RFC: Kill off addRamDisk

[Olaf Hering]

 On Thu, Jul 28, David Gibson wrote:

> Can anyone think of any problems with abolishing addRamDisk - the icky
> iSeries-specific method of loading an initial ramdisk.  If not, I'll
> push this to Andrew.
> 
> iSeries has a home-built method of attaching an initial ramdisk, which
> relies on an addRamDisk helper program poking the data, location and
> offset into the vmlinux after build.  Both SLES and RHEL now use the
> normal initramfs mechanisms instead of the iSeries specific initrd on
> 2.6 kernels, so let's get rid of this obsolete mechanism.

How is the initramfs vs. loop mounted file related?
Maybe I miss the point, how is the initrd content passed to the kernel
during boot after your change? pseries has its own section in zImage.initrd,
and yaboot passes just the initrd memrange in r3+r4. 

Can the $O/usr/initramfs_data.cpio.gz file linked later into the vmlinux
file? I see an '.init.ramfs' section there, perhaps you want to replace
the addRamdisk binary with an addRamdisk shell script which calls
objcopy. But this way we lose the overlay feature.
A simple 'find . | cpio -H newc --create | gzip -9 > tmp_initrd.gz'
would not work anymore as unprivileged user because mknod cant be done
that way.