net/mlx5: Introduce port selection namespace
authorMaor Gottlieb <maorg@nvidia.com>
Sun, 23 May 2021 10:34:28 +0000 (13:34 +0300)
committerSaeed Mahameed <saeedm@nvidia.com>
Tue, 19 Oct 2021 03:18:07 +0000 (20:18 -0700)
Add new port selection flow steering namespace. Flow steering rules in
this namespaceare are used to determine the physical port for egress
packets.

Signed-off-by: Maor Gottlieb <maorg@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
drivers/net/ethernet/mellanox/mlx5/core/fs_cmd.c
drivers/net/ethernet/mellanox/mlx5/core/fs_core.c
drivers/net/ethernet/mellanox/mlx5/core/fs_core.h
drivers/net/ethernet/mellanox/mlx5/core/fw.c
drivers/net/ethernet/mellanox/mlx5/core/main.c
include/linux/mlx5/device.h
include/linux/mlx5/fs.h
include/linux/mlx5/mlx5_ifc.h

index 7db8df6..caefdb7 100644 (file)
@@ -969,6 +969,7 @@ const struct mlx5_flow_cmds *mlx5_fs_cmd_get_default(enum fs_flow_table_type typ
        case FS_FT_NIC_TX:
        case FS_FT_RDMA_RX:
        case FS_FT_RDMA_TX:
+       case FS_FT_PORT_SEL:
                return mlx5_fs_cmd_get_fw_cmds();
        default:
                return mlx5_fs_cmd_get_stub_cmds();
index fe501ba..8d8696f 100644 (file)
@@ -2191,6 +2191,10 @@ struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
                if (steering->fdb_root_ns)
                        return &steering->fdb_root_ns->ns;
                return NULL;
+       case MLX5_FLOW_NAMESPACE_PORT_SEL:
+               if (steering->port_sel_root_ns)
+                       return &steering->port_sel_root_ns->ns;
+               return NULL;
        case MLX5_FLOW_NAMESPACE_SNIFFER_RX:
                if (steering->sniffer_rx_root_ns)
                        return &steering->sniffer_rx_root_ns->ns;
@@ -2596,6 +2600,7 @@ void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
        steering->fdb_root_ns = NULL;
        kfree(steering->fdb_sub_ns);
        steering->fdb_sub_ns = NULL;
+       cleanup_root_ns(steering->port_sel_root_ns);
        cleanup_root_ns(steering->sniffer_rx_root_ns);
        cleanup_root_ns(steering->sniffer_tx_root_ns);
        cleanup_root_ns(steering->rdma_rx_root_ns);
@@ -2634,6 +2639,21 @@ static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)
        return PTR_ERR_OR_ZERO(prio);
 }
 
+#define PORT_SEL_NUM_LEVELS 3
+static int init_port_sel_root_ns(struct mlx5_flow_steering *steering)
+{
+       struct fs_prio *prio;
+
+       steering->port_sel_root_ns = create_root_ns(steering, FS_FT_PORT_SEL);
+       if (!steering->port_sel_root_ns)
+               return -ENOMEM;
+
+       /* Create single prio */
+       prio = fs_create_prio(&steering->port_sel_root_ns->ns, 0,
+                             PORT_SEL_NUM_LEVELS);
+       return PTR_ERR_OR_ZERO(prio);
+}
+
 static int init_rdma_rx_root_ns(struct mlx5_flow_steering *steering)
 {
        int err;
@@ -3020,6 +3040,12 @@ int mlx5_init_fs(struct mlx5_core_dev *dev)
                        goto err;
        }
 
