[Libguestfs] [PATCH 7/7] v2v: add support for SUSE VMDP drivers

[Roman Kagan]

On Tue, Apr 05, 2016 at 05:38:40PM +0200, Cedric Bosdonnat wrote:
> On Tue, 2016-04-05 at 16:16 +0300, Roman Kagan wrote:
> > On Tue, Apr 05, 2016 at 03:08:48PM +0200, Cedric Bosdonnat wrote:
> > > On Tue, 2016-04-05 at 15:52 +0300, Roman Kagan wrote:
> > > > On Tue, Apr 05, 2016 at 01:47:33PM +0200, Cédric Bosdonnat wrote:
> > > > > Note that VMDP's block driver pvvxblk depends on the ballooning
> > > > > driver
> > > > > pvvxbn.
> > > > 
> > > > Can you please elaborate on this?  Is the balloon driver *used*
> > > > by
> > > > the
> > > > storage driver so that the latter won't work without the former? 
> 
> That's correct.  The current design requires that the balloon driver be
> loaded. The pvvxblk driver is a single binary that runs on both KVM and
> Xen.  It determines at runtime which environment it is running on and
> acts accordingly.  The dependency is required for Xen.  This also
> creates the dependency for KVM. 
> 
> > > >  What
> > > > will happen if the virtio-balloon device isn't configured in the 
> > > > VM?
> 
> The balloon driver will not load and hence the block driver will not
> load.
> 
> > > It's just that without the virtio balloon driver installed, the
> > > virtio
> > > block one won't start properly, resulting in the famous 0x7B error.
> > 
> > Does it also require the presence of the balloon device?
> 
> Yes.  This the default installation.  The only way to remove the
> balloon device is to launch the vm by constructing a manually
> configured command line.

I don't know what management software you use on top of qemu; in libvirt
it's certainly possible to remove a balloon device.

> Weather the balloon device is present or not,
> the user is able to set the current and maximum memory allocations
> differently.  If different and the balloon device/driver is not
> present, the vm will undoubtedly crash at some point. 

I see no reason for that, but that's out of scope of this discussion.

I got your point.  Looks like the patchset to cut down the amount of
regedits which I was about to post (I'm finishing testing it on various
Windows versions) won't work for you as is because it assumes only
single-file no-dependency storage drivers.  I'll post it anyway; we'll
see if it can be adjusted to cover your case, too.

Roman.