Energy Monitoring

[chan]

There's a default application to monitor energy for XU3, and its sources are on https://github.com/hardkernel/EnergyMonitor. I needed to log monitoring data like XU+E script.
I wrote a simple c++ program using the given source. It shows data every 5 second in default.

Compile:
g++ main.cpp getnode.cpp -o sensors -std=c++11
Execute:
sudo sensors or sudo sensors 2
In /etc/rc.local
/usr/local/bin/sensors >> /var/log/sensors.log 2>&1

I hope this could be useful to someone.

[odroid]

Thank for the great energy logger!

[jmawer]

Hi
I believe there is a minor bug in git version of EnergyMonitor. When you read the CPU frequency on you only read the first 3 characters from the file and then make assumptions about the rest. This means that if the clock frequency is less than 1GHz displayed frequency is wrong. For example if I set frequency of CPU7 to 200MHz (by setting scaling_max_freq = 200,000) the Energy monitor still displays 2000MHz, it was a bit worrying when I set it to 900MHz and it said 9000MHz!

As an aside is there a reason that EnergyMonitor uses cpuinfo_cur_freq for current frequency (default only readable) rather than scaling_cur_freq? This means by default standard system users can't see the operating frequency of the CPUs is scaling_cur_freq invalid at some point e.g. if some feature is enabled?

Cheers

John

[odroid]

I don't like the 9000Mhz either 9Ghz..
We will fix the wrong 3 digits issue soon.

But I don't understand the difference between cpuinfo_cur_freq and scaling_cur_freq clearly.
Please help me to make sense.

[jmawer]

By default (as shipped)
/sys/devices/system/cpu/cpu<0-7>/cpufreq cpuinfo_cur_freq/cpuinfo_cur_freq has permissions 400 so only root can read
/sys/devices/system/cpu/cpu<0-7>/cpufreq cpuinfo_cur_freq/scaling_cur_freq has permissions 444 so anyone can read it
Energy monitor uses cpuinfo_cur_freq so if you are not root you can not see values for frequency. As far as I know the values should always be the same so it would make more sense to use the one which defaults to use world read but I'm not sure. Not a big deal just a thought gives a better user experience out of the box.

Thanks
John

[odroid]

Okay... make sense.

[odroid]

I don't like the 9000Mhz either 9Ghz..
We will fix the wrong 3 digits issue soon.

But I don't understand the difference between cpuinfo_cur_freq and scaling_cur_freq clearly.
Please help me to make sense.

The 3 digits assinging is related to only GPU. So it seems not to be an issue.

[jmawer]

Sorry about that I was running the one I got on system install and reading the code in git as I use Qt5 and did't want to install Qt4 and wasn't reading carefully enough. mdjr actually fixed it on his August14 commit by reading 8 characters instead of 5 from the file. I installed Qt4-dev built and all is fine

[seungju.son]

At first, thank you for sharing your logging code.
I downloaded and compiled the source but it did't work.

The message said

"No sensors found!
Make sure you loaded all the kernel drivers you need.
Try sensors-detect to find out which these are."

Then I tried the command 'sensors-detect', but no sensors were detected.
I also tried installing lm-sensors.

What should I do for next?

I am using ODROID XU3, by the way.

[odroid]

You need to do "modprobe" a driver module of INA231 energy sensor.
I will let you know the module name tomorrow because it's already 10:00 PM in Korea.

[odroid]

This!

Code: Select all

modprobe ina231-sensor

[seungju.son]

Thanks, but it doesn't work.

The message is like

modprobe: FATAL: Module ina231-sensor not found.

I tried

modprobe ina231-sensors
modprobe i2c-isa

as well, but still not working.

[seungju.son]

There are some directories which seems like i2c and INA231 drivers are installed.

Code: Select all

odroid@odroid:/sys/bus/i2c/drivers/INA231/3-0041/driver$ ls -al
total 0
drwxr-xr-x  2 root root    0 Feb 24 15:53 .
drwxr-xr-x 10 root root    0 Feb 24 15:53 ..
lrwxrwxrwx  1 root root    0 Feb 24 16:20 3-0041 -> ../../../../devices/12c60000.i2c/i2c-3/3-0041
lrwxrwxrwx  1 root root    0 Feb 24 16:20 3-0044 -> ../../../../devices/12c60000.i2c/i2c-3/3-0044
lrwxrwxrwx  1 root root    0 Feb 24 16:20 3-0045 -> ../../../../devices/12c60000.i2c/i2c-3/3-0045
--w-------  1 root root 4096 Feb 24 16:20 bind
--w-------  1 root root 4096 Feb 24 16:20 uevent
--w-------  1 root root 4096 Feb 24 16:20 unbind

but with lsmod command

Code: Select all

odroid@odroid:/sys/bus/i2c/drivers/INA231/3-0041/driver$ lsmod
Module                  Size  Used by
8192cu                531585  0 
hid_apple               5674  0 
rfcomm                 33046  0 
bnep                   10854  2 
bluetooth             228012  10 bnep,rfcomm
rfkill                 12820  3 bluetooth
fuse                   74749  2 
nls_cp437               5128  1 
vfat                    9563  1 
fat                    54008  1 vfat

nothing.

[odroid]

Which OS image do you use?
Let me know the file name.

[seungju.son]

I use Linux 3.10.63+

[odroid]

Did you installed this image?
ubuntu-14.04.1lts-lubuntu-odroid-xu3-20150212.img

[seungju.son]

I don't remember which file exactly I use but the Kernel is earlier than 20150212 version.

