BSP Customization
You can customize NVIDIA® Jetson™ Linux Driver Package (L4T) by:
• Manually configuring and setting up the software drivers before use.
• Using NVIDIA® SDK Manager to perform a customized setup.
Note: | NVIDIA SDK Manager performs the following: • Installs NVIDIA® Jetson™ Linux Driver Package
• Configures or flashes your Jetson device
• Runs samples If you installed using SDK Manager, skip these topics. For more information, see the section “Installing JetPack” in the Jetson Developer Kit User Guide for your platform. |
Boot Options
On a NVIDIA® Jetson™ reference board, Linux boots from a root file system (rootfs) on integrated memory (eMMC or QSPI), attached memory (SD card or USB device), or network-accessible storage.
The bootloader must be loaded from the internal eMMC.
The kernel and DTB can be loaded from:
• An SD card (formatted to GPT)
• A USB drive (formatted to GPT)
• A network (via DHCP/TFTP)
• An NVMe SSD (formatted to GPT)
Choosing a Boot Device for Jetson AGX Xavier Series Platforms
Applies to: Jetson AGX Xavier series
CBoot Boot Options (CBO) is a device tree that can be used to set certain boot configuration options, such as boot device priority and IP addresses used to boot from a network.
The name of the CBO device tree node is boot-configuration.
Node Properties
boot-order
The boot-order property contains a list of boot devices in order of priority. Each device is specified by its name and controller information. Possible entries are:
• sd: Represents an SD card on any controller.
• usb: Represents an external USB mass storage device on any controller and port.
• net: Represents network boot.
• emmc: Represents built-in MMC storage.
• nvme…: Represents built-in NVMe SSD storage. The format of the entry is one of:
• nvme: C<n> | To NVMe storage on PCIe device <n> |
• nvme:pcie@<addr> | To use NVMe storage at PCIe address <addr> |
• nvme | To use any available NVMe device |
For example:
boot-order = "sd", "usb", "nvme", "net", "emmc";
tftp-server-ip
The tftp-server-ip property specifies the IP address of the TFTP server.
The property is specified as:
tftp-server-ip = /bits/ 8 <a b c d>
Where <a b c d> are the four parts of a V4 IP address.
dhcp-enabled
The dhcp-enabled property has no value. If it is present, the device obtains its IP configuration from a DHCP server. If it is absent, the device obtains its IP configuration from the static-ip, ip-netmask, and ip-gateway properties.
static-ip, ip-netmask, and ip-gateway
The static-ip property specifies the static IP address in the IP configuration.
The ip-netmask property specifies the IP address’s subnet mask in the IP configuration.
The ip-gateway property specifies the gateway’s IP address in the IP configuration.
All three properties are required if dhcp-enabled is false or unspecified; otherwise they are not needed and are ignored if present.
The properties are specified as:
static-ip = /bits/ 8 <a b c d>
ip-netmask = /bits/ 8 <a b c d>
ip-gateway = /bits/ 8 <a b c d>
Where <a b c d> are the four parts of a V4 IP address or subnet mask.
Examples
This topic presents examples of complete boot-configuration definitions.
Example: DHCP Enabled
/dts-v1/;
/ {
compatible = "nvidia,cboot-options-v1";
boot-configuration {
boot-order = "sd", "usb", "nvme", "emmc", "net";
tftp-server-ip = /bits/ 8 <10 1 2 3>;
dhcp-enabled;
};
};
Example: Static IP Configuration Specified
/dts-v1/;
/ {
compatible = "nvidia,cboot-options-v1";
boot-configuration {
boot-order = "sd", "usb", "nvme", "net", "emmc";
tftp-server-ip = /bits/ 8 <10 1 2 3>;
static-ip = /bits/ 8 <10 1 2 2>;
ip-netmask = /bits/ 8 <255 255 255 0>;
ip-gateway = /bits/ 8 <10 1 2 1>;
};
};
Rebuilding the DTB
If you modify the DTS, you must rebuild the DTB.
To modify the DTS and rebuild the DTB
1. Locate the cbo.dts file in this directory:
$ cd <bsp>/Linux_for_Tegra/bootloader/
Where <bsp> is the location of Jetson Board Support Package BSP.
3. Enter this command to convert the DTS to a DTB:
dtc -I dts -O dtb -o cbo.dtb cbo.dts
The dtc executable is available in:
<bsp>/Linux_for_tegra/kernel/dtc
Choosing a Boot Device for Other Platforms
Applies to: Jetson Nano devices, Jetson TX2 series, and Jetson TX1
On platforms other than NVIDIA® Jetson AGX Xavier™ series, CBoot uses U‑Boot to perform the booting process. Boot device priority is determined by U‑Boot.
