Re: Native CD test results

[Pavel Roskin <proski@gnu.org>]

On Sat, 2008-03-29 at 13:13 +0200, Vesa Jääskeläinen wrote:

> Yeah... I noticed that too when I was playing with IDT's some time ago 
> and Bean found out the reason. I think we need to update this so that it 
> will never happen. Perhaps divide code to sections or so? We could 
> provide this in LD variable... hmm... I could investigate what would be 
> required for it... Good place to learn more about LD scripts.

If I understand correctly, kernel.img is a binary, so grub-mkimage won't
be able to get any linker data.

I think we could use one of the fields in the beginning of startup.S to
store the size of the part that should not be compressed.  For instance,
grub_compressed_size could be used for that, and grub-mkimage read it,
use it instead of GRUB_KERNEL_MACHINE_RAW_SIZE and then replace it with
the value it would normally put there.

That's the change in startup.S

--- kern/i386/pc/startup.S
+++ kern/i386/pc/startup.S
@@ -91,7 +91,7 @@
 VARIABLE(grub_kernel_image_size)
        .long   0
 VARIABLE(grub_compressed_size)
-       .long   0
+       .long   grub_stop - _start
 VARIABLE(grub_install_dos_part)
        .long   0xFFFFFFFF
 VARIABLE(grub_install_bsd_part)

Of course, we could put another label next to grub_stop with a more
descriptive name, like grub_uncompressed_end.

lnxboot.S needs to be changed to remove GRUB_KERNEL_MACHINE_RAW_SIZE as
well.  I don't understand at the first glance why lnxboot.S needs it.
Perhaps it could be calculated in a different way.

-- 
Regards,
Pavel Roskin