On Mon, 2009-03-23 at 21:25 -0300, Otavio Salvador wrote: > On Mon, Mar 23, 2009 at 7:38 PM, Frans Pop <elendil@planet.nl> wrote: > > On Monday 23 March 2009, Ian Campbell wrote: > >> On Mon, 2009-03-23 at 23:05 +0100, Frans Pop wrote: > >> > On Monday 23 March 2009, Ian Campbell wrote: > >> > > IIRC I was asked to add direct to i386 instead of adding to kernel > >> > > wedge. Maybe that was because i386 was the only one back then > >> > > though and for two arches kernel-wedge makes sense. The only other > >> > > (currently) potential arch is ia64 though. > >> > > >> > IMO adding them to kernel wedge does not make sense as long as that > >> > results in the modules being added in generic kernel udebs (such as > >> > nic-modules) for arches where they are sure to be unused. > >> > >> The modules aren't even built on architectures where Xen is not > >> supported or enabled in the kernel .config, so there is no danger of > >> them getting into any udebs where they don't belong. The modules are > >> marked as optional for this reason, see for example the 486 kernel > >> which doesn't have Xen enabled in it, and as expected there are no Xen > >> modules in any 486 udeb despite them being present in the 386 specific > >> udeb lists. > > > > You mentioned IA64. As long as we have no D-I support for Xen for that, > > adding the udebs in kernel-wedge would result in them being uselessly > > being included there. I'm unable to test IA64 myself but having the Xen modules available isn't useless to people who can. AFAIK there is no reason to expect it won't work as well as it does on i386/amd64. > > Also, for i386 you have the split between the 486 kernel and the Xen > > kernel for D-I, but that's not true for amd64. So for amd64 the Xen > > modules would end up in all images. > > Sure we really care about them being included on amd64 images? > > amd64 machines are powerful enough to the increase of memory > due those module be mostly ignored and about ia64 I do believe > we'll end up supporting them once kernel does it properly. > > I see i386 case mostly a workaround due our backward compatibility > for older machines. I do agree that this is really important, so > makes sense to provde a specific image for xen. This isn't the case > for amd64 it doesn't makes sense for me. > > Only con I see about having it in all amd64 images is that we'll > need to be carefully how to proper document it. Besides that, I > see no problem with that. Agreed, there should be no need for a separate Xen variant on amd64. I'd still like to add the netboot/xen directory but populate it with symlinks to the relevant ../ paths to get the right generic files, this keeps the path to a Xen kernel is consistent across i386 and amd64 (since tools are being built which know about the path) and there is somewhere to put the xm-debian.cfg file. I still need to work out how to do that in the d-i build system though. In terms of having the modules in all images, I've recently begun looking into what it would take to support Xen domU installation from CD (and/or maybe DVD). I'm still in the early stages (waiting for my mirror to download so I can investigate how debian-cd works) so I don't really have concrete proposal yet. It should be easy for amd64 since no extra files are required above the 2 driver modules. For i386 things might be a little trickier due to the different kernel variant being required which certainly won't fit on the normal CD-1 or DVD-1. Ian. -- Ian Campbell Decorate your home. It gives the illusion that your life is more interesting than it really is. -- C. Schulz
Attachment:
signature.asc
Description: This is a digitally signed message part