Linux Host System Prerequisites
To use L4T on a Linux host system, the following hardware and software prerequisites must be met:
• Host PC running Linux
• A kernel image
L4T contains a kernel image for your use. Alternatively, you can download and rebuild the kernel image from source.
• Bootloader
Flashing on a NVIDIA® Jetson™ developer board requires a bootloader, which is NVIDIA T-Boot (nvtboot).
• Network file system
If you intend to boot Linux on the reference board from your Linux host system or a network-accessible server.
• A USB cable to plug into the recovery port.
Setting Up a TFTP Server on the Host System
The host system must provide a TFTP server for use by the target.
To set up a TFTP server
1. Install the tftp package and its dependencies by entering this command:
$ sudo apt-get install xinetd tftpd tftp
2. Create the text file /etc/xinetd.d/tftp, for example by entering the command:
$ sudo vim /etc/xinetd.d/tftp
3. Put this content in the file, then save it:
service tftp
{
protocol = udp
port = 69
socket_type = dgram
wait = yes
user = nobody
server = /usr/sbin/in.tftpd
server_args = ~/tftpboot
disable = no
}
4. Generate the signed kernel image. Create the directory ~/tftpboot (specify the pathname defined by server_args) and copy the kernel image to it:
$ sudo ./flash.sh -k kernel --no-flash --sign <board> mmcblk0p1
$ mkdir ~/tftpboot
$ cp /Linux_for_Tegra/bootloader/boot_sigheader.img.encrypt ~/tftpboot/boot.img
5. Generate the signed .dtb and copy the .dtb file to the same directory.
• For NVIDIA Jetson Xavier™ NX:
$ sudo ./flash.sh -k kernel-dtb --no-flash --sign jetson-xavier-nx-devkit mmcblk0p1
$ cp /Linux_for_Tegra/bootloader/tegra194-p3668-p3509-0000_sigheader.dtb.encrypt ~/tftpboot/jetson.dtb
• For Jetson AGX Xavier series:
$ sudo ./flash.sh -k kernel-dtb --no-flash --sign jetson-xavier mmcblk0p1
$ cp /Linux_for_Tegra/bootloader/tegra194-p2888-0001-p2822-0000_sigheader.dtb.encrypt ~/tftpboot/jetson.dtb
6. Restart the xinetd server:
$ sudo service xinetd restart
Extracting Jetson Linux Driver Package
Use these procedures to extract the Jetson Linux Driver Package (L4T). Commands in the examples assume you extracted the release package in ~/.
To extract Jetson Linux Driver Package
• Extract the package manually by executing the command:
$ sudo tar -vxjf Jetson_Linux_R<release_num>_aarch64.tbz2
Where <release_num> is the release number of the current release.
Login Credentials
L4T does not provide default log-in credentials. Create your own log-in credentials the first time you boot.
Installing Additional Packages
NVIDIA provides additional L4T software components and updates in APT (Debian) repositories, accessible through the apt utility.
NVIDIA maintains the following APT repositories:
• For NVIDIA® Jetson™ Nano devices and Jetson TX1: https://repo.download.nvidia.com/jetson/t210
• For Jetson TX2 series: https://repo.download.nvidia.com/jetson/t186
• For Jetson AGX Xavier series: https://repo.download.nvidia.com/jetson/t194
• Packages used on all Jetson platforms: https://repo.download.nvidia.com/jetson/common
The package nvidia-l4t-apt-source is pre-installed in the L4T root filesystem. It identifies the platform it is running on and adds the appropriate repositories to the software source list.
The packages in the APT repositories are signed with GPG keys. The corresponding public key is pre-installed in the L4T root filesystem. Once the repositories are added to the source list, apt can download and install packages.
Note: | The APT repositories described here are also used to upgrade existing packages and install packages that NVIDIA adds to the set initially installed with L4T. For more information, see the topic Updating the Jetson Device and Host. |
Repackaging Debian Packages
You can use the script nvdebrepack.sh to repackage the existing L4T Debian packages. The script can be found in Linux_for_Tegra/tools/Debian/. See Linux_for_Tegra/tools/Debian/nvdebrepack.txt for usage and examples.
Building the Debian Bootloader Package Yourself
This procedure is an alternative to the procedure
Repackaging Debian Packages, above.
The Debian Bootloader package installs one or more payload files, depending on the type of processor in your Jetson device. The package’s post-install script automatically updates Bootloader from the payload files.
The distributed versions of the Bootloader package support only Jetson reference carrier boards. If you are using a custom carrier board you may have to customize this package to change ODM data, pinmux assignments, etc. To customize Bootloader:
1. Generate customized versions of the BUPs you need to change.
2. Unpack the package.
3. Replace the applicable BUP or BUPs with your customized versions.
4. Repack the BUPs into the package.
5. Install the customized package.
This section describes the package dependencies and scripts that NVIDIA uses to build the Bootloader package. You may use it as a reference to create your own Bootloader package.
