Re: Command extension suggestions

[Marco Gerards]

Bean <address@hidden> writes:

> On Thu, Jun 14, 2007 at 12:49:26PM +0200, Marco Gerards wrote:
>> Well, I see the problem but I do not agree with the solution.
>> 
>> The problem for GRUB 2 is that initrd is very linux specific.  It's
>> part of a loader.  Perhaps we either have to extend loopback to load a
>> file into memory on beforehand.  Or add a memdisk disk or so.
>> 
>> One problem with initrd is that it is very architecture specific.
>> Another problem is that the initrd is unloaded when you load another
>> kernel or OS.  Besides that, reusing initrd appears hackish to me :-).
>
> I have another idea on this subject. First, We can extend the function of
> grub-mkimage so that it can handle data files. The data files embed in
> core.img can be accessed using a special device such as (ed). Then, we can
> put all necessary files, such as modules, grub.cfg and other data files in
> a single core.img. This kernel is self-sustaining, no extra file is needed
> for it to function properly.

This sounds good.  In what way do you want to embed them?  A
filesystem image (for example minixfs)?  Some simple archive?

You don't have to put modules in there, they can be made part of
core.img.

> Even if we don't create mono kernel, embeded data files can be usefully. For
> example, moddep.lst, fs.lst and command.lst, these files are best placed in
> core.img. Also, we can check the presence of embeded grub.cfg before loading
> external one, this is similar to the preset menu concept in GRUB Legacy.

The problem with a too big core.img is that it won't fit in the first
cylinder.  So we need to think carefully about what should go in there
and what shouldn't.

But making it possible to pack everything so it can be booted using
PXE or so sounds good.  There was a feature for GRUB Legacy to add a
menu.lst at compile-time.  But having something more generic and not
requiring a recompile is way better.

--
Marco