net/mlx5: Vectorize the low level core hairpin object

Enhance the hairpin setup code at the core to support a set of N
(RQ,SQ) pairs. This will be later used by the caller to set RSS
spreading among the different RQs.
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

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

@@ -269,7 +269,7 @@ static int mlx5e_hairpin_create_transport(struct mlx5e_hairpin *hp)
 	tirc = MLX5_ADDR_OF(create_tir_in, in, ctx);
 
 	MLX5_SET(tirc, tirc, disp_type, MLX5_TIRC_DISP_TYPE_DIRECT);
-	MLX5_SET(tirc, tirc, inline_rqn, hp->pair->rqn);
+	MLX5_SET(tirc, tirc, inline_rqn, hp->pair->rqn[0]);
 	MLX5_SET(tirc, tirc, transport_domain, hp->tdn);
 
 	err = mlx5_core_create_tir(hp->func_mdev, in, MLX5_ST_SZ_BYTES(create_tir_in), &hp->tirn);
@@ -440,8 +440,8 @@ static int mlx5e_hairpin_flow_add(struct mlx5e_priv *priv,
 	}
 
 	netdev_dbg(priv->netdev, "add hairpin: tirn %x rqn %x peer %s sqn %x prio %d log data size %d\n",
-		   hp->tirn, hp->pair->rqn, hp->pair->peer_mdev->priv.name,
-		   hp->pair->sqn, match_prio, params.log_data_size);
+		   hp->tirn, hp->pair->rqn[0], hp->pair->peer_mdev->priv.name,
+		   hp->pair->sqn[0], match_prio, params.log_data_size);
 
 	hpe->hp = hp;
 	hash_add(priv->fs.tc.hairpin_tbl, &hpe->hairpin_hlist,

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

@@ -437,28 +437,40 @@ static int mlx5_hairpin_create_sq(struct mlx5_core_dev *mdev,
 static int mlx5_hairpin_create_queues(struct mlx5_hairpin *hp,
 				      struct mlx5_hairpin_params *params)
 {
-	int err;
+	int i, j, err;
 
-	err = mlx5_hairpin_create_rq(hp->func_mdev, params, &hp->rqn);
-	if (err)
-		goto out_err_rq;
+	for (i = 0; i < hp->num_channels; i++) {
+		err = mlx5_hairpin_create_rq(hp->func_mdev, params, &hp->rqn[i]);
+		if (err)
+			goto out_err_rq;
+	}
 
-	err = mlx5_hairpin_create_sq(hp->peer_mdev, params, &hp->sqn);
-	if (err)
-		goto out_err_sq;
+	for (i = 0; i < hp->num_channels; i++) {
+		err = mlx5_hairpin_create_sq(hp->peer_mdev, params, &hp->sqn[i]);
+		if (err)
+			goto out_err_sq;
+	}
 
 	return 0;
 
 out_err_sq:
-	mlx5_core_destroy_rq(hp->func_mdev, hp->rqn);
+	for (j = 0; j < i; j++)
+		mlx5_core_destroy_sq(hp->peer_mdev, hp->sqn[j]);
+	i = hp->num_channels;
 out_err_rq:
+	for (j = 0; j < i; j++)
+		mlx5_core_destroy_rq(hp->func_mdev, hp->rqn[j]);
 	return err;
 }
 
 static void mlx5_hairpin_destroy_queues(struct mlx5_hairpin *hp)
 {
-	mlx5_core_destroy_rq(hp->func_mdev, hp->rqn);
-	mlx5_core_destroy_sq(hp->peer_mdev, hp->sqn);
+	int i;
+
+	for (i = 0; i < hp->num_channels; i++) {
+		mlx5_core_destroy_rq(hp->func_mdev, hp->rqn[i]);
+		mlx5_core_destroy_sq(hp->peer_mdev, hp->sqn[i]);
+	}
 }
 
 static int mlx5_hairpin_modify_rq(struct mlx5_core_dev *func_mdev, u32 rqn,
@@ -505,41 +517,53 @@ static int mlx5_hairpin_modify_sq(struct mlx5_core_dev *peer_mdev, u32 sqn,
 
 static int mlx5_hairpin_pair_queues(struct mlx5_hairpin *hp)
 {
-	int err;
+	int i, j, err;
 
