Description of problem:

If an attempt is made to unbind a device from vfio-pci while it is in use by a guest, the kernel will block until the device is released by the guest.  Kernel bug 1206275 adds a notification mechanism when such an attempt is made to relay the request to userspace.  This change makes the corresponding change in QEMU to receive and act on that notification.  With this change, a guest that supports hot-unplug of devices should automatically release the device when an unbind attempt is made.

47cbe50 vfio-pci: Enable device request notification support

Version-Release number of selected component (if applicable):


How reproducible:


Steps to Reproduce:
1.
2.
3.

Actual results:


Expected results:


Additional info:

For QE, I would try the following tests:

 - With an 82576 VF assigned to a guest, attempt to remove the igb driver in the host, via modprobe -r.  The VF should automatically be unplugged from the VM to allow this to complete.
 - Same as above except modify the number of SR-IOV VFs.  When set to 0, the VF should be automatically removed from VM.
 - Manually echo the VF PCI device to the vfio-pci driver unbind file in sysfs.

Please note that this is a cooperative interface, if the guest is running a guest that does not support hotplug, is unwilling to release the device, or is sitting in firmware, the existing behavior to block the unbind/remove will continue.