Re: grub-probe seems to be having problems

[Vladimir 'φ-coder/phcoder' Serbinenko]

On 05.03.2012 21:07, Lennart Sorensen wrote:
> On Sat, Mar 03, 2012 at 11:00:28PM +0100, Vladimir 'φ-coder/phcoder' Serbinenko 
> wrote:
> > On 01.03.2012 20:47, Lennart Sorensen wrote:
> > > address@hidden:~# /usr/sbin/grub-probe --device /dev/md1 --target=partmap
> > > /usr/sbin/grub-probe: error: disk `mduuid/dc00bd3a0e9491bb0b66fa8b9246c2e8' not 
> > > found.
> > Attached patch eliminates some idiosyncracies problems with mdraid
> > 1.x. Also fixes partial import problem and raid5 partial array
> > problem.
> No improvement that I can see.
>
> Is it possible there is endianess bugs in the code?
Yes, I've just fixed one. grub-fstest works on big-endian now. Couldn't 
test grub-probe since the Ubuntu on my PPC test machine is outdated.
> 1.x format is ALWAYS little endian, but I am on a big endian machine.
> 0.9 format is host endianess.
GRUB always assumes that RAID metadata is in little-endian. We don't 
have any "native-endian parsers" since we assume that grub-probe results 
are valid for GRUB runtime as well even if they run on different 
endianness. If we need any support for structures which can be either le 
or be we compile code twice, once for le, once for be (look at e.g. cpio 
code). I couldn't test the images created on big-endian platform due to 
ancient distro.
>


--
Regards
Vladimir 'φ-coder/phcoder' Serbinenko