What is nvgpu.ko file?

Robotics & Edge Computing Jetson Systems Jetson Orin NX
1

Hello,

I have two questions about nvgpu.ko file

First, According to the build kernel of kernel customization in the Jetson Linux Developer Guide, it is guided to replace nvgpu.ko. What changes if I replace it?

link : Kernel Customization — Jetson Linux Developer Guide documentation

Is it similar in function to installing the L4T 32.7.x version of the module on the rootfs after building, as demonstrated in the following command?

$ sudo make ARCH=arm64 O=$TEGRA_KERNEL_OUT modules_install
INSTALL_MOD_PATH=/Linux_for_Tegra/rootfs/

Second, Why is there a significant difference in size between the nvgpu.ko file generated after running apply_binary.sh on the sample rootfs provided by Jetson Linux 35.4.1 and the nvgpu.ko file generated after building the default kernel source?

2

For your first question, yes, it is similar thing as what you did on rel-32.7.x. You can forget about just nvgpu.ko and install all of the ko files.

For your 2nd issue, it has answer in rel-32 document too. The built out file didn’t get strip.

image
image929×212 22.2 KB

3

Thank you very much for your kind reply.

I understand the second question.
But in the first question, I didn’t understand for sure because I lacked knowledge.
Could you tell me exactly what file “nvgpu.ko” is?

4

It is just one of the ko file built out from nvidia driver source code. Rel-32 install every ko files (including nvgpu.ko) to the rootfs. Rel-35 only gives simplified steps.

Kernel has form as kernel image and kernel modules (.ko files). Ko files are also part of the kernel drivers.

5

Thank you !
I’m really sorry, but let me check one last time, the nvgpu.ko file contains all the drivers from the kernel build, not the module type driver for gpu, right?

6

If not, is there a separate command to install all drivers from kernel build like Rel-32?
$ sudo make ARCH=arm64 O=$TEGRA_KERNEL_OUT modules_install
INSTALL_MOD_PATH=/Linux_for_Tegra/rootfs/

7

Of course it does not include all the driver. It is just one file from one source code.

8

$ sudo make ARCH=arm64 O=$TEGRA_KERNEL_OUT modules_install INSTALL_MOD_PATH=/Linux_for_Tegra/rootfs/

this shall still work.

9

I thought it wouldn’t work because only the Real Time kernel was written in the manual. Thank you very much for your answers.

10

That’s because building real-time kernel changes the kernel suffix, which would definitely cause kernel modules to fail to load because the version/name does not match. Otherwise, you usually don’t need to replace/re-install kernel modules after re-building kernel images if the kernel suffix is the same.

11

You might find this of use:
https://forums.developer.nvidia.com/t/topic/262647/12

Note that the RT kernel is a modification of the existing kernel, and so this counts as needing both the base kernel Image and the modules to all need install, preferably with a new CONFIG_LOCALVERSION.

12

I’m thankful to each and every one of you for your support. Based on what you’ve all shared, I will give it a try.