You are not logged in.

#1 2020-02-02 16:33:09

Alec76
Member
Registered: 2016-12-28
Posts: 9

lm-sensors fails with LTS kernel

I'm having no trouble monitoring my hardware with the Linux kernel.
However, when I use the Linux-lts kernel sensors-detect fails.

I'm using a Threadripper 3960x and a ASRock TRX40 creator motherboard.
I'm presuming the reason that Linux works and Linux-lts doesn't relates to the newness of the hardware. But I wonder if it isn't possible to add a module to the lts kernel as workaround until the lts kernel is updated? Or if there is any other possible workaround? Or maybe my presumption is wrong?

Here's the error from sensors-detect:

Driver `k10temp':
  * Chip `AMD Family 17h thermal sensors' (confidence: 9)

Do you want to overwrite /etc/conf.d/lm_sensors? (YES/no): y
Job for lm_sensors.service failed because the control process exited with error code.
See "systemctl status lm_sensors.service" and "journalctl -xe" for details.
Unloading cpuid... OK

Here's the output of systemctl status lm_sensors.service:

lm_sensors.service - Initialize hardware monitoring sensors
     Loaded: loaded (/usr/lib/systemd/system/lm_sensors.service; enabled; vendor preset: disabled)
     Active: failed (Result: exit-code) since Sun 2020-02-02 11:11:35 EST; 38s ago
    Process: 14457 ExecStart=/sbin/modprobe -qab $BUS_MODULES $HWMON_MODULES (code=exited, status=0/SUCCESS)
    Process: 14458 ExecStart=/usr/bin/sensors -s (code=exited, status=1/FAILURE)
   Main PID: 14458 (code=exited, status=1/FAILURE)

Feb 02 11:11:35 ebonizer systemd[1]: Starting Initialize hardware monitoring sensors...
Feb 02 11:11:35 ebonizer sensors[14458]: No sensors found!
Feb 02 11:11:35 ebonizer sensors[14458]: Make sure you loaded all the kernel drivers you need.
Feb 02 11:11:35 ebonizer sensors[14458]: Try sensors-detect to find out which these are.
Feb 02 11:11:35 ebonizer systemd[1]: lm_sensors.service: Main process exited, code=exited, status=1/FAILURE
Feb 02 11:11:35 ebonizer systemd[1]: lm_sensors.service: Failed with result 'exit-code'.
Feb 02 11:11:35 ebonizer systemd[1]: Failed to start Initialize hardware monitoring sensors.

Offline

#2 2020-02-02 19:28:35

Xabre
Member
From: Serbia
Registered: 2009-03-19
Posts: 757

Re: lm-sensors fails with LTS kernel

Probably should use newer kernel, since new Threadripper 2 and supporting kernel code were released only recently.

Offline

#3 2020-02-02 20:15:07

Alec76
Member
Registered: 2016-12-28
Posts: 9

Re: lm-sensors fails with LTS kernel

Good point.
The only thing is I'm experiencing problems with the newer kernel (I think related to this bug )
So if possible I'd like to use LTS, and since that cpu is a particularly hot potato, I'd love to have access to lm_sensors if there's some way to add it.

Offline

#4 2020-02-02 20:27:18

loqs
Member
Registered: 2014-03-06
Posts: 18,928

Re: lm-sensors fails with LTS kernel

linux-lts is going to move to 5.4 once linux has moved to 5.5.  Is the issue also present under 5.5?

Offline

#5 2020-02-02 21:54:16

Alec76
Member
Registered: 2016-12-28
Posts: 9

Re: lm-sensors fails with LTS kernel

I'm currently running 5.4.15.

Offline

#6 2020-02-02 22:30:56

loqs
Member
Registered: 2014-03-06
Posts: 18,928

Re: lm-sensors fails with LTS kernel

If you do not use any out of tree modules so do not need linux-headers the following updates linux to 5.5.1

# pacman -U https://mirrors.edge.kernel.org/archlinux/testing/os/x86_64/linux-5.5.1.arch1-1-x86_64.pkg.tar.zst

I assume you have linux-lts installed as well so you have a reserve kernel to boot into.

Last edited by loqs (2020-02-02 22:31:14)

Offline

#7 2020-02-03 12:36:36

Alec76
Member
Registered: 2016-12-28
Posts: 9

Re: lm-sensors fails with LTS kernel

Thanks!

I got Linux 5.5.1 loaded, but it doesn't look like my Nvidia package works with it as startx fails.

At this point, I think I will just wait for the update.

Offline

#8 2020-02-03 12:45:56

loqs
Member
Registered: 2014-03-06
Posts: 18,928

Re: lm-sensors fails with LTS kernel

The nvidia package is an out of tree module.

# pacman -U https://mirrors.edge.kernel.org/archlinux/testing/os/x86_64/nvidia-440.48.02-1-x86_64.pkg.tar.zst

Offline

#9 2020-02-03 13:45:33

Alec76
Member
Registered: 2016-12-28
Posts: 9

Re: lm-sensors fails with LTS kernel

Thanks again. I was able to get 5.5.1 running. Unfortunately, my issue persisted with 5.5.1.

I downgraded to 5.4.1 (and this time downgraded Nvidia accordingly--thanks for that), and now everything is working to my liking. But I'm still looking for a long-term solution.

My problem has to do with freezing while starting GPU rendering in Blender, but I will open a separate topic to deal with that if future kernel/Blender updates don't fix it.

Offline

#10 2020-02-03 20:28:37

Ropid
Member
Registered: 2015-03-09
Posts: 1,069

Re: lm-sensors fails with LTS kernel

In that status output for your "lm_sensors.service", it says that the "sensors -s" command has failed. According to the man-page, that command makes the sensors tool apply the "set" lines in its config files. Maybe see if you can find out what's going wrong there.

If I look here in the "/etc/sensors3.conf" config file, then the "set" commands in that file all look boring. I'm guessing all the "set" statements are just configuring the min and max range for voltage readings. I think that kind of stuff doesn't actually matter and is just to configure the "alarm" feature?

The config file has sections for the different chips that lm_sensors supports. To find the section for your board, look inside the "/etc/conf.d/lm_sensors" file that sensors-detect has written.

I'd then try commenting out the "set" statements in the section for your board and see what happens. You could also experiment with the values that are used in the set statements. Maybe there's a change you can do to make the broken "set" run without error. You could then put that into your own file in "/etc/sensors.d/" so that you can leave the main "/etc/sensors3.conf" file unchanged.


EDIT:

I think I might have misunderstood what's going on for you. That error message that sensors prints in the service status output, that means it couldn't find a driver/chip for your board? If that's the case, does that actually matter? Doesn't the "sensors" command still show the temperature from the "k10temp" driver? Isn't that enough to get the CPU temperature?

If that's what's happening, I guess you should just disable the lm_sensors service.

Last edited by Ropid (2020-02-03 20:35:30)

Offline

Board footer

Powered by FluxBB