I am working with the TX2 and trying to do most development work on an Ubuntu host machine for modularity between multiple TX2’s.
Currently the process is to make updates to filesystem in rootfs/ and flash to eMMC using the command:
$ sudo ./flash.sh jetson-tx2 mmcblk0p1
However, “flash.sh” doesn’t seem to install the Jetpack features (i.e. Cuda or Visionworks). Is there a process to install these features without using the Jetpack installer which overrides customized work in the rootfs/ directory?
I hesitate to answer because I can’t actually test from my Fedora host, but understand that JetPack has essentially two stages of operation: Flashing a Jetson through the USB cable, or adding packages through wired ethernet. During flash the driver package and sample rootfs are used (the rootfs subdirectory you mentioned), and this goes to the Jetson without those additional packages. Command line flash and JetPack behave the same way for the flash step.
After flash the Jetson reboots (you don’t actually have to flash, you could uncheck flash and go for just the packages) and extra packages (such as CUDA) are installed over wired ethernet. Those packages never touch the rootfs subdirectory.
Many of the extra packages get added somewhere in “/usr/local” of the Jetson, and won’t be visible in the rootfs directory of the host. Try running Jetpack with wired ethernet connected, disable the flash feature, and check for what additional packages you want installed.
Note that most of those additional packages were meant to be installed only via JetPack, but once you’ve run JetPack (with or without actual package installs) there will be a “repository.json” file which has the URLs of most of those packages…you can use wget on those packages and then manually install (not necessarily recommended because install order might take some experimentation).