IB/mlx4: Add mechanism to support flow steering over IB links

The mlx4 device requires adding IB flow spec to rules that apply over
infiniband link layer.  This patch adds a mechanism to add such a rule.

If higher levels e.g. IP/UDP/TCP flow specs are provided, the device
requires us to add an empty wild-carded IB rule. Furthermore, the device
requires the QPN to be put in the rule.

Add here specific parsing support for IB empty rules and the ability
to self-generate missing specs based on existing ones.

Signed-off-by: Matan Barak <matanb@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>

1 files changed, 134 insertions, 1 deletions

diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index e8d0c5592fd..6b7f227ca9e 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -55,6 +55,7 @@
 #define DRV_RELDATE	"April 4, 2008"
 
 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
+#define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
 
 MODULE_AUTHOR("Roland Dreier");
 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
@@ -825,6 +826,7 @@ struct mlx4_ib_steering {
 };
 
 static int parse_flow_attr(struct mlx4_dev *dev,
+			   u32 qp_num,
 			   union ib_flow_spec *ib_spec,
 			   struct _rule_hw *mlx4_spec)
 {
@@ -840,6 +842,14 @@ static int parse_flow_attr(struct mlx4_dev *dev,
 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
 		break;
+	case IB_FLOW_SPEC_IB:
+		type = MLX4_NET_TRANS_RULE_ID_IB;
+		mlx4_spec->ib.l3_qpn =
+			cpu_to_be32(qp_num);
+		mlx4_spec->ib.qpn_mask =
+			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
+		break;
+
 
 	case IB_FLOW_SPEC_IPV4:
 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
@@ -871,6 +881,115 @@ static int parse_flow_attr(struct mlx4_dev *dev,
 	return mlx4_hw_rule_sz(dev, type);
 }
 
+struct default_rules {
+	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
+	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
+	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
+	__u8  link_layer;
+};
+static const struct default_rules default_table[] = {
+	{
+		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
+		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
+		.rules_create_list = {IB_FLOW_SPEC_IB},
+		.link_layer = IB_LINK_LAYER_INFINIBAND
+	}
+};
+
+static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
+					 struct ib_flow_attr *flow_attr)
+{
+	int i, j, k;
+	void *ib_flow;
+	const struct default_rules *pdefault_rules = default_table;
+	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
+
+	for (i = 0; i < sizeof(default_table)/sizeof(default_table[0]); i++,
+	     pdefault_rules++) {
+		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
+		memset(&field_types, 0, sizeof(field_types));
+
+		if (link_layer != pdefault_rules->link_layer)
+			continue;
+
+		ib_flow = flow_attr + 1;
+		/* we assume the specs are sorted */
+		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
+		     j < flow_attr->num_of_specs; k++) {
+			union ib_flow_spec *current_flow =
+				(union ib_flow_spec *)ib_flow;
+
+			/* same layer but different type */
+			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
+			     (pdefault_rules->mandatory_fields[k] &
+			      IB_FLOW_SPEC_LAYER_MASK)) &&
+			    (current_flow->type !=
+			     pdefault_rules->mandatory_fields[k]))
+				goto out;
+
+			/* same layer, try match next one */
+			if (current_flow->type ==
+			    pdefault_rules->mandatory_fields[k]) {
+				j++;
+				ib_flow +=
+					((union ib_flow_spec *)ib_flow)->size;
+			}
+		}
+
+		ib_flow = flow_attr + 1;
+		for (j = 0; j < flow_attr->num_of_specs;
+		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
+			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
+				/* same layer and same type */
+				if (((union ib_flow_spec *)ib_flow)->type ==
+				    pdefault_rules->mandatory_not_fields[k])
+					goto out;
+
+		return i;
+	}
+out:
+	return -1;
+}
+
+static int __mlx4_ib_create_default_rules(
+		struct mlx4_ib_dev *mdev,
+		struct ib_qp *qp,
+		const struct default_rules *pdefault_rules,
+		struct _rule_hw *mlx4_spec) {
+	int size = 0;
+	int i;
+
+	for (i = 0; i < sizeof(pdefault_rules->rules_create_list)/
+			sizeof(pdefault_rules->rules_create_list[0]); i++) {
+		int ret;
+		union ib_flow_spec ib_spec;
+		switch (pdefault_rules->rules_create_list[i]) {
+		case 0:
+			/* no rule */
+			continue;
+		case IB_FLOW_SPEC_IB:
+			ib_spec.type = IB_FLOW_SPEC_IB;
+			ib_spec.size = sizeof(struct ib_flow_spec_ib);
+
+			break;
+		default:
+			/* invalid rule */
+			return -EINVAL;
+		}
+		/* We must put empty rule, qpn is being ignored */
+		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
+				      mlx4_spec);
+		if (ret < 0) {
+			pr_info("invalid parsing\n");
+			return -EINVAL;
+		}
+
+		mlx4_spec = (void *)mlx4_spec + ret;
+		size += ret;
+	}
+	return size;
+}
+
 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
 			  int domain,
 			  enum mlx4_net_trans_promisc_mode flow_type,
@@ -882,6 +1001,7 @@ static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_att
 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
 	struct mlx4_cmd_mailbox *mailbox;
 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
+	int default_flow;
 
 	static const u16 __mlx4_domain[] = {
 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
@@ -916,8 +1036,21 @@ static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_att
 
 	ib_flow = flow_attr + 1;
 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
+	/* Add default flows */
+	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
+	if (default_flow >= 0) {
+		ret = __mlx4_ib_create_default_rules(
+				mdev, qp, default_table + default_flow,
+				mailbox->buf + size);
+		if (ret < 0) {
+			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
+			return -EINVAL;
+		}
+		size += ret;
+	}
 	for (i = 0; i < flow_attr->num_of_specs; i++) {
-		ret = parse_flow_attr(mdev->dev, ib_flow, mailbox->buf + size);
+		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
+				      mailbox->buf + size);
 		if (ret < 0) {
 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
 			return -EINVAL;