Error in device opening /dev/apex_0 on kernel 6.6.28

[luchmedia]

Description

Upgraded my Rasbperry Pi 5 to kernel Linux raspberrypi 6.6.28+rpt-rpi-v8 #1 SMP PREEMPT Debian 1:6.6.28-1+rpt1 (2024-04-22) aarch64 GNU/Linux.
Trying to launch an edgetpu example in docker accessing /dev/apex_0 device causes problems.
Build gasket-dkms as in #808, worked fine with 6.5 kernel.

run container
sudo docker run -it --device /dev/apex_0:/dev/apex_0 coral /bin/bash

run test in container
python3 /usr/share/edgetpu/examples/classify_image.py --model /usr/share/edgetpu/examples/models/mobilenet_v2_1.0_224_inat_bird_quant_edgetpu.tflite --label /usr/share/edgetpu/examples/models/inat_bird_labels.txt --image /usr/share/edgetpu/examples/images/bird.bmp

crash

Traceback (most recent call last):
File "/usr/share/edgetpu/examples/classify_image.py", line 54, in
main()
File "/usr/share/edgetpu/examples/classify_image.py", line 44, in main
engine = ClassificationEngine(args.model)
File "/usr/lib/python3/dist-packages/edgetpu/classification/engine.py", line 48, in init
super().init(model_path)
File "/usr/lib/python3/dist-packages/edgetpu/basic/basic_engine.py", line 92, in init
self._engine = BasicEnginePythonWrapper.CreateFromFile(model_path)
RuntimeError: Error in device opening (/dev/apex_0)!

also tested with Frigate NVR and /dev/apex_0 can't be reached also if device is mapped.

version: "3"
services:
  frigate:
    container_name: frigate
    privileged: true
    restart: unless-stopped
    image: ghcr.io/blakeblackshear/frigate:stable
    shm_size: "256mb"
    devices:
      - /dev/apex_0:/dev/apex_0
      - /dev/video19:/dev/video19
    volumes:
      - /etc/localtime:/etc/localtime:ro
      - /home/pi/frigate-nvr/config.yml:/config/config.yml
      - /media/Hdd/recordings:/media/frigate
      - type: tmpfs
        target: /tmp/cache
        tmpfs:
          size: 1500000000
    ports:
      - "5000:5000"
      - "8554:8554"
      - "8555:8555/tcp"
      - "8555:8555/udp"

frigate.detectors.plugins.edgetpu_tfl ERROR : No EdgeTPU was detected

As you can see /dev/apex_0 is available and visible as pi user

lsmod | grep apex

apex 20480 0
gasket 102400 1 apex

lspci -nn | grep 089a

0000:01:00.0 System peripheral [0880]: Global Unichip Corp. Coral Edge TPU [1ac1:089a]

ls /dev/apex_0

/dev/apex_0

Click to expand!

Issue Type

Bug

Operating System

Linux

Coral Device

M.2 Accelerator with dual Edge TPU

Other Devices

No response

Programming Language

Python 3.9

Relevant Log Output

No response

[Stefan2483]

I have the same issue after upgrading to 6.6.28+rpt-rpi-v8
lspci -nn | grep 089a
0000:01:00.0 System peripheral [0880]: Global Unichip Corp. Coral Edge TPU [1ac1:089a]
ls /dev/apex*
/dev/apex_0
it is visible in docker after mounting it but frigate is not able to access it.
Python3.9

[TheNoobInventor]

I had the same issue but after hours of reading issues and forum posts I think I have found a fix for it.

I'm using this dual Nvme shield by Geekworm. I'm currently using only one slot for the Coral TPU pcie and booting from an SD card. But I will test booting from an SSD and update my post with any findings.

UPDATE: It still works when I boot from an SSD on one of the NVMe slots with the Coral TPU in the other one.

My kernel version is:
Linux pi5 6.6.31+rpt-rpi-v8 #1 SMP PREEMPT Debian 1:6.6.31-1+rpt1 (2024-05-29) aarch64 GNU/Linux

I followed the suggestion from jdb in his forum post and added this line to the /boot/firmware/config.txt file:

dtoverlay=pciex1-compat-pi5,no-mip

After rebooting the Pi5, I was able to run the example PyCoral container to test the TPU using this Pineboards blog post. Then I tested Frigate and it worked well.

I hope this helps someone out.