Jetson-rdma-picoevb not working for PC build

Autonomous Machines Jetson & Embedded Systems Jetson AGX Xavier
1

I am trying to build and install the linux PC version of the jetson-rdma-picoevb driver GitHub - NVIDIA/jetson-rdma-picoevb: Minimal HW-based demo of GPUDirect RDMA on NVIDIA Jetson AGX Xavier running L4T on my linux PC with Rtx A4500 and Ubuntu 20.04.6 LTS. I followed the instructions to build it, but it says undefined symbols for nvidia_p2p* functions. I also tried to build the Nvidia driver 525.125.06 locally, which is the driver I’m running my graphics card, and link against it. It finished build, but when I did insmod it says “Unknown symbol nvidia_p2p*, module using GPL-only symbols uses symbols from proprietary module nvidia.”

How can I fix those errors to build and run it on my PC? Thanks!

2

I’m not faimilar with this project, but that seems to missing some CUDA libraries from that message.
Do you have CUDA installed on your host PC?

3

Yes I have installed CUDA on my host PC.
The README says to run ./build-for-pc-native.sh, but it got this error:

“ERROR: modpost: “nvidia_p2p_get_pages” undefined!”

I think it may be because those function names are missing in Module.symvers in /usr/src/linux.
How can I build it? Thanks!

4

I think those nvidia functions need to be built into /usr/src/linux/Module.symvers in order to register those symbols with the kernel. How can I do it? Do I need to use GitHub - NVIDIA/open-gpu-kernel-modules: NVIDIA Linux open GPU kernel module source? Thanks!

6

Hi,

RDMA is a hardware-dependent feature.
The sample is for the Jetson platform.

For desktop GPU, please check the below document:

Thanks.

8

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