Video

https://youtu.be/m0Q632otdrA

Attendees

7

Topics

• Tales from the Crypt at ELC Europe. See full list of videos at https://elinux.org/ELC_Europe_2022_Presentations
  • See Tim’s presentation slides link at https://elinux.org/images/1/14/Tales_from_the_crypt_v2.pdf
    • See video at https://youtu.be/8VADXvHTx_k
  • Related presentation in https://elinux.org/images/4/42/ELCE2022-UEFISecureBootOTAUpdatesOnARM.pdf
    • See video at https://youtu.be/H_dBnwkTAbw
• Matt Updates on Jetpack 5:
  • Matt now has Orin hardware, fixed one issue with edk2 in https://github.com/OE4T/meta-tegra/issues/1055
  • Investigation into why OP-TEE from source build is failing in https://github.com/OE4T/meta-tegra/pull/1049
    • GCC 9 or earlier works, GCC 10 or later doesn’t, reason unclear.
    • Workaround is to make it to depend on GCC 9 or earlier toolchain and building that way.
    • No response from NVIDIA other than to use NVIDIA toolchains.
    • Don’t want to include a toolchain in the meta-tegra layer because it’s a lot of work, interferences with other layers. Would rather figure out why it doesn’t work with the GCC toolchains.
    • Matt will follow up on the forum with Ilies’ post regarding OP-TEE build at https://forums.developer.nvidia.com/t/optee-vs-gcc-12-x-y/226747
• Had a response in the NVIDIA forum regarding upstream, have a path for upstreaming
  • Detail about submissions now on the EDK repo page - see https://github.com/NVIDIA/edk2-nvidia/wiki/CONTRIBUTING
• Discussed update plans (mender/swupdate) and A/B Support
  • Kurt has everything working with swupdate and A/B
    • Update bootloader dtb and CPU bootloader, these are the only two bootloader partitions updating now
    • Using extlinux in the rootfs, that’s the only thing changing
    • Bootloader with default, firmware A/B and rootfs A/B - keying everything off the firmware to keep in sync. This loads the firmware and initramfs from correct A/B partition based on firmware A/B selection
    • Because UEFI has its own DTB and replacement for the plugin manager implementation - bootloader overlays - this is the reason to update the dtbs through the bootloader. Expects to find those in the partition keyed by firmware A/B. This keeps everything in sync.
    • In one instance saw the rootfs slot differ from the firmware slot, something strange happened with UEFI variables, something got corrupted. Only noticed this once.
    • Mostly leaving partitions alone other than UEFI and UEFI DTB
    • This is different from initramfs from oe4t - it is looking at the rootfs slot. If something gets out of sync with the bootloader there could be an issue there.
  • The intention is to allow them to be different - change in EDK repo to keep them independent.
  • Changing a variable that unties the A/B from the rootfs - this gets complicated.
  • Selection of where it gets for extlinux loader… whether it’s tied to firmware selection or rootfs selection or not… default which has those tied together is simpler to understand.
  • With Nvidia tools updating individual partitions, just selecting the two partitions and whether A/B.
• QT6 support
  • Qt company typically takes care of this.
• Jetpack 5.1 Upcoming release
  • No real details in the roadmap update.
• Orin Devices
  • No HDMI
  • Devicetree support for Monitors is difficult
• Running into issues with container support on Jetpack 5
  • Discussed support for multiple versions of CUDA.
    • Will take some work to support this on meta-tegra. To use, need to modify library paths, etc.
• Installing onto NVMe
  • Have looked into this, haven’t come up with a process that will map well to meta-tegra or typical use case for yocto users.
    • Need to treat the NVMe as a mass storage device from build host.
  • Would prefer to do something a bit more automated.
  • Instead of the doflash script, boot with an sdcard payload.
    • Could run from the rescue partition
    • Torodex does something similar for their easy installer.
    • RAUC using NBD, Network Block Device, for similar functionality