> On 30-Jun-2023, at 2:02 PM, Michael S. Tsirkin <m...@redhat.com> wrote:
>
> On Fri, Jun 30, 2023 at 01:11:33PM +0530, Ani Sinha wrote:
>>>
>>> Thus the check for unoccupied function 0 needs to use pci_is_vf() instead
>>> of checking ARI capability, and that can happen in do_pci_register_device().
>>>
>>>> Also where do you propose we move the check?
>>>
>>> In pci_qdev_realize(), somewhere after pc->realize() and before option ROM
>>> loading.
>>
>> Hmm, I tried this. The issue here is something like this would be now
>> allowed since the PF has ARI capability:
>>
>> -device pcie-root-port,id=p -device igb,bus=p,addr=0x2.0x0
>>
>> The above should not be allowed and when used, we do not see the igb
>> ethernet device from the guest OS.
>
> I think it's allowed because it expects you to hotplug function 0 later,
This is about the igb device being plugged into the non-zero slot of the
pci-root-port. The guest OS ignores it.
> no?
>
> I am quite worried about all this work going into blocking
> what we think is disallowed configurations. We should have
> maybe blocked them originally, but now that we didn't
> there's a non zero chance of regressions,
Sigh, no medals here for being brave :-)
> and the benefit
> is not guaranteed.
>
> --
> MST
>
