Good morning all,

a few years ago, when ARM was still mostly powering smartphones, I made
my employer purchase an ARM64 server for research purposes: a Cavium
ThunderX based Gigabyte R120-T35 (mainboard MT30-GS2).

After endless fights and attempts to run kernels newer than the latest
Buster one (4.19.0-27) on that machine (all of them ending up with silent
or cryptic kernel failures), not being able to upgrade user-space, I had
asked for help on the debian-arm64 list, but apparently was too late
(or one of too few people who got such a thing at all?).

I eventually was reminded of my old but still trusted RasPi 0ws - which
had undergone in-place "full-upgrade"s while not upgrading the kernel
with the rest of the distro, due to their "special" handling of DTBs etc.
by Raspbian. (Raspbian suggest to preform a full install instead of a
full upgrade.)

In short, even with regular Debian, it should be possible to keep the old
(known to run) kernel, put the crucial kernel packages on APT hold
(linux-image*, linux-headers-*, linux-libc-dev?, linux-kbuild-*, anything
else?) and run an in-place "full-upgrade" (aka "dist-upgrade"), for proof
of the concept.
Has anyone done this yet and can share ideas?

If this "Raspbian-Frankenstein" approach works, how to translate it to a
working FAI setup?
I guess I'd have to provide a "repository" with the old (but still present
in some Debian archives) packages, build a NFSroot using that, and tweak
the package_config to keep the versions fixed - again: anything else?

Thanks for listening :) any suggestion, link, idea ... is appreciated!

- Steffen

Antwort per Email an