OE4T Meeting Notes 2025‐10‐10 - OE4T/meta-tegra GitHub Wiki

Attendees

10

Video

https://youtu.be/ve7csDqTFeA

Discussion

r38 Release Status

  • r38 project page at https://github.com/orgs/OE4T/projects/2
  • demo-image-full, sato, weston all booting.
  • Some drivers not quite ready. Hard to tell whether errors/warnings from the kernel are meaningful.
  • gstreamer working
  • All boot firmware with sources
  • Have a secure partition manager, new thing based on google partition manager which ARM has adopted.
  • optee and uefi both building from source
  • Standalone memory manager is building successfully, needed instructions updated. Jetson specific version which works.
  • USB Keyboard/mouse input is very laggy in weston, sometimes get 100 copies of whatever you type
  • CUDA, deepstream all in good shape.
  • Container support implemented.
  • Tested without demo-distro with oe-core plus demo build and without systemd.
  • Next task is to fix up and put some thought into flashing procedure
  • Likely will remove initrd-flash to make it clear this isn’t the same process. Mulling over whether that should be the default or if it should presign binaries during the build and have the tegraflash generate uefi flashing process instead.
  • May be something needed at runtime that we can’t provide at build presign.
  • Even if not the case now might happen in the future
  • Signing process takes a long time, changed the way the RTM boot gets signed, need to go through the process 2-3 times.
  • Will speed up the process to only flash the QSPI when the bootloader changes.
  • Minimal patch of flashing scripts, attempting to avoid modifying these, look fragile.
  • Trying to make the sequence match L4T flow as closely as possible.
  • May move OE4T flash scripts out to a repo, have those referenced from the repo.
  • Goal is to align with next LTS from yocto.
  • Don’t provide a flash layout for redundant filesystem for R38 yet, won’t be an option for A/B updates until later releases.
  • SBSA compliance for arm server baseline architecture - looking more like a PC. BIOS is fixed over time, only thing updated on regular basis is the OS.
  • Still firmware buried in QSPI flash, plan to discuss with NVIDIA.
  • Reduce the pain/need to do capsule update.
  • With R38 l4T launcher is only available outside UEFI.
  • Hardware itself implements newer algorithms for signing. More things to sign but overall process is the same.
  • Using adb with some custom software instead of exposing mass storage device for flashing. Not aligned with software update.
  • L4T flashing scripts prepare all binaries currently, but this may change over time.
  • A/B is on the roadmap for Thor, as is secureboot and disk encryption.
  • Have done a lot of testing from holoscan, deepstream 8 working but a bit hacky.
  • Not planning to merge into master until there is support for Orin
  • Looks like capsule builds still exist in R38, still using bup to capsule sequence.
  • Can’t use Linux 6.16 with Thor at the moment, only 6.08, hopefully will be in the next L4T release.

Building L4TLauncher into UEFI

  • Have this working for Jetpack 6, Backport to Jetpack 5 is working, not sure if it can be shared upstream.
  • What NVIDIA calls the Minimal UEFI build.
  • Had problems with secureboot signing, allocate 2K for signature, that space was too small. Had to change that to change secureboot signing.
  • ESP isn’t actually redundant in base implementation.
  • Firmware loaded in QSPI flash may still have an impact on other drivers in the OS, getting that matched up properly may still be an issue.

swupdate resolution for unbootable slot