Re: [PATCH net] virtio_net: fix alignment for virtio_net_hdr_v1_hash

["Michael S. Tsirkin" <mst@redhat.com>]

On Thu, Oct 30, 2025 at 10:40:13AM +0800, Jason Wang wrote:
> On Wed, Oct 29, 2025 at 4:20 PM Paolo Abeni <pabeni@redhat.com> wrote:
> >
> > On 10/29/25 2:24 AM, Jason Wang wrote:
> > > From: "Michael S. Tsirkin" <mst@redhat.com>
> > >
> > > Changing alignment of header would mean it's no longer safe to cast a
> > > 2 byte aligned pointer between formats. Use two 16 bit fields to make
> > > it 2 byte aligned as previously.
> > >
> > > This fixes the performance regression since
> > > commit ("virtio_net: enable gso over UDP tunnel support.") as it uses
> > > virtio_net_hdr_v1_hash_tunnel which embeds
> > > virtio_net_hdr_v1_hash. Pktgen in guest + XDP_DROP on TAP + vhost_net
> > > shows the TX PPS is recovered from 2.4Mpps to 4.45Mpps.
> > >
> > > Fixes: 56a06bd40fab ("virtio_net: enable gso over UDP tunnel support.")
> > > Cc: stable@vger.kernel.org
> > > Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
> > > Signed-off-by: Jason Wang <jasowang@redhat.com>
> >
> > Whoops, I replied to the older thread before reading this one.
> >
> > Acked-by: Paolo Abeni <pabeni@redhat.com>
> 
> I apologize, build will be broken since
> 
> commit b2284768c6b32aa224ca7d0ef0741beb434f03aa
> Author: Jason Wang <jasowang@redhat.com>
> Date:   Wed Oct 22 11:44:21 2025 +0800
> 
>     virtio-net: zero unused hash fields
> 
> I will prepare a new version.
> 
> Btw, it looks like there's an uAPI change that may break builds of the
> userspace:

No, I think this has not been out long enough to matter.
QEMU imports linux headers extremely quickly but it can adapt.


> diff --git a/include/uapi/linux/virtio_net.h b/include/uapi/linux/virtio_net.h
> index 8bf27ab8bcb4..1db45b01532b 100644
> --- a/include/uapi/linux/virtio_net.h
> +++ b/include/uapi/linux/virtio_net.h
> @@ -193,7 +193,8 @@ struct virtio_net_hdr_v1 {
> 
>  struct virtio_net_hdr_v1_hash {
>         struct virtio_net_hdr_v1 hdr;
> -       __le32 hash_value;
> +       __le16 hash_value_lo;
> +       __le16 hash_value_hi;
> 
> We can have a kernel only version for this but it probably means we
> need a kernel only version for all the future extension of vnet
> header?
> 
> Thanks
> 

Let's not complicate things too much please.

-- 
MST