[Jetson Orin AGX | CUDA 12.6] cuda-gdb causes SIGSEGV in libcudadebugger.so.1 when entering kernel

Developer Tools CUDA Developer Tools CUDA-GDB
1

Hello,

I’m encountering a segmentation fault when debugging even the simplest CUDA program using cuda-gdb on Jetson Orin AGX.
The program crashes immediately when entering the kernel function.

Reproducible code

#include <cuda_runtime.h>
#include <iostream>

__global__ void hello() {
    printf("Hello CUDA World !!\n");
}

int main() {
    hello<<<2, 4>>>();
    cudaDeviceSynchronize();
    return 0;
}

Behavior

When launching the program under cuda-gdb, execution stops inside
/usr/local/cuda/include/cuda_runtime.h at this line:

return ::cudaLaunchKernel((const void *)func, gridDim, blockDim, args, sharedMem, stream);

and then immediately receives a segmentation fault:

Thread 1 "00_hello_world" received signal SIGSEGV, Segmentation fault.
0x0000fffff46c168c in ?? () from /lib/aarch64-linux-gnu/libcudadebugger.so.1

This happens even without any breakpoints.
CPU-side code runs fine, but as soon as the kernel is launched, cuda-gdb crashes.

image
image1844×1075 71.5 KB

image
image1861×1099 339 KB

Environment

Hardware: Jetson Orin AGX 64GB Developer Kit
JetPack: 6.1.2 (L4T 36.4.7)
CUDA Toolkit: 12.6 (build V12.6.68)
cuda-gdb version:

NVIDIA (R) cuda-gdb 12.6
Based on GNU gdb 13.2
Configured as "aarch64-elf-linux-gnu"

nvcc build command:

nvcc -G -g -O0 hello_world.cu -o hello_world

launch.json (VS Code / Nsight VS Code Edition):

{
  "type": "cuda-gdb",
  "request": "launch",
  "program": "${fileDirname}/${fileBasenameNoExtension}",
  "cwd": "${fileDirname}",
  "args": [],
  "stopAtEntry": false
}

What I’ve confirmed

  • The same code and setup work correctly on x86_64 Ubuntu with CUDA 12.6 (no crash).

  • The crash only occurs on Jetson Orin (aarch64).

  • Happens regardless of breakpoints — just running under cuda-gdb triggers it.

Question

Is this a known issue with cuda-gdb or libcudadebugger.so.1 on JetPack 6.1.2 / CUDA 12.6 (aarch64)?
Are there any workarounds or specific flags required for debugging kernels on Jetson?

Thanks in advance for your help.

2

Hi @yuki.akimoto.sub
Thank you for reporting the issue! Could you please collect and share additional information to help us identify the issue?

  • Output of the groups command on target.
  • Do you see the same issue when running the debugging session as root?
  • Please try running the debugging session (as normal user) with additional logging enabled:
    • Add NVLOG_CONFIG_FILE variable pointing the nvlog.config file (attached
      nvlog.config (539 Bytes)). E.g.: NVLOG_CONFIG_FILE=${HOME}/nvlog.config
    • Run the debugging session.
    • You should see the /tmp/debugger.log file created - could you share it with us?

Also any chance you can try the same scenario with Jetpack 7?