Updated to 5.15, fancontrol only able to control first fan

[teythoon]

After upgrading to the 5.15 kernel, I noticed that the fan control is only able to control one of the fans, while the other spins at full speed.  Experimentally I found out that the PWM control for the second fan is ineffective:

 

helios4# uname -r
5.15.86-mvebu
helios4# echo 0 > /dev/fan-j10/pwm1 # works, fan is off
helios4# echo 255 > /dev/fan-j10/pwm1 # works, fan at full speed
helios4# echo 0 > /dev/fan-j17/pwm1 # does nothing, full speed
helios4# echo 255 > /dev/fan-j17/pwm1 # does nothing, full speed

 

Some digging around I found that 92-mvebu-gpio-remove-hardcoded-timer-assignment.patch has been ported to 5.15, but it doesn't seem to work for me.

[Heisath]

Ok I will try to reproduce and look at the patch. Does anyone else experience the same? 5.15 has been out for a while.

[Heisath]

Hi,

 

I just tested this on my helios4 with linux kernel 5.15.80 

uname -a:
Linux helios4 5.15.80-mvebu #22.11.1

 

and both pwm controllers work just fine.

 

Have you tried going back to make sure it is not a HW problem? Or tried swapping fans around? 

 

Also make sure your /dev devices symlink to the correct path:
 

/dev/fan-j10 -> /sys/devices/platform/j10-pwm/hwmon/hwmon3
/dev/fan-j17 -> /sys/devices/platform/j17-pwm/hwmon/hwmon4

 

EDIT:

I just did a fresh build from armbian master, 23.02.0-trunk. PWM on both channels working fine.

uname -a
Linux helios4 5.15.88-mvebu #trunk SMP Sat Jan 14 13:47:38 CET 2023 armv7l GNU/Linux

 

So you'd need to provide more info on how you built your kernel version to get to this problem.

 

[lexming]

I have the same issue as OP. Only the fan connected to connector J17 changes speed according to PWM settings in kernel. Fan at J10 is always at full speed.

 

On my side this has happened for a long time, at least for 6 months. So I can say that it's not related to 5.15.86-mvebu. Now I am on 5.15.80-mvebu (current linux-image-current-mvebu in bullseye). I swapped and replaced fans to verify a hardware issue on the affected fan as well, but in all cases the fan connected to J10 spins at full speed. If there is a hardware error, it must be on the connector / board.

 

On system's side all seems fine though, setting the fan speed through `/dev/fan-j10/pwm1` works without error and the value sticks.

[teythoon]

Thanks for looking into this, and sorry for the late response.

 

> I just tested this on my helios4 with linux kernel 5.15.80 and both pwm controllers work just fine.

 

Odd.

 

> Have you tried going back to make sure it is not a HW problem? Or tried swapping fans around? 

>

> Also make sure your /dev devices symlink to the correct path:

 

Yes, I went back to the kernel provided by the package linux-image-legacy-mvebu which AFAICS is roughly the same version as was used with Buster-based Armbian.  Going back immediately fixed the problem.  The symlinks look fine.

 

helios4# uname -r
4.19.167-mvebu
helios4# ls -l /dev/fan*
lrwxrwxrwx 1 root root 42 Jan  8 16:13 /dev/fan-j10 -> /sys/devices/platform/j10-pwm/hwmon/hwmon3
lrwxrwxrwx 1 root root 42 Jan  8 16:13 /dev/fan-j17 -> /sys/devices/platform/j17-pwm/hwmon/hwmon4

 

> So you'd need to provide more info on how you built your kernel version to get to this problem.

 

First I installed the kernel from the stock Armbian package linux-image-current-mvebu.  Then, after noticing the problem, I built the kernel images from Armbian master to make sure the problem still persists.  And it did.  I did not change the configuration or add patches during my build, I mostly tried to see whether the patch was indeed in.

[teythoon]

I found a Product Change Notification in the Kobol wiki that may be of interest: https://wiki.kobol.io/helios4/files/pcn/Helios4_PCN_201906060001.pdf

 

> All Helios4 rev 1.2 and further will be using the new ARMADA 388 Rev-2.1 SoM.

 

I haven't looked at the board, but I bought my Helios in 2018, so I'm pretty sure that it uses the ARMADA 388 Rev-2.0 SOM.

[lexming]

My board is an Helios4 rev 1.1. If only early revisions of the board are affected, that could explain why not everybody is experiencing this issue.