Can't run cuda samples in jetson xavier nx

I build a simple debian rootfs and used below ways to support the cuda drivers in my rootfs under L4T 32.5:

  • Copy my rootfs to Linux_for_Tegra/rootfs/
  • Use to install nvidia releated packages(–target-overlay)
  • Install cuda-repo-l4t-10-2-local-10.2.89_1.0-1_arm64.deb manually(download by sdkmanage)
  • build the cuda samples in /usr/loacal/cuda/samples

When I running the samples, I find below errors:

# ./asyncAPI 
[./asyncAPI] - Starting...
CUDA error at ../../common/inc/helper_cuda.h:775 code=100(cudaErrorNoDevice) "cudaGetDeviceCount(&device_count)"


It seems that some issue from your rootfs update.
We try the sample with r32.5, and it can work correctly.

$ ./asyncAPI
[./asyncAPI] - Starting...
GPU Device 0: "Xavier" with compute capability 7.2

CUDA device [Xavier]
time spent executing by the GPU: 8.84
time spent by CPU in CUDA calls: 0.14
CPU executed 31665 iterations while waiting for GPU to finish

Could you share the details on how do you update the rootfs?