On Wednesday, April 8, 2015, 5:28:17 AM, Laurent Vivier wrote:
Le 8 avril 2015 à 10:57, Peter Robinson <pbrobinson(a)gmail.com>
a écrit :
Hi Laurent,
Welcome, firstly please most to the list, not to me. I'm likely not
the person to answer your questions ;-)
> I'm new on the mailing list. I've a powermac G5 and I'd really like to
> improve the installer as it does not work correctly for Fedora 19/20/21.
Could you elaborate? Is it device G5 specific
I've been collecting a few PowerMac G5s and iMac G5s over the last
year, with the intent of helping improve support in Fedora.
I've also got some PowerMac G4s, an iMac G4, and PowerBook G4... but
with F21 the Power secondary arch decided to drop ppc32 support
(including the ppc64 arch multilib support for ppc32) because there
were simply not enough folks with the time and skills to handle
ppc32-specific issues (especially since these failures also caused the
ppc64 builds to fail).
I started working on the updates required for the Fedora Mac G5
install in late November. I contacted Brian Lane in the install
team, and he provided some useful direction.
It's been a slow process. I hoped to find time over the Christmas
break to work on this, but I've had less time than I anticipated,
between my "real job" being busier than expected plus "real
life"
things that needed doing. While I've been using Fedora since FC5, I am
relatively new to PPC Macs and very new to Python, and the Fedora
development process.
We've also hit some bumps on the ppc64 arch that prevented new Mac G5
installs from about mid-November until just this week - so I only had
one working setup (and needed to focus on fixing that first).
C'est la vie.
----
The Mac install story has not been healthy for a while. Around F16,
support for installing on ppc32 machines was dropped. F8 was the last
release with live CD/DVD support - which is _really_ useful for debug
of boot and video issues.
There has been a lot of churn in the primary arch installer (anaconda
rewrite, split of blivet, etc.) since release F18. At the same time,
the main ppc64 target (IBM POWER) wanted to move the boot processing
from yaboot (which has very few upstream activity) to grub2.
The install and boot process on Newworld Macs (G5 on ppc64 arch) is
different in a number of areas than the main IBM POWER process. The
boot volume must use Apple Partition Map, and a special HFS+ format
bootstrap partition that contains the bootloader. There are
also a number of special actions that need to be done to make the
final ISO disk act as a Mac boot disk.
In the F18 timeframe, the partitioning and bootstrap related code was
in anaconda and used pyparted, parted, hfsutils and other tools and
libraries. With the split into blivet and subsequent (very) active
development, there has been a lot of churn in the combined codebase.
Before F20, there were two boot stages, both handled by yaboot. In F20
the configuration was changed to move stage1 to grub2, but kept yaboot
for the Mac-specific stage2. An issue with lorax (used to build the
ISO) was found by Eric Larson, who proposed a fix and made a modified
Netinst ISO available from SourceForge (see
https://bugzilla.redhat.com/show_bug.cgi?id=1020112).
This install relies on the F20 repositories being in good shape, but
normally is quite workable. There is no working DVD ISO (for a full
offline install) for G5 Macs at this time.
F20 is the release used to build F21, and thanks to Eric's
contribution is (relatively easy) to install on Mac G5s - except for
some big-endian video driver issues (which means NVidia often needs to
be set up with acceleration turned off).
Unfortunately, there are a lot of issues that rose later.
- Yaboot is ppc32-only. That's one of the reasons that it needed to
be replaced by grub2. It depends on the ppc32 glibc and nss
libraries.
- Anaconda added the internal dependency for yaboot, but does not
check it up front with the rest of the components. It starts the
install, and if there is a dependency issue it only comes to light
when the installer tries to do the actual bootstrap setup stage.
There is no recovery if this happens.
- yaboot, and these ppc32 libraries are not present in the F21 ISO,
but the boot configuration still generates the yaboot stage 2 config
files. This means the F21 netinst is no longer bootable on a Mac
G5.
- One of the reasons Mac folks liked yaboot was that it supported dual
boot with OS X. It's unclear if grub2 does... we'll find out when
the installer is updated to correctly support grub2. I
- You can install F21 as follows:
- Install F20 (so your Apple Bootstrap partition and /boot are set up
and contain the F20 yaboot and configuration.
- Remove the F20 ppc rpms (yaboot, glibc and nss).
