Right now we basically let the user attempt to assign any PCI device to a guest and hope it works, when there is actually quite a few caveats that we can warn them about. Unfortunately much of this requires info that libvirt doesn't currently expose. https://bugzilla.redhat.com/show_bug.cgi?id=967231 And the libvirt device assignment default is likely changing to VFIO for libvirt 1.1.4 or soon after, and there are differences between what VFIO and traditional assignment can handle.
libvirt reports iommuGroup info in the nodedev XML. To support that, we would need to tweak the UI to only allow attaching all devices in the same iommuGroup as one logical block of devices. This won't work for hotplug though so we'd need to make that bit clear.
libvirt has some reporting here with iommu in capabilities, but in practice this has never really been a problem. If someone shows up with a patch for it I'll accept it but it's not worth explicitly tracking IMO