[xen-tools-discuss] Defining Partitions

[Stéphane Jourdois]

Hi everybody,

Le 31/10/2010 12:01, Axel Beckert a écrit :
> JFTR: Stephane and me are both at MiniDebConf at Paris this weekend
> discussing these issues in real life. :-)
> 
> On Sun, Oct 31, 2010 at 10:13:47AM +0100, St?phane Jourdois wrote:
>> So... 2 things :
>> - the fix is wrong,
> 
> Well, not complete and a bit too strict, but not wrong. :-)

So to inform everybody, Axel and myself did see yesterday that the fix
is indeed perfectly wrong, as the sort thing was already taking place
elsewhere in the code. The fix does not harm in any way, it is only code
duplication.

Axel did manage to reproduce, and we then just understood that the
problem sits with pygrub, which we both do not know very well.

So the commit has to be reverted, and I will look further into pygrub to
understand what happens exactly. First hints :
 - removing "/boot" in the kernel line of menu.lst doest help to boot,
but it should not, as (hd0,0) is the swap partition, and /dev/xvda2 is
the root one. So pygrub is never told to use the boot partition (hd0,2
or xvda3), and still it does, which is really surprising at the moment.
 - perhaps pygrub just does not use any of this information but the
first argument to the kernel parameter ? Then how does it know which
partition to use ?
 - we have to guess in xen-create-image which partition will be used to
boot ("boot" partition if it does exist, "root" else) and give that
information to hooks (particularly 80-install-kernel but perhaps
somewhere else).

I'm off for the day and will begin to look further into this this evening.

Sorry about the mess !

Stephane.

-- 
 ///  Stephane Jourdois     /"\  ASCII RIBBON CAMPAIGN \\\
(((    Consultant securite  \ /    AGAINST HTML MAIL    )))
 \\\   157 Bd Davout         X                         ///
  \\\  75020  Paris         / \    +33 6 8643 3085    ///