On Mon, Oct 21, 2019 at 09:19:20AM -0600, Alex Williamson wrote:
On Fri, 18 Oct 2019 22:20:40 +0200
Jens Freimann <jfreim...@redhat.com> wrote:
[...]
+ if (!pdev->net_failover_pair_id) {
+ error_setg(&vdev->migration_blocker,
+ "VFIO device doesn't support migration");
+ ret = migrate_add_blocker(vdev->migration_blocker, &err);
+ if (err) {
+ error_propagate(errp, err);
+ goto error;
+ }
+ } else {
+ pdev->qdev.allow_unplug_during_migration = true;
Why is this unique to vfio-pci, shouldn't any device set as the primary
for a failover allow unplug during migration, and therefore should it
be done in the core code rather than device driver? With the
net_failover_pair_id set in PCIDevice, I should be able to test with an
e1000e NIC as primary, but this suggests they wouldn't be handled
identically elsewhere. Thanks,
I assume you're talking about pci core code not qdev core. Failover is
pci specific at this time (only the part to hide devices is more
generic and is in qdev code). I can set the flag to allow migration in
pci_qdev_realize().
regards,
Jens
