it's the patches or if I'm doing something wrong. I'm not even sure if I'm supposed to use 5.5 or whatever is in the 5.6 git at the moment. In any case it's over my head to debug it, I'm just reporting my results. If someone wants to point me in the right direction I'll be happy to try it again though. I have an R7 3700x.

[playenball]

It didn't compile on Manjaro linux-mainline 5.5-rc6. After applying the patches here's what I got when trying to compile:

AR drivers/gpu/built-in.a
make: *** [Makefile:1693: drivers] Error 2
==> ERROR: A failure occurred in build().
Aborting...

I don't know if it's the patches or if I'm doing something wrong. I'm not even sure if I'm supposed to use 5.5 or whatever is in the 5.6 git at the moment. In any case it's over my head to debug it, I'm just reporting my results. If someone wants to point me in the right direction I'll be happy to try it again though. I have an R7 3700x.

Did you apply the patches in the right order? It compiled without any issues for me by just adding the patches and updating Manjaro's linux55 PKGBUILD file to apply them.

[NotMine999]

The main question is, why can't AMD provide patches for for k10temp with all this? It's totally something they should have been doing even before CPUs came out.

I think the general Linux community still acknowledges that AMD support for Linux is "behind" relative to the work that is done by Intel.

The fact that AMD cut back so deeply a number of years ago is not "unreported news". As anyone knows that has held a budget-responsible role in a business should know, it takes more M-O-N-E-Y to hire new talent than it does to retain new talent. Good Linux programming talent, especially those programmers that are skilled in writing "down next to the hardware" device drivers, can be really difficult.

Why should AMD hire these people and not the distributions or even independent developers? Access ... access to the internal bits of info that lawyers don't want released to the public; a despicable idea to some in these forums but it's true.

Sure, a specification can be released, but that's only part of the solution. Knowing how that specification was/is implemented in the hardware ('internal bits of info') can help a lot in developing good reliable code (and hardware) to work with that hardware. Just ask those folks that work for Intel (or AMD) that contribute Linux code to the kernel; it's sometimes easier to ask questions when you all work for the same company than it is when you are an outsider to that company.

It all comes down to internal resources and internal funding; M-O-N-E-Y.

"No bucks, no Buck Rogers." https://www.youtube.com/watch?v=a7rGA0Zv8R4

Perhaps in the near future we will see AMD spend some of the M-O-N-E-Y that it made on these new Zen family architecture CPUs on internal development resources for Linux driver development.