Black screen after kernel recompilation with new Jetson Linux 36.3

Robotics & Edge Computing Jetson & Embedded Systems Jetson AGX Orin
1

Hi NV team and other community members

I’m working on Jetson AGX Orin (32G) kit and recompiling the kernel (the newest Jetson Linux 36.3 from JetPack 6.0). But nvidia-modest.ko doesn’t work after installing the new kernel and some extra in-tree (SCTP) modules.

I tried the following steps, all natively done on Jetson AGX Orin Kit

  1. I flashed the kit with the new JetPack 6.0 with SDKmanager. The default kernel doesn’t have the SCTP module.
  2. I downloaded the new BSP files from the Jetpack 6.0 site and got source code./source_sync.sh -k -t jetson_36.3. And I made slight changes to defconfig to add SCTP support with make menuconfig and make savedefconfig, and the modified defconfig is moved to arch/arm64/configs/.
  3. I used the provided script ./nvbuild.sh -o DIR to compile the kernel and then ./nvbuild.sh -i -o DIRto install the new Image and kernel modules, including the OOT Nvidia modules.
  4. Then, I rebooted but got a black screen, but I can login with SSH. I check dmesg, it shows several nvidia_modeset: Unknown symbol load errors. But other Nvidia modules are loaded and working well (maybe, I can run nvidia-smi and get the output). The SCTP module is also enabled and loaded.
  5. I restore the original defconfig and follow Step 3 to re-re-compile and install. After rebooting, I got the display and all GUI back.

I compared the two config files; the only difference was the SCTP-related CONFIG.

image
image2699×497 35.6 KB

I wonder if there is any conflict between nvidia-modest and other modules or if there is anything I missed during compilation.

Thanks!

3

Put the full log here, please.
I think if you enable some kernel drivers as standalone modules (=m),
and there are other changes to the kernel image (=y), then you can just install only new modules instead of all kernel stuff.

4

Thanks for your reply.
I put the full log of dmesg here.
dmesg.txt (87.4 KB)

If I intend only to compile and install the modules, should I go with the one located in /usr/src/.../linux-jammy/kernel-source?

5

I feel like you did not handle well with OOT modules. They are just built but not installed.
Actually our document for JetPack 6 does not use nvbuild.sh anymore.
We have a new Makefile handling this.

It’s a in-tree module so you still need the complete kernel source.

6

Hi Dave,
Thanks for your guidance.
I followed the Jetpack guide to build the kernel with modified defconfig but still got something wrong.
I build the kernel natively on Jetson AGX Orin kit, with the steps in documentation and just ignoring the ENV settings for cross-compile. All in Linux_for_Tegra/source folder.

  1. run make -C kernel
  2. run sudo -E make install -C kernel
  3. run make modules
  4. run sudo -E make modules_install
  5. run sudo nv-update-initrd

Then I rebooted, and there was no display output after the boot messages. Here is the full dmesg log.
dmesg.txt (72.7 KB)

P.S.: After getting the black screen, I restored the original defconfig and re-run the above steps. (also to verify the above steps) and the display works well.

7

Then will it work if you build this module into the kernel image? (=y)

8

No, it’s still not working.
I built SCTP into the image.

image
image1123×281 10.3 KB

But the display is still not working.
dmesg here.
dmesg.txt (72.6 KB)

9

I just tried and display is working fine:

nvidia@tegra-ubuntu:~$ cat /proc/config.gz | gunzip | grep CONFIG_IP_SCTP
CONFIG_IP_SCTP=m
nvidia@tegra-ubuntu:~$ lsmod | grep nvidia_modeset
nvidia_modeset 1302528 6
nvidia 1589248 13 nvidia_modeset
host1x_nvhost 40960 8 nvhost_isp5,nvhost_nvcsi_t194,nvidia,tegra_camera,nvhost_capture,nvhost_nvcsi,nvhost_vi5,nvidia_modeset
host1x 208896 5 host1x_nvhost,host1x_fence,nvgpu,tegra_drm,nvidia_modeset

Is your INSTALL_MOD_PATH set correctly?
For native compilation, it should be your root directory.

10

Sorry for disturbing you again, but I still cannot make it.

Here are the steps I applied.

  1. run make menuconfig and make savedefconfig in Linux_for_Tegra/source/kernel/kernel-jammy-src/ to get the modified defconfig, and move it to arch/arm64/configs/ folder
  2. got back to Linux_for_Tegra/source/ and run make -C kernel
  3. run export INSTALL_MOD_PATH=/ and sudo -E make install -C kernel
  4. run export KERNEL_HEADERS=$PWD/kernel/kernel-jammy-src and make modules
  5. run sudo -E make modules_install
  6. run sudo nv-update-initrd

Still got the same errors as before.

Does the “root directory” you mentioned before mean / or /root?

Would you mind sharing your steps?

11

I was cross-compiling on my host PC, but native compilation should also do it.
Maybe you can change the kernel suffix string (LOCALVERSION=...) to make sure all the kernel image and modules do take effect.

I did:

Under Linux_for_Tegra/source/kernel/kernel-jammy-src/:

make ARCH=arm64 LOCALVERSION=-davey-tegra defconfig
make ARCH=arm64 LOCALVERSION=-davey-tegra menuconfig
make ARCH=arm64 LOCALVERSION=-davey-tegra -j8 Image
make ARCH=arm64 LOCALVERSION=-davey-tegra -j8 modules
sudo make modules_install INSTALL_MOD_PATH=/Linux_for_Tegra/rootfs
(Manually copy Image into Linux_for_Tegra/kernel/Image and Linux_for_Tegra/rootfs/boot/Image)

Under Linux_for_Tegra/source/:
export KERNEL_HEADERS=$PWD/kernel/kernel-jammy-src
make modules
export INSTALL_MOD_PATH=Linux_for_Tegra/rootfs
sudo -E make modules_install

Under Linux_for_Tegra/:
sudo ./tools/l4t_update_initrd.sh

Then flash the device.

I mean / on your device.

12

Hi Dave,
Thanks for your patience and kindness. I finally made it through cross-compiling.

1 Like
13

Great!
Glad you made it.

15

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