Skip to contentSkip to conversation

add new PHY affinity support in MLX5 PMD

PATCH Superseded Jiawei Wang 2 participants 9 messages Feb 3, 2023 – Feb 26, 2023 Network drivers

lore
Share & tools
lore
#0 lore raw

cover add new PHY affinity support in MLX5 PMD

This patch series implement PHY affinity item support in MLX5 PMD and the support for TIS creation with tx_phy_affinity value of Tx queue.

This patch series relies on the preceding ethdev API:
http://patches.dpdk.org/project/dpdk/patch/[email protected]/

Jiawei Wang (3):
drivers: add lag Rx port affinity in PRM
net/mlx5: add PHY affinity item support
drivers: enhance the Tx queue affinity

 doc/guides/nics/features/default.ini |   1 +
 doc/guides/nics/features/mlx5.ini    |   1 +
 doc/guides/nics/mlx5.rst             |   8 ++-
 drivers/common/mlx5/mlx5_devx_cmds.c |   3 +
 drivers/common/mlx5/mlx5_devx_cmds.h |   1 +
 drivers/common/mlx5/mlx5_prm.h       |  15 ++--
 drivers/net/mlx5/linux/mlx5_os.c     |   6 ++
 drivers/net/mlx5/mlx5.c              |  43 +++++-------
 drivers/net/mlx5/mlx5.h              |   3 +
 drivers/net/mlx5/mlx5_devx.c         |  24 ++++---
 drivers/net/mlx5/mlx5_ethdev.c       |   1 +
 drivers/net/mlx5/mlx5_flow.h         |   3 +
 drivers/net/mlx5/mlx5_flow_dv.c      | 100 ++++++++++++++++++++++++++-
 drivers/net/mlx5/mlx5_flow_hw.c      |  14 ++++
 drivers/net/mlx5/mlx5_tx.h           |   1 +
 drivers/net/mlx5/mlx5_txq.c          |   8 +++
 16 files changed, 184 insertions(+), 48 deletions(-)
Signature · 1 line
2.18.1
#1 lore raw

PATCH 1/3 drivers: add lag Rx port affinity in PRM

This patch adds function to query hca capability via Devx for lag_rx_port_affinity.

---
 drivers/common/mlx5/mlx5_devx_cmds.c | 3 +++
 drivers/common/mlx5/mlx5_devx_cmds.h | 1 +
 drivers/common/mlx5/mlx5_prm.h       | 7 +++++--
 drivers/net/mlx5/linux/mlx5_os.c     | 4 ++++
 drivers/net/mlx5/mlx5.h              | 2 ++
 5 files changed, 15 insertions(+), 2 deletions(-)
drivers/common/mlx5/mlx5_devx_cmds.c +3 -0
index e3a4927d0f..a157ab4b6c 100644
@@ -1141,6 +1141,9 @@ mlx5_devx_cmd_query_hca_attr(void *ctx,
 	attr->outer_ipv4_ihl = MLX5_GET
 		(flow_table_nic_cap, hcattr,
 		 ft_field_support_2_nic_receive.outer_ipv4_ihl);
+	attr->lag_rx_port_affinity = MLX5_GET
+		(flow_table_nic_cap, hcattr,
+		 ft_field_support_2_nic_receive.lag_rx_port_affinity);
 	/* Query HCA offloads for Ethernet protocol. */
 	hcattr = mlx5_devx_get_hca_cap(ctx, in, out, &rc,
 			MLX5_GET_HCA_CAP_OP_MOD_ETHERNET_OFFLOAD_CAPS |
drivers/common/mlx5/mlx5_devx_cmds.h +1 -0
index c94b9eac06..9cf045ccaa 100644
@@ -288,6 +288,7 @@ struct mlx5_hca_attr {
 	uint32_t alloc_flow_counter_pd:1;
 	uint32_t flow_counter_access_aso:1;
 	uint32_t flow_access_aso_opc_mod:8;
+	uint32_t lag_rx_port_affinity:1;
 };
 
