IB/mlx4: Add support for the don't trap rule

Add support for receiving multicast/unicast traffic with
the don't trap rule.

Sniffing these packets requires a flow steering rule of type NORMAL
at priority 0 with flag IB_FLOW_ATTR_FLAGS_DONT_TRAP set.
Choosing between multicast or unicast is done via ethernet L2 dest_mac
mask and value:
- If mask is all zeros - unicast and multicast are set.
- If mask non zero - only mask with multicast bit 1 and rest 0 is
                     supported, the mac value will choose if it is
                     multicast or unicast rule.

If the mask multicast bit is on and some other bits are on too, it means
a request for specific multicast or unicast, this is not supported,
either receive all multicast or all unicast.

Only when limitations are met registered QP will receive requested type
but other QPs can receive same traffic if registered for it.
Otherwise, if limitations are not met, an error will be returned.

Limitations:
- Rule must be with priority 0.
- A0 mode is not supported.
- Sniffer QP cannot appear in any other flow steering rule.
Signed-off-by: Marina Varshaver <marinav@mellanox.com>
Reviewed-by: Matan Barak <matanb@mellanox.com>
Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

drivers/infiniband/hw/mlx4/main.c

@@ -1643,6 +1643,56 @@ static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_
 	return err;
 }
 
+static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
+				      struct ib_flow_attr *flow_attr,
+				      enum mlx4_net_trans_promisc_mode *type)
+{
+	int err = 0;
+
+	if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
+	    (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
+	    (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
+		return -EOPNOTSUPP;
+	}
+
+	if (flow_attr->num_of_specs == 0) {
+		type[0] = MLX4_FS_MC_SNIFFER;
+		type[1] = MLX4_FS_UC_SNIFFER;
+	} else {
+		union ib_flow_spec *ib_spec;
+
+		ib_spec = (union ib_flow_spec *)(flow_attr + 1);
+		if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
+			return -EINVAL;
+
+		/* if all is zero than MC and UC */
+		if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
+			type[0] = MLX4_FS_MC_SNIFFER;
+			type[1] = MLX4_FS_UC_SNIFFER;
+		} else {
+			u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
+					    ib_spec->eth.mask.dst_mac[1],
+					    ib_spec->eth.mask.dst_mac[2],
+					    ib_spec->eth.mask.dst_mac[3],
+					    ib_spec->eth.mask.dst_mac[4],
+					    ib_spec->eth.mask.dst_mac[5]};
+
+			/* Above xor was only on MC bit, non empty mask is valid
+			 * only if this bit is set and rest are zero.
+			 */
+			if (!is_zero_ether_addr(&mac[0]))
+				return -EINVAL;
+
+			if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
+				type[0] = MLX4_FS_MC_SNIFFER;
+			else
+				type[0] = MLX4_FS_UC_SNIFFER;
+		}
+	}
+
+	return err;
+}
+
 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
 				    struct ib_flow_attr *flow_attr,
 				    int domain)
@@ -1653,7 +1703,8 @@ static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
 	struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
 	int is_bonded = mlx4_is_bonded(dev);
 
-	if (flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)
+	if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
+	    (flow_attr->type != IB_FLOW_ATTR_NORMAL))
 		return ERR_PTR(-EOPNOTSUPP);
 
 	memset(type, 0, sizeof(type));
@@ -1666,7 +1717,19 @@ static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
 
 	switch (flow_attr->type) {
 	case IB_FLOW_ATTR_NORMAL:
-		type[0] = MLX4_FS_REGULAR;
+		/* If dont trap flag (continue match) is set, under specific
+		 * condition traffic be replicated to given qp,
+		 * without stealing it
+		 */
+		if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
+			err = mlx4_ib_add_dont_trap_rule(dev,
+							 flow_attr,
+							 type);
+			if (err)
+				goto err_free;
+		} else {
+			type[0] = MLX4_FS_REGULAR;
+		}
 		break;
 
 	case IB_FLOW_ATTR_ALL_DEFAULT:
@@ -1678,8 +1741,8 @@ static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
 		break;
 
 	case IB_FLOW_ATTR_SNIFFER:
-		type[0] = MLX4_FS_UC_SNIFFER;
-		type[1] = MLX4_FS_MC_SNIFFER;
+		type[0] = MLX4_FS_MIRROR_RX_PORT;
+		type[1] = MLX4_FS_MIRROR_SX_PORT;
 		break;
 
 	default:

drivers/net/ethernet/mellanox/mlx4/fw.c

@@ -157,7 +157,8 @@ static void dump_dev_cap_flags2(struct mlx4_dev *dev, u64 flags)
 		[29] = "802.1ad offload support",
 		[31] = "Modifying loopback source checks using UPDATE_QP support",
 		[32] = "Loopback source checks support",
-		[33] = "RoCEv2 support"
+		[33] = "RoCEv2 support",
+		[34] = "DMFS Sniffer support (UC & MC)"
 	};
 	int i;
 
@@ -810,6 +811,8 @@ int mlx4_QUERY_DEV_CAP(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
 	if (field & 0x80)
 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_FS_EN;
 	dev_cap->fs_log_max_ucast_qp_range_size = field & 0x1f;
+	if (field & 0x20)
+		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER;
 	MLX4_GET(field, outbox, QUERY_DEV_CAP_PORT_BEACON_OFFSET);
 	if (field & 0x80)
 		dev_cap->flags2 |= MLX4_DEV_CAP_FLAG2_PORT_BEACON;

drivers/net/ethernet/mellanox/mlx4/mcg.c

@@ -752,8 +752,10 @@ static const u8 __promisc_mode[] = {
 	[MLX4_FS_REGULAR]   = 0x0,
 	[MLX4_FS_ALL_DEFAULT] = 0x1,
 	[MLX4_FS_MC_DEFAULT] = 0x3,
-	[MLX4_FS_UC_SNIFFER] = 0x4,
-	[MLX4_FS_MC_SNIFFER] = 0x5,
+	[MLX4_FS_MIRROR_RX_PORT] = 0x4,
+	[MLX4_FS_MIRROR_SX_PORT] = 0x5,
+	[MLX4_FS_UC_SNIFFER] = 0x6,
+	[MLX4_FS_MC_SNIFFER] = 0x7,
 };
 
 int mlx4_map_sw_to_hw_steering_mode(struct mlx4_dev *dev,

include/linux/mlx4/device.h

@@ -219,6 +219,7 @@ enum {
 	MLX4_DEV_CAP_FLAG2_UPDATE_QP_SRC_CHECK_LB = 1ULL << 31,
 	MLX4_DEV_CAP_FLAG2_LB_SRC_CHK           = 1ULL << 32,
 	MLX4_DEV_CAP_FLAG2_ROCE_V1_V2		= 1ULL <<  33,
+	MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER   = 1ULL <<  34,
 };
 
 enum {
@@ -1160,6 +1161,8 @@ enum mlx4_net_trans_promisc_mode {
 	MLX4_FS_REGULAR = 1,
 	MLX4_FS_ALL_DEFAULT,
 	MLX4_FS_MC_DEFAULT,
+	MLX4_FS_MIRROR_RX_PORT,
+	MLX4_FS_MIRROR_SX_PORT,
 	MLX4_FS_UC_SNIFFER,
 	MLX4_FS_MC_SNIFFER,
 	MLX4_FS_MODE_NUM, /* should be last */