4

I have a (very old) computer (with an ASUS A8N-SLI DELUXE motherboard and an AMD Opteron(tm) Processor 180) that I monitor with munin, and the Vcore Voltage as read by lm-sensors is always in a Critical state because the input volt is always below the min threshold. I never bothered much for years, but today I tried to fix that misreporting. However, after several hours of troubleshooting I was unable to address the issue because I cannot change the min/max thresholds that are read-only.

Here are my findings so far:

The Critical state comes because the input Vcore voltage value is outside the min/max values.

A sensors -u reports the following:

...
atk0110-acpi-0
Adapter: ACPI interface
Vcore Voltage:
  in0_input: 1.104
  in0_min: 1.450
  in0_max: 1.750
...

As you can see the in0_min is 1.45V but the in0_input is less than min 1.104V.

After reading the documentation of lm-sensors config file (https://linux.die.net/man/5/sensors.conf) I found out that I can correct the min/max values with some proper configuration. So I created a file /etc/sensors.d/atk0110.conf with the following contents:

chip "atk0110-*"

   label in0 "Vcore Voltage"

   set in0_min 1.05

Then I executed sensors -s (Read the section Set Statement in the man page for more info) in order for the settings to take effect and boom! I got the following error:

Error: File /etc/sensors.d/atk0110.conf, line 5: Failed to set value
atk0110-acpi-0: At least one "set" statement failed

After some Googling I found out that this error occurs because these values are hardcoded in the BIOS and cannot be changed. They are exposed through the sysfs filesystem to the kernel once the module asus_atk0110 is loaded, and the values are read from the path /sys/bus/acpi/drivers/ATK0110/ATK0110:00/hwmon/hwmon1/. In particular for in0_min the value is read from this file: /sys/bus/acpi/drivers/ATK0110/ATK0110:00/hwmon/hwmon1/in0_min.

My second attempt was to try to try the compute statement in sensors conf file but I soon found out that I cannot just compute the min/max. When one uses the compute statement in sensors.conf file all sub-features of an input are affected by the computation. From the sensors.conf man page:

A compute statement describes how a feature's raw value should be translated to a real-world value, and how a real-world value should be translated back to a raw value again. This is most useful for voltage sensors, because in general sensor chips have a limited range and voltages outside this range must be divided (using resistors) before they can be monitored...
...
...
A compute statement applies to all sub-features of the target feature for which it makes sense.
...

That means in my case, all three values, in0_input,in0_min,in0_max will be affected. So if I use something like this in the config:

chip "atk0110-*"

   label in0 "Vcore Voltage"

   compute in0 @-0.4,@

what I end up reading is this:

atk0110-acpi-0
Adapter: ACPI interface
Vcore Voltage:
  in0_input: 0.992
  in0_min: 1.050
  in0_max: 1.350

As you can see all values decreased by 0.4 volts, so again the input value is lower than min and the Critical munin status remains.

Is there any way to soft-tune ONLY the min/max values in lm-sensors if these values are read-only?

Another option would be to try to fix the issue from the munin side if possible, but the source of the problem in this case is lm-sensors, so I believe that's where the problem should be addressed.

Vangelis Tasoulas
  • 1,637
  • 1
  • 16
  • 16

3 Answers3

3

This answer solves the issue at the munin side by changing the warning/critical thresholds that munin uses. So although my problem is solved, I will not mark the answer as the solution. Ideally the problem should be solved around lm-sensors.

Go to the node that uses the sensors_volt plugin and execute the command munin-run sensors_volt config. The command will return a key/value list that is recognized by munin:

# munin-run sensors_volt config
graph_title Voltages
graph_vlabel Volt
graph_args --base 1000 --logarithmic
graph_category sensors
volt1.label Vcore Voltage
volt1.warning 1.51:1.69
volt1.critical 1.45:1.75
volt2.label  +3.3 Voltage
volt2.warning 3.12:3.48
volt2.critical 3.00:3.60
volt3.label  +5.0 Voltage
volt3.warning 4.70:5.30
volt3.critical 4.50:5.50
volt4.label +12.0 Voltage
volt4.warning 11.60:12.80
volt4.critical 11.20:13.20

These key/values can be altered in the munin.conf at your master node.

In my case I want to change the volt1.warning and volt1.critical values so in the section where the host is defined I added the following lines:

[host.localdomain]
    address 1.2.3.4
    use_node_name yes
    sensors_volt.volt1.warning 1.05:1.40
    sensors_volt.volt1.critical 1.00:1.45

So now the min/max warning values have been set to 1.05V/1.40V and the critical values to 1V/1.45V.

Vangelis Tasoulas
  • 1,637
  • 1
  • 16
  • 16
0

In case you do not over- or undervolt your system, you might condsider ignoring the Vcore:

chip "atk0110-*"
 ignore in0
Oliver R.
  • 131
  • 4
-1

You have to be root to do the set command so you need to use sudo sensors -s and then it should change the limits. When you run sensors again the limits should be now as you set them.

sebasth
  • 14,332
  • 4
  • 50
  • 68
BonnieTwin
  • 1
  • 1