 /* LAG Context. */
drivers/common/mlx5/mlx5_prm.h +5 -2
index 9294f65e24..8bbb800206 100644
@@ -905,7 +905,8 @@ struct mlx5_ifc_fte_match_set_misc_bits {
 	u8 vxlan_vni[0x18];
 	u8 reserved_at_b8[0x8];
 	u8 geneve_vni[0x18];
-	u8 reserved_at_e4[0x6];
+	u8 lag_rx_port_affinity[0x4];
+	u8 reserved_at_e8[0x2];
 	u8 geneve_tlv_option_0_exist[0x1];
 	u8 geneve_oam[0x1];
 	u8 reserved_at_e0[0xc];
@@ -2044,7 +2045,9 @@ struct mlx5_ifc_ft_fields_support_bits {
  * Table 1872 - Flow Table Fields Supported 2 Format
  */
 struct mlx5_ifc_ft_fields_support_2_bits {
-	u8 reserved_at_0[0xd];
+	u8 reserved_at_0[0xa];
+	u8 lag_rx_port_affinity[0x1];
+	u8 reserved_at_c[0x2];
 	u8 hash_result[0x1];
 	u8 reserved_at_e[0x1];
 	u8 tunnel_header_2_3[0x1];
drivers/net/mlx5/linux/mlx5_os.c +4 -0
index a71474c90a..60462da39d 100644
@@ -1390,6 +1390,10 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
 				DRV_LOG(DEBUG, "DV flow is not supported!");
 		}
 #endif
+		if (hca_attr->lag_rx_port_affinity) {
+			sh->lag_rx_port_affinity_en = 1;
+			DRV_LOG(DEBUG, "LAG Rx Port Affinity enabled");
+		}
 	}
 	/* Process parameters and store port configuration on priv structure. */
 	err = mlx5_port_args_config(priv, mkvlist, &priv->config);
drivers/net/mlx5/mlx5.h +2 -0
index 16b33e1548..bbd7262a51 100644
@@ -1374,6 +1374,8 @@ struct mlx5_dev_ctx_shared {
 	uint32_t hws_tags:1; /* Check if tags info for HWS initialized. */
 	uint32_t shared_mark_enabled:1;
 	/* If mark action is enabled on Rxqs (shared E-Switch domain). */
+	uint32_t lag_rx_port_affinity_en:1;
+	/* lag_rx_port_affinity is supported. */
 	uint32_t hws_max_log_bulk_sz:5;
 	/* Log of minimal HWS counters created hard coded. */
 	uint32_t hws_max_nb_counters; /* Maximal number for HWS counters. */
Signature · 1 line
2.18.1
#2 lore raw

PATCH 2/3 net/mlx5: add PHY affinity item support

In reply to Jiawei Wang

This patch adds the new phy affinity item support in PMD:
RTE_FLOW_ITEM_TYPE_PHY_AFFINITY.

This patch adds the validation function for the new item, it works for NIC-RX and FDB rule on ROOT-table only.

---
 doc/guides/nics/features/default.ini |   1 +
 doc/guides/nics/features/mlx5.ini    |   1 +
 doc/guides/nics/mlx5.rst             |   8 ++-
 drivers/net/mlx5/linux/mlx5_os.c     |   2 +
 drivers/net/mlx5/mlx5.h              |   1 +
 drivers/net/mlx5/mlx5_flow.h         |   3 +
 drivers/net/mlx5/mlx5_flow_dv.c      | 100 ++++++++++++++++++++++++++-
 drivers/net/mlx5/mlx5_flow_hw.c      |  14 ++++
 8 files changed, 127 insertions(+), 3 deletions(-)
doc/guides/nics/features/default.ini +1 -0
index 510cc6679d..fd5edd11c8 100644
@@ -141,6 +141,7 @@ udp                  =
 vlan                 =
 vxlan                =
 vxlan_gpe            =
+phy_affinity         =
 
 [rte_flow actions]
 age                  =
doc/guides/nics/features/mlx5.ini +1 -0
index 62fd330e2b..9142f04f93 100644
@@ -87,6 +87,7 @@ vlan                 = Y
 vxlan                = Y
 vxlan_gpe            = Y
 represented_port     = Y
+phy_affinity         = Y
 
 [rte_flow actions]
 age                  = I
doc/guides/nics/mlx5.rst +7 -1
index f137f156f9..5569b045d7 100644
@@ -106,6 +106,7 @@ Features
 - Sub-Function representors.
 - Sub-Function.
 - Matching on represented port.
+- Matching on phy affinity.
 
 
 Limitations
@@ -595,13 +596,18 @@ Limitations
   - key
   - sequence
 
-  Matching on checksum and sequence needs MLNX_OFED 5.6+.
+- Matching on checksum and sequence needs MLNX_OFED 5.6+.
 
 - The NIC egress flow rules on representor port are not supported.
 
 - When using DV/verbs flow engine (``dv_flow_en`` = 1/0 respectively), Match on SPI field
   in ESP header for group 0 needs MLNX_OFED 5.6+.
 
+- Match on phy affinity:
+
+  - Supports NIC ingress flow in group 0.
+  - Supports E-Switch flow in group 0 and depends on
+    device-managed flow steering (DMFS) mode.
 
 Statistics
 ----------
drivers/net/mlx5/linux/mlx5_os.c +2 -0
index 60462da39d..1c26b30702 100644
@@ -1394,6 +1394,8 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
 			sh->lag_rx_port_affinity_en = 1;
 			DRV_LOG(DEBUG, "LAG Rx Port Affinity enabled");
 		}
+		priv->num_lag_ports = hca_attr->num_lag_ports;
+		DRV_LOG(DEBUG, "The number of lag ports is %d", priv->num_lag_ports);
 	}
 	/* Process parameters and store port configuration on priv structure. */
 	err = mlx5_port_args_config(priv, mkvlist, &priv->config);
drivers/net/mlx5/mlx5.h +1 -0
index bbd7262a51..c7a7b176b8 100644
@@ -1668,6 +1668,7 @@ struct mlx5_priv {
 	unsigned int mtr_reg_share:1; /* Whether support meter REG_C share. */
 	unsigned int lb_used:1; /* Loopback queue is referred to. */
 	uint32_t mark_enabled:1; /* If mark action is enabled on rxqs. */
+	uint32_t num_lag_ports:4; /* Number of ports can be bonded. */
 	uint16_t domain_id; /* Switch domain identifier. */
 	uint16_t vport_id; /* Associated VF vport index (if any). */
 	uint32_t vport_meta_tag; /* Used for vport index match ove VF LAG. */
drivers/net/mlx5/mlx5_flow.h +3 -0
index e376dcae93..64b1b7c37c 100644
@@ -219,6 +219,9 @@ enum mlx5_feature_name {
 /* Meter color item */
 #define MLX5_FLOW_ITEM_METER_COLOR (UINT64_C(1) << 44)
 
+/* PHY affinity item */
+#define MLX5_FLOW_ITEM_PHY_AFFINITY (UINT64_C(1) << 49)
+
 /* Outer Masks. */
 #define MLX5_FLOW_LAYER_OUTER_L3 \
 	(MLX5_FLOW_LAYER_OUTER_L3_IPV4 | MLX5_FLOW_LAYER_OUTER_L3_IPV6)
drivers/net/mlx5/mlx5_flow_dv.c +98 -2

136 lines not on this page · raw message

drivers/net/mlx5/mlx5_flow_hw.c +14 -0
index 20c71ff7f0..e5ca86ca4b 100644
@@ -4715,6 +4715,20 @@ flow_hw_pattern_validate(struct rte_eth_dev *dev,
 							  "Unsupported meter color register");
 			break;
 		}
+		case RTE_FLOW_ITEM_TYPE_PHY_AFFINITY:
+		{
+			if (!priv->sh->lag_rx_port_affinity_en)
+				return rte_flow_error_set(error, EINVAL,
+							  RTE_FLOW_ERROR_TYPE_ITEM, NULL,
+							  "Unsupported phy affinity with Older FW");
+			if ((attr->transfer && priv->fdb_def_rule) || attr->egress)
+				return rte_flow_error_set(error, EINVAL,
+							  RTE_FLOW_ERROR_TYPE_ITEM, NULL,
+							  "Phy affinity item not supported"
+							  " with egress or transfer"
+							  " attribute");
+			break;
+		}
 		case RTE_FLOW_ITEM_TYPE_VOID:
 		case RTE_FLOW_ITEM_TYPE_ETH:
 		case RTE_FLOW_ITEM_TYPE_VLAN:
Signature · 1 line
2.18.1
#3 lore raw

PATCH 3/3 drivers: enhance the Tx queue affinity

In reply to Jiawei Wang

Previous patch supports the tx phy affinity configuration in the Tx queue API, it supports to set the affinity value per Queue.

This patch updates TIS creation with tx_phy_affinity value of Tx queue, TIS index 1 goes to hardware port 0, TIS index 2 goes to hardware port 1, and TIS index 0 is reserved for default HWS hash mode.

---
 drivers/common/mlx5/mlx5_prm.h |  8 -------
 drivers/net/mlx5/mlx5.c        | 43 +++++++++++++++-------------------
 drivers/net/mlx5/mlx5_devx.c   | 24 ++++++++++---------
 drivers/net/mlx5/mlx5_ethdev.c |  1 +
 drivers/net/mlx5/mlx5_tx.h     |  1 +
 drivers/net/mlx5/mlx5_txq.c    |  8 +++++++
 6 files changed, 42 insertions(+), 43 deletions(-)
drivers/common/mlx5/mlx5_prm.h +0 -8
index 8bbb800206..ded001d0b2 100644
@@ -2331,14 +2331,6 @@ struct mlx5_ifc_query_nic_vport_context_in_bits {
 	u8 reserved_at_68[0x18];
 };
 
-/*
- * lag_tx_port_affinity: 0 auto-selection, 1 PF1, 2 PF2 vice versa.
- * Each TIS binds to one PF by setting lag_tx_port_affinity (>0).
- * Once LAG enabled, we create multiple TISs and bind each one to
- * different PFs, then TIS[i] gets affinity i+1 and goes to PF i+1.
- */
-#define MLX5_IFC_LAG_MAP_TIS_AFFINITY(index, num) ((num) ? \
-						    (index) % (num) + 1 : 0)
 struct mlx5_ifc_tisc_bits {
 	u8 strict_lag_tx_port_affinity[0x1];
 	u8 reserved_at_1[0x3];
drivers/net/mlx5/mlx5.c +19 -24