I installed the OS on an eMMC card and boot with it.

[erik]

I also have trouble running it. I get the following error: /dev/sensor_armOpen FailOpenINA231
It used to work. Maybe it has something to do that I yesterday installed suggested updates?

[odroid]

@erik
Do you use this image?
ubuntu-14.04.1lts-lubuntu-odroid-xu3-20150212.img

If you use it, you should find the "ina231-sensor" in the result of "lsmod".

Anyway, follow this instruction and let me know the result.

1. add the ina231-sensor in the blacklist (/etc/modprobe.d/blacklist )
2. after rebooting, check the ina231-sensor from the "lsmod" command.
3. modprobe ina231-sensor and run the monitor program.

[seungju.son]

I tried your instructions, but it didn't work.


when I do "lsmod", the result is as follows.

Code: Select all

odroid@odroid:~$ lsmod
Module                  Size  Used by
8192cu                531585  0 
rfcomm                 33046  0 
bnep                   10854  2 
bluetooth             228012  10 bnep,rfcomm
rfkill                 12820  3 bluetooth
fuse                   74749  2 
nls_cp437               5128  1 
vfat                    9563  1 
fat                    54008  1 vfat

Do you think the previous version of kernel causes the problem?

[seungju.son]

omg
There is no 'ina231-sensor' in the result of 'lsmod', but the sensor logging binary works!!!

I don't understand how it was fixed.

[rooted]

The module must be compiled into the kernel.

[seungju.son]

Now I know what was wrong.
(I was stupid... all my faults)

I had been doing

Code: Select all

sudo sensors

instead of

Code: Select all

sudo ./sensors

There is generally used command 'sensors' , and the binary name is same with the command.

but I still have question.

when logging, A7_W and MEM_W data seems like converted.

Those logging data is different with monitoring data(display version).

[erik]

I use the the preinstalled Ubuntu and then I have been installing the suggested updates regularly. Now I have: Ubuntu 14.04.2 LTS (GNU/Linux 3.10.54+ armv7l)

I don't know why I got the error. After restarting the device I no longer get the error. However, lsmod shows the following
Module Size Used by
rfcomm 32722 0
bnep 10882 2
joydev 9066 0
bluetooth 226463 10 bnep,rfcomm
rfkill 12917 2 bluetooth
fuse 75320 2
nls_cp437 5132 1
vfat 9526 1
fat 54089 1 vfat


Thanks for your help!

[odroid]

The latest kernel should be 3.10.69 if you ran Kernel update via ODROID-Utility.

Code: Select all

#uname –a 
Linux odroid 3.10.69 #1 SMP PREEMPT Thu Feb 12 15:22:14 BRST 2015 armv7l armv7l armv7l GNU/Linux

[erik]

OK. I see. Now, I updated it using ODROID-Utility. Before, I only updated the packages suggest by Ubuntu. Thanks!

[seungju.son]

I have another question about power sensor frequency.

I am using the logging code and modified time interval by using usleep instead of sleep.

It seems like the power sensor value is updated 4 time in a second(4Hz), and the unit is Watt.

Is the logged value representing power in 250ms? or 1sec?

I wonder if I get mean value of 4 logged value for real power value in one second.

[saigomathi]

Hi odroid i am using odroid xu4 SCB. is it possible to allocate workloads to per-core basis and also is it possible to measure energy after each task execution?

[odroid]

XU4 has no energy sensor on the board.

[raditya160053]

I'm getting the following error message on running "sudo ./sensor"

Code: Select all

Time|CPU0_Freq(MHz)|CPU1_Freq|CPU2_Freq|CPU3_Freq|CPU4_Freq|CPU5_Freq|CPU6_Freq|CPU7_Freq|GPU_Freq|CPU0_Temp(C)|CPU1_Temp|CPU2_Temp|CPU3_Temp|GPU_Temp|A7_W|A15W|MEM_W|GPU_W|Watt_Sum
This loops forever until press <Ctrl-C>!
terminate called after throwing an instance of 'std::logic_error'
  what():  basic_string::_M_construct null not valid
[1]    1939 abort      sudo ./sensors

Does anybody know why this might be happening? I'm using a xu3 board with ubuntu 18.04.

[odroid]

Check your dmesg output if your board was detected as XU3 or XU4?

[raditya160053]

Check your dmesg output if your board was detected as XU3 or XU4?

dmesg output:
OF: fdt: Machine model: Hardkernel Odroid XU3

[odroid]

What is your current kernel version and Ubuntu version?
Could you run "sudo modprobe ina231-sensor"?

[raditya160053]

uname -a:
Linux odroid 4.14.85-152 #1 SMP PREEMPT Mon Dec 3 03:00:02 -02 2018 armv7l armv7l armv7l GNU/Linux
Ubuntu 18.04
I did run the modprobe command, threw no errors and no output.

This is on a micro-sd card, When I use the emmc that comes with the board the above code works just fine.

[raditya160053]

Check your dmesg output if your board was detected as XU3 or XU4?

I installed kernel 4.14 on an eMMC and tried it on XU3 and it's getting recognized as XU4

dmesg output:

[ 0.000000] OF: fdt: Machine model: Hardkernel Odroid XU4

Is there a way to fix this? I'm trying to use the sensors on XU3, but if it gets recognized as XU4, I can't do that.

Thanks

[mad_ady]

You can change boot.ini and force loading of the xu3 dtb, instead of relying on autodetection.