Re: Guix System for embedded systems roadmap.

["Ludovic Courtès" <ludo@gnu.org>]

Hello Mathieu,

Mathieu Othacehe <m.othacehe@gmail.com> skribis:

> At Fosdem 2020, I gave a talk about using GNU Guix as an alternative to
> Yocto, focusing on cross-compiling a Guix System. Here's a status and
> my personal roadmap on this topic.

I’m not much of an embedded person, but this looks like a great
application and it’s great to see progress being made!

> * Somehow related to the previous point, producing a disk-image,
>   currently means spawning a virtual machine. This can be very slow, and
>   using --system, we currently emulate the execution of a virtual
>   machine for a foreign architecture.
>
>   I'd like to propose an alternative mechanism which would be faster and
>   not involving virtual machines. Maybe producing the disk-image in a
>   container?

Unfortunately, I don’t think that’s possible.  The reason we resort to
VMs is that the Linux kernel doesn’t allow you, for instance, to mount a
file system without being root.  So doing things like running Parted,
mounting a file system, and populating it typically requires root
privileges.  (In some cases, there are tools like mksquashfs that can do
that from user-space, but it’s very ad-hoc.)

Thoughts?

> * Increase board support catalog, even if it's tricky because many
>   boards need proprietary blobs to boot (such as Raspberry Pis). The
>   effort started by Danny on wip-buildroot could be resumed.

Speaking of which, I’d love to get my A20 OLinuXino running Guix
System.  :-)

I checked what Buildroot had about that board, but I found the
information to be rather scattered and of varying levels of abstraction,
which made it hard to see if there was anything different from what we
do.

Thanks,
Ludo’.