Re: [PATCH v2] net/mlx5e: don't assume psp tx skbs are ipv6 csum handling

[Cosmin Ratiu <cratiu@nvidia.com>]

On Mon, 2026-01-26 at 11:38 -0800, Daniel Zahka wrote:
> mlx5e_psp_handle_tx_skb() assumes skbs are ipv6 when doing a partial
> TCP checksum with tso. Make correctly mlx5e_psp_handle_tx_skb()
> handle
> ipv4 packets.
> 
> Fixes: e5a1861a298e ("net/mlx5e: Implement PSP Tx data path")
> Signed-off-by: Daniel Zahka <daniel.zahka@gmail.com>
> ---
> This is a bug when an ipv4 tx skb passes through
> mlx5e_psp_handle_tx_skb() and tso is requested. It was previously
> undetected in my testing because my setup involves cx7's on both
> ends,
> and mlx5e_handle_csum() marks PSP rx skb's with csum_unnecessary.
> 
> To reproduce the problem just turn off NETIF_F_RXCSUM and observe:
> nstat -a | grep TcpInCsumErrors
> ---
> Changes in v2:
> - move declarations down into branches where they are used.
> - Link to v1:
> https://lore.kernel.org/r/20260123-dzahka-fix-tx-csum-partial-v1-1-7b0107693883@gmail.com
> ---
>  .../net/ethernet/mellanox/mlx5/core/en_accel/psp_rxtx.c | 17
> +++++++++++------
>  1 file changed, 11 insertions(+), 6 deletions(-)
> 
> diff --git
> a/drivers/net/ethernet/mellanox/mlx5/core/en_accel/psp_rxtx.c
> b/drivers/net/ethernet/mellanox/mlx5/core/en_accel/psp_rxtx.c
> index c17ea0fcd8ef..ef7f5338540f 100644
> --- a/drivers/net/ethernet/mellanox/mlx5/core/en_accel/psp_rxtx.c
> +++ b/drivers/net/ethernet/mellanox/mlx5/core/en_accel/psp_rxtx.c
> @@ -177,8 +177,6 @@ bool mlx5e_psp_handle_tx_skb(struct net_device
> *netdev,
>  {
>  	struct mlx5e_priv *priv = netdev_priv(netdev);
>  	struct net *net = sock_net(skb->sk);
> -	const struct ipv6hdr *ip6;
> -	struct tcphdr *th;
>  
>  	if (!mlx5e_psp_set_state(priv, skb, psp_st))
>  		return true;
> @@ -190,11 +188,18 @@ bool mlx5e_psp_handle_tx_skb(struct net_device
> *netdev,
>  		return false;
>  	}
>  	if (skb_is_gso(skb)) {
> -		ip6 = ipv6_hdr(skb);
> -		th = inner_tcp_hdr(skb);
> +		int len = skb_shinfo(skb)->gso_size +
> inner_tcp_hdrlen(skb);
> +		struct tcphdr *th = inner_tcp_hdr(skb);
>  
> -		th->check = ~tcp_v6_check(skb_shinfo(skb)->gso_size
> + inner_tcp_hdrlen(skb), &ip6->saddr,
> -					  &ip6->daddr, 0);
> +		if (skb->protocol == htons(ETH_P_IP)) {
> +			const struct iphdr *ip = ip_hdr(skb);
> +
> +			th->check = ~tcp_v4_check(len, ip->saddr,
> ip->daddr, 0);
> +		} else {
> +			const struct ipv6hdr *ip6 = ipv6_hdr(skb);
> +
> +			th->check = ~tcp_v6_check(len, &ip6->saddr,
> &ip6->daddr, 0);
> +		}
>  	}
>  
>  	return true;
> 
> ---
> base-commit: 709bbb015538dfd5c97308b77c950d41a4d95cd3
> change-id: 20260122-dzahka-fix-tx-csum-partial-952e8dc28375
> 
> Best regards,

Reviewed-by: Cosmin Ratiu <cratiu@nvidia.com>