Re: status of PV drivers in Xen 4.0

[Guillaume Rousse <Guillaume.Rousse@inria.fr>]

Le 12/04/2010 17:40, Andrew Lyon a écrit :
> On Mon, Apr 12, 2010 at 12:27 PM, Guillaume Rousse
> <Guillaume.Rousse@inria.fr> wrote:
>> Le 10/04/2010 14:58, Pasi Kärkkäinen a écrit :
>>>> Have you been successful building the drivers for .31 or .32?
>> I just tried 2.6.33 so far, without success.
>>
>>>> I'd like to make a set of patches available to apply to Xen 3.4.2 or
>>>> 4.0 to enable building the drivers with recent kernels and the forward
>>>> ported kernels I usually use, but while I've been able to rebuild them
>>>> on a opensuse 11.2 system running 2.6.31.12 I am finding it hard to
>>>> work out which patches need to be applied and what else the rpm build
>>>> process does to enable compilation.
>>>>
>>>> Any pointers would be much appreciated.
>>>>
>>>
>>> Did you take a look at novell's sles11 sp1 kernel? it's 2.6.32 and
>>> should have pv-on-hvm drivers included?
>> Yes, that's Jan Beulich said on this list. However, I've buid a
>> 2.6.32.11 kernel, with all xen patches available from SLE11-SP1 (as well
>> as a few others needed for applying xen patches properly), and used the
>> corresponding defconfig file. But I can't find any reference to those
>> drivers, at least with the names given in README file of xen tarball
>> (xen-platform-pci.ko, xen-vbd.ko, xen-vnif.ko and xen-balloon.ko).
>> That's probably just because of a name change.
>>
>> Anyway (unless I'm wrong), the actual usefulness of pv-on-hvm is on
>> standard linux kernels, not on dom0 itself, and the issue seems to be on
>> xen Makefile side.
>> --
> 
> I've been looking at this over the weekend and I've made some
> progress, you are correct about the usefulness of the pv-on-hvm
> drivers, however you need to bear in mind that the openSUSE/SLES11
> kernels *always* include the xen patches even if they are not
> activated for a particular kernel flavour (e.g. desktop, debug etc),
> so there are some differences in how the drivers are built,
> specifically the build process uses header files from the xenified
> kernel instead of from Xen itself.
OK, I understand the issue: you can't build those drivers against a
standard kernel, you need to have some xen patches applied, even if the
resulting options are  not activated.

Indeed, I have different build results when trying to build those
drivers against a xenified 2.6.32.11 kernel:
/home/guillomovitch/cooker/xen/xen-4.0.0/unmodified_drivers/linux-2.6/balloon/balloon.c:
In function ‘balloon_init’:
/home/guillomovitch/cooker/xen/xen-4.0.0/unmodified_drivers/linux-2.6/balloon/balloon.c:583:
erreur: implicit declaration of function ‘pagevec_init’
/home/guillomovitch/cooker/xen/xen-4.0.0/unmodified_drivers/linux-2.6/balloon/balloon.c:583:
erreur: ‘free_pagevec’ undeclared (first use in this function)
/home/guillomovitch/cooker/xen/xen-4.0.0/unmodified_drivers/linux-2.6/balloon/balloon.c:583:
erreur: (Each undeclared identifier is reported only once
/home/guillomovitch/cooker/xen/xen-4.0.0/unmodified_drivers/linux-2.6/balloon/balloon.c:583:
erreur: for each function it appears in.)
make[2]: ***
[/home/guillomovitch/cooker/xen/xen-4.0.0/unmodified_drivers/linux-2.6/balloon/balloon.o]
Erreur 1

Including "<linux/pagevec.h>" in ballon.c fixe the first error, but not
the second one.

Anyway, I can't ask the mandriva kernel team to include the whole xen
patches in all kernel just for this. I probably have to isolate the
strict subset of patches needed to make a kernel a suitable target for
those drivers.

-- 
BOFH excuse #223:

The lines are all busy (busied out, that is -- why let them in to begin
with?).