Thank you for that, where my confusion comes from, is why do we have these LZMA files in the MR? Can we not use standard lzma library and tools? So I just learned, that this is the LZMA loader, and it lives in quite a few archtectures. Thats sad. Can I assume that it's just a special 'shrink-wrapper' around our regular payload (an uncompress linux kernel) then? Which also (unless we want to double-bag) restricts us to LZMA as compression? (we could do a self-extracting XZ kernel, wrapped as an lzma payload :p)
As for the CVIMG format, I think that's not a huge thing, and you mostly solved that, and we can still do what we want to do by faking things if needed at all? afaik openwrt expects a single kernel+rootfs, so if we have a 0 byte rootfs (or whatever it is needed to please the cvimg), we can make that work too, right?
Worst case, we re-compile the bootloader and modify it slightly 
Yeah, I know; but I thought I read that (at least the rtl8196e) actually had those 4 instructions!
The toolchain changes are not that big, and I doubt openwrt devs would mind them that much ... so going that route is fine with me.
That is good to know! But also on the 'lower layer'. e.g. prom.c 'doesn't care' if you will?
I'm first gonna work out the 'absolute minimal set of changes needed' e.g. init() (and uart, cause we need uart :p) + those lexra changes.
I'll extract them from your branch, combine them with what I learn from https://github.com/shibajee/binutils though I doubt anything new is there.
Since you haven't pushed anything in 2 years, I take it nothing is needed/new there?
I'll probably have to go back one step anyway, and just manage to compile and boot what you have in that branch anyway 