[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: [Stretch] Status for architecture qualification



Hi Niels!

On 06/05/2016 12:01 PM, Niels Thykier wrote:
> Beyond mips64el, we are not aware of any new architectures for Stretch.
> 
> I kindly ask you to:
> 
>  * Porters, please assert if your architecture is targeting Stretch.

To give some insight what's happening in Debian Ports. We have two candidates which
I think would qualify for inclusion in a stable release. There is also a third
potential candidate for future releases of Debian once the hardware has become
available.

ppc64:

This architecture is basically on par with the release architectures. We have over
11.000 packages installed with some fluctuation due to the frequent necessary binNMUs
for the Haskell packages. Moreover, yaboot, the bootloader used on many powerpc
machines is now supporting ppc64, too. So building usable debian-installer CD images
should be possible:

> https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=322540

Although yaboot still currently FTBFS on ppc64, so someone has to look into that.

Both openSUSE and Fedora provide official installation images for ppc64, upstream
development in the toolchain and kernel is active, too.

> http://dl.fedoraproject.org/pub/fedora-secondary/releases/23/Server/ppc64/iso/
> http://download.opensuse.org/ports/ppc/distribution/13.2/iso/

To set up buildds and porterboxes, virtual machines could be set up on the same
POWER servers that are used to provide powerpc and ppc64el machines, provided
that there are still enough resources available.

sparc64:

Since sparc (32-bit userland, 64-bit kernel) was dropped after Wheezy, development
focus shifted to sparc64 which is fully 64-bit. Upstream development was recently
picked up again and both kernel and toolchain are again in active development with
many developers being employed by Oracle. They have released a reference platform
last fall:

> https://oss.oracle.com/projects/linux-sparc/

I have invested lots of time and effort to get sparc64 into a usable state in Debian.
We are close to 11.000 installed packages. Missing packages include Firefox,
Thunderbird/Icedove, golang and LibreOffice to name the most important ones. I
haven't looked into golang yet, but Firefox, Thunderbird and LibreOffice have good
chances to be working soon. LibreOffice has just some tests failing in the testsuite
thanks to James Clarke who has ported the architecture-dependent code from sparc to
sparc64. Firefox and Thunderbird require some fixing in the JavaScript engine:

> https://bugzilla.mozilla.org/show_bug.cgi?id=1275204

There are also some pending patches which fix binutils/gold and gcc-5/6 after which
even more packages should build without problems:

> https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=809509
> https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=818934
> https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=792921

Otherwise, many packages mostly fail to build from source because of some bad
programming practices which provokes unaligned access:

> https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=806208

To set up buildds and porterboxes, we would need new hardware. We have some older
SPARC Fire T2000 machines running as buildds plus one beefy SPARC-T5 (192 GiB RAM,
192 CPU threads), but those are not really suitable as official DSA machines. I
have reached out to Oracle and asked for hardware donations for that matter.

sh4:

This architecture is also known as SuperH and is actually no longer actively
developed and marketed by its developing company, Renesas.

However, there have been interesting developments for this architecture, it is
becoming open source and therefore potentially interesting for many Debian users
and developers who are concerned with privacy and free computing:

> https://lwn.net/Articles/647636/

This has become possible since - as explained in the article above - all patents
related to SuperH are expiring one after another meaning that the hardware can be
reimplemented without infringing any patents.

The current result of this effort is the J-Core project:

> http://j-core.org/

The big advantage of Super-H/J-Core over existing open source architectures like
OpenRISC is that both kernel and toolchain are already fully supported so that
Debian can be installed on these machines without any limitations.

We are currently at around 9100 installed packages, a large number of packages will
soon be able to build once I have finished bootstrapping GHC (the Haskell compiler)
which takes some time on my older SH-7785LCR hardware (still building for 14 days).


The two biggest issues with sh4 are currently with binutils and the kernel. binutils
has problems when building Qt5:

> https://sourceware.org/bugzilla/show_bug.cgi?id=17739

While the kernel needs some additional syscalls to be wired up in the interface which
prevents gtk+3.0_3.20.x being built on sh4:

> https://bugzilla.kernel.org/show_bug.cgi?id=119121

As the SH port of the Linux kernel was recently turned into maintained state again
after two developers involved with the J-Core project picked up the code, the latter
issue will be hopefully resolved soon. It's already been put onto their TODO list.

gcc upstream for SH is also still active, so both gcc-5 and gcc-6 work fine on sh4.
Before I picked up sh4, there were actually many issues in the SH backend in gcc which
prevented gcc-4.9/5/6 from being built on this architecture. With lots of patience,
debugging, bug reporting and patching, gcc-5/6 are both in a healthy and usable state.

Currently available SuperH hardware is rather slow and has usually not more than 512 MiB
of RAM, so it wouldn't make sense to set up buildds and porterboxes with what is currently
available. However, since the J-Core project is currently working on open source hardware,
it might be possible that in the future, faster hardware becomes available at affordable
prices. So it makes sense to keep an eye on sh4 and J-Core for future Debian releases.

Thanks,
Adrian

PS: If other Debian people are interested in joining our efforts to work on the sparc64
port or making a Debian port for the J-Core happen, I would be happy to provide access
to porterboxes.

-- 
 .''`.  John Paul Adrian Glaubitz
: :' :  Debian Developer - glaubitz@debian.org
`. `'   Freie Universitaet Berlin - glaubitz@physik.fu-berlin.de
  `-    GPG: 62FF 8A75 84E0 2956 9546  0006 7426 3B37 F5B5 F913

Attachment: signature.asc
Description: OpenPGP digital signature


Reply to: