net/mlx5e: Fix checksum handling for non-stripped vlan packets

Now as rx-vlan offload can be disabled, packets can be received
with vlan tag not stripped, which means is_first_ethertype_ip will
return false, for that we need to check if the hardware reported
csum OK so we will report CHECKSUM_UNNECESSARY for those packets.
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/mellanox/mlx5/core/en_main.c

@@ -109,6 +109,7 @@ static void mlx5e_update_sw_counters(struct mlx5e_priv *priv)
 		s->lro_bytes	+= rq_stats->lro_bytes;
 		s->rx_csum_none	+= rq_stats->csum_none;
 		s->rx_csum_sw	+= rq_stats->csum_sw;
+		s->rx_csum_inner += rq_stats->csum_inner;
 		s->rx_wqe_err   += rq_stats->wqe_err;
 		s->rx_mpwqe_filler += rq_stats->mpwqe_filler;
 		s->rx_mpwqe_frag   += rq_stats->mpwqe_frag;

drivers/net/ethernet/mellanox/mlx5/core/en_rx.c

@@ -543,16 +543,26 @@ static inline void mlx5e_handle_csum(struct net_device *netdev,
 
 	if (lro) {
 		skb->ip_summed = CHECKSUM_UNNECESSARY;
-	} else if (likely(is_first_ethertype_ip(skb))) {
+		return;
+	}
+
+	if (is_first_ethertype_ip(skb)) {
 		skb->ip_summed = CHECKSUM_COMPLETE;
 		skb->csum = csum_unfold((__force __sum16)cqe->check_sum);
 		rq->stats.csum_sw++;
-	} else {
-		goto csum_none;
+		return;
 	}
 
+	if (likely((cqe->hds_ip_ext & CQE_L3_OK) &&
+		   (cqe->hds_ip_ext & CQE_L4_OK))) {
+		skb->ip_summed = CHECKSUM_UNNECESSARY;
+		if (cqe_is_tunneled(cqe)) {
+			skb->csum_level = 1;
+			skb->encapsulation = 1;
+			rq->stats.csum_inner++;
+		}
 		return;
-
+	}
 csum_none:
 	skb->ip_summed = CHECKSUM_NONE;
 	rq->stats.csum_none++;

drivers/net/ethernet/mellanox/mlx5/core/en_stats.h

@@ -62,6 +62,7 @@ struct mlx5e_sw_stats {
 	u64 rx_csum_good;
 	u64 rx_csum_none;
 	u64 rx_csum_sw;
+	u64 rx_csum_inner;
 	u64 tx_csum_offload;
 	u64 tx_csum_inner;
 	u64 tx_queue_stopped;
@@ -90,6 +91,7 @@ static const struct counter_desc sw_stats_desc[] = {
 	{ MLX5E_DECLARE_STAT(struct mlx5e_sw_stats, rx_csum_good) },
 	{ MLX5E_DECLARE_STAT(struct mlx5e_sw_stats, rx_csum_none) },
 	{ MLX5E_DECLARE_STAT(struct mlx5e_sw_stats, rx_csum_sw) },
+	{ MLX5E_DECLARE_STAT(struct mlx5e_sw_stats, rx_csum_inner) },
 	{ MLX5E_DECLARE_STAT(struct mlx5e_sw_stats, tx_csum_offload) },
 	{ MLX5E_DECLARE_STAT(struct mlx5e_sw_stats, tx_csum_inner) },
 	{ MLX5E_DECLARE_STAT(struct mlx5e_sw_stats, tx_queue_stopped) },
@@ -272,8 +274,9 @@ static const struct counter_desc pport_per_prio_pfc_stats_desc[] = {
 struct mlx5e_rq_stats {
 	u64 packets;
 	u64 bytes;
-	u64 csum_none;
 	u64 csum_sw;
+	u64 csum_inner;
+	u64 csum_none;
 	u64 lro_packets;
 	u64 lro_bytes;
 	u64 wqe_err;
@@ -285,8 +288,9 @@ struct mlx5e_rq_stats {
 static const struct counter_desc rq_stats_desc[] = {
 	{ MLX5E_DECLARE_STAT(struct mlx5e_rq_stats, packets) },
 	{ MLX5E_DECLARE_STAT(struct mlx5e_rq_stats, bytes) },
-	{ MLX5E_DECLARE_STAT(struct mlx5e_rq_stats, csum_none) },
 	{ MLX5E_DECLARE_STAT(struct mlx5e_rq_stats, csum_sw) },
+	{ MLX5E_DECLARE_STAT(struct mlx5e_rq_stats, csum_inner) },
+	{ MLX5E_DECLARE_STAT(struct mlx5e_rq_stats, csum_none) },
 	{ MLX5E_DECLARE_STAT(struct mlx5e_rq_stats, lro_packets) },
 	{ MLX5E_DECLARE_STAT(struct mlx5e_rq_stats, lro_bytes) },
 	{ MLX5E_DECLARE_STAT(struct mlx5e_rq_stats, wqe_err) },

include/linux/mlx5/device.h

@@ -645,7 +645,8 @@ struct mlx5_err_cqe {
 };
 
 struct mlx5_cqe64 {
-	u8              rsvd0[2];
+	u8		outer_l3_tunneled;
+	u8		rsvd0;
 	__be16		wqe_id;
 	u8		lro_tcppsh_abort_dupack;
 	u8		lro_min_ttl;
@@ -659,7 +660,7 @@ struct mlx5_cqe64 {
 	__be16		slid;
 	__be32		flags_rqpn;
 	u8		hds_ip_ext;
-	u8		l4_hdr_type_etc;
+	u8		l4_l3_hdr_type;
 	__be16		vlan_info;
 	__be32		srqn; /* [31:24]: lro_num_seg, [23:0]: srqn */
 	__be32		imm_inval_pkey;
@@ -680,12 +681,22 @@ static inline int get_cqe_lro_tcppsh(struct mlx5_cqe64 *cqe)
 
 static inline u8 get_cqe_l4_hdr_type(struct mlx5_cqe64 *cqe)
 {
-	return (cqe->l4_hdr_type_etc >> 4) & 0x7;
+	return (cqe->l4_l3_hdr_type >> 4) & 0x7;
+}
+
+static inline u8 get_cqe_l3_hdr_type(struct mlx5_cqe64 *cqe)
+{
+	return (cqe->l4_l3_hdr_type >> 2) & 0x3;
+}
+
+static inline u8 cqe_is_tunneled(struct mlx5_cqe64 *cqe)
+{
+	return cqe->outer_l3_tunneled & 0x1;
 }
 
 static inline int cqe_has_vlan(struct mlx5_cqe64 *cqe)
 {
-	return !!(cqe->l4_hdr_type_etc & 0x1);
+	return !!(cqe->l4_l3_hdr_type & 0x1);
 }
 
 static inline u64 get_cqe_ts(struct mlx5_cqe64 *cqe)