It Doesn't Look Like A Ryzen/EPYC Thermal Driver Will Make It For Linux 4.14

[phoronix]

Phoronix: It Doesn't Look Like A Ryzen/EPYC Thermal Driver Will Make It For Linux 4.14

While the Ryzen CPUs have been available for a few months now and the higher-wattage Threadripper and EPYC processors are now available too, the Linux thermal driver remains missing in action and it's looking less likely that it will materialize for Linux 4.14...

http://www.phoronix.com/scan.php?page=news_item&px=Zen-Thermal-4.14-Where

[JPFSanders]

For most users a Zen thermal driver isn't necessary

No, it is a must, regardless of who ends using it, why the heck do we have to be treated like second class every single time, something as basic as the thermal readings should have been publicly available from day 0.

It is such a stupid thing it is irritating.

[torsionbar28]

No, it is a must, regardless of who ends using it, why the heck do we have to be treated like second class every single time, something as basic as the thermal readings should have been publicly available from day 0.

It is such a stupid thing it is irritating.

Meh, it's not so critical any more nowadays. Years ago, it wasn't hard at all to overheat and fail a CPU, so monitoring temps was essential. Modern processors however, have effective thermal management, and will throttle performance to stay within temp envelope, or even shutdown altogether to prevent hardware damage. I too like to monitor my CPU temp, even though it's little more than trivia these days, but playing the 2nd hand citizen card seems extreme. Go have a beer and relax. Do you even own a Zen based computer?

[ramrod]

Meh, it's not so critical any more nowadays. Years ago, it wasn't hard at all to overheat and fail a CPU, so monitoring temps was essential. Modern processors however, have effective thermal management, and will throttle performance to stay within temp envelope, or even shutdown altogether to prevent hardware damage. I too like to monitor my CPU temp, even though it's little more than trivia these days, but playing the 2nd hand citizen card seems extreme. Go have a beer and relax. Do you even own a Zen based computer?

It would have been very helpful a couple months ago for me when I was troubleshooting my newly built Ryzen PC which was shutting down unexpectedly every other day... I really hope AMD gets their shit together when it comes to stuff like this.

[lem79]

Aren't some of the motherboard sensor chips supported, so CPU temps can be monitored through them?

[Space Heater]

This is embarrassing at this point, no thermal driver and no vega support in mainline.

[boxie]

Use bios to control and monitor temps and fans. Disable corresponding kernel features. Stress your pc, reboot to bios and check temps from bios.

That certainly is the hard way of doing it - most of us use https://github.com/groeck/it87

[Adarion]

Oh well. It certainly isn't awesome for sure. But I wonder if there wasn't some discussion if maybe maybe some sensors are hooked up via onboard SIO / EC chips and those have often become a pain lately.
I mean, I wouldn't have expected that a termal sensor would involve a lot of coding, if it doesn't influence fans or other things directly but rather just gives out a temperature value (plus minus an offset maybe).

[DrYak]

I am not familiar with the internal of Linux. But I think it supports dynamically loaded modules. I don't understand why does the kernel have to include these drivers' complete source code. Can people just compile the latest GPU module as .ko and then load it into the kernel?

regarding the GPU : no you can't easily.
(you're at risk of getting error message regarding wrong version of kernel API when you try to compile).

The GPU is a nice example of why not.
A modern system like Linux is an extremely complex beast with a tons of component all in interplay.
regularily when you edit is small part somewhere it can have repercussion on all the parts that rely on it.

the graphical stack in Linux is in a constant evolution, there are constantly new elements and new capabilities added, meaning the way all the modules talk to each other evolves and the drivers must be adapted to that.

if a driver's source is part of the kernel tree, that means it's accessible to all the developers. if somebody change the way some thing are handled and it has an impact on GPU drivers, they can fix/update the GPU drivers, that's why all the drivers that are part of the kernel (e.g.: Radeon, Intel) always "just work" as the kernel updates - they are modified to follow the latest upgrades.

if a driver is out of tree, the developers responsible of the kernel that upgrade it can't upgrade the driver too. It's up to the devs of the driver to fix their code to follow the latest evolutions of the linux driver.
That why 3rd party proprietary drivers like Nvidia's are a bit problematic on "rolling" distros, and why there's a delay between a new linux kernel and when these driver finally get to work.

----

Now in the specific case of a CPU thermal device :
these are much simpler, there's a lot less thing interplaying with each other, chances are high that you'll be able to grab some beta code somewhere, sucessfully compile a .ko driver and load it into the kernel.

I think that since the last overhaul of the "i2c" driver and "hwmon" stacks, we haven't seen much modification impacting the way thermal works.

Also :
- some motherboard are able to monitor the CPU temperature themselves, and you could ask the motherboard through ACPI.
- some motherboard feature extra sensors, some of which could also measure the CPU temperature separately. You can either ask that sensors for the temperature directly, e.g. using an appropriate i2c driver. Or check if that sensors is available on one of the standard API of the motherboard (like ACPI).

[abi_dalzim]

While a Zen thermal driver isn't really a 'must' for most people in the sense that you need one to have a working system or be able to see cpu temps (a lot of motherboards have their own sensors which are supported) it is a 'must' in the sense that a thermal driver is a pretty simple piece of code for AMD to write and makes your product look much more mature.

And in b4 someone says 'if it's that simple write it yourself' buy be a few ryzen based computers (you know more than one for testing)... like 1 Ryzen 7 1700, one Threadripper 1950x and one dual cpu Epyc (with however the 32c/64t cpus are called).