Hi Thomas, On Fri, 2025-01-17 at 12:00:21 +0100, Thomas Lange wrote: > Hi, > > I've already successfully booted an ARM64 server via UEFI/PXE using > FAI. No installation was done, because this was a special FAI > environment, were no installation is needed, but some hardware testing > tools are used.
I had installed Stretch and Buster on a ThunderX machine, via FAI, quite a while ago. Unfortunately I'm now stuck with old kernels as the box won't boot any recent one. > FAI 6.2.5 also includes some new code for creating a FAI-CD for > arm64. I could not test it myself because of missing hardware. I never used a FAI-CD on any architecture... > I've read that the M1 CPU needs a kernel using 16k pages. That's something that looks important, nicely spotted! > I guess you mean Asahi not Avahi which is something different. Indeed. This must have been autocorrect in action, replacing Japanese beer with something more computer-related ;) > Not sure if Apple can do UEFI or if it needs a signed kernel (and > initrd) image. That's actually what I wanted to know. Others may help, perhaps? (Maybe, as a first step, there's a hidden means to make the boot process more verbose and non-graphic?) > I've just created a generic nfsroot using a 16k pages kernel from > Debian backports. It's available from: > > https://fai-project.org/download/misc/ Thanks, downloaded, will go and analyze it. > A short info is in /root/README inside the nfsroot. > You may try to PXE boot this nfsroot. Will do as soon as I know how... Random thought: Would this also run on a more "normal" ARM64 machine? > Add this to dhcpd.conf: > > if substring(option vendor-class-identifier, 0, 20) = "PXEClient:Arch:00011" { > filename "boot/grub/grubnetaa64.efi"; > } I've been using this for years now ;) but I'm using the signed one, should I revert this? > Then you need a grub.cfg in /srv/tftp/grub/grub.cfg: > > insmod all_video > insmod efi_gop > terminal_output gfxterm > > set timeout=3 > > echo === GRUB grub.cfg geladen > sleep 3 > > > menuentry 'FAI network boot' { > echo 'Loading FAI kernel ...' > sleep 1 > linux $prefix/vmlinuz-6.5.0-5-arm64 ip=dhcp > root=192.168.33.250:/srv/fai/nfsroot-arm:vers=3 rootovl > FAI_FLAGS=verbose,sshd,createvt FAI_CONFIG_SRC=nfs://faiserver/srv/fai/config > FAI_ACTION=install > echo 'Loading FAI initrd ...' > sleep 1 > initrd $prefix/initrd.img-6.5.0-5-arm64 > } > > Here you must adjust the kernel and initrd names/versions. Perhaps I'll go and try this with the ThunderX. > Copy kernel and initrd into this dir: > # cp -pv /srv/fai/nfsroot/boot/vmlinu* /srv/fai/nfsroot/boot/initrd* > /srv/tftp/grub/ > # chmod a+r /srv/fai/nfsroot/boot/initrd* My directory structure is a bit different but I understand what you mean. Shouldn't the second command change the copy, not the source of the initrd? > It' more than a year ago, that I've test this. Please sent me any feeback. Will do as soon as I've got time to address the machine. Maybe next week, maybe after mid-February... > I will try to build a FAI-CD for ARM64 in the next days. Your choice; I won't use it ;) Thanks so far, Steffen -- Steffen Grunewald, Cluster Administrator Max Planck Institute for Gravitational Physics (Albert Einstein Institute) Am Mühlenberg 1 * D-14476 Potsdam-Golm * Germany ~~~ Fon: +49-331-567 7274 Mail: steffen.grunewald(at)aei.mpg.de ~~~
