net/mlx5e: Enable CQE based moderation on TX CQ

By using CQE based moderation on TX CQ we can reduce the number of TX
interrupt rate. Besides the benefit of less interrupts, this also
allows the kernel to better utilize TSO. Since TSO has some CPU overhead,
it might not aggregate when CPU is under high stress. By reducing the
interrupt rate and the CPU utilization, we can get better aggregation
and better overall throughput.
The feature is enabled by default and has a private flag in ethtool
for control.

Throughput, interrupt rate and TSO utilization improvements:
(ConnectX-4Lx 40GbE, unidirectional, 1/16 TCP streams, 64B packets)
---------------------------------------------------------
Metric   | Streams | CQE Based | EQE Based | improvement
---------------------------------------------------------
BW       |    1    |  2.4Gb/s  | 2.15Gb/s  |  +11.6%
IR       |    1    |  27Kips   | 50.6Kips  |  -46.7%
TSO Util |    1    |  74.6%    | 71%       |  +5%
BW       |    16   |  29Gb/s   | 25.85Gb/s |  +12.2%
IR       |    16   |  482Kips  | 745Kips   |  -35.3%
TSO Util |    16   |  69.1%    | 49%       |  +41.1%

*BW = Bandwidth, IR = Interrupt rate, ips = interrupt per second.
TSO Util = bytes in TSO sessions / all bytes transferred

Signed-off-by: Tal Gilboa <talgi@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

1 files changed, 27 insertions, 11 deletions

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index 73d7c672c4ff..d1c3dc946486 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -681,7 +681,7 @@ static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 	}
 
 	INIT_WORK(&rq->am.work, mlx5e_rx_am_work);
-	rq->am.mode = params->rx_cq_period_mode;
+	rq->am.mode = params->rx_cq_moderation.cq_period_mode;
 	rq->page_cache.head = 0;
 	rq->page_cache.tail = 0;
 
@@ -1974,7 +1974,7 @@ static void mlx5e_build_rx_cq_param(struct mlx5e_priv *priv,
 	}
 
 	mlx5e_build_common_cq_param(priv, param);
-	param->cq_period_mode = params->rx_cq_period_mode;
+	param->cq_period_mode = params->rx_cq_moderation.cq_period_mode;
 }
 
 static void mlx5e_build_tx_cq_param(struct mlx5e_priv *priv,
@@ -1986,8 +1986,7 @@ static void mlx5e_build_tx_cq_param(struct mlx5e_priv *priv,
 	MLX5_SET(cqc, cqc, log_cq_size, params->log_sq_size);
 
 	mlx5e_build_common_cq_param(priv, param);
-
-	param->cq_period_mode = MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
+	param->cq_period_mode = params->tx_cq_moderation.cq_period_mode;
 }
 
 static void mlx5e_build_ico_cq_param(struct mlx5e_priv *priv,
@@ -3987,14 +3986,32 @@ static bool hw_lro_heuristic(u32 link_speed, u32 pci_bw)
 		 (pci_bw <= 16000) && (pci_bw < link_speed));
 }
 
+void mlx5e_set_tx_cq_mode_params(struct mlx5e_params *params, u8 cq_period_mode)
+{
+	params->tx_cq_moderation.cq_period_mode = cq_period_mode;
+
+	params->tx_cq_moderation.pkts =
+		MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_PKTS;
+	params->tx_cq_moderation.usec =
+		MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_USEC;
+
+	if (cq_period_mode == MLX5_CQ_PERIOD_MODE_START_FROM_CQE)
+		params->tx_cq_moderation.usec =
+			MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_USEC_FROM_CQE;
+
+	MLX5E_SET_PFLAG(params, MLX5E_PFLAG_TX_CQE_BASED_MODER,
+			params->tx_cq_moderation.cq_period_mode ==
+				MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
+}
+
 void mlx5e_set_rx_cq_mode_params(struct mlx5e_params *params, u8 cq_period_mode)
 {
-	params->rx_cq_period_mode = cq_period_mode;
+	params->rx_cq_moderation.cq_period_mode = cq_period_mode;
 
 	params->rx_cq_moderation.pkts =
 		MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_PKTS;
 	params->rx_cq_moderation.usec =
-			MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC;
+		MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC;
 
 	if (cq_period_mode == MLX5_CQ_PERIOD_MODE_START_FROM_CQE)
 		params->rx_cq_moderation.usec =
@@ -4002,10 +4019,11 @@ void mlx5e_set_rx_cq_mode_params(struct mlx5e_params *params, u8 cq_period_mode)
 
 	if (params->rx_am_enabled)
 		params->rx_cq_moderation =
-			mlx5e_am_get_def_profile(params->rx_cq_period_mode);
+			mlx5e_am_get_def_profile(cq_period_mode);
 
 	MLX5E_SET_PFLAG(params, MLX5E_PFLAG_RX_CQE_BASED_MODER,
-			params->rx_cq_period_mode == MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
+			params->rx_cq_moderation.cq_period_mode ==
+				MLX5_CQ_PERIOD_MODE_START_FROM_CQE);
 }
 
 u32 mlx5e_choose_lro_timeout(struct mlx5_core_dev *mdev, u32 wanted_timeout)
@@ -4065,9 +4083,7 @@ void mlx5e_build_nic_params(struct mlx5_core_dev *mdev,
 			MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
 	params->rx_am_enabled = MLX5_CAP_GEN(mdev, cq_moderation);
 	mlx5e_set_rx_cq_mode_params(params, cq_period_mode);
-
-	params->tx_cq_moderation.usec = MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_USEC;
-	params->tx_cq_moderation.pkts = MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_PKTS;
+	mlx5e_set_tx_cq_mode_params(params, cq_period_mode);
 
 	/* TX inline */
 	params->tx_max_inline = mlx5e_get_max_inline_cap(mdev);