Re: [PATCH net-next v7] virtio_net: Support RX hash XDP hint

[Jesper Dangaard Brouer <hawk@kernel.org>]

On 13/04/2024 06.10, Liang Chen wrote:
> The RSS hash report is a feature that's part of the virtio specification.
> Currently, virtio backends like qemu, vdpa (mlx5), and potentially vhost
> (still a work in progress as per [1]) support this feature. While the
> capability to obtain the RSS hash has been enabled in the normal path,
> it's currently missing in the XDP path. Therefore, we are introducing
> XDP hints through kfuncs to allow XDP programs to access the RSS hash.
> 
> 1.
> https://lore.kernel.org/all/20231015141644.260646-1-akihiko.odaki@daynix.com/#r
> 
> Signed-off-by: Liang Chen <liangchen.linux@gmail.com>
> ---
>    Changes from v6:
> - fix a coding style issue
>    Changes from v5:
> - Preservation of the hash value has been dropped, following the conclusion
>    from discussions in V3 reviews. The virtio_net driver doesn't
>    accessing/using the virtio_net_hdr after the XDP program execution, so
>    nothing tragic should happen. As to the xdp program, if it smashes the
>    entry in virtio header, it is likely buggy anyways. Additionally, looking
>    up the Intel IGC driver,  it also does not bother with this particular
>    aspect.
> ---
>   drivers/net/virtio_net.c | 55 ++++++++++++++++++++++++++++++++++++++++
>   1 file changed, 55 insertions(+)
> 
> diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> index c22d1118a133..2a1892b7b8d3 100644
> --- a/drivers/net/virtio_net.c
> +++ b/drivers/net/virtio_net.c
> @@ -4621,6 +4621,60 @@ static void virtnet_set_big_packets(struct virtnet_info *vi, const int mtu)
>   	}
>   }
>   
> +static int virtnet_xdp_rx_hash(const struct xdp_md *_ctx, u32 *hash,
> +			       enum xdp_rss_hash_type *rss_type)
> +{
> +	const struct xdp_buff *xdp = (void *)_ctx;
> +	struct virtio_net_hdr_v1_hash *hdr_hash;
> +	struct virtnet_info *vi;
> +
> +	if (!(xdp->rxq->dev->features & NETIF_F_RXHASH))
> +		return -ENODATA;
> +
> +	vi = netdev_priv(xdp->rxq->dev);
> +	hdr_hash = (struct virtio_net_hdr_v1_hash *)(xdp->data - vi->hdr_len);
> +
> +	switch (__le16_to_cpu(hdr_hash->hash_report)) {
> +	case VIRTIO_NET_HASH_REPORT_TCPv4:
> +		*rss_type = XDP_RSS_TYPE_L4_IPV4_TCP;
> +		break;
> +	case VIRTIO_NET_HASH_REPORT_UDPv4:
> +		*rss_type = XDP_RSS_TYPE_L4_IPV4_UDP;
> +		break;
> +	case VIRTIO_NET_HASH_REPORT_TCPv6:
> +		*rss_type = XDP_RSS_TYPE_L4_IPV6_TCP;
> +		break;
> +	case VIRTIO_NET_HASH_REPORT_UDPv6:
> +		*rss_type = XDP_RSS_TYPE_L4_IPV6_UDP;
> +		break;
> +	case VIRTIO_NET_HASH_REPORT_TCPv6_EX:
> +		*rss_type = XDP_RSS_TYPE_L4_IPV6_TCP_EX;
> +		break;
> +	case VIRTIO_NET_HASH_REPORT_UDPv6_EX:
> +		*rss_type = XDP_RSS_TYPE_L4_IPV6_UDP_EX;
> +		break;
> +	case VIRTIO_NET_HASH_REPORT_IPv4:
> +		*rss_type = XDP_RSS_TYPE_L3_IPV4;
> +		break;
> +	case VIRTIO_NET_HASH_REPORT_IPv6:
> +		*rss_type = XDP_RSS_TYPE_L3_IPV6;
> +		break;
> +	case VIRTIO_NET_HASH_REPORT_IPv6_EX:
> +		*rss_type = XDP_RSS_TYPE_L3_IPV6_EX;
> +		break;
> +	case VIRTIO_NET_HASH_REPORT_NONE:
> +	default:
> +		*rss_type = XDP_RSS_TYPE_NONE;
> +	}

Why is this not implemented as a table lookup?

Like:
 
https://elixir.bootlin.com/linux/v6.9-rc4/source/drivers/net/ethernet/intel/igc/igc_main.c#L6652
  https://elixir.bootlin.com/linux/latest/A/ident/xdp_rss_hash_type

--Jesper

> +
> +	*hash = __le32_to_cpu(hdr_hash->hash_value);
> +	return 0;
> +}
> +
> +static const struct xdp_metadata_ops virtnet_xdp_metadata_ops = {
> +	.xmo_rx_hash			= virtnet_xdp_rx_hash,
> +};
> +
>   static int virtnet_probe(struct virtio_device *vdev)
>   {
>   	int i, err = -ENOMEM;
> @@ -4747,6 +4801,7 @@ static int virtnet_probe(struct virtio_device *vdev)
>   				  VIRTIO_NET_RSS_HASH_TYPE_UDP_EX);
>   
>   		dev->hw_features |= NETIF_F_RXHASH;
> +		dev->xdp_metadata_ops = &virtnet_xdp_metadata_ops;
>   	}
>   
>   	if (vi->has_rss_hash_report)