Re: [net] cxgb4: advertise NETIF_F_HW_CSUM

[Alexander Duyck <alexander.duyck@gmail.com>]

On Tue, Jan 5, 2021 at 9:01 AM Rohit Maheshwari <rohitm@chelsio.com> wrote:
>
> advertise NETIF_F_HW_CSUM instead of protocol specific values of
> NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM. This change is added long
> back in other drivers. This issue is seen recently when TLS offload
> made it mandatory to enable NETIF_F_HW_CSUM.
>
> Fixes: 2ed28baa7076 ("net: cxgb4{,vf}: convert to hw_features")
> Signed-off-by: Rohit Maheshwari <rohitm@chelsio.com>
> ---
>  drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c | 5 ++---
>  1 file changed, 2 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
> index 7fd264a6d085..f99f43570d41 100644
> --- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
> +++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
> @@ -6831,14 +6831,13 @@ static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
>                 netdev->irq = pdev->irq;
>
>                 netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
> -                       NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
> +                       NETIF_F_HW_CSUM |
>                         NETIF_F_RXCSUM | NETIF_F_RXHASH | NETIF_F_GRO |
>                         NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
>                         NETIF_F_HW_TC | NETIF_F_NTUPLE;
>
>                 if (chip_ver > CHELSIO_T5) {
> -                       netdev->hw_enc_features |= NETIF_F_IP_CSUM |
> -                                                  NETIF_F_IPV6_CSUM |
> +                       netdev->hw_enc_features |= NETIF_F_HW_CSUM |
>                                                    NETIF_F_RXCSUM |
>                                                    NETIF_F_GSO_UDP_TUNNEL |
>                                                    NETIF_F_GSO_UDP_TUNNEL_CSUM |

If you are going to enable the feature you should fully enable the
feature. My concern is the "nocsum:" label in hwcsum(). If you are
going to say you support the feature you should at least look at
dealing with the exception case and process a software checksum via
skb_checksum_help() rather than just not bothering and "hope a bad
packet is detected".

- Alex