Re: [PATCH] net/vhost: get csum offload capabilities of vhost backend

[Maxime Coquelin <maxime.coquelin@redhat.com>]

Hi Wenwu,

On 2/17/22 16:16, Wenwu Ma wrote:
> The current vhost backend lacks csum offloads information,
> which will cause testpmd command such as "csum set tcp hw
> <port_id>" to fail. This patch adds the information according
> to the device features.
> 
> Signed-off-by: Wenwu Ma <wenwux.ma@intel.com>
> ---
>   drivers/net/vhost/rte_eth_vhost.c | 18 ++++++++++++++++++
>   1 file changed, 18 insertions(+)
> 
> diff --git a/drivers/net/vhost/rte_eth_vhost.c b/drivers/net/vhost/rte_eth_vhost.c
> index 070f0e6dfd..7593d5a9ae 100644
> --- a/drivers/net/vhost/rte_eth_vhost.c
> +++ b/drivers/net/vhost/rte_eth_vhost.c
> @@ -1281,6 +1281,24 @@ eth_dev_info(struct rte_eth_dev *dev,
>   				RTE_ETH_TX_OFFLOAD_VLAN_INSERT;
>   	dev_info->rx_offload_capa = RTE_ETH_RX_OFFLOAD_VLAN_STRIP;
>   
> +	if (internal->vid != -1) {
> +		uint64_t features = 0;
> +		if (rte_vhost_get_negotiated_features(internal->vid, &features) != 0)
> +			return 0;
> +
> +		if (features & (1ULL << VIRTIO_NET_F_CSUM)) {
> +			dev_info->tx_offload_capa |= RTE_ETH_TX_OFFLOAD_TCP_CKSUM |
> +						RTE_ETH_TX_OFFLOAD_UDP_CKSUM |
> +						RTE_ETH_TX_OFFLOAD_IPV4_CKSUM;
> +		}
> +
> +		if (features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
> +			dev_info->rx_offload_capa |= RTE_ETH_RX_OFFLOAD_TCP_CKSUM |
> +						RTE_ETH_RX_OFFLOAD_UDP_CKSUM |
> +						RTE_ETH_RX_OFFLOAD_IPV4_CKSUM;
> +		}
> +	}
> +
>   	return 0;
>   }
>   
> 

This patch has lots of gaps, since the negotiated Virtio features can
change if the guest driver decides so, so the exposed ethdev offload
capabilities may not represent what is really supported by the guest
driver.

I have done a series that handles this issue by implementing SW
fallbacks in case of misalignment between host application and guest
application:

http://patches.dpdk.org/project/dpdk/cover/20220505102729.821075-1-maxime.coquelin@redhat.com/

Please help reviewing & testing the series if possible.

Thanks,
Maxime