[Clfs-support] ppc kernel stack overflow

[Ken Moffat]

----- Forwarded message from Ken Moffat <zarniwhoop at ntlworld.com> -----
 The fscking list list-headers got the better of me this time.
Forwarding a copy to the list.

On Mon, Dec 10, 2007 at 05:20:21PM +0100, Antonio Bulgheroni wrote:
> Dear all,
> 
>     first of all, I would like to thank all the developers for making cross
> builbing linux from scratch an easy task!  Unfortunately I'm here to report
> a problem I'm not able to fix.
> 
> A brief intro first. I'm using CLFS because I have a single board computer
> called MVME6100 with a Motorola ppc processor and I want to run linux on top
> of that. I don't have any other ppc computer, so the only duable way is
> cross-compiling. Motorola is (unofficially) distributing a patch for the
> linux kernel (2.6.14) to make it compatible with MVME6100 and using CFLS I
> managed to boot up into the temporary system and start refining the
> installation toward the final system. Instead of compiling the kernel you
> suggest, I'm using the Motorola patched one and also I'm not using UDEV.
> 
 You already have more experience than I do (otherwise you wouldn't
have got this far), so please forgive me if I suggest things you
have already looked at.

 The problems seem to happen as soon as perl's configure system gets
into the kernel headers.  I think you are probably using ARCH=ppc in
the kernel (2.6.14 was, I think, before ARCH=powerpc).
Unfortunately, your kernel is too old to provide its own sanitised
headers, so I can't suggest that you use the headers from the
patched kernel sources.

 The clfs linux-headers have had little usage on embedded ppc, so
it's possible there is a breakage hiding amongst them.  But equally,
I have to say the same thing about the older linux-libc-headers.
What is absolutely certain is that you cannot safely use newer
headers than the kernel.  So in this case, headers from 2.6.14 or
older.

 You should also look _in_detail_ at the kernel patch to see if any
of the userspace headers need to be altered, e.g. perhaps there are
new #ifdef options in the kernel headers to support this processor.
Probably, start by looking at /usr/include/{,linux/,sys/}time.h to
see what they all reference and look to see if any of it is affected
by the Motorola patch.  If it was me, I'd probably give up :-(

 Hmm, it's also possible that gcc is doing the wrong thing -
mainstream 32-bit powerpc is reasonably well tested against recent
versions of gcc-4.1 and 4.2, but a recent compiler with an old
kernel could open up a load of new problems.  Is this board merged,
or close to being merged, in recent kernels or did public
development stop at 2.6.14 ?  Maybe you could try an older version
of gcc - with the appropriate patches.  You haven't mentioned which
version of the book you started from, but I suppose clfs-1.0 is
probably as old as I can recommend for this.

ĸen
-- 
das eine Mal als Tragödie, das andere Mal als Farce

----- End forwarded message -----

-- 
das eine Mal als Tragödie, das andere Mal als Farce