Re: [PATCH v2 2/4] virtio-iommu: Default to bypass during boot

[Cornelia Huck <cohuck@redhat.com>]

On Wed, Feb 09 2022, Eric Auger <eric.auger@redhat.com> wrote:

> Hi,
>
> On 2/9/22 12:10 PM, Cornelia Huck wrote:
>> On Tue, Feb 08 2022, "Michael S. Tsirkin" <mst@redhat.com> wrote:
>>
>>> On Tue, Feb 08, 2022 at 06:42:57PM +0100, Cornelia Huck wrote:
>>>> On Thu, Jan 27 2022, Jean-Philippe Brucker <jean-philippe@linaro.org> wrote:
>>>>
>>>>> @@ -988,9 +1025,9 @@ static void virtio_iommu_device_realize(DeviceState *dev, Error **errp)
>>>>>      virtio_add_feature(&s->features, VIRTIO_IOMMU_F_INPUT_RANGE);
>>>>>      virtio_add_feature(&s->features, VIRTIO_IOMMU_F_DOMAIN_RANGE);
>>>>>      virtio_add_feature(&s->features, VIRTIO_IOMMU_F_MAP_UNMAP);
>>>>> -    virtio_add_feature(&s->features, VIRTIO_IOMMU_F_BYPASS);
>>>>>      virtio_add_feature(&s->features, VIRTIO_IOMMU_F_MMIO);
>>>>>      virtio_add_feature(&s->features, VIRTIO_IOMMU_F_PROBE);
>>>>> +    virtio_add_feature(&s->features, VIRTIO_IOMMU_F_BYPASS_CONFIG);
>>>> Hm. In the other patch, you say that you don't support cross-version
>>>> migration (I assume across QEMU versions?)
>>> I missed that ... where does it say this?
>> In bf447d9b-c039-ccdc-f24f-ab8b56c1b196@redhat.com and follow-ups
>> (unless I misread that; maybe it's more about this concrete boundary and
>> not generally?)
>
> We were considering the virtio-iommu QEMU device currently isn't used
> for production yet, as far as we know, because we were missing the ACPI
> integration.
> So we envisionned to not care about mig subsections and just add the new
> field in the VMState.
>
> would that make sense?

If people are currently mostly playing with this device, it would not
hurt too much, I guess. (Should we document expectations somewhere?)

Adding a compat prop for this feature would be easy enough, but juggling
subsections is a bit more involved, so I see the argument for not caring
about them.

>
> Thanks
>
> Eric
>
>>
>>>> Because changing the feature
>>>> set will be guest-visible, and would need some compat handling if you
>>>> plan to support this.