Note: This article is for an older version of L4T 24.2.1. A newer version of this article is now available. Please see: Build Kernel for Jetson TX1 .
In this article, we cover building a kernel onboard the NVIDIA Jetson TX1. Looky here:
Background and Motivation
Note: This article is for intermediate users, and the methods within are somewhat experimental. You should be familiar with the purpose of the kernel. You should be able to read shell scripts to understand the steps described.
When the Jetson TX1 was first shipped the operating system, L4T 23.1, was a hybrid 64-bit kernel, 32-bit user space affair. The only practical route to rebuild the kernel was to use a host computer because two different development toolchains are needed.
With the introduction of the L4T 24.X releases (currently L4T 24.2), both the kernel and the user space are now 64-bit. NVIDIA gives detailed instructions on how to build the system using a host computer. There are other good sets of instructions around, including “Compiling Tegra X1 source code” over at RidgeRun.
If you are building systems which require generating the entirety of Jetson TX1 system, those are good options. For a person like me, it’s a little overkill. Most of the time I just want to compile an extra driver or two as modules to support some extra hardware with the TX1. What to do, what to do …
As it turns out the 24.X kernels are mostly 64-bit, but there’s a sprinkle of 32-bit object files here and there. So here’s an idea: why not just grab the compiled 32-bit sprinkles from a host build, and put them into the kernel build process? That way the kernel and modules can be built on the Jetson TX1 itself without the need for a host development environment.
Now normally I am against developing on host machines when I can develop on a target machine. The TX1 is certainly competent enough for development. But in this case, it seemed worth it to build the needed 32-bit object files on a development host. Once the 32-bit files are built, it is a pretty straightforward task to build the rest of the kernel (with the associated modules) entirely on the TX1 itself.
The script files to build the kernel on the Jetson TX1 are available on the JetsonHacks Github account in the buildJetsonTX1 repository.
$ git clone https://github.com/jetsonhacks/buildJetsonTX1Kernel.git
$ cd buildJetsonTX1Kernel
There are three main scripts. The first script, getKernelSources.sh gets the kernel sources from the NVIDIA developer website, then unpacks the sources into /usr/src/kernel.
After the sources are installed, the script opens an editor on the kernel configuration file. In the video, the local version of the kernel is set. The stock kernel uses -tegra as its local version identifier. Make sure to save the configuration file when done editing.
The second script, patchAndBuildKernel.sh, patches one of the sources files to more easily compile the kernel.
Then the script copies over the 32-bit object files, and proceeds to build the kernel and modules using make. The modules are then installed in /lib/modules/
The third script, copyImage.sh, copies over the newly built Image and zImage files into the /boot directory.
Once the images have been copied over to the /boot directory, the machine must be restarted for the new kernel to take effect.
The kernel and module sources, along with the compressed versions of the source, are located in /usr/src
After building the kernel, you may want to save the sources off-board to save some space (they take up about 3GB) You can also save the boot images and modules for later use, and to flash other Jetsons from the PC host.
For a lot of use cases, it makes sense to be able to compile the kernel and add modules from the device itself. This particular technique is the first attempt at that. Note that it is new, and not thoroughly tested at this point. Use it at your own risk.
The video above was made directly after flashing the Jetson TX1 with L4T 24.2 using JetPack 2.3.