On 2014-10-25 13:36, Jan Kiszka wrote:
> On 2014-10-21 00:34, Knut Omang wrote:
>> diff --git a/hw/pci/pci_bridge.c b/hw/pci/pci_bridge.c
>> index 40c97b1..e6832c4 100644
>> --- a/hw/pci/pci_bridge.c
>> +++ b/hw/pci/pci_bridge.c
>> @@ -376,8 +376,14 @@ int pci_bridge_initfn(PCIDevice *dev, const char *typename)
>>      sec_bus->address_space_io = &br->address_space_io;
>>      memory_region_init(&br->address_space_io, OBJECT(br), "pci_bridge_io", 65536);
>>      br->windows = pci_bridge_region_init(br);
>> +
>>      QLIST_INIT(&sec_bus->child);
>>      QLIST_INSERT_HEAD(&parent->child, sec_bus, sibling);
>> +
>> +    if (dev->bus->iommu_opaque) {
>> +        pci_setup_iommu(sec_bus, dev->bus->iommu_fn, dev->bus->iommu_opaque);
>> +    }
>> +
> 
> So, if I followed the discussion in the cover-letter thread correctly,
> this should rather move into the bridge device init functions because
> the PCI[e]-PCI bridge ("pci-bridge") would not call it, right?

Not right. We need the setup in any case (except for the virtio bridges
I'm currently thinking of for encapsulating non-translatable virtio
devices). But something still has to change to reflect the requester ID
aliasing of the PCIe-PCI bridge, no?

Jan