63 lines not on this page · raw message

drivers/net/mlx5/mlx5_devx.c +13 -11
index f6e1943fd7..a3fe0b3b98 100644
@@ -1190,17 +1190,19 @@ static uint32_t
 mlx5_get_txq_tis_num(struct rte_eth_dev *dev, uint16_t queue_idx)
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
-	int tis_idx;
-
-	if (priv->sh->bond.n_port && priv->sh->lag.affinity_mode ==
-			MLX5_LAG_MODE_TIS) {
-		tis_idx = (priv->lag_affinity_idx + queue_idx) %
-			priv->sh->bond.n_port;
-		DRV_LOG(INFO, "port %d txq %d gets affinity %d and maps to PF %d.",
-			dev->data->port_id, queue_idx, tis_idx + 1,
-			priv->sh->lag.tx_remap_affinity[tis_idx]);
-	} else {
-		tis_idx = 0;
+	struct mlx5_txq_data *txq_data = (*priv->txqs)[queue_idx];
+	int tis_idx = 0;
+
+	if (priv->sh->bond.n_port) {
+		if (txq_data->tx_phy_affinity) {
+			tis_idx = txq_data->tx_phy_affinity;
+		} else if (priv->sh->lag.affinity_mode == MLX5_LAG_MODE_TIS) {
+			tis_idx = (priv->lag_affinity_idx + queue_idx) %
+				priv->sh->bond.n_port + 1;
+			DRV_LOG(INFO, "port %d txq %d gets affinity %d and maps to PF %d.",
+				dev->data->port_id, queue_idx, tis_idx,
+				priv->sh->lag.tx_remap_affinity[tis_idx - 1]);
+		}
 	}
 	MLX5_ASSERT(priv->sh->tis[tis_idx]);
 	return priv->sh->tis[tis_idx]->id;
drivers/net/mlx5/mlx5_ethdev.c +1 -0
index 4a85415ff3..354bb4f022 100644
@@ -352,6 +352,7 @@ mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
 	info->switch_info.domain_id = priv->domain_id;
 	info->switch_info.port_id = priv->representor_id;
 	info->switch_info.rx_domain = 0; /* No sub Rx domains. */
+	info->nb_phy_ports = priv->sh->bond.n_port;
 	if (priv->representor) {
 		uint16_t port_id;
 
drivers/net/mlx5/mlx5_tx.h +1 -0
index a056be7ca8..674c2aebe5 100644
@@ -144,6 +144,7 @@ struct mlx5_txq_data {
 	uint16_t inlen_send; /* Ordinary send data inline size. */
 	uint16_t inlen_empw; /* eMPW max packet size to inline. */
 	uint16_t inlen_mode; /* Minimal data length to inline. */
+	uint8_t tx_phy_affinity; /* TxQ affinity configuration. */
 	uint32_t qp_num_8s; /* QP number shifted by 8. */
 	uint64_t offloads; /* Offloads for Tx Queue. */
 	struct mlx5_mr_ctrl mr_ctrl; /* MR control descriptor. */
drivers/net/mlx5/mlx5_txq.c +8 -0
index 5543f2c570..4e53706807 100644
@@ -392,6 +392,13 @@ mlx5_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
 		container_of(txq, struct mlx5_txq_ctrl, txq);
 	int res;
 
+	if (conf->tx_phy_affinity > priv->num_lag_ports) {
+		rte_errno = EINVAL;
+		DRV_LOG(ERR, "port %u unable to setup Tx queue index %u"
+			" affinity is %u exceeds the maximum %u", dev->data->port_id,
+			idx, conf->tx_phy_affinity, priv->num_lag_ports);
+		return -rte_errno;
+	}
 	res = mlx5_tx_queue_pre_setup(dev, idx, &desc);
 	if (res)
 		return res;
@@ -1095,6 +1102,7 @@ mlx5_txq_new(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
 	tmpl->txq.elts_m = desc - 1;
 	tmpl->txq.port_id = dev->data->port_id;
 	tmpl->txq.idx = idx;
+	tmpl->txq.tx_phy_affinity = conf->tx_phy_affinity;
 	txq_set_params(tmpl);
 	if (txq_adjust_params(tmpl))
 		goto error;
Signature · 1 line
2.18.1
#4 lore raw

v2 cover Add Tx queue mapping of aggregated ports in MLX5 PMD

In reply to Jiawei Wang

This patch implements RTE_FLOW_ITEM_TYPE_AGGR_AFFINITY support in MLX5 PMD.

This patch adds the MLX5 PMD support for two device ops:
- map_aggr_tx_affinity
- count_aggr_ports

This patch series relies on the preceding RTE API:
http://patches.dpdk.org/project/dpdk/list/?series=27064

v2:
* update the PMD code based on the new RTE API.

