1

I currently have the trouble of some hardware overheating regularly. Looking through the Linux kernel, I found that the thermal sensors are currently fully supported, however the preconfigured limits seem much too high. I am currently looking to modify the kernel module, but I am not sure what would be the best way to modify it. I would like the following:

  • Have a file somewhere, where I can read the current temperature
  • Have a couple of files I can use from userspace to modify the current limits

Now I am not sure what virtual file interface I should use for this. So far I see the following options:

  • Expose all data via the /dev/ filesystem, use ioctl() to modify limits
  • Add the files in sysfs
  • Add the files in debugfs

I am not sure what the pros and cons of each of these methods would be. So far I have looked at sysfs, and it seems that this is not the right way to go. The number of sensors is highly hardware dependent, and I would have to add various hacks to get one file per sensor (or one directory per sensor), possibly adding another kobject for each sensor so I can add files in the resulting directory. Previously the sensor data seemed to have been available through /sys/class/hwmon/hwmon0/device/temp1_input, however it was removed there.

debugfs also does not seem like the right way to go, since I am not so much interested in kernel debugging, but rather providing a nicer way to configure and read the sensors.

devfs however seems like a lot of overkill for such a simple setup.

What would be the best way to expose the sensor data and make the limits configurable through user space?

LiKao
  • 10,408
  • 6
  • 53
  • 91
  • The question seems Too broad/Primary opinion-based for me... There is no well-formed border even between `sysfs`/`debugfs`. As for `/dev`, it is dependent, whether *device* may represent the sensor, which is very subjective. As for me, `sysfs` is the most suitable place for expose limit of the sensor. – Tsyvarev Nov 23 '15 at 16:54
  • @Tsyvarev: Yeah, I wasn't sure if it might not be too broad or elicit too much discussion. I was hoping for some general rules, but as you say I am not even sure about the clear line between `sysfs` and `debugfs`. I was not sure if sysfs is the right place because of http://stackoverflow.com/questions/19529018/how-does-linux-kernel-creates-sysfs/19530495#19530495 and http://stackoverflow.com/a/19671817/663957. – LiKao Nov 23 '15 at 20:03
  • Questions you refer to don't disallow such things under `/sys`, they just recommend to use device representation for them. (And this device is something another essence then one under `/dev`). – Tsyvarev Nov 23 '15 at 20:48

2 Answers2

2

According to Documentation/hwmon/sysfs-interface, temp[1-*]_max and temp[1-*]_crit are already defined for that pupose and should be implemented read-write when possible. If that is not already the case for your driver, then that is what you want to implement.

Alexandre Belloni
  • 2,244
  • 13
  • 12
  • Thanks for the tip, that really looks like the best place to put that data. As far as I can see for this I have to somehow register the device with the hwmon driver, is that correct? – LiKao Nov 25 '15 at 21:45
  • That would be preferable but it is not strictly necessary. – Alexandre Belloni Nov 26 '15 at 10:30
1

No need to expose the file system for the temperature sensors. Register your temperature sensor as thermal zone using thermal framework and you give get /sysfs interface. which can be used from user-space. Go through the below link to know how to access it from user-space.

https://www.lookup2learn.com/post/thermal-management-in-linux

asking questions
  • 66
  • 6