Can we perform the initial install from an ARM host?

[Shamino0]

I'm using meta-tegra to develop a BSP for a custom board using an Orin Nano module and NVMe storage. We've got it mostly working, including A/B booting, which is great. But in order to install new images using the doflash.sh and initrd_flash scripts, we have to run those scripts from an x86_64 Linux host. This is fine for our development process, but....

The board we're using has a second processor with an ARM (aarch64) core, also running Linux. We would like to be able to install the Orin Nano's firmware (including the initial install onto a new/blank module) from the ARM64 processor, using a USB link between the two processors.

It seems like this should be possible, but the scripts are using a few NVIDIA proprietary binaries (e.g. tegrabct_v2, tegraparser_v2 and tegrarcm_v2). Are these binaries necessary? Are there open source equivalents that could be compiled for an ARM64 host? Is their function documented well enough so we could write our own?

Ultimately, we want to develop an original module-imager utility customized for our board (to avoid all the overhead of the NVIDIA tools, that must support a wide variety of modules, carriers and configurations), but as a first step, we'd like to be able to just see if we can image our module from an ARM64 host (maybe a Jetson developer kit or Seeed Studio computer), and progress from there.

Can anyone help out here? Maybe pointers to documentation or work that has already been done by the community?

[madisongh]

This topic has come up several times... here is a semi-recent NVIDIA dev forum thread on the topic. Short answer: not supported, stick to x86.

The tools are necessary, are binary-only, and there are no sources available. I have used binfmt/qemu-user-static to run some of the x86 binaries on an arm64 host, but only for signing binaries during a build, not for the flashing process. Even then, I wouldn't count on that working for production purposes.