Re: [GNU-linux-libre] ppc64el support in endorsed distros

[Matias Fonzo]

Hello Denis,

The maintainer of Dragora GNU/Linux-Libre here.

Since 2012 I have started to re-do the distribution from scratch, the
method that I propose is that the distribution can be built from
scratch.  The method consists in simple shell scripts but well-organized
in stages.  For example, the stage 0 build a cross-compiler, the other
current stages are designed as a previous steps to get the final
system.  Something like LFS and CLFS does, but I consider the method is
a bit beyond of those (because I understand how LFS/CLFS works).
Unlike LFS/CLFS, there two important aspects, 1) all the components are
100% free software. 2) the method is agnostic, you can modify, create or
add the steps that you want to achieve for a stage (for example, to
create other gnu/linux distributions).  When I said that the method is
agnostic, this is especially for the stage 0, the cross-compiler.  This
does not contain hard coded paths.

The independence of the cross compiler is such that I have managed to
produce several cross compilers that use a non-convenient form.  The
first phase, where a cross-compiler for the native architecture is
created; and the second phase, where the produced cross-compiler is
used to create other compilers for the different supported
architectures.  The only cons maybe for you or other people, is that
the set is based on the musl C library.

I've published a recent version, which contains support for the
ppc64le, if you want to give a try:

http://lists.nongnu.org/archive/html/dragora-users/2018-05/msg00000.html

I can provide more information about the bootstrap method, etc. Let me
know!

On Thu, 31 May 2018 12:46:32 +0200
Denis 'GNUtoo' Carikli <address@hidden> wrote:

> On Fri, 18 May 2018 10:41:41 +0200
> Andreas Grapentin <address@hidden> wrote:
> 
> Hi,
>  
> > I have been involved in that a bit, and I think the current problem
> > is creating a suitable cross toolchain to start the port in
> > earnest. If we had people with experience and ideas in that regard,
> > that might me very helpful.  
> I'm very interested in being able to create arbitrary toolchains from
> an FSDG distribution, as I need it to be able to compile older
> (ARM) kernels, compile lm32 code for some AMD processor (SMU) present
> in AMD CPUs, etc.
> 
> == Cross toolchains ==
> In Parabola we have crosstool-ng but I didn't spend enough time
> on it to successfully create a toolchain.
> > $ pacman -sS crosstool
> > pcr/crosstool-ng 1.23.0-1.parabola1
> >     Versatile (cross-)toolchain generator, with Linux-libre kernel
> > support pcr/crosstool-ng-git 1.22.0.r21.g2d3c70d-1.parabola2
> >     Versatile cross-toolchain generator, with Linux-libre kernel
> > support  
> 
> Another approach would be to add support for that architecture in Guix
> and build the toolchain through that.
> 
> I remember seeing compatibility the results of test in tables with
> the combination of different versions of glibc, gcc, and binutils for
> each architecture. Typcally, with x86, combining most of the
> glibc, gcc, and binutils versions will work, whereas you might hit
> problematic bugs when you do that for other less-scrutinized
> architectures.
> 
> In any case, building a toolchain is not enough, you also need it to
> work enough to bootstrap enough userspace to be able to boot (and
> then be able to build packages and the toolchain on the booted
> system).
> 
> == Using another distro ==
> Another approach would be to use any GNU/Linux distribution (which is
> not necessarily FSDG compliant) that supports ppc64 little-endian, run
> it in qemu, and with it bootstrap a Parabola for ppc64 little-endian.
> 
> In one hand this may seem ethically acceptable[1]:
> > But there is one special case where using some nonfree software, and
> > even urging others to use it, can be a positive thing. That's when
> > the use of the nonfree software aims directly at putting an end to
> > the use of that very same nonfree software.  
> 
> But if you see it with another point of view it might be problematic
> as it would be preferable not have to run non-FSDG distributions to
> port FSDG distributions to other architectures in general. And porting
> Parabola to ppc64 little-endian will not enable to avoid running
> non-FSDG distributions when porting to a different architecture.
> 
> It would also be nice to be able to create arbitrary toolchains with
> FSDG compliant software.
> 
> References:
> -----------
> [1]https://www.gnu.org/philosophy/is-ever-good-use-nonfree-program.html
> 
> Denis.