Andrea Bolognani <abologna@redhat.com> writes:

> On Mon, 2019-07-29 at 14:57 +0200, Sergio Lopez wrote:
> [...]
>>  /* virtio-mmio device */
>>  
>>  static Property virtio_mmio_properties[] = {
>>      DEFINE_PROP_BOOL("format_transport_address", VirtIOMMIOProxy,
>>                       format_transport_address, true),
>> +    DEFINE_PROP_BOOL("modern", VirtIOMMIOProxy, modern, false),
>>      DEFINE_PROP_END_OF_LIST(),
>>  };
>
> Not a QEMU developer so forgive me if I say something silly, but IIUC
> you'd be able to opt into the new feature by using eg.
>
>   -device virtio-net-device,modern=on
>
> However, virtio-pci devices already have a mechanism to control the
> VirtIO protocol version, where you use
>
>   -device virtio-net-pci,disable-modern=no,disable-legacy=yes
>
> to get a VirtIO 1.x-only device and
>
>   -device virtio-net-pci,disable-modern=no,disable-legacy=no
>
> for a transitional device.
>
> Can you please make sure virtio-mmio uses the existing interface
> instead of introducing a new one?

The problem here is that virtio-pci devices register an specific type
for each kind of supported device (virtio-net-pci, virtio-blk-pci...),
while virtio-mmio doesn't. This saves a lot of boilerplate, but also
implies that bus properties can't be passed through the attached device
(virtio-blk-device can't carry properties for it's virtio-mmio parent
bus).

Sergio.