Current CPU temperature file detection is inaccurate

[Vdragon]

Currently temp-throttle uses the first found file in TEMPERATURE_FILES as CPU temperature, however it is pretty inaccurate as:

• temperature is not always CPU temperature(Motherboard etc.)
• CPU temperature is not always the most accurate one(per-core temperature)
• temperature file of the first CPU is not always starts from 1(don't know why)

Maybe we should put files that are known to be CPU temperature over generic ones(like from specific hardware monitor driver).

[Sepero]

Yeah, good idea. When I created it, I simply threw in a bunch of locations where the system temperature can be (possibly) accessed. Different systems seem to access temperatures in different locations. I have no idea what the criteria is for how the kernel determines what file location to use. I just assume it is a function of the hardware vendor. There may be more locations to access the temperature that I don't even know about. I don't know how to validate which ones are most likely to be CPU temperature.

[Sepero]

If you know how to validate which file locations are most likely to access the CPU temperature, then I'm willing to implement it. Sorry if that wasn't clear from my last post. I just need the information. haha :)

[Djfe]

Just found this script: amazing! thx for writing it

what I found out: for me cpu temperatures are located in /sys/class/hwmon/hwmon0/ hwmon0 is the only one that exists for me I'm using Fedora 21/22

The CPU temperatures are in temp2, temp3,temp4 and temp5 (4 cores) temp1 doesn't exist (I have no clue why)

you can find out what it is if you output name first of all (/sys/class/hwmon/hwmon0/name) it contains "coretemp" for me tempx_label outputs "Core x" for me x is a number here it's temp_2-5 and Core 0-3 for me

maybe labels allow a correct detection of the cpu temperatures but there is probably a better way it probably should also decide depending on the used OS

[Sepero]

Thanks for the report, Djfe. Please tell me this info about the hardware you are running:

• Computer Manufacturer
• CPU
• Architecture

Also what is:

• Your kernel version: uname -a
• The exact location of for reading the temperature of your first CPU core ("/sys/class/hwmon/hwmon0/hwmon0/temp2 ?")

[Djfe]

Yes it's temp2 like for @vdragon it doesn't start from 1 My notebook: http://www.toshiba.de/discontinued-products/satellite-a660-11m/

I'll post the kernel later And I'll zip the hwmon for you, so you can take a look yourself ;)

[Sepero]

This appears to be what I have so far :)

• Toshiba
• Intel i7
• 64bit
• Linux bau.nb 4.1.12-101.fc21.x86_64 #1 SMP
• /sys/class/hwmon/hwmon0/hwmon0/temp2

[Sepero]

Additional question- Are there ever times when the temperature switches to the file /sys/class/hwmon/hwmon0/hwmon0/temp1 or /sys/class/hwmon/hwmon0/hwmon0/temp0?

[Sepero]

Note: Support for /sys/class/hwmon/hwmon0/hwmon0/temp2 has been added to the "testing" branch.

[Djfe]

ups it's only once hwmon0 not twice /sys/class/hwmon/hwmon0/

the content: http://www.file-upload.net/download-11039390/hwmon0.zip.html (I didn't add the symlinks device and subsystem, because they contain symlinks back to hwmon0 and they are not important)

The notebook specs are correct and yes I never noticed being core 0 something else than temp2

[Djfe]

totally forgot: Linux bau.nb 4.1.12-101.fc21.x86_64 #1 SMP Wed Oct 28 15:18:44 UTC 2015 x86_64 x86_64 x86_64 GNU/Linux

[Sepero]

ups it's only once hwmon0 not twice

Ok, I corrected for this in the testing branch

[Sepero]

Until someone can discover a better method for discovering CPU temperatures, I think that takes care of everything for now.

[findepi]

Just to add. In my case /sys/class/hwmon/hwmon0/ contains a bunch of ACPI temperatures with /sys/class/hwmon/hwmon0/temp8_input being close to what I believe is CPU temperature as reported by sensors command. But, the CPU temperatures are in /sys/class/hwmon/hwmon1/temp[24]_input in my case (dual core i7). Oh,cat /sys/class/hwmon/hwmon1/name says "coretemp" while cat /sys/class/hwmon/hwmon0/name says "acpitz". Maybe this can be used to differentiate?

It seems, ability to specify custom temperature file may be handy (as in pull request #25). So may be ability to watch in many temperatures at once (like all cores.)

[Sepero]

My apologies for the long delay, findepi. I'm currently working on including that ability to specify a custom temperature location. You'll find it in the testing branch.

[Sepero]

Resolved in testing branch