Building kernels on a different machine

[Nenad Antonic <nenad at math dot hr>]

	It is quite convenient to build debian kernel packages on one machine,
which has all the neccessary code, sources and automation, while they
are being installed on different machines. 

	The described fglrx-build.sh script, a well as the procedure suggested
by ATI, is geared towards building on the target machine.
	It might be possible to simply NFS mount /usr/src from the building
machine to target, and then apply such a procedure. However, it is not a
clean and reliable solution.

	My main problem was chasing the right kernel version. As I use Debian
kernel-package, I use (say)
VERSION = 2
PATCHLEVEL = 6
SUBLEVEL = 6
EXTRAVERSION = -mm4
and then my --append_to_version and --revision (following make-kpkg
conventions).
	In fglrx-build.sh script, EXTRAVERSION is not taken care of, while
FLAVOUR contains both that and --append_to_version argument from the
main kernel build, while later it is used as --append_to_version. 

	This gets quite confused within the scripts, where "uname" and
"/lib/modules" from the building (not target!) are being used.

	Could it be possible to explicitly give these parameters at the
beginning (in some config file?), and to skip all such tests later? It
would greatly simplify the task of automation in such a setting. 

	Does anybody (Flavio?) knows which are the variables used in ATI
provided scripts that should be set by other means, while building on a
different machine?

	Actually, I would like to modify Flavio's script to work in such a
situation. Understanding the ATI scripts is beyond me :-)

	Any help would be appreciated.

- Nenad.