Jiawei Wang (3):
net/mlx5: add lag Rx port affinity in PRM
net/mlx5: add aggregated affinity item support
net/mlx5: enhance the Tx queue affinity

 doc/guides/nics/features/default.ini |   1 +
 doc/guides/nics/features/mlx5.ini    |   1 +
 doc/guides/nics/mlx5.rst             |   6 ++
 drivers/common/mlx5/mlx5_devx_cmds.c |   3 +
 drivers/common/mlx5/mlx5_devx_cmds.h |   1 +
 drivers/common/mlx5/mlx5_prm.h       |  15 ++--
 drivers/net/mlx5/linux/mlx5_os.c     |   6 ++
 drivers/net/mlx5/mlx5.c              |  49 ++++++-------
 drivers/net/mlx5/mlx5.h              |   3 +
 drivers/net/mlx5/mlx5_devx.c         |  24 ++++---
 drivers/net/mlx5/mlx5_flow.h         |   3 +
 drivers/net/mlx5/mlx5_flow_dv.c      | 100 ++++++++++++++++++++++++++-
 drivers/net/mlx5/mlx5_flow_hw.c      |  14 ++++
 drivers/net/mlx5/mlx5_tx.h           |   4 ++
 drivers/net/mlx5/mlx5_txq.c          |  38 ++++++++++
 15 files changed, 221 insertions(+), 47 deletions(-)
Signature · 1 line
2.18.1
#5 lore raw

PATCH v2 1/3 net/mlx5: add lag Rx port affinity in PRM

This patch adds function to query hca capability via Devx for lag_rx_port_affinity.

---
 drivers/common/mlx5/mlx5_devx_cmds.c | 3 +++
 drivers/common/mlx5/mlx5_devx_cmds.h | 1 +
 drivers/common/mlx5/mlx5_prm.h       | 7 +++++--
 drivers/net/mlx5/linux/mlx5_os.c     | 4 ++++
 drivers/net/mlx5/mlx5.h              | 2 ++
 5 files changed, 15 insertions(+), 2 deletions(-)
drivers/common/mlx5/mlx5_devx_cmds.c +3 -0
index fb670950ef..bfc6e09eac 100644
@@ -1163,6 +1163,9 @@ mlx5_devx_cmd_query_hca_attr(void *ctx,
 	attr->outer_ipv4_ihl = MLX5_GET
 		(flow_table_nic_cap, hcattr,
 		 ft_field_support_2_nic_receive.outer_ipv4_ihl);
+	attr->lag_rx_port_affinity = MLX5_GET
+		(flow_table_nic_cap, hcattr,
+		 ft_field_support_2_nic_receive.lag_rx_port_affinity);
 	/* Query HCA offloads for Ethernet protocol. */
 	hcattr = mlx5_devx_get_hca_cap(ctx, in, out, &rc,
 			MLX5_GET_HCA_CAP_OP_MOD_ETHERNET_OFFLOAD_CAPS |
drivers/common/mlx5/mlx5_devx_cmds.h +1 -0
index 6b5a6b8667..8e68eeaf37 100644
@@ -293,6 +293,7 @@ struct mlx5_hca_attr {
 	uint32_t flow_counter_access_aso:1;
 	uint32_t flow_access_aso_opc_mod:8;
 	uint32_t cross_vhca:1;
+	uint32_t lag_rx_port_affinity:1;
 };
 
