Re: [PATCH] pcie: Release references of virtual functions

[Akihiko Odaki]

On 2023/04/11 19:11, Philippe Mathieu-Daudé wrote:
> On 11/4/23 11:04, Akihiko Odaki wrote:
> > pci_new() automatically retains a reference to a virtual function when
> > registering it so we need to release the reference when unregistering.
> >
> > Fixes: 7c0fa8dff8 ("pcie: Add support for Single Root I/O 
> > Virtualization (SR/IOV)")
> > Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
> > ---
> >   hw/pci/pcie_sriov.c | 1 +
> >   1 file changed, 1 insertion(+)
> >
> > diff --git a/hw/pci/pcie_sriov.c b/hw/pci/pcie_sriov.c
> > index aa5a757b11..76a3b6917e 100644
> > --- a/hw/pci/pcie_sriov.c
> > +++ b/hw/pci/pcie_sriov.c
> > @@ -211,6 +211,7 @@ static void unregister_vfs(PCIDevice *dev)
> >               error_free(local_err);
> >           }
> >           object_unparent(OBJECT(vf));
> > +        object_unref(OBJECT(vf));
> >       }
> >       g_free(dev->exp.sriov_pf.vf);
> >       dev->exp.sriov_pf.vf = NULL;
>
> It feels the issue is at the device creation.

I added object_unref() to unregister_vfs() because dev->exp.sriov_pf.vf 
actually holds the reference. Practically there should be no difference 
as the parent bus also keeps the reference until unregister_vfs() calls 
object_unparent(), but I think it is semantically more correct to 
object_unref() in unregister_vfs().

> [/me looking at the code]
>
> What about:
>
> -- >8 --
> diff --git a/hw/pci/pcie_sriov.c b/hw/pci/pcie_sriov.c
> index aa5a757b11..fca3bf6e72 100644
> --- a/hw/pci/pcie_sriov.c
> +++ b/hw/pci/pcie_sriov.c
> @@ -150,7 +150,7 @@ static PCIDevice *register_vf(PCIDevice *pf, int 
> devfn, const char *name,
>       PCIBus *bus = pci_get_bus(pf);
>       Error *local_err = NULL;
>
> -    qdev_realize(&dev->qdev, &bus->qbus, &local_err);
> +    pci_realize_and_unref(dev, bus, &local_err);
>       if (local_err) {
>           error_report_err(local_err);
>           return NULL;
> ---