Re: Who has had success installing a Guix system on arm?

[Vagrant Cascadian <vagrant@debian.org>]

[-- Attachment #1: Type: text/plain, Size: 3142 bytes --]

On 2020-11-05, Timothy Sample wrote:
> Jesse Gibbons <jgibbons2357@gmail.com> writes:
>
>> Has anyone in this mailing list successfully used the Guix system on
>> an armhf or aarch64 computer?

I'll also add that I have used Guix System on Pinebook Pro, Pinebook,
rock64, rockpro64, pine64+, veyron speedy (a.k.a. ASUS C201), wandboard
solo and overdrive1000... maybe others. :)

Most of these I have started by booting to another OS, installing the
guix binary distribution, and then using "guix system init" to install a
Guix System to another partition and/or device.

On some of them I've managed to install a single image that only
requires changing the bootloader, as they can share a common kernel.

It's possible, for sure, but not all of them support all features;
e.g. video acceleration is only recently being added for some of these
platforms and some only support a headless system with a serial
console...

I also tend to prefer the linux-libre-arm*-generic kernel variants, as
the "linux-libre" kernels usually require adding specific modules to the
initrd configuration in your config.scm, which can be time consuming and
difficult to track down when you can't boot the system, and for extra
fun, the exact set of modules needed sometimes changes between major
kernel versions.


> For the BeagleBone, it required using an older version of U-Boot.
> Upstream U-Boot has stopped supporting it.  Our package patches support
> back in, but it didn’t work for me.  I believe I’ve tried to use Guix on
> it three times and only succeeded once.  If you have a BeagleBone Black
> and want to use Guix, let me know.  I could track down my notes and see
> if I can still produce a usable image.

I haven't tested recently, but at the time I made those patches it did
boot. Also, the am335x_evm should boot, you just have to make sure your
first partition starts at 4MB+ rather than the typical 2M.


> For the Veyron boards, it is starting to work quite well.  The current
> version of U-Boot can boot these boards, but you can also boot U-Boot
> from the stock firmware (which is Depthcharge).  This is needed because
> the Guix initial RAM disk is too big for Depthcharge to boot directly.

Curious; I've never been able to get the U-Boot working on the veyron
speedy, but use an old libreboot to boot guix (but only linux-libre
versions < 5.5, 5.6+ gets stuck in some loop). The initrd size appears
to be a non-issue for me.


> The point I’m trying to make is that Guix can be made to work on armhf,
> but it is not at all smooth sailing.  There are usually one or two
> things that need fixing before it will work.  IME, we are close to
> having a decent experience, but we’re definitely not there yet.

Yes, I'll definitely agree that it is improving, but maybe not that it
is close. :)


Most of this is not specific to guix per se, it is generally a problem
with arm systems in general; there is much less consistancy across
platforms which requires fiddling with bootloaders and kernels more
often than some other platforms.


live well,
  vagrant

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 227 bytes --]