deviceQuery succeed in host, but failed in docker

Hi Guys,
I installed cuda on xavier platfrom(host), and run “deviceQuery” succesfully with returning PASS.
but in my docker(docker image in host), it’s return:
cudaGetDeviceCount returned 38
-> no CUDA-capable device is detected
Result = FAIL

both docker img and host run “tegrastatus”, return normally.
nvidia@in_dev_docker:/usr/local/cuda/samples/1_Utilities/deviceQuery$ /apollo/tegrastats
RAM 2796/15690MB (lfb 7x2MB) SWAP 720/7845MB (cached 36MB) CPU [13%@1190,0%@1190,0%@1190,0%@1190,0%@1190,0%@1190,0%@1190,0%@1190] EMC_FREQ 0% GR3D_FREQ 0% AO@40.5C GPU@40C Tboard@41C Tdiode@43.5C AUX@40.5C CPU@40.5C thermal@40.2C PMIC@100C GPU 155/155 CPU 465/465 SOC 931/931 CV 0/0 VDDRQ 0/0 SYS5V 1620/1620

docker mapping directory from host lists here,and also run ldconfig to config the path:
/usr/local/cuda-10.0/
/usr/lib/aarch64-linux-gnu/tegra
/usr/lib/aarch64-linux-gnu/tegra-egl

“/usr/lib/aarch64-linux-gnu” is not mapping, use docker its own, but did some library linking, as below:
/usr/lib/aarch64-linux-gnu/libdrm_nvdc.so -> tegra/libdrm.so.2
/usr/lib/aarch64-linux-gnu/libcuda.so -> tegra/libcuda.so
/usr/lib/aarch64-linux-gnu/libdrm_tegra.so -> libdrm_tegra.so.0.0.0(copy from host)

Do I miss something? host can run devicequery, it says the cuda driver is correct, but, not clear failed in docker.

Dear asnail,

Could you please get some hints from https://devtalk.nvidia.com/default/topic/1030912/faq/installing-driveworks-within-a-docker-image/?
Unfortunately, we do not officially support docker.

update:
sudo ./deviceQuery PASSED!
./deviceQuery always Failed!

chmod 777 /dev/nvidia* not works!

what’s the problem?

Dear asnail,

It seems like a permission issue.
Could you try to give sudo permission with below? Thanks.
sudo chmod 777

Hi Steve
Thanks,
I tried sudo, forgot put it on, it’s not working, but I found sudo chmod 777 /dev/*, it works!
I am troubleshooting this issue which node caused the permission issue.

Hi asnail,

Any result can be shared?

issue resolved with sudo chmod 666 /dev/nv*
default permission is “-660 root:video” afer installed.