Mailing List Archive

Summary of PPC team meeting
On Fri, March 18, 20UTC until 22UTC the ppc-team had a meeting about those

1) project management elections
2) future goals
3) AltiVec in portage
4) QEMU ABI wrapper
5) Kernel issues
6) LiveCD
7) ppc64 and ppc merger

This is a summary of the topics:

1) project management elections

There should be lead-elections every six months, say after a release, or when
it is needed. You can volunteer for the Operational or Strategical Lead (or
both). You have to write your plans and tell them to the team. Then a voting
will take place via email. You can vote for one or several candidates. If
there are two with the same ammount of votes, there will be a further election
between those two. If you ran for both Operational and Strategical Lead you
need to declare which one is your first choice.

2) future goals

Instead of project goals posted with the beginning of a year, the ppc team
want to release quarterly updates about the goals. An example of those could
be found at the Gentoo Documentation Project. The Strategical Lead has to
track the goals and is responsible for their publication.

The current goals for 2005.0 are:
- OldWorld Support for 2005.0
- XLiveDVD for ppc
- Support of more machines
- compatibility list of PowerPCs with Gentoo running
- merge pegasos-sources into gentoo-(dev-)sources
- (better) support for CHRP/PReP
- support the installer
- better AltiVec support

And the all-time goals are:
- stable portage-tree
- minimized buglist
- at least two releases in a year
- having a nice time in the team :)

To keep a minimized buglist we want to introduce ppc-bugdays again. They
should take place monthly or even more often when it's needed. Best for devs
would be around Saturday afternoon (depending on their timezone).

The next ppc-bugday is on the same day like the usual bugday: April 2. To
find good users who may become a dev, we will talk about the bugs in

3) AltiVec in portage

Some poitns about AltiVec:
1. gcc-4 will have an autovectorizer
2. the linux kernel will have an improved AltiVec support (avoiding vrsave on
non AltiVec tasks)
3. AltiVec abi for non AltiVec code is relatively slow

And there are two plans to resolve the problems:
1. keep wide AltiVec support and suggest to use -mabi=altivec everywhere and
maybe use the autovectorizer
2. restrict altivec to the AltiVec enable packages and make the others use the
non-AltiVec abi (which includes work on proper flag-o-matic support)
3. work on an AltiVec enabled glibc

We decided to work on 2 and 3.

4) QEMU ABI wrapper

qemu is nice. lu_zero wants to a binfmt hack, so that we can run
non-ppc-applications on ppc. But he can't do that alone and so he is
searching for other devs. It's a nice attempt, but we have higher prioritated
situations to solve.

5) Kernel issues

We found out recently, that the 2.6.11-kernel produce several problems on ppc.
At least three could be named:
- RTC/hwclock hickups on several machines
- linker problems with special kernel configs (a guess might be the highmem
- compile-problems on G5 32bit

These made us to use a 2.6.10-kernel with 2.6.11-patches backported for the
release. In the long term the G5 32bit-kernel support will vanish, so that we
only support ppc32 on 32-bit-machines.

6) LiveCD

The release's LiveCD is about to be finished. Currently only the kernel
causes problems, as not all patches has been commited to CVS (nb: at time of
writing the summary this isn't an issue any more). G3G4-support works well,
and the G5 has also been tested on two machines. The Pegasos-kernel causes
some headaches, but there is at least one compiled kernel which can boot quite
properly. Stages has been tested quite well, GRP not, but that isn't such an

Thanks go to all supporters for this release!

7) ppc64 and ppc

With the dwindling support for the 32-bit G5 kernel becoming more apparent and
the 64-bit kernel support under rapid development tgall and dostrow would
like to propose a partial merger of the ppc and ppc64 teams for the following
purposes: all ppc64 hardware is able to run with 3 different userlands, a
pure 64-bit environment as we have now, a 64-bit kernel with a 32-bit
userland, which is very easy to create, and a multilib environment, which as
soon as dostrow gets finished, making hardened work on ppc64. Two of the
three of these configs allow regular 32-bit ppc binaries to run on the system
and the ppc64 team lacks the manpower to handle troubleshooting both 64 and 32
bit issues.

There was no objection during the meeting from noone of the attendees by the
ppc and ppc64 team.

That means, we will work an a common-profile with sub-profiles for ppc and
ppc64. A first target-date would be the 2005.1 release. The masterplan[tm]
for the team-merge should be presented at the next meeting (in about four