Versions of the Bootloader Package
Jetson Linux can obtain either of two versions of the Bootloader package, depending on which Jetson device you are using. The Bootloader packages install the following payloads:
• The package for Jetson Nano devices and Jetson TX1:
/opt/ota_package/t21x/bl_update_payload
• The package for other Jetson devices:
• BUPs for Jetson Xavier NX series and Jetson AGX Xavier series:
• /opt/ota_package/t19x/bl_only_payload
• /opt/ota_package/t19x/bl_update_payload
• BUPs for Jetson TX2 series:
• /opt/ota_package/t18x/bl_only_payload
• /opt/ota_package/t18x/bl_update_payload
Pre-Dependencies and Dependencies
The Bootloader package’s pre-dependencies are:
• nvidia-l4t-core (must match this package’s major release)
The package’s dependencies are:
• nvidia-l4t-tools
Package Scripts
The package has a post-installation script, which performs the BUP installation. You can get it by extracting the .deb file.
Building Kernel Debian Packages Yourself
You can customize the L4T kernel by getting the kernel source packages, making your changes, and building the Debian packages.
This section describes the package dependencies and scripts that NVIDIA uses to build the kernel packages. You may use it as a reference to create your own Debian packages.
Working with the Packages
The kernel packages are all open source. Three of the four kernel packages are in public_sources.tbz2. You can download this archive from the NVIDIA Developer Center with the following links:
This archive contains another archive named kernel_src.tbz2, which in turn contains three directories of header files:
• nvidia-l4t-kernel/
• nvidia-l4t-kernel-dtbs/
• nvidia-l4t-kernel-headers/
The debian.org
Guide for Debian Maintainers gives guidelines for modifying the open source files and creating new Debian packages from them.
The fourth kernel package,
nvidia-l4t-jetson-io, is discussed in
nvidia-l4t-jetson-io, below.
Package Dependencies
Most L4T Debian packages pre-depend on nvidia-l4t-core. This package prevents package installation on an incompatible Jetson platform, e.g., installing a Jetson TX2 kernel on a Jetson Nano device. nvidia-l4t-core does not perform the installation if it detects an incompatible platform.
nvidia-l4t-core also prevents a partial upgrade, in which one L4T package upgrades to a new major release (e.g. release 32.4 to release 32.5), but other L4T packages that depend on it are not upgraded as well. Partial upgrades can cause compatibility issues between firmware, programs, and libraries that have been upgraded and ones that have not.
nvidia-l4t-kernel
nvidia-l4t-kernel contains files for the L4T kernel itself.
Pre-Dependencies and Dependencies
This package’s pre-dependencies are:
• nvidia-l4t-core (must match this package’s major release)
This package’s dependencies are:
• None
Package Scripts
This package has a post-installation script, which you can get by extracting the .deb file. It performs these actions:
1. Executes depmod -a.
2. Creates a dpkg trigger file named /usr/lib/linux/triggers/<release>. The trigger invokes actions defined in /etc/kernel/postinst.d to update initramfs/grub configs/… when the kernel is updated. This conforms to the standard Ubuntu kernel update procedure.
nvidia-l4t-kernel-dtbs
nvidia-l4t-kernel-dtbs contains files for L4T’s device tree blobs (DTBs).
The package installs all of the .dtb files in /boot/. When you flash the board, it installs the .dtb file used by the board in /boot/dtb/ by checking the board specification against the DTBs’ compatibility information.
Pre-Dependencies and Dependencies
This package’s pre-dependencies are:
• nvidia-l4t-core (must match this package’s major release)
This package’s dependencies are:
• device-tree-compiler
• nvidia-l4t-kernel
Package Scripts
This package has a post-installation script, which you can get by extracting the .deb file.
The post-installation script performs these actions:
1. Decompiles the .dtb file used by the board in /boot/dtb/ and gets the bootargs property in the /chosen node.
2. Decompiles the corresponding
.dtb file in
/boot/ and substitutes the
bootargs property from step
1 in the resulting
.dts file.
3. Recompiles the .dts file to a .dtb file and puts it in /boot/dtb/.
nvidia-l4t-kernel-headers
nvidia-l4t-kernel-headers contains L4T kernel header files.
Pre-Dependencies and Dependencies
This package’s pre-dependencies are:
• nvidia-l4t-core (must match this package’s major release)
This package’s dependencies are:
• nvidia-l4t-kernel
Package Scripts
This package has a post-installation script, which you can get by extracting the .deb file. The script performs these actions:
1. Extracts the kernel header tarball to /usr/src/.
2. Updates two symbolic link files, build and source, to point to the correct kernel header files directory.
nvidia-l4t-jetson-io
nvidia-l4t-jetson-io contains Python scripts concerned with Jetson I/O functions.
