Commit 939de57d authored by Daniel Jurgens's avatar Daniel Jurgens Committed by Saeed Mahameed
Browse files

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: default avatarDaniel Jurgens <danielj@mellanox.com>
Reviewed-by: default avatarTariq Toukan <tariqt@mellanox.com>
Signed-off-by: default avatarSaeed Mahameed <saeedm@mellanox.com>
parent 8c4dc42b

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

+2 −0
Original line number Diff line number Diff line
@@ -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,

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

+2 −1
Original line number Diff line number Diff line
@@ -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;

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

+6 −1
Original line number Diff line number Diff line
@@ -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)

include/linux/mlx5/cq.h

+5 −5
Original line number Diff line number Diff line
@@ -125,9 +125,9 @@ 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)
@@ -135,8 +135,8 @@ enum { 


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)

CRA Git | Maintained and supported by SUSTech CRA and CCSE