- Use Fedup to upgrade to F21 over the network. This gives you a
system with a working bootstrap (but where the actual F21 install
boot configuration is not used).
It's ugly, but workable until the Mac G5 boot setup gets fixed to
purely use grub2 in Anaconda and blivet.
- F20 Anaconda appears to have a bug where even if you ask to install
without updates, it still uses the updates repository.
- Issues arose with the F20 repository contents, and the process used
to create/update the F20 repository.
F20 primary arch updated Perl to 5.18.4 (security fixes), but there
was no F20 ppc64-specific build. Any primary-built noarch packages
are not rebuilt by the seconary arch, but are just copied over. The
utility used to do this does not check for noarch packages depending
on an arch-specific package.
F20 ppc64 would no longer install via Netinst due to the dependency
errors.
I opened
https://bugzilla.redhat.com/show_bug.cgi?id=1200176 and
a F20 ppc64 Perl was built and shadowed.
Just before this fix was pushed, the F20 primary glibc was updated
with security fixes. There was a new build of F20 ppc64 and ppc32
glibc and nss included in the updates repository. The netinst
process still failed with dependency errors on my G5s, this time
complaining about ppc32 glibc.
I opened
https://bugzilla.redhat.com/show_bug.cgi?id=1192824 and
after a few iterations (including fixing the F20 ppc64 repository
multilib setup). Thanks to Peter and Dan for working through this
very messy issue.
F20 being sane for multilib is important not just for the Mac G5
install, but for IBM POWER too - F20 is the last release to support
ppc32 (via multlib), and will be switched to EOL when F22 comes out.
At least now, it will be workable.
On F19, I think it was : yaboot was not installed correctly and I was
unable to boot from HDD.
On 20, too, during the install there was a dependency on yaboot that cannot be resolved
See the long tale above. You should be able to do a F20 network
install using Eric's modified netinst ISO.
On 21, a dependency on glib that cannot be resolved, which is
strange if it works on other platforms.
See above. The _ppc32_ glibc dependency of yaboot interferes with the
update of the _ppc64_ rpms to F21. Remove the yaboot-related rpms,
while leaving the actual boot files in place and you should be able to
update to F21 via fedup.
Moreover, it seems behaviour differs between text installer and
graphical installer (VNC, as "nouveau" driver seems totally broken
on powermac, an endian issue, I guess)
Web search shows that installing nouveau without acceleration works.
There are other issues that affect some Radeon video setups as well.
For example, The ATI X1900 in my G5 quad is not recognized because it
expects the video card to have an AtomBIOS in ROM. The GPU is little
endian, the CPU big endian... but the Mac X1900 ROM is Mac OS/X
specific and not AtomBIOS. There are reported to be many endian fixes
that allow the current big-endian AtomBIOS interpreter to work, but
one would have to come up with a suitable file (and load it from disk)
to get that card fully functional. There are reports by an Ubuntu user
that if one installs a recent PCI-X video card alongside the X1900,
one can run Linux on that card (and let OS/X use the X1900 in dual
boot). That report was for an X4500-vintage card - I picked up an
X6670 series card, but have had no time to play (and the F20 install
issue totally got in the way).
> How many time I have to test this Beta ?
As many times as you feel necessary on the test matrix?
F21 had a number of issues (related to ModemManager) that prevented
a clean boot. You can disable that component after install, update
and test. I ran into problems with F21 ppc64's Mate.
Try installing F22 (using fedup from F20), and see how it goes.
Opening bzs for bugs you encounter is important. Posting here about
issues is also helpful. The IRC #fedora-ppc is also useful, but there
has not been much activity lately.
> Do you know if the kernel on the ISO supports the PowerMac G5 SMU
? It is
> very unpleasant to work with 5 fan at full speed behind the desk ;-)
I've no idea what that is, that's definitely for someone
else.
Is someone else has the answer ? ;-)
For F20, you want to issue the following command as root:
modprobe i2c-powermac
You should be able to set up /etc/modprobe.d to load the correct
driver (I've not done that, as I've been focused on other areas).
Apparently the Mac G5 cooling fan support was updated in more recent
kernels with a new "Windfarm" driver. Since it's under active
development you should be able to determine the driver name and then
try to explicitly load it. If that works, then the issue is to
determine why it's not being automatically loaded. I believe other
posts talked about that - so hopefully you can make progress there.
I'm going to start work again on the install updates (using F22 as a
base), so hopefully F23 will be in good shape.
Al