Skip to content

OE4T Meeting Notes 2022 04 14

Jump to bottom Edit New page
Matt Madison edited this page Apr 15, 2022 · 2 revisions

Video

https://youtu.be/TWQEC_LeeW8

Attendees

9

Topics

  • Jetpack 5
    • List of latest status and things which still need to be done at https://github.com/OE4T/meta-tegra/wiki/JetPack-5-Developer-Preview-L4T-R34.1.0-Notes
    • Have branches in meta-tegra, meta-tegra-community and tegra-demo-distro repos, autobuilder is setup do to checks on each commit. Running builds on JP5 developer preview only for xavier platforms today.
    • Still need to get orin MACHINE definitions in place.
    • Have a sato image up and running, have a weston image up and running.
    • Have a dedicated slack channel at https://oe4t.slack.com/archives/C03AEG2PD7H/p1649521822950479
    • Prebuilt gstreamer plugins are working with 1.20 gstreamer. Can’t build from sources yet due to other new APIs which are binary only libraries, don’t ship headers for those.
      • Haven’t tried nvcompositor yet. (Update: looks like that plugin has been removed altogether.)
    • Bootloader change to switch to UEFI.
      • Have sources and a recipe for building. Default to using prebuilt copy but can build from source if desired
    • Secure OS
      • Have the trusted firmware building from source. Need to guess what version of upstream they’ve used for sources, believe he’s figured that out.
      • Started with OPTEE, more difficult to get building. Started looking at taking recipe from meta-arm. May add dependency on meta-arm and use recipes or make private copy in tegra layers. NVIDIA is based on slightly older version than what’s currently in meta-arm.
    • Will impact mender integration.
    • Discussion about future upgrade plans, consensus is most attendees are likely to be moving toward swupdate.
    • LG has stepped out as maintainer as yocto openembedded - stuck on dunfell. Kirkstone has lots of old references to licenses.
    • Still support bootloader update package mechanism in jetpack 5. Interactions of bootloader to kernel - will they stick to device tree, switch to ACPI, how to retain state (presumably UEFI boot variables). Lots of questions around these which only NVIDIA will answer for us. No secure boot support yet. For xavier will need to maintain some compatibility. Orin potentially will be very different. Got rid of tegra DRM mechanism, new mechanism for Orin. Not sure if this is more closely aligned with discrete GPU. Look at L43 34.1.0 documentation regarding weston documentation, specifically call out different drivers for Xavier vs Orin. Matt is looking at swupdate going forward,
    • Are they going to release Xavier modules which won’t be usable with Jetpack 4.x and if so when?
    • Some discussion in NVIDIA meetings regarding a plan to continue to support on existing hardware with new releases of Jetpack 4.x. Made a comment about being able to move to newer kernels on Jetpack 4.x releases. Would be good to get more clarification on the plan for Jetpack 4.x releases going forward, including kernel upgrades.
    • Matt's dev forum post about missing library header files is here
    • Haven’t looked at container support yet on Jetpack 5, needs to be on the todo list.
  • CI-Testing strategies
  • Disk Encryption
    • Got device with fuses burned - put automated steps into secureboot tegra repo at https://github.com/moto-timo/secureboot-tegra
    • Planning to share utilities in the dunfell branch. Would ideally like cli tools which could talk to both trusty for for JP4 and optee for JP5.
    • Would help to try to align with more NVIDIA tools.
  • Power/Thermal
    • On dev board the fan comes on full on initial boot. In an image built from meta-tegra this doesn’t happen.
    • ina3221 driver must be included.
    • Compare kernel config and services to board not working.
    • Might be related to a fork on uboot that works with AGX. Important that uboot carries forward the in-memory devicetree from cboot.
  • Wayland/weston/qt5 issues with latest dunfell.
  • Clara AGX Devkit - working to get Discrete GPU enabled on that platform. Have started with meta-tegra and added discrete GPU recipes for CUDA, cudnn. New to openembedded and yocto. Would like to start sharing going forward.
    • Specific changes to support clara might be nice to have in meta-tegra.
    • If trying to generalize to be discrete GPU across other platforms like x86- may be nice to have in another layer.
      • No goals for x86 today.
    • Download source URIs need to point to locations which aren’t behind an authentication firewall.
      • dGPU and CUDA are public downloads
      • cuDNN is behind a credential server.
    • Already have some things working, just need to file work with legal to put on github. Assuming it can be its own layer now, can discuss moving into meta-tegra later.
Add a custom footer
Add a custom sidebar
Clone this wiki locally