Re: [Xen-devel] build suggestion

[Brian Wolfe <ahzz@xxxxxxxxxxx>]

On Tue, 2004-09-21 at 02:25, Ian Pratt wrote:
> > What if the Makefile looped through all config-* files found in
> > install/boot and built a kernel based on each config? That way you could
> > mix and match kernel versions simply by having prepared configs, eg
> > 2.6.8.1-xen0, 2.6.8.1-xenU, 2.4.27-xenU.
> 
> I'd certainly like to nuke the LINUX_RELEASE mechanism and have a
> list of image types that should be built e.g. 
> "linux-2.4 linux-2.6 netbsd-2.0 freebsd".

This is sort of what I am working towards with xenlinux-builder. I've
based it off of make, so it should be easy to integrate it into the main
package as a utility program. The idea was to create and manage
individual config files for discretely separate kernel images during and
post xen compilation.

Right now it's only in a deb format, and only works with the debian
xen-kernel-patch .deb package for applying the sparse tree.
(xen-kernel-patch is a unified diff of the sparse tree patches.) I would
like to argue for the benifit of converting the sparse patches to diffs
in the build process since the sparse patch uses symlinks, which get
damaged once you start applying 3rd party patches to the resulting
kernel source tree.

 Due to the size of the diffs, I don't think we want to distribute 4+
diffs with lots of duplicate patching. Maybe we can split the diff into
the globally common portion and the kernel and version specific
portions. This would make it feasible to generate unified diffs instead
of sparse patches without bulking up the xen source tree too much.

The second reason for the unified diffs vs sparse patches is that
everyone understands them. True, everyone understands the sparse
patches, but if a distro builds both a 2.4.27 AND a 2.6.8.1 kernel, then
the sparse patching with symlinks WILL walk on each other when the
distro applies it's own patches (see the debianizing stuff of the xen
source diff for debian for the temp solution of converting to unified
diff prior to creating the kernel images).

> 
> For each of these, I guess we could see whether there are
> appropriate config file(s) present in install/boot and then loop
> building them all. If there are no suitable config files, it
> could generate some defaults.  However, this scheme wouldn't deal
> with applying different patches to each kernel.
> 
> We'd certainly welcome suggestions (and patches!) as to how to
> improve the build process and better integrate with building
> deb/rpm packages. However, we should avoid "over-automization":
> it should still be straightforward for someone who's familiar
> with running 'make xconfig' and customising their own Linux
> kernel to still do so.
> 
> I haven't had a chance to look at Brian's working in building deb
> packages, but is there stuff here which we should be pulling into
> the top-level makefile?

IMHO yes. I'm still fairly new at packaging, so don't take anythign that
I create as debian gospel. ;) I rely on Adam to brow beat me anytime I
fsck things up. *grin* Though I am getting better.  I would suggest the
first step of examining the sparse patch conversion to unified diff and
xenlinux-builder package. These are the primary weak points in making
xen portable between distros and work cleaner with packaging. It is NOT
cleanly integrated -yet-.  I need to study BK some more and determine
how to easilly manage importing the main tree patch sets into a local
repository until I get something decently stable for inclusion into the
main tree.

> 
> > Also, whatever happened to the windows XP port?
> 
> The existing windows XP port was done within the University using
> windows source code, which has meant that even binaries of the
> port can only be released to people that have signed the same
> source licence. Since there wasn't any likelihood of a more
> general release of Xen-XP being possible, we haven't kept the
> port up to date with current releases of Xen.
> 
> We'd obviously like to get XP running on Xen in a form that can
> be generally released, and are looking into a number of
> options. We hope to add Xen support for the forthcoming Intel VT
> hardware extensions, which we believe should enable Xen to
> support vanilla unmodified operating systems images.  Even with
> the extra hardware support, there'll still be an inevitable
> performance penalty of not doing proper para-virtualisation, but
> at least it will enable legacy images to be supported. We'd hope
> that all new OS installationss would use kernels containing
> para-virtualised extensions. 

This is going to take someone that is familiar with the business process
of getting MS to play ball. 8-( I don't think that I would trust them to
do anything internally just yet. In my experience and info from contacts
they aren't that great at making their stuff play ball with other
people's work. They prefer the reverse by far and thus have far more
experience (and tendancy to do it this way). 8-P

> 
> Ian
> 
> 
> -------------------------------------------------------
> This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170
> Project Admins to receive an Apple iPod Mini FREE for your judgement on
> who ports your project to Linux PPC the best. Sponsored by IBM.
> Deadline: Sept. 24. Go here: http://sf.net/ppc_contest.php
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@xxxxxxxxxxxxxxxxxxxxx
> https://lists.sourceforge.net/lists/listinfo/xen-devel



-------------------------------------------------------
This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170
Project Admins to receive an Apple iPod Mini FREE for your judgement on
who ports your project to Linux PPC the best. Sponsored by IBM.
Deadline: Sept. 24. Go here: http://sf.net/ppc_contest.php
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxxx
https://lists.sourceforge.net/lists/listinfo/xen-devel