RE: [PATCH v11 3/4] virtio-net: define UDP tunnel segmentation offload feature

[Cornelia Huck <cohuck@redhat.com>]

On Sun, Jan 19 2025, Parav Pandit <parav@nvidia.com> wrote:

> Hi Paolo, Michael,
>
>> From: Paolo Abeni <pabeni@redhat.com>
>> Sent: Wednesday, November 27, 2024 3:07 PM
>> 
>> 
>> +\item[VIRTIO_NET_F_GUEST_UDP_TUNNEL_GSO (46)] Driver can receive
>> GSO
>> +packets
>> +  carried by a UDP tunnel.
>> +
>> +\item[VIRTIO_NET_F_HOST_UDP_TUNNEL_GSO (48)] Device can receive
>> GSO
>> +packets
>> +  carried by a UDP tunnel.
>
> Feature bits 42 to 49 are not applicable to device specific type. They are reserved.
> Please see the note. 
>
> Current spec description is:
>
> 0 to 23, and 50 to 127 Feature bits for the specific device type
> 24 to 41 Feature bits reserved for extensions to the queue and feature negotiation mechanisms, see 6
> 42 to 49, and 128 and above Feature bits reserved for future extensions.
>
> So 2 bits of these patch and of patch 4, 46 to 49 to be shifted to bits 65 to 68.
>
> Or an alternative would be to amend,
>
> Bits 46 to 127 for specific device type,
> And 42 to 45 and 128 reserved for future extensions.
>
> Michael,
> Please let me know so that I can have the follow up patch to this one as voting is started.
> And the fix is required for it.
>
> Is there any historic reason for 42-49 as reserved, or it can be used as proposed in this patch?
> If it was only kept as elastic buffer for device generic, and device type to grow it is probably ok.
> Using 46 to 49 has small advantage to reuse in the offloads bit as below.
> Please confirm.

See the reasoning given in 88895f56e642 ("Reserve more feature bits for
device type usage") -- the idea was to push out the point in time when
generic feature bits will have to be in the bit area beyond 63. Given
that we only used one of those generic bits in the meantime, I assume it
will still be some time before we run out of bits, even if we let the
net driver grab some bits there. Still, I'd probably prefer that the net
device used different bits. [And issues like this are easy to miss
during review unfortunately.]

However, if the device was to use different bits, the clean solution
would be to withdraw, respin, and revote. If we only changed the
reserved ranges, we'd be fine with a change on top, so that would be
less hassle in the short team. If we decide that we can live with the
smaller feature bit space below 63, I'd be fine with that solution.

>
>
>> @@ -1862,6 +2068,7 @@ \subsubsection{Control
>> Virtqueue}\label{sec:Device Types / Network Device / Devi
>>  #define VIRTIO_NET_F_GUEST_TSO6       8
>>  #define VIRTIO_NET_F_GUEST_ECN        9
>>  #define VIRTIO_NET_F_GUEST_UFO        10
>> +#define VIRTIO_NET_F_GUEST_UDP_TUNNEL_GSO  46
>>  #define VIRTIO_NET_F_GUEST_USO4       54
>>  #define VIRTIO_NET_F_GUEST_USO6       55