Re: not releasing powerpc (Re: beta3 released)
On Wed, Mar 17, 2004 at 12:21:42PM +0000, Colin Watson wrote:
> On Wed, Mar 17, 2004 at 11:51:08AM +0100, Sven Luther wrote:
> > On Wed, Mar 17, 2004 at 09:31:06AM +0000, Colin Watson wrote:
> > > On Tue, Mar 16, 2004 at 09:59:14PM +0100, Sven Luther wrote:
> > > > Also, Colin said he would have a look at this this evening, so we will
> > > > know tomorrow about it.
> > >
> > > OK, I built kernel-patch-2.4.25-powerpc without your postinst hack.
> > > (Incidentally, this is just how it builds by default. If you want to
> > > have a postinst hack, it *must* be in the source package or in a
> >
> > Yeah. I had been struggling with this stuff for more than 48 hours
> > though, and also usually submit the patch to Manoj for inclusion in
> > kernel-package.
> >
> > > build-dependency, not just a local change on your system, otherwise
> > > people won't be able to rebuild it correctly. It would be very
> > > embarrassing if a kernel security update broke d-i!)
> >
> > Well, as i usually build powerpc security kernels, ...
>
> You also make it more difficult for people like me to try to help. A
> patch in debian/ that gets applied to the postinst after it's installed
Yeah, but will break as soon as kernel-package is updated, no ?
So you gain some facility in the short run, for brokeness in the long
run.
Look at : http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=230251
For example.
> would do fine if hacks are needed, although not all that robust against
> kernel-package changes: maybe just keeping your own fork of
> image.postinst in debian/ would be more sensible. That's logically what
> you're doing anyway.
Yeah. Getting ride of kernel-package would be more sensible though, at
least until kernel-package gets fixed to work in our case (post sarge
timeframe though, said Manoj).
> > > It seems to install successfully:
> >
> > Ok, can you put them somewhere i can download them (i need the modules
> > and the -chrp image) and i will give it a try.
>
> http://riva.ucam.org/~cjwatson/tmp/kernel-patch-2.4.25-powerpc/
>
> (please be nice to that, it's behind ADSL ...)
Ok downloading, will test, and if it is ok, i will prepare a new
version.
> > > Setting up kernel-image-2.4.25-powerpc-pmac (2.4.25-2) ...
> > > depmod: *** Unresolved symbols in /lib/modules/2.4.25-powerpc/kernel/drivers/sound/kahlua.o
> > > /boot/vmlinux does not exist. Installing from scratch, eh?
> > > Or maybe you don't want a symbolic link here. Hmm? Lets See.
> > >
> > > root@cairhien:/tmp# ls -l /boot/vmlinux
> > > lrwxrwxrwx 1 root root 22 Mar 17 09:11 /boot/vmlinux -> vmlinux-2.4.25-powerpc
> > >
> > > Can you clarify the problem with leaving the symlink code in? I don't
> > > seem to be seeing it. If I can reproduce it then I stand more chance of
> > > fixing it.
> >
> > This is in a d-i chroot ?
>
> Yes.
>
> Is the changelog for kernel-package 8.083 relevant here?
>
> * Bug fix: "kernel-package: silent_modules should be YES on powerpc even
> on first install of official images.", thanks to Sven
> Luther. Accomodate the official powerpc kernel images to not prompt
> even on initial install when /lib/modules/$version/ already
> exists. This allows the separation of the modules from the kernel
> image, since different powerpc subarches all require different images,
> but can share the modules. (Closes: #230251).
Yeah, it might, not entirely sure though, will check.
> > > This image puts the symlink in /boot/vmlinux. I think just /vmlinux
> >
> > No, i think not. It is often the case that /boot and / are not on the
> > same partition, especially as some OF seem not to be able to boot from
> > ext3/whatever file systems you use for /.
>
> Hmm. Yes, I forgot that yaboot doesn't look this up when you run ybin in
> the sort of way lilo does. How does grub handle this?
No idea really, i use my own build kernels on x86, and use the real
kernel name, so ...
> > > would be better, since that's what kernel-package does by default so
> > > it'll be more compatible with people's locally-built kernels: my
> > > locally-built kernels have been using /vmlinux without me giving any
> > > special directions. However, if it stays as /boot/vmlinux then
> > > yaboot-installer needs to be told about it.
> >
> > Do you know what filesystems yaboot is able to boot from.
>
> [cjwatson@cairhien ~/src/packages/yaboot/yaboot-1.3.11/second]$ ls fs_*
> fs_ext2.c fs_iso.c fs_of.c fs_reiserfs.c fs_xfs.c
Ok. do you know if fs_ext2 supports also ext3 filesystems not cleanly
unmounted ?
> (fs_of.c says "OpenFirmware supported filesystems".)
>
> > Also remember that not everyone uses yaboot. What about quik for
> > example ?
>
> The fact that other boot loader installers may have to be changed to
> have kernel symlinks in /boot doesn't alter the fact that
> yaboot-installer has to be changed. :)
>
> We don't have a quik-installer, anyway ...
We should though.
> > > > I think maybe it would be more interesting to fix yaboot-installer
> > > > instead so it doesn't rely on the symlink being present, but fall back
> > > > to using the real kernel in case the symlink is missing, this would
> > > > add more robustness to the whole process.
> > >
> > > I'm not keen on that idea. The resulting system wouldn't be upgradeable
> > > as smoothly if its yaboot.conf hardcoded the kernel-image version rather
> > > than using the symlink, which would be an unexpected kick in the teeth
> > > for people doing kernel security updates.
> >
> > Well, you could naturally have kernel-image postinst upgrade that file,
> > as i guess it is done on native grub/lilo systems on x86. Not sure
> > though.
>
> No, it's not. lilo.conf uses /vmlinuz and is not altered by the
> kernel-image postinst. grub is the same. Currently, yaboot works this
> way too (although with /vmlinux rather than /vmlinuz, of course), and
> it's excellent.
>
> I'm not aware of any architecture where the boot loader's configuration
> file is rewritten at install time. It would be pretty ugly.
>
> > Anyway, current kernel-package is utterly broken with regard to powerpc,
> > for example it doesn't know how to detect if you are on a old world or a
> > new world machine, and thus if you need yaboot or quik.
>
> Right, it detects this at build-time rather than run-time, which is
> wrong since the pmac kernels work on both (as I understand it). How come
> there's no bug filed about this?
because nobody cares, and kernel-package stuff is utterly broken on
powerpc right now.
> > What are the exact requirement of yaboot in this area ?
>
> What do you need beyond the above?
No idea, which is why i am asking.
Also keep in mind that yaboot also is used on IBM RS6K boxes.
> > And as said, i would very much like to get ride of kernel-package, and
> > do it by hand.
>
> Ugh. I don't think getting out of step with all the other architectures
> is the way forward.
Well, we have a kernel-module package, which should do all the module
related postinst things, and a kernel-image which should set the kernel
symlink. I don't really know we need all the rest of the complex stuff
in kernel-package, and kernel-package is too complex to easily fix for
me. I have been askind Manoj about this since i first took over the
powerpc kernel package, and he said he won't do this until the post
sarge time. So, i have been forced to do ugly workarounds.
And i don't speak perl, but you are welcome to give a hand to fix this
issue properly, if you feel like that.
Also, kernel-package is not debconfized, so it will break in d-i on
problematic cases.
Friendly,
Sven Luther
Reply to: