Hi everyone, i’m trying to install debian 11/12/13 4th day in a row. But even installer not launching. Custom U-boot for debian is installed, but even any other kernel than 4.9.337-tegra not loading. Please, someone, give me your dd of rootfs of APP partition dump, or any instructions. I hate ubuntu 18, it;s old af and all packages are too old
Hi,
Do you mean you would like to use later version of debian? We would suggest use default version since it is supposed to be more stable.
i’ve tried debian bullseye, bookwork, and trixie
I’m kind of being the bearer of bad news, but until you get to the Orin series running L4T R36.x (L4T is what you call the flash content after NVIDIA drivers are added; see “head -n 1 /etc/nv_tegra_release”), then only the Ubuntu provided is supported. Beyond that, even if you have expert knowledge, it can get a lot more difficult trying to use something else, which I’ll explain below. Here is the supported L4T release list given specific hardware:
https://developer.nvidia.com/linux-tegra
Jetsons do not have an actual BIOS, it is all in software. Thus, the entire boot chain and BIOS equivalent are custom. Every flash is essentially also a BIOS flash. This chain is entirely custom up until the point when UEFI became fully implemented (which is in L4T R36.x). Thus, if boot was not designed to detect some other device or setup, then you would be looking to customize an initrd to reach the o/s being booted to. Up to here, this isn’t necessarily that difficult if you use the NVIDIA kernel.
The NVIDIA kernel has out-of-tree content. Mainline does not have the boot support until you get to the L4T R36.x release. If you can stick to the NVIDIA kernel for the L4T release, even if adapting to another o/s, then you have a chance. You’d have to move to L4T R36.x to use mainline, but that’s an enormous convenience when porting. Not only does R36.x support any mainline kernel, it also implements UEFI.
UEFI has an initial stage which is custom. Once that software is running UEFI presents an abstracted interface which is standardized. Thus, if you use NVIDIA’s UEFI to get to that abstracted stage, then the rest of boot is more or less not much different than an ordinary PC. The UEFI device support might be more limited in scope, but it can be worked with.
Once the kernel is booted (which must be custom if the boot chain is not UEFI), in the older TX1, the GPU driver (used also with CUDA, not just with display) can be bound to a specific Xorg X server release (it is dynamically loaded into the X server; this means the ABI must match and you would have to recompile against a different X server, but in those earlier drivers it is closed source).
A shorter explanation is to say you can only switch distribution or release if the boot chain can boot the kernel (which is custom using out-of-tree content), and you can only use the GPU if it is paired with a specific Xorg release (and that Xorg release is tied to the specific Linux kernel).
I’m not really a VM guy, but it might be easier to run a VM and put what you really want to work with in that. I’m sure there are a lot of similar difficult issues to run a VM, but if you are expert with that, then it might work.
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.