XAPI on Ubuntu 12.04
Stefano Stabellini
stefano.stabellini at eu.citrix.com
Thu Jun 9 12:43:47 UTC 2011
On Thu, 9 Jun 2011, Stefan Bader wrote:
> Hi Stefano, Mike, Jonathan,
>
> On 08.06.2011 17:36, Stefano Stabellini wrote:
> > Hi Stefan, John,
> > I would like to ask you a question regarding out of tree kernel modules
> > packaging on Ubuntu, I am not sure if you are the right people to ask
> > this question, if you are not I would appreciate if you could point me
> > in the right direction.
> >
> > Let me give you a little bit of background first: I am contacting you on
> > behalf of the main developers of XAPI (CC'ed). XAPI is an Open Source
> > enterprise management toolstack for Xen, part of the XCP project
> > (http://www.xen.org/products/cloudxen.html). They are interested in
> > having XAPI packaged on both Debian and Ubuntu, in particular Ubuntu
> > 12.04.
> > Correct me if I am wrong, but from what I have heard, thanks to the fact
> > that dom0 support is upstream in the Linux kernel, there is a good
> > possibility that Xen dom0 support will be enabled again in the next
> > Ubuntu releases.
>
> Right, will be trying to have it working with 11.10 with blkback now mainline in
> 3.0 as well.
Great to hear!
Feel free to contact me, Konrad or xen-devel if you find any issues.
> > Meanwhile the XCP guys are working hard to get XAPI packaged for Debian
> > and, if all goes well, XAPI should be in Debian Sid in time for the
> > Ubuntu 12.04 freeze.
> > As a consequence XAPI should come for free in Ubuntu 12.04 and should
> > work out of the box. Users would just need to apt-get install xen and
> > apt-get install xapi to have it working.
> >
> > However there is an issue left to be solved, and it is the reason I am
> > contacting you now.
> > In order to have good block performances XAPI relies on an Open Source
> > kernel module called "blktap". "blktap" is not currently upstream in the
> > Linux kernel and probably won't be in the foreseeable future.
> > The XCP guys have plans to move "blktap" completely in userspace,
> > removing the need for this module, but probably they won't make it in
> > time for Ubuntu 12.04.
> > So the question is: would it be possible to get "blktap" packaged in
> > Ubuntu 12.04, even if it is not upstream in Linux? I am not familiar
> > with your policy regarding kernel modules packaging, but I guess it
> > could be packaged in its own separate deb in the "universe" repository?
> > If it is possible, what are the steps involved in getting this done?
> >
> In some cases we added drivers to the main kernel package but this can lead to
> drivers getting out of sync with the changes in the kernel or user-space parts
> using it. So we rather try to keep those changes at the minimum (like drivers
> required for installation).
>
> So IMO, the preferable way would be to make use of DKMS if that works out.
> Basically the XAPI source package would create an additional DKMS binary package
> that contains the blktap kernel source, which then gets (re-)compiled on
> installation and every time a new kernel with a different ABI gets installed.
> The advantages I would see from this are:
> - the source package has control over migration. So whenever there is a user-
> space solution, the DKMS package could be made obsolete and upgrading the
> user-space side could remove it.
> - bug fixes are independent of kernel changes and can easily be synced with
> required user-space changes (if those are needed).
> - the DKMS solution can be the same for Debian and Ubuntu. So both benefit.
>
> There are already several packages out there, that use DKMS: like
> bcmwl-kernel-source, the nvidia and ati binary drivers, virtualbox-ose-dkms,
> open-vm-dkms. Those could server as an example. And there is also
>
> https://wiki.ubuntu.com/Kernel/Dev/DKMSPackaging
>
> which should be of help. At the moment I am looking at fixing up iscsitarget,
> which is exactly the case where you got a user-space side and a kernel module.
> Unfortunately it is slightly broken right now. But hopefully I have it improved
> soon.
Thank you very much, this seems like a very good solution!
We'll start looking into it ASAP.
More information about the kernel-team
mailing list