Add support for receive side scaling stride, RSSS, in mlx5en(4).

The receive side scaling stride parameter is a value which define the interval
between active receive side queues. The traffic for the inactive queues is
redirected to the nearest active queue by use of modulus. The default value
of this parameter is one, which means all receive side queues are used.

The point of this feature is to redirect more traffic to fewer receive side
queues in order to take more advantage of sorted large receive offload,
sorted LRO. The sorted LRO works better when more packets are accumulated
per service interval.

MFC after:		3 days
Approved by:		re (marius)
Sponsored by:		Mellanox Technologies

3 files changed, 28 insertions, 4 deletions

diff --git a/sys/dev/mlx5/mlx5_en/en.h b/sys/dev/mlx5/mlx5_en/en.h
index eff172b55cdc..73f0268ca270 100644
--- a/sys/dev/mlx5/mlx5_en/en.h
+++ b/sys/dev/mlx5/mlx5_en/en.h
@@ -454,6 +454,7 @@ struct mlx5e_params {
 	u32	rx_priority_flow_control __aligned(4);
 	u16	tx_max_inline;
 	u8	tx_min_inline_mode;
+	u8	channels_rsss;
 };
 
 #define	MLX5E_PARAMS(m)							\
@@ -462,6 +463,7 @@ struct mlx5e_params {
   m(+1, u64 tx_queue_size, "tx_queue_size", "Default send queue size")	\
   m(+1, u64 rx_queue_size, "rx_queue_size", "Default receive queue size") \
   m(+1, u64 channels, "channels", "Default number of channels")		\
+  m(+1, u64 channels_rsss, "channels_rsss", "Default channels receive side scaling stride") \
   m(+1, u64 coalesce_usecs_max, "coalesce_usecs_max", "Maximum usecs for joining packets") \
   m(+1, u64 coalesce_pkts_max, "coalesce_pkts_max", "Maximum packets to join") \
   m(+1, u64 rx_coalesce_usecs, "rx_coalesce_usecs", "Limit in usec for joining rx packets") \
diff --git a/sys/dev/mlx5/mlx5_en/mlx5_en_ethtool.c b/sys/dev/mlx5/mlx5_en/mlx5_en_ethtool.c
index d85afb570872..85b1fe85617f 100644
--- a/sys/dev/mlx5/mlx5_en/mlx5_en_ethtool.c
+++ b/sys/dev/mlx5/mlx5_en/mlx5_en_ethtool.c
@@ -493,6 +493,24 @@ mlx5e_ethtool_handler(SYSCTL_HANDLER_ARGS)
 			mlx5e_open_locked(priv->ifp);
 		break;
 
+	case MLX5_PARAM_OFFSET(channels_rsss):
+		/* network interface must be down */
+		if (was_opened)
+			mlx5e_close_locked(priv->ifp);
+
+		/* import number of channels */
+		if (priv->params_ethtool.channels_rsss < 1)
+			priv->params_ethtool.channels_rsss = 1;
+		else if (priv->params_ethtool.channels_rsss > 128)
+			priv->params_ethtool.channels_rsss = 128;
+
+		priv->params.channels_rsss = priv->params_ethtool.channels_rsss;
+
+		/* restart network interface, if any */
+		if (was_opened)
+			mlx5e_open_locked(priv->ifp);
+		break;
+
 	case MLX5_PARAM_OFFSET(channels):
 		/* network interface must be down */
 		if (was_opened)
@@ -1041,6 +1059,7 @@ mlx5e_create_ethtool(struct mlx5e_priv *priv)
 	priv->params_ethtool.tx_queue_size = 1 << priv->params.log_sq_size;
 	priv->params_ethtool.rx_queue_size = 1 << priv->params.log_rq_size;
 	priv->params_ethtool.channels = priv->params.num_channels;
+	priv->params_ethtool.channels_rsss = priv->params.channels_rsss;
 	priv->params_ethtool.coalesce_pkts_max = MLX5E_FLD_MAX(cqc, cq_max_count);
 	priv->params_ethtool.coalesce_usecs_max = MLX5E_FLD_MAX(cqc, cq_period);
 	priv->params_ethtool.rx_coalesce_mode = priv->params.rx_cq_moderation_mode;
diff --git a/sys/dev/mlx5/mlx5_en/mlx5_en_main.c b/sys/dev/mlx5/mlx5_en/mlx5_en_main.c
index 08a0e98fa0fb..916ebe72c46c 100644
--- a/sys/dev/mlx5/mlx5_en/mlx5_en_main.c
+++ b/sys/dev/mlx5/mlx5_en/mlx5_en_main.c
@@ -2204,14 +2204,16 @@ mlx5e_open_rqt(struct mlx5e_priv *priv)
 	MLX5_SET(rqtc, rqtc, rqt_max_size, sz);
 
 	for (i = 0; i < sz; i++) {
-		int ix;
+		int ix = i;
 #ifdef RSS
-		ix = rss_get_indirection_to_bucket(i);
-#else
-		ix = i;
+		ix = rss_get_indirection_to_bucket(ix);
 #endif
 		/* ensure we don't overflow */
 		ix %= priv->params.num_channels;
+
+		/* apply receive side scaling stride, if any */
+		ix -= ix % (int)priv->params.channels_rsss;
+
 		MLX5_SET(rqtc, rqtc, rq_num[i], priv->channel[ix]->rq.rqn);
 	}
 
@@ -3083,6 +3085,7 @@ mlx5e_build_ifp_priv(struct mlx5_core_dev *mdev,
 
 	priv->mdev = mdev;
 	priv->params.num_channels = num_comp_vectors;
+	priv->params.channels_rsss = 1;
 	priv->order_base_2_num_channels = order_base_2(num_comp_vectors);
 	priv->queue_mapping_channel_mask =
 	    roundup_pow_of_two(num_comp_vectors) - 1;