On Tue, Mar 06, 2018 at 11:19:13AM +0100, Paolo Bonzini wrote:
> On 05/03/2018 11:43, Peter Xu wrote:
> > On Mon, Mar 05, 2018 at 04:43:09PM +0800, Liu, Yi L wrote:
> >> On Fri, Mar 02, 2018 at 05:06:56PM +0100, Paolo Bonzini wrote:
> >>> On 01/03/2018 11:33, Liu, Yi L wrote:
> >>>> + pci_device_notify_iommu(pdev, PCI_NTY_DEV_ADD);
> >>>> +
> >>>> pci_setup_sva_ops(pdev, &vfio_pci_sva_ops);
> >>>>
> >>>> return;
> >>>> @@ -3134,6 +3136,7 @@ static void vfio_exitfn(PCIDevice *pdev)
> >>>> {
> >>>> VFIOPCIDevice *vdev = DO_UPCAST(VFIOPCIDevice, pdev, pdev);
> >>>>
> >>>> + pci_device_notify_iommu(pdev, PCI_NTY_DEV_DEL);
> >>>
> >>> Please make the names longer: PCI_IOMMU_NOTIFY_DEVICE_ADDED and
> >>> PCI_IOMMU_NOTIFY_DEVICE_REMOVED. (This is independent of my other
> >>> remark, about doing this in generic PCI code for all devices that
> >>> register SVA ops).
> >>
> >> Thanks for the suggestion, will appply.
> >
> > Isn't the name too generic if it's tailored for VFIO only? Would
> > something like PCI_IOMMU_NOTIFY_VFIO_ADD be a bit better?
>
> I don't think it's for VFIO only. It's just that VFIO is the only
> caller of pci_setup_sva_ops.
Indeed. E.g., we can have emulated devices that also want to provide
the SVA ops.
--
Peter Xu
