net/mlx5e: Use CQE padding for Ethernet CQs

Writing 64B CQEs to 128B cache lines results in a RMW operation. Padding the CQEs to 128B if possible improves performance on 128B cache line systems like PPC. Testing on PPC showed up to a 24% improvement in small packet throughput vs the default behavior, depending on the workload and system topology. Signed-off-by: Daniel Jurgens <danielj@mellanox.com> Reviewed-by: Tariq Toukan <tariqt@mellanox.com> Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
mariux64 · Dec 11, 2018 · 939de57 · 939de57
1 parent 8c4dc42
commit 939de57
Show file tree

Hide file tree

Showing 4 changed files with 15 additions and 7 deletions.
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -2224,6 +2224,8 @@ static void mlx5e_build_common_cq_param(struct mlx5e_priv *priv,
 	void *cqc = param->cqc;
 
 	MLX5_SET(cqc, cqc, uar_page, priv->mdev->priv.uar->index);
+	if (MLX5_CAP_GEN(priv->mdev, cqe_128_always) && cache_line_size() >= 128)
+		MLX5_SET(cqc, cqc, cqe_sz, CQE_STRIDE_128_PAD);
 }
 
 static void mlx5e_build_rx_cq_param(struct mlx5e_priv *priv,

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/wq.c b/drivers/net/ethernet/mellanox/mlx5/core/wq.c
@@ -155,7 +155,8 @@ int mlx5_cqwq_create(struct mlx5_core_dev *mdev, struct mlx5_wq_param *param,
 		     void *cqc, struct mlx5_cqwq *wq,
 		     struct mlx5_wq_ctrl *wq_ctrl)
 {
-	u8 log_wq_stride = MLX5_GET(cqc, cqc, cqe_sz) + 6;
+	/* CQE_STRIDE_128 and CQE_STRIDE_128_PAD both mean 128B stride */
+	u8 log_wq_stride = MLX5_GET(cqc, cqc, cqe_sz) == CQE_STRIDE_64 ? 6 : 7;
 	u8 log_wq_sz     = MLX5_GET(cqc, cqc, log_cq_size);
 	int err;
 

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/wq.h b/drivers/net/ethernet/mellanox/mlx5/core/wq.h
@@ -179,7 +179,12 @@ static inline u32 mlx5_cqwq_get_ci(struct mlx5_cqwq *wq)
 
 static inline struct mlx5_cqe64 *mlx5_cqwq_get_wqe(struct mlx5_cqwq *wq, u32 ix)
 {
-	return mlx5_frag_buf_get_wqe(&wq->fbc, ix);
+	struct mlx5_cqe64 *cqe = mlx5_frag_buf_get_wqe(&wq->fbc, ix);
+
+	/* For 128B CQEs the data is in the last 64B */
+	cqe += wq->fbc.log_stride == 7;
+
+	return cqe;
 }
 
 static inline u32 mlx5_cqwq_get_ctr_wrap_cnt(struct mlx5_cqwq *wq, u32 ctr)

diff --git a/include/linux/mlx5/cq.h b/include/linux/mlx5/cq.h
@@ -125,18 +125,18 @@ struct mlx5_cq_modify_params {
 };
 
 enum {
-	CQE_SIZE_64 = 0,
-	CQE_SIZE_128 = 1,
-	CQE_SIZE_128_PAD = 2,
+	CQE_STRIDE_64 = 0,
+	CQE_STRIDE_128 = 1,
+	CQE_STRIDE_128_PAD = 2,
 };
 
 #define MLX5_MAX_CQ_PERIOD (BIT(__mlx5_bit_sz(cqc, cq_period)) - 1)
 #define MLX5_MAX_CQ_COUNT (BIT(__mlx5_bit_sz(cqc, cq_max_count)) - 1)
 
 static inline int cqe_sz_to_mlx_sz(u8 size, int padding_128_en)
 {
-	return padding_128_en ? CQE_SIZE_128_PAD :
-				size == 64 ? CQE_SIZE_64 : CQE_SIZE_128;
+	return padding_128_en ? CQE_STRIDE_128_PAD :
+				size == 64 ? CQE_STRIDE_64 : CQE_STRIDE_128;
 }
 
 static inline void mlx5_cq_set_ci(struct mlx5_core_cq *cq)