On 11/30/2015 05:07 AM, Shmulik Ladkani wrote: > Hi, > > On Wed, 25 Nov 2015 16:24:39 +0800 Jason Wang <jasow...@redhat.com> wrote: >>>>> @@ -2568,6 +2572,7 @@ static void vmxnet3_class_init(ObjectClass *class, >>>>> void *data) >>>>> c->class_id = PCI_CLASS_NETWORK_ETHERNET; >>>>> c->subsystem_vendor_id = PCI_VENDOR_ID_VMWARE; >>>>> c->subsystem_id = PCI_DEVICE_ID_VMWARE_VMXNET3; >>>>> + c->is_express = 1; >>>> Should we do this conditionally? And how about the migration >>>> compatibility? Looks like pcie device is using vmstate_pcie_device >>>> instead of vmstate_pci_device, maybe need a new property bit for this. >>> (Responding for the entire series) >>> >>> Agreed. Will limit these changes for new versions. >>> >>> What's your suggested plan? >>> Does it make sense to have a property for each change (as they are not >>> necessarily related), or is it too tedious and one property will suffice? >> Since they are not necessarily related, we'd better use a property for >> each change. > Would it make sense if we expose a new vmxnet3 type to differenciate > pcie vs pci instances of vmxnet3? > > Otherwise, migration gets more complicated, as we need to use either > vmstate_pci_device or vmstate_pcie_device; also, upon vm load, we need > to preserve the semantics saved (whether the instance was pci or pcie). > > I have managed to do so, but is a bit tedious; Exposing a new type seems > cleaner.
Yes, it's a good idea to have a new type. Thanks
