Re: [Qemu-devel] [PATCH 00/30] complete conversion to hotplug-handler API

[Igor Mammedov]

On Wed, 24 Sep 2014 15:01:44 +0200
Cornelia Huck <address@hidden> wrote:

> On Wed, 24 Sep 2014 11:47:49 +0000
> Igor Mammedov <address@hidden> wrote:
> 
> > Series unifies different hotplug mechanisms to a recent
> > hotplug-handler API and does shallow conversion of
> > devices that still use legacy qdev hotplug to it dropping
> > not used after that legacy hotplug path [29/30].
> > It also relaces SCSI's own way to do hotplug/unplug with
> > hotplug-handler callbacks leaving it the only method
> > perform hotplug tasks.
> > And the last patch [30/30] allows to unplug of BUS-less
> > devices using hotplug-handler API.
> > 
> > Converted devices are covered with new hotplug
> > unit-tests, except of:
> >  s390x-*: I have no idea how or means to test it, but code
> >           is close to virtio, so it's converted using
> >           virtio template
> >  pvscsi: is broken, so no means to test it
> > 
> > Git tree for testing:
> >   https://github.com/imammedo/qemu/commits/hp_ctrl_conversion_v1
> 
> Gave it a quick test with the s390-ccw-virtio machine.
> 
> device_add virtio-balloon-ccw,id=xxx,devno=fe.2.1111
> 
> works (device added in qemu and crws injected to guest). However,
> 
> device_del xxx
> 
> removes the device in qemu but no crws are injected to the guest. That
> seems to indicate that virtio_ccw_busdev_unplug() isn't called?
> 
> <runs under gdb>
> 
> Indeed, the callback isn't invoked for hotunplug.
Cornelia,
Thanks for testing, could you share your QEMU CLI and commands you use
for testing. Since it's surprise removal, I'll probably would be able
to debug it without working guest.

BTW:
Is there a way to test it without s390 host+kvm, I've googled for
howto, but haven't found a working CLI+guest combination yet on
x86 fedora host?