[parisc-linux] Creating a bootable image.

[parisc-linux] Creating a bootable image.

[Alex deVries]

I'm mailing this because I'm not entirely sure what I'm doing is correct.

The way I understand it, the kernel we cross compile on Linux is in fact
a broken ELF32 file, which doesn't really matter, because at this point we
don't need the kernel to have any loader information itself.  My theory is
that the kernel itself can be cross compiled properly regardless of how
messed our loader formats are.

However, to actually boot you need IPL, which requires SOM linking, so
that much must be done on HPUX.

What I'm doing is to add a binary prebuilt version of IPL into the kernel
source tree, and enough glue so you can just do:

make bootable

and it'll create a kernel and boot header that should just boot right
away.  That way to do kernel work (and not bootstrap work), you don't need
HPUX at all.

Does that make sense?

- Alex

-- 
Alex deVries <puffin@redhat.com>     
Red Hat Contrib Net maintainer          
Ottawa, Canada  
PA RISC Linux Porting Team

Re: [parisc-linux] Creating a bootable image.

[Philipp Rumpf]

> I'm mailing this because I'm not entirely sure what I'm doing is correct.

Neither am I, but arch/parisc/hp/boot/README seems to say something different.

> The way I understand it, the kernel we cross compile on Linux is in fact
> a broken ELF32 file, which doesn't really matter, because at this point we
> don't need the kernel to have any loader information itself. 

I think we should build a binary image of the kernel (that is not ELF) like
we do for x86.

> However, to actually boot you need IPL, which requires SOM linking, so
> that much must be done on HPUX.

This I do not understand. According to arch/parisc/hp/boot/README, the IPL is
stored in a special binary format which we could generate from ELF like we
do for the kernel image.

> and it'll create a kernel and boot header that should just boot right
> away.  That way to do kernel work (and not bootstrap work), you don't need
> HPUX at all.

I think the vast majority of parisc boxes runs HP-UX at the moment so it
probably would not be a big win.

	Philipp Rumpf

Re: [parisc-linux] Creating a bootable image.

[Alex deVries]

On Tue, 1 Jun 1999, Philipp Rumpf wrote:

> > I'm mailing this because I'm not entirely sure what I'm doing is correct.
> Neither am I, but arch/parisc/hp/boot/README seems to say something different.

That file is actually from Jason Eckhardt's boot loader.  How do you think
it's incorrect?

> > The way I understand it, the kernel we cross compile on Linux is in fact
> > a broken ELF32 file, which doesn't really matter, because at this point we
> > don't need the kernel to have any loader information itself. 
> 
> I think we should build a binary image of the kernel (that is not ELF) like
> we do for x86.

You're right, raw binary should work just fine.

> > However, to actually boot you need IPL, which requires SOM linking, so
> > that much must be done on HPUX.
> This I do not understand. According to arch/parisc/hp/boot/README, the IPL is
> stored in a special binary format which we could generate from ELF like we
> do for the kernel image.

Hm.  Okay then.  So what we need to do is fix the cross compiling of
mkipl.  I'll have a look at this.

> > and it'll create a kernel and boot header that should just boot right
> > away.  That way to do kernel work (and not bootstrap work), you don't need
> > HPUX at all.
> I think the vast majority of parisc boxes runs HP-UX at the moment so it
> probably would not be a big win.

Being able to cross compile really is a big win.  There's a lot of people
with slower 700 series boxes.  Also, building and booting is a lot easier
if you can xcompile on your PIII.  Booting into HPUX is pretty slow.

- Alex

Re: [parisc-linux] Creating a bootable image.

[Mike Shaver]

Philipp Rumpf wrote:
> > and it'll create a kernel and boot header that should just boot right
> > away.  That way to do kernel work (and not bootstrap work), you don't need
> > HPUX at all.
> 
> I think the vast majority of parisc boxes runs HP-UX at the moment so it
> probably would not be a big win.

I think that being able to cross your kernel is a big win, since it
makes the edit-compile-test cycle much shorter if you've got a spare
machine off of which to netboot.

I've got the linker script documentation printed and waiting for my move
to safely complete, but someone who's got their hands in already should
feel free to pre-empt me.

Mike

-- 
40858.97 38157.23