-	/* set peer SQ */
-	err = mlx5_hairpin_modify_sq(hp->peer_mdev, hp->sqn,
-				     MLX5_SQC_STATE_RST, MLX5_SQC_STATE_RDY,
-				     MLX5_CAP_GEN(hp->func_mdev, vhca_id), hp->rqn);
-	if (err)
-		goto err_modify_sq;
-
-	/* set func RQ */
-	err = mlx5_hairpin_modify_rq(hp->func_mdev, hp->rqn,
-				     MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY,
-				     MLX5_CAP_GEN(hp->peer_mdev, vhca_id), hp->sqn);
+	/* set peer SQs */
+	for (i = 0; i < hp->num_channels; i++) {
+		err = mlx5_hairpin_modify_sq(hp->peer_mdev, hp->sqn[i],
+					     MLX5_SQC_STATE_RST, MLX5_SQC_STATE_RDY,
+					     MLX5_CAP_GEN(hp->func_mdev, vhca_id), hp->rqn[i]);
+		if (err)
+			goto err_modify_sq;
+	}
 
-	if (err)
-		goto err_modify_rq;
+	/* set func RQs */
+	for (i = 0; i < hp->num_channels; i++) {
+		err = mlx5_hairpin_modify_rq(hp->func_mdev, hp->rqn[i],
+					     MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY,
+					     MLX5_CAP_GEN(hp->peer_mdev, vhca_id), hp->sqn[i]);
+		if (err)
+			goto err_modify_rq;
+	}
 
 	return 0;
 
 err_modify_rq:
-	mlx5_hairpin_modify_sq(hp->peer_mdev, hp->sqn, MLX5_SQC_STATE_RDY,
-			       MLX5_SQC_STATE_RST, 0, 0);
+	for (j = 0; j < i; j++)
+		mlx5_hairpin_modify_rq(hp->func_mdev, hp->rqn[j], MLX5_RQC_STATE_RDY,
+				       MLX5_RQC_STATE_RST, 0, 0);
+	i = hp->num_channels;
 err_modify_sq:
+	for (j = 0; j < i; j++)
+		mlx5_hairpin_modify_sq(hp->peer_mdev, hp->sqn[j], MLX5_SQC_STATE_RDY,
+				       MLX5_SQC_STATE_RST, 0, 0);
 	return err;
 }
 
 static void mlx5_hairpin_unpair_queues(struct mlx5_hairpin *hp)
 {
-	/* unset func RQ */
-	mlx5_hairpin_modify_rq(hp->func_mdev, hp->rqn, MLX5_RQC_STATE_RDY,
-			       MLX5_RQC_STATE_RST, 0, 0);
+	int i;
+
+	/* unset func RQs */
+	for (i = 0; i < hp->num_channels; i++)
+		mlx5_hairpin_modify_rq(hp->func_mdev, hp->rqn[i], MLX5_RQC_STATE_RDY,
+				       MLX5_RQC_STATE_RST, 0, 0);
 
-	/* unset peer SQ */
-	mlx5_hairpin_modify_sq(hp->peer_mdev, hp->sqn, MLX5_SQC_STATE_RDY,
-			       MLX5_SQC_STATE_RST, 0, 0);
+	/* unset peer SQs */
+	for (i = 0; i < hp->num_channels; i++)
+		mlx5_hairpin_modify_sq(hp->peer_mdev, hp->sqn[i], MLX5_SQC_STATE_RDY,
+				       MLX5_SQC_STATE_RST, 0, 0);
 }
 
 struct mlx5_hairpin *
@@ -550,13 +574,17 @@ mlx5_core_hairpin_create(struct mlx5_core_dev *func_mdev,
 	struct mlx5_hairpin *hp;
 	int size, err;
 
-	size = sizeof(*hp);
+	size = sizeof(*hp) + params->num_channels * 2 * sizeof(u32);
 	hp = kzalloc(size, GFP_KERNEL);
 	if (!hp)
 		return ERR_PTR(-ENOMEM);
 
 	hp->func_mdev = func_mdev;
 	hp->peer_mdev = peer_mdev;
+	hp->num_channels = params->num_channels;
+
+	hp->rqn = (void *)hp + sizeof(*hp);
+	hp->sqn = hp->rqn + params->num_channels;
 
 	/* alloc and pair func --> peer hairpin */
 	err = mlx5_hairpin_create_queues(hp, params);

include/linux/mlx5/transobj.h

@@ -78,14 +78,17 @@ void mlx5_core_destroy_rqt(struct mlx5_core_dev *dev, u32 rqtn);
 struct mlx5_hairpin_params {
 	u8  log_data_size;
 	u16 q_counter;
+	int num_channels;
 };
 
 struct mlx5_hairpin {
 	struct mlx5_core_dev *func_mdev;
 	struct mlx5_core_dev *peer_mdev;
 
-	u32 rqn;
-	u32 sqn;
+	int num_channels;
+
+	u32 *rqn;
+	u32 *sqn;
 };
 
 struct mlx5_hairpin *