+       if (MLX5_CAP_FLOWTABLE_PORT_SELECTION(dev, ft_support)) {
+               err = init_port_sel_root_ns(steering);
+               if (err)
+                       goto err;
+       }
+
        if (MLX5_CAP_FLOWTABLE_RDMA_RX(dev, ft_support) &&
            MLX5_CAP_FLOWTABLE_RDMA_RX(dev, table_miss_action_domain)) {
                err = init_rdma_rx_root_ns(steering);
index 98240ba..79d3753 100644 (file)
@@ -97,7 +97,8 @@ enum fs_flow_table_type {
        FS_FT_SNIFFER_TX        = 0X6,
        FS_FT_RDMA_RX           = 0X7,
        FS_FT_RDMA_TX           = 0X8,
-       FS_FT_MAX_TYPE = FS_FT_RDMA_TX,
+       FS_FT_PORT_SEL          = 0X9,
+       FS_FT_MAX_TYPE = FS_FT_PORT_SEL,
 };
 
 enum fs_flow_table_op_mod {
@@ -129,6 +130,7 @@ struct mlx5_flow_steering {
        struct mlx5_flow_root_namespace *rdma_rx_root_ns;
        struct mlx5_flow_root_namespace *rdma_tx_root_ns;
        struct mlx5_flow_root_namespace *egress_root_ns;
+       struct mlx5_flow_root_namespace *port_sel_root_ns;
        int esw_egress_acl_vports;
        int esw_ingress_acl_vports;
 };
@@ -341,7 +343,8 @@ struct mlx5_flow_root_namespace *find_root(struct fs_node *node);
        (type == FS_FT_SNIFFER_TX) ? MLX5_CAP_FLOWTABLE_SNIFFER_TX(mdev, cap) :         \
        (type == FS_FT_RDMA_RX) ? MLX5_CAP_FLOWTABLE_RDMA_RX(mdev, cap) :               \
        (type == FS_FT_RDMA_TX) ? MLX5_CAP_FLOWTABLE_RDMA_TX(mdev, cap) :      \
-       (BUILD_BUG_ON_ZERO(FS_FT_RDMA_TX != FS_FT_MAX_TYPE))\
+       (type == FS_FT_PORT_SEL) ? MLX5_CAP_FLOWTABLE_PORT_SELECTION(mdev, cap) :      \
+       (BUILD_BUG_ON_ZERO(FS_FT_PORT_SEL != FS_FT_MAX_TYPE))\
        )
 
 #endif
index f4f8993..1037e36 100644 (file)
@@ -149,6 +149,12 @@ int mlx5_query_hca_caps(struct mlx5_core_dev *dev)
        if (err)
                return err;
 
+       if (MLX5_CAP_GEN(dev, port_selection_cap)) {
+               err = mlx5_core_get_caps(dev, MLX5_CAP_PORT_SELECTION);
+               if (err)
+                       return err;
+       }
+
        if (MLX5_CAP_GEN(dev, hca_cap_2)) {
                err = mlx5_core_get_caps(dev, MLX5_CAP_GENERAL_2);
                if (err)
index 47d92fb..f844639 100644 (file)
@@ -1416,6 +1416,7 @@ static const int types[] = {
        MLX5_CAP_TLS,
        MLX5_CAP_VDPA_EMULATION,
        MLX5_CAP_IPSEC,
+       MLX5_CAP_PORT_SELECTION,
 };
 
 static void mlx5_hca_caps_free(struct mlx5_core_dev *dev)
index 109cc81..347167c 100644 (file)
@@ -1185,6 +1185,7 @@ enum mlx5_cap_type {
        MLX5_CAP_DEV_EVENT = 0x14,
        MLX5_CAP_IPSEC,
        MLX5_CAP_GENERAL_2 = 0x20,
+       MLX5_CAP_PORT_SELECTION = 0x25,
        /* NUM OF CAP Types */
        MLX5_CAP_NUM
 };
@@ -1342,6 +1343,20 @@ enum mlx5_qcam_feature_groups {
        MLX5_GET(e_switch_cap, \
                 mdev->caps.hca[MLX5_CAP_ESWITCH]->max, cap)
 
