Re: Fwd: Re: Force location of GRUB's boot and core images ?

[Pascal Hambourg]

Le 09/04/2017 à 16:46, Xen a écrit :
> I forgot to CC this to you.

Thanks for resending. It makes me easier to reply.

> Pascal Hambourg schreef op 09-04-2017 13:44:
>
> > I just tried it. grub-install executed without any error, but
> > chainloading the partition boot sector leaves me with a blank screen
> > after the "boot" command. bootinfoscript reports that grub2 is
> > installed in the boot sector of the partition and looks at sector 1 of
> > the same hard disk (which contains the GPT header) for core.img.
>
> I guess you mean that boot.img will have (contain) an offset relative to
> what it thinks is the entire disk (that was actually a partition
> loopback mounted to appear as a single disk) and so upon activation will
> not know that it was installed on a partition and will not know how to
> use a relative offset from where boot.img was installed?

I guess so.

> > I guess that proper booting would require that the boot image looks at
> > the partition offset+1 instead.
>
> Which it can't know because it thought it was sitting on a complete
> disk, instead, unless it uses relative offsets.

Well, grub-install could have been smart enough and convert the virtual 
positions in the loop device into the real sector positions on the 
underlying disk. After all, this it what it does when installing on a 
regular partition. A partition is just another kind of virtual device.

But, thinking of it, that would prevent using a loop device to create a 
bootable image and write that image on a real boot device. I guess we 
can't have both.