Re: Regression on powerpc-ieee2375

[Pavel Roskin]

On Thu, 2008-04-17 at 16:24 +0200, Jordi Mallach wrote:
> The menu loads correctly, and when an entry gets selected, I get:
> 
> Booting Debian GNU/Linux, linux 2.6.25-rc8-powerpc (single-user mode)
> Invalid memory access a t %SRR0:  00139d4    %SRR1: 00003030 load-size=5a474 
> adler32=50409797

I suspect this change:

* kern/elf.c (grub_elf32_load): Get the physical address after the hook
function is called, so that it's possible to change it inside the hook.

It's just because it involves ELF files, and PowerPC kernels are ELF
files.

You may want to use the git clone of the repository for bisection:
git://repo.or.cz/grub2.git

That commit is 3e29769b6e7d9dce668e719f87fd3508bdef3bcf

> I've found out the last version that works is 1.96+20080219-2, with
> latest changelog entry:
> 
> 
> 2008-02-19  Robert Millan  <address@hidden>

That's 53094f1fa0d674e155fe137eff5c8b6a4f7d33d2

> The next Debian version available is 20080228-1, with the following
> changelog entry:
> 
> 2008-02-28  Robert Millan  <address@hidden>

That's 3532b06122258d8551b7e7f37cd03c0c5d3633c8

Happy bisecting :-)

> I mentioned the grub-probe "(hd)" problem back in February, but I
> don't
> remember reaching any conclusion on what the correct behaviour should
> be. I think Pavel suggested that grub translated these unusual device
> names to add a number between OF and grub, but that's all I can
> remember. Any help to get grub-probe functional on Apple would be
> cool.

Actually, I would prefer that we retire device.map and use BIOS/firmware
data, relative paths and labels whenever possible.  That should work for
the absolute majority of the users.

Sophisticated cross-device installs should provide the needed mappings
on the command line.  But I don't think I have ever suggested that
publicly.

-- 
Regards,
Pavel Roskin