+#define MLX5_CAP_PORT_SELECTION(mdev, cap) \
+       MLX5_GET(port_selection_cap, \
+                mdev->caps.hca[MLX5_CAP_PORT_SELECTION]->cur, cap)
+
+#define MLX5_CAP_PORT_SELECTION_MAX(mdev, cap) \
+       MLX5_GET(port_selection_cap, \
+                mdev->caps.hca[MLX5_CAP_PORT_SELECTION]->max, cap)
+
+#define MLX5_CAP_FLOWTABLE_PORT_SELECTION(mdev, cap) \
+       MLX5_CAP_PORT_SELECTION(mdev, flow_table_properties_port_selection.cap)
+
+#define MLX5_CAP_FLOWTABLE_PORT_SELECTION_MAX(mdev, cap) \
+       MLX5_CAP_PORT_SELECTION_MAX(mdev, flow_table_properties_port_selection.cap)
+
 #define MLX5_CAP_ODP(mdev, cap)\
        MLX5_GET(odp_cap, mdev->caps.hca[MLX5_CAP_ODP]->cur, cap)
 
index 0106c67..259fcc1 100644 (file)
@@ -83,6 +83,7 @@ enum mlx5_flow_namespace_type {
        MLX5_FLOW_NAMESPACE_RDMA_RX,
        MLX5_FLOW_NAMESPACE_RDMA_RX_KERNEL,
        MLX5_FLOW_NAMESPACE_RDMA_TX,
+       MLX5_FLOW_NAMESPACE_PORT_SEL,
 };
 
 enum {
index c614ad1..db1d9c6 100644 (file)
@@ -767,6 +767,18 @@ struct mlx5_ifc_flow_table_nic_cap_bits {
        u8         reserved_at_20c0[0x5f40];
 };
 
+struct mlx5_ifc_port_selection_cap_bits {
+       u8         reserved_at_0[0x10];
+       u8         port_select_flow_table[0x1];
+       u8         reserved_at_11[0xf];
+
+       u8         reserved_at_20[0x1e0];
+
+       struct mlx5_ifc_flow_table_prop_layout_bits flow_table_properties_port_selection;
+
+       u8         reserved_at_400[0x7c00];
+};
+
 enum {
        MLX5_FDB_TO_VPORT_REG_C_0 = 0x01,
        MLX5_FDB_TO_VPORT_REG_C_1 = 0x02,
@@ -1515,7 +1527,8 @@ struct mlx5_ifc_cmd_hca_cap_bits {
        u8         uar_4k[0x1];
        u8         reserved_at_241[0x9];
        u8         uar_sz[0x6];
-       u8         reserved_at_248[0x2];
+       u8         port_selection_cap[0x1];
+       u8         reserved_at_248[0x1];
        u8         umem_uid_0[0x1];
        u8         reserved_at_250[0x5];
        u8         log_pg_sz[0x8];
@@ -3164,6 +3177,7 @@ union mlx5_ifc_hca_cap_union_bits {
        struct mlx5_ifc_flow_table_nic_cap_bits flow_table_nic_cap;
        struct mlx5_ifc_flow_table_eswitch_cap_bits flow_table_eswitch_cap;
        struct mlx5_ifc_e_switch_cap_bits e_switch_cap;
+       struct mlx5_ifc_port_selection_cap_bits port_selection_cap;
        struct mlx5_ifc_vector_calc_cap_bits vector_calc_cap;
        struct mlx5_ifc_qos_cap_bits qos_cap;
        struct mlx5_ifc_debug_cap_bits debug_cap;
@@ -10434,9 +10448,16 @@ struct mlx5_ifc_dcbx_param_bits {
        u8         reserved_at_a0[0x160];
 };
 
+enum {
+       MLX5_LAG_PORT_SELECT_MODE_QUEUE_AFFINITY = 0,
+       MLX5_LAG_PORT_SELECT_MODE_PORT_SELECT_FT,
+};
+
 struct mlx5_ifc_lagc_bits {
        u8         fdb_selection_mode[0x1];
-       u8         reserved_at_1[0x1c];
+       u8         reserved_at_1[0x14];
+       u8         port_select_mode[0x3];
+       u8         reserved_at_18[0x5];
        u8         lag_state[0x3];
 
        u8         reserved_at_20[0x14];