RDMA/mlx5: Set lag tx affinity according to slave

The patch sets the lag tx affinity of the data QPs and the GSI QPs
according to the LAG xmit slave.

For GSI QPs, in case the link layer is Ethenet (RoCE) we create two GSI
QPs, one for each physical port. When the driver selects the GSI QP, it
will consider the port affinity result.  For connected QPs, the driver
sets the affinity of the xmit slave.

The above, ensures that RC QP and it's corresponding GSI QP will transmit
from the same physical port.

Link: https://lore.kernel.org/r/20200430192146.12863-17-maorg@mellanox.com
Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

1 files changed, 26 insertions, 7 deletions

diff --git a/drivers/infiniband/hw/mlx5/gsi.c b/drivers/infiniband/hw/mlx5/gsi.c
index 1afbf03d1a98..40d418153891 100644
--- a/drivers/infiniband/hw/mlx5/gsi.c
+++ b/drivers/infiniband/hw/mlx5/gsi.c
@@ -119,10 +119,17 @@ struct ib_qp *mlx5_ib_gsi_create_qp(struct ib_pd *pd,
 	struct mlx5_ib_gsi_qp *gsi;
 	struct ib_qp_init_attr hw_init_attr = *init_attr;
 	const u8 port_num = init_attr->port_num;
-	const int num_pkeys = pd->device->attrs.max_pkeys;
-	const int num_qps = mlx5_ib_deth_sqpn_cap(dev) ? num_pkeys : 0;
+	int num_qps = 0;
 	int ret;
 
+	if (mlx5_ib_deth_sqpn_cap(dev)) {
+		if (MLX5_CAP_GEN(dev->mdev,
+				 port_type) == MLX5_CAP_PORT_TYPE_IB)
+			num_qps = pd->device->attrs.max_pkeys;
+		else if (dev->lag_active)
+			num_qps = MLX5_MAX_PORTS;
+	}
+
 	gsi = kzalloc(sizeof(*gsi), GFP_KERNEL);
 	if (!gsi)
 		return ERR_PTR(-ENOMEM);
@@ -261,7 +268,7 @@ static struct ib_qp *create_gsi_ud_qp(struct mlx5_ib_gsi_qp *gsi)
 }
 
 static int modify_to_rts(struct mlx5_ib_gsi_qp *gsi, struct ib_qp *qp,
-			 u16 qp_index)
+			 u16 pkey_index)
 {
 	struct mlx5_ib_dev *dev = to_mdev(qp->device);
 	struct ib_qp_attr attr;
@@ -270,7 +277,7 @@ static int modify_to_rts(struct mlx5_ib_gsi_qp *gsi, struct ib_qp *qp,
 
 	mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_QKEY | IB_QP_PORT;
 	attr.qp_state = IB_QPS_INIT;
-	attr.pkey_index = qp_index;
+	attr.pkey_index = pkey_index;
 	attr.qkey = IB_QP1_QKEY;
 	attr.port_num = gsi->port_num;
 	ret = ib_modify_qp(qp, &attr, mask);
@@ -304,12 +311,17 @@ static void setup_qp(struct mlx5_ib_gsi_qp *gsi, u16 qp_index)
 {
 	struct ib_device *device = gsi->rx_qp->device;
 	struct mlx5_ib_dev *dev = to_mdev(device);
+	int pkey_index = qp_index;
+	struct mlx5_ib_qp *mqp;
 	struct ib_qp *qp;
 	unsigned long flags;
 	u16 pkey;
 	int ret;
 
-	ret = ib_query_pkey(device, gsi->port_num, qp_index, &pkey);
+	if (MLX5_CAP_GEN(dev->mdev,  port_type) != MLX5_CAP_PORT_TYPE_IB)
+		pkey_index = 0;
+
+	ret = ib_query_pkey(device, gsi->port_num, pkey_index, &pkey);
 	if (ret) {
 		mlx5_ib_warn(dev, "unable to read P_Key at port %d, index %d\n",
 			     gsi->port_num, qp_index);
@@ -338,7 +350,10 @@ static void setup_qp(struct mlx5_ib_gsi_qp *gsi, u16 qp_index)
 		return;
 	}
 
-	ret = modify_to_rts(gsi, qp, qp_index);
+	mqp = to_mqp(qp);
+	if (dev->lag_active)
+		mqp->gsi_lag_port = qp_index + 1;
+	ret = modify_to_rts(gsi, qp, pkey_index);
 	if (ret)
 		goto err_destroy_qp;
 
@@ -457,11 +472,15 @@ static int mlx5_ib_gsi_silent_drop(struct mlx5_ib_gsi_qp *gsi,
 static struct ib_qp *get_tx_qp(struct mlx5_ib_gsi_qp *gsi, struct ib_ud_wr *wr)
 {
 	struct mlx5_ib_dev *dev = to_mdev(gsi->rx_qp->device);
+	struct mlx5_ib_ah *ah = to_mah(wr->ah);
 	int qp_index = wr->pkey_index;
 
-	if (!mlx5_ib_deth_sqpn_cap(dev))
+	if (!gsi->num_qps)
 		return gsi->rx_qp;
 
+	if (dev->lag_active && ah->xmit_port)
+		qp_index = ah->xmit_port - 1;
+
 	if (qp_index >= gsi->num_qps)
 		return NULL;