On Thu, Jan 10, 2008 at 08:48:24PM +0000, David Laight wrote:
> On Wed, Jan 09, 2008 at 08:34:12PM +0100, Manuel Bouyer wrote:
> > Hi,
> > I'm almost done merging xen support back in arch/i386, in the same way it
> > was done for amd64 Xen supoort (there's much, much less duplicate code this
> > way). There's only spl.S and vector.S remaining, which I hope to handle this
> > evening, as well as reworking a bit the xen/i386 config files to match amd64
> > (especially, introduce a arch/xen/include/i386 directory and move some files
> > here).
> 
> When looking at the amd64 locore.S and vector.S I found that they were
> largely unreadably because they seemed to contain the amd64 and xen
> implementations of certain functions that were 98% different.
> I actually considered splitting out some of the xen code into a different
> file!

AFAIK only start and trap02 are completely different (and of course the
interrupt stubs). Others functions share a lot of code. I think, out of
locore.S and vector.S only start and the interrupt stubs could be
moved in a xen file.
One problem with this is that, when the native files are changed,
the xen files are occasionally forgottten. If all the code is in the same
file, this is less likely to happen.

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
     NetBSD: 26 ans d'experience feront toujours la difference
--