----- Original Message -----
From: "Alex Perez" <aperez(a)alexperez.com>
Sent: Saturday, May 9, 2020 2:49:17 AM
Subject: Re: Fedora CoreOS on multi-arch
How difficult would it be to add armhfp to the CoreOS builds? If you're
interested in the use case, it's for the third and fourth generation
OLPC laptops, the XO-1.75 and XO-4, both of which are based on Marvell
MMP, and are 32-bit, with ~1GB RAM. As of Fedora 32, stock Fedora armhfp
kernels boot and userland runs, unmodified, on the XO-1.75.
I think that it is kind of technically possible, but not that simple. IMHO possibly
harder, at least from my perspective, than ppc64le, aarch64 and s390x together. To
elaborate even aarch64 will work currently only on UEFI(qemu somewhat tested only).
I'm not really sure if there is even some that generic way to do it for armhfp(VMs or
boards). My current understanding is that even for aarch64 there will be need to alter the
produced "bare-metal" images for the target "non standard" HW/board.
Something similar to the arm-image-installer(hopefully just patch for it).
For starting to enable the FCOS you will have to look in the
. Either fully build it on
armhfp(IMO easier), although I believe we aren't building Fedora base container image
for it or cross build from(aarch64,hard or other system via full emulation hardest). You
should be able just start by following the README(IMO it is extremely easy to build Fedore
CoreOS locally, on x86_64) and see what will break. One thing is sure that the armhfp will
need patches for
other place at minimum.
Actually one blocking point might be that you need nested virt working due to the way how
the assembler works currently. So not really that easy and I have definitively forgotten
some other possible pain points.
For me, from arm side, I hope that I will be able to have time to look in to running FCOS
on baremetal aarch64 RPI4(VM runs on rawhide Fedora fine for me ;) ) and RockPro64.