OE4T / meta-tegra

BSP layer for NVIDIA Jetson platforms, based on L4T
MIT License
418 stars 230 forks source link

TX2 support for Langdale and newer #1259

Closed rustand2 closed 1 year ago

rustand2 commented 1 year ago

From what I can see Kirkstone is the latest Yocto release that is supported for Jetson TX2 by this layer. Are there any plans to add support for TX2 in the newer meta-tegra branches?

ichergui commented 1 year ago

Hello @rustand2 you can use kirkstone-l4t-r32.7.x branch

Please see our wiki page to check the active branches

dwalkes commented 1 year ago

I think we briefly discussed this at the end of the Feb Meeting and then I forgot to add to the meeting agenda in March. The challenge as I recall will be to try to keep the 4.9 kernel working with future layers, especially as Nvidia support sunsets for Jetpack 4. Pull requests or investigation/reports are welcome.

rustand2 commented 1 year ago

So Nvidia has no plans to keep supporting TX2 in newer Jetpacks? And meta-tegra is dependent on Nvidia to release newer Jetpacks in order to update the kernel?

madisongh commented 1 year ago

So Nvidia has no plans to keep supporting TX2 in newer Jetpacks? And meta-tegra is dependent on Nvidia to release newer Jetpacks in order to update the kernel?

That's correct, TX1/Nano and TX2 platforms are not supported in JetPack 5/L4T R35.x. It might be possible to switch to a mainline kernel, but that wouldn't be compatible with any of the JetPack packages, since those rely on the downstream kernel changes and drivers that NVIDIA supplies.

rustand2 commented 1 year ago

Well, as far as I know I do not use any Jetpack packages unless they are used by the system in some way. I just use it as a normal Linux machine. I was under the impression that these things were somehow required in order for Linux to work on these machines.

rustand2 commented 1 year ago

If I switch to a mainline kernel do I still need meta-tegra (and can I still use meta-tegra)? I'm pretty scared of setting up all this Nvidia specific stuff on my own without meta-tegra, I feel like it is extreme overcomplicated and proprietary compared to just booting something on a normal pc. I especially like the four chainloaded bootloaders (MB1, MB2, cboot, uboot)...

I already use a separate devicetree. Is there anything else crucial that meta-tegra does? I assume it is mostly in the boot process?

madisongh commented 1 year ago

You might be able to reuse the tegraflash packaging and scripts from here, but otherwise this layer is pretty closely coupled with NVIDIA's BSP, and would try to install their userland libraries, which are tied to their downstream kernel and GPU driver and wouldn't be usable with a mainline kernel.

I'd expect this to be a bunch of work, and unfortunately I can't even dig up any links to provide as references to start with.

dwalkes commented 1 year ago

I'd expect this to be a bunch of work, and unfortunately I can't even dig up any links to provide as references to start with.

See https://github.com/OE4T/meta-tegra/discussions/965 for a thread on nano, and https://elinux.org/Jetson/Nano/Upstream for a page on an upstream kernel build for nano.

I do not use any Jetpack packages unless they are used by the system in some way. I just use it as a normal Linux machine

If you don't need the GPU and can consider another platform you might just use an RPi 4 and save yourself some headaches and cash.