Pre-Dependencies and Dependencies
This package’s pre-dependencies are:
• nvidia-l4t-core (must match this package’s major release)
This package’s dependencies are:
• mount
• python3
• util-linux
• nvidia-l4t-kernel
• device-tree-compiler
You can also get the dependencies by extracting the Debian file.
Package Scripts
This package has no package scripts.
Configuring NFS Root on the Linux Host
To boot the target device from NFS, you must provide an NFS root mount point on your Linux host system.
Prerequisites
• An Ethernet connection to install packages on the host.
• An Ethernet connection on the target.
To configure NFS root on the Linux host
1. Install the NFS components on your host machine:
$ sudo apt-get install nfs-common nfs-kernel-server
2. The NFS server must know which directories you want to export for clients. This information is specified in the /etc/exports file.
• Modify /etc/exports to look somewhat like this:
$ /nfsroot *(rw,nohide,insecure,no_subtree_check,async,no_root_squash)
• After adding the entry, restart with the command:
$ sudo /etc/init.d/nfs-kernel-server restart
3. Create an /nfsroot directory on your Linux host system:
$ sudo mkdir /nfsroot
4. Set the permissions of the /nfsroot directory like this:
$ sudo chmod 755 /nfsroot
$ sudo chown root.root /nfsroot
5. Copy the file system to the nfsroot directory:
$ cd ./rootfs
$ sudo cp –a * /nfsroot
6. Export the root point:
$ sudo exportfs -a
Alternatively, you can export or un-export all directories by using the -a and -u flags. The following command un-exports all directories:
$ sudo exportfs -au
7. Optionally, if the Ubuntu firewall blocks NFS root access, it must be disabled, depending on your configuration, with the following command:
$ sudo ufw disable
8. If there are issues performing the NFS boot, verify that everything on the host system is configured properly by executing the following step on a booted target board through USB/SD/internal eMMC. It should be possible to mount the host NFS root point on the target device:
$ mkdir rootfs
$ sudo mount -v -o nfsvers=3 <IP-ADDR>:/nfsroot rootfs
Where <IP-ADDR> is the IP address of the Linux Host machine as taken from the ifconfig command. This proves that the host configuration is correct.
Note: | Before executing the mount command on the target system, install the nfs-common package with the command: $ sudo apt-get install nfs-common |
To boot the target with the NFS root point, see
Flashing and Booting the Target Device. Be sure to include the
-N option for the NFS root point.
Determining Version and Platform Information
Use these procedures to determine:
• The flashed BSP version
• The kernel version
• Other platform information
To determine the BSP version and other platform information
• Execute command:
head -1 /etc/nv_tegra_release
Output is similar to the following:
# R31, REVISION: 0.0, GCID: , BOARD: t186ref, EABI: aarch64, DATE: Wed Aug 1 23:57:14 UTC 2018"
To determine the kernel version
• Execute this command in the kernel directory:
head -22 Makefile | tail -3
Output is similar to the following:
VERSION = 4
PATCHLEVEL = 9
SUBLEVEL = 140
• If the system is running, determine the kernel version with the command:
uname -a
Output is similar to the following:
Linux tegra-ubuntu 4.9.140-tegra #16 SMP PREEMPT Mon Jun 3 12:08:39 PDT 2019 aarch64 GNU/Linux
To determine boot configuration
All boot configuration is present in the flashing configuration. For example, jetson-xavier.conf provides the following information:
CHIPID=0x19;
EMMC_CFG=flash_t194_sdmmc.xml;
BOOTPARTSIZE=8388608;
EMMCSIZE=31276924928;
ITS_FILE=;
SYSBOOTFILE=p2972-0000/extlinux.conf;
BPFDTB_FILE=tegra194-a01-bpmp-p2888-a01.dtb;
DTB_FILE=tegra194-p2888-0001-p2822-0000.dtb;
TBCDTB_FILE=tegra194-p2888-0001-p2822-0000.dtb;
The full definition of the boot configuration is in p2792-0000.conf.common. Files like jetson-xavier.conf reference that file, then specify additions and changes.
The default kernel build configuration is as follows:
arch/arm64/configs/tegra_defconfig
Improving System Performance
This section discusses types of customization that you can use to improve the performance of a Jetson platform.
Lightweight Window Manager Alternatives
The window system installed with L4T is Gnome, the standard Ubuntu window manager. In many cases you can reduce boot time, response time, memory consumption, and CPU utilization by replacing Gnome with a lightweight window manager. NVIDIA recommends the LXDE environment with Compton compositing manager for this purpose.
For LXDE installation instructions and other information, see the section
LXDE in topic
Window Systems.