On 12/01/2016 11:44, Knut Omang wrote: > On Fri, 2016-01-01 at 23:37 +0100, Paolo Bonzini wrote: >> >> On 31/12/2015 19:13, Ilya Lesokhin wrote: >>> I was able to overcome this issue by calling object_unparent on my >>> device but I’m not sure that the correct way of fixing it. >> >> Yes, it's definitely the right way to fix it. > > Sorry for the late follow-up on this one, but I had to find some more > time to spend with the code (and with valgrind too) to understand > better/verify what was going on in the qdev/qom layers. > > In the SR/IOV patch the object is created by pci_create. > Since there is no corresponding pci_delete, I assume this means that > the correct way to clean up from pci_create is simply a call to > object_unparent() as you indicate, and this is what is missing from the > patch set. > > So the full setup/teardown sequence per VF then becomes: > > pci_create(...) > <realize> > > <unrealize> > object_unparent(...)
Almost: object_unparent takes care of unrealizing. Paolo
