AGX ORIN使用cuda coredump方案无法生成.nvcudmp文件

Autonomous Machines Jetson & Embedded Systems Jetson AGX Orin
, ,
1

JETPACK版本：5.1.1
硬件版本：AGX ORIN
故障现象：
部署了获取gpu coredump信息的debug方案，但是当出现nvgpu错误的时候，系统没有生成nvcudmp文件
复现步骤：

  1. sudo usermod -aG debug $USER
  2. ulimit -c unlimited
  3. export CUDA_ENABLE_COREDUMP_ON_EXCEPTION=1
  4. 进入到/usr/local/cuda/samples/0_Simple/目录下
  5. 将附件：cudaOpenMP.zip复制到该目录下并解压到当前目录
  6. 解压后进入到/usr/local/cuda/samples/0_Simple/cudaOpenMP.目录
  7. 执行sudo ./cudaOpenMP，此时通过dmesg -w可以观察到内核已经上报nvgpu相关错误，见附件dmesg.log
  8. 使用sudo find / -name .nvcu，发现系统中并没有生成该文件
2

cudaOpenMP.zip (336.8 KB)
dmesg.log (111.7 KB)

4

Hi,

We will try if we can see similar behavior first.
Will get back to you later.

Thanks.

5

Hi,

Based on your command, the global variable is set to the $USER account but the app is run with root.
Could you copy the app to $HOME and run it without sudo?

Thanks.

6

谢谢，调整用户后，目前可以正常生成.nvcudmp文件

8

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.