Re: 2.6.28-rc3-git1: spitz still won't boot

[Richard Purdie <rpurdie@rpsys.net>]

On Fri, 2008-11-07 at 18:23 +0000, Russell King wrote:
> On Fri, Nov 07, 2008 at 05:57:33PM +0300, Dmitry wrote:
> > 2008/11/7 Pavel Machek <pavel@suse.cz>:
> > > On Fri 2008-11-07 21:23:41, Eric Miao wrote:
> > >> Well, IIRC spitz still needs the patch to change the vmlinux.ld.S.
> > >> Did you guys ever try that?
> > >
> > > I never heard about that patch, do you have it handy?
> > 
> > http://rpsys.net/openzaurus/patches/archive/pxa-linking-bug.patch
> > 
> > I plan to submit a bit modified version of this patch later.
> 
> That doesn't look like something that should be accepted.  Take a moment
> to put some thought into the question.  Why should we _allocate_ and
> contain the stack in the resulting image?  Does the stack contain any
> data that must be pre-initialized?
> 
> Obviously not.

Firstly, I don't think that patch should ever make it into a mainline
kernel. I can perhaps give some clues why its required though. I think
the bootloader on the zaurus truncates the image when writing the kernel
into flash using the standard flashing process. By having that much
extra padding on the end of the kernel, nothing important is lost.

How did the original 2.4 kernels ever work? Binutils used to be buggy
and left this padding in. Nobody therefore ever noticed the bug in the
bootloader.

The above theory is guesswork based on the kernels I've seen work and
not work and I wrote the patch mentioned above as a workaround a long
time ago and more pressing issues meant I never got back to it. I'd love
to see someone work out the problem for sure!

Cheers,

Richard