I’ve observed a few instances of complete device resets, potentially kernel panics, when using the desktop test image. This coincided with using the chromium-browser navigation of the khadas github page.
Please let me know if you would like any specific logs, or files if it happens again
If your device is connected to the internet, and you notice any sorts of glitching or kernel panics,
try running dmesg | nc termbin.com 9999 on the command, it will provide a link which we can use to read the logs.
Also,
you can see the application logs when you launch the application from the command line instead of clicking the desktop icon and running, you can share those as well, as logs, screenshots or pictures.
It happened again, when exercising the NPU
As requested I have triggered the logs upload https://termbin.com/z19u
This is happening several times a day, so easily reproducible
Was this done just after noticing the issue or after rebooting ?
If the device reboots after the issue, then the provided method of using the command will not work for capturing logs, it requires the use of the serial debugging tool.
If you have text files you can share them with pastebin, for other files any other file sharing service can work like mediafire, Google drive public link etc.
Generally I can get it to reboot within 15 minutes of light usage.
It most reliably crashes when running a program almost carbon copied from the face_recognition demo npu demo.
So I would expect you to be able to reproduce using the face_recognition_cap demo
I have narrowed this down further, I believe the problem related to network activity,
Exercising the npu alone doesn’t reproduce the issue
Steps that appear to make it quickly reproducible
(1) Fresh install of the Ubuntu 5.15 Desktop Image
(2) apt update and apt upgrade
(3) apt install chromium-browser
(4) download the vim4_npu_application demos from github
(5) build the face_recognition demo
(6) script repeated runs of the demo, such as
#!/bin/bash
counter=1
while [ $counter -le 1000000000 ]
do
echo $counter
sudo ./face_recognition -M …/data/model/retinaface_int8.adla -m …/data/model/facenet_int8.adla -p …/data/img/lin_1.jpg
((counter++))
done
Yes, we have reproduce this issue on our side, we are working on it, but still have no solution at this moment yet, we will update you when we make progress.