Re: [Qemu-devel] [Qemu-block] Behavior of QMP "query-block"

[Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>]

On Fri, May 26, 2017 at 10:02:32PM +0200, Kevin Wolf wrote:
> Am 26.05.2017 um 18:55 hat John Snow geschrieben:
> > On 05/26/2017 07:40 AM, Kevin Wolf wrote:
> > > There is one completely crazy thing that Xen does with respect to disks.
> > > Instead of having support for their PV disks (i.e. virtio-blk, just
> > > different) in the BIOS, they add _both_ an IDE disk and a PV disk to the
> > > VM, so that the bootloader or non-PV-aware guest OSes can access the IDE
> > > disk, for which they most certainly do have drivers. As soon as a driver
> > > for the PV disk is loaded, however, that driver calls a hypervisor
> > > function that removes all the IDE disks from the VM and leaves only the

Not hypervisor. It calls the PCI device which is emulated in QEMU. That
"unplugs" the IDE.

> > > PV ones there, so that the PV-aware guest doesn't see two same disks.
> > > 
> > > I suspect that what you're seeing initially is the IDE disks, and when
> > > the PV driver is loaded, they disappear.
> > 
> > Sure, but does that mean that you'd see *no* block devices via a query
> > afterwards?
> 
> I believe xen_disk creates an anonymous BlockBackend, so it wouldn't
> appear in 'info block'. Though remembering that Xen does weird things is
> enough for me, I don't really want to know all the details that I am
> lucky enough to have forgotten. And trust me, neither do you.

Keep in mind that PV drivers are kernel drivers. Unless the qdisk driver
is used - and then you should see that as well.
> 
> Kevin
>