 /* LAG Context. */
drivers/common/mlx5/mlx5_prm.h +5 -2
index 613cc6face..26a1f0717d 100644
@@ -906,7 +906,8 @@ struct mlx5_ifc_fte_match_set_misc_bits {
 	u8 vxlan_vni[0x18];
 	u8 reserved_at_b8[0x8];
 	u8 geneve_vni[0x18];
-	u8 reserved_at_e4[0x6];
+	u8 lag_rx_port_affinity[0x4];
+	u8 reserved_at_e8[0x2];
 	u8 geneve_tlv_option_0_exist[0x1];
 	u8 geneve_oam[0x1];
 	u8 reserved_at_e0[0xc];
@@ -2069,7 +2070,9 @@ struct mlx5_ifc_ft_fields_support_bits {
  * Table 1872 - Flow Table Fields Supported 2 Format
  */
 struct mlx5_ifc_ft_fields_support_2_bits {
-	u8 reserved_at_0[0xd];
+	u8 reserved_at_0[0xa];
+	u8 lag_rx_port_affinity[0x1];
+	u8 reserved_at_c[0x2];
 	u8 hash_result[0x1];
 	u8 reserved_at_e[0x1];
 	u8 tunnel_header_2_3[0x1];
drivers/net/mlx5/linux/mlx5_os.c +4 -0
index a71474c90a..60462da39d 100644
@@ -1390,6 +1390,10 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
 				DRV_LOG(DEBUG, "DV flow is not supported!");
 		}
 #endif
+		if (hca_attr->lag_rx_port_affinity) {
+			sh->lag_rx_port_affinity_en = 1;
+			DRV_LOG(DEBUG, "LAG Rx Port Affinity enabled");
+		}
 	}
 	/* Process parameters and store port configuration on priv structure. */
 	err = mlx5_port_args_config(priv, mkvlist, &priv->config);
drivers/net/mlx5/mlx5.h +2 -0
index a766fb408e..32797008c1 100644
@@ -1386,6 +1386,8 @@ struct mlx5_dev_ctx_shared {
 	uint32_t hws_tags:1; /* Check if tags info for HWS initialized. */
 	uint32_t shared_mark_enabled:1;
 	/* If mark action is enabled on Rxqs (shared E-Switch domain). */
+	uint32_t lag_rx_port_affinity_en:1;
+	/* lag_rx_port_affinity is supported. */
 	uint32_t hws_max_log_bulk_sz:5;
 	/* Log of minimal HWS counters created hard coded. */
 	uint32_t hws_max_nb_counters; /* Maximal number for HWS counters. */
Signature · 1 line
2.18.1
#6 lore raw

PATCH v2 2/3 net/mlx5: add aggregated affinity item support

In reply to Jiawei Wang

This patch adds the new aggregated affinity item support in PMD:
RTE_FLOW_ITEM_TYPE_AGGR_AFFINITY.

This patch adds the validation function for the new item, it works for NIC-RX and FDB rule on ROOT-table only.

---
 doc/guides/nics/features/default.ini |   1 +
 doc/guides/nics/features/mlx5.ini    |   1 +
 doc/guides/nics/mlx5.rst             |   6 ++
 drivers/net/mlx5/linux/mlx5_os.c     |   2 +
 drivers/net/mlx5/mlx5.h              |   1 +
 drivers/net/mlx5/mlx5_flow.h         |   3 +
 drivers/net/mlx5/mlx5_flow_dv.c      | 100 ++++++++++++++++++++++++++-
 drivers/net/mlx5/mlx5_flow_hw.c      |  14 ++++
 8 files changed, 126 insertions(+), 2 deletions(-)
doc/guides/nics/features/default.ini +1 -0
index e249f62f31..0834bdc300 100644
@@ -145,6 +145,7 @@ udp                  =
 vlan                 =
 vxlan                =
 vxlan_gpe            =
+aggr_affinity         =
 
 [rte_flow actions]
 age                  =
doc/guides/nics/features/mlx5.ini +1 -0
index c58e1934e9..7276e5bd1a 100644
@@ -90,6 +90,7 @@ vlan                 = Y
 vxlan                = Y
 vxlan_gpe            = Y
 represented_port     = Y
+aggr_affinity         = Y
 
 [rte_flow actions]
 age                  = I
doc/guides/nics/mlx5.rst +6 -0
index 6510e74fb9..dbf0c7a4bc 100644
@@ -107,6 +107,7 @@ Features
 - Sub-Function representors.
 - Sub-Function.
 - Matching on represented port.
+- Matching on aggregated affinity.
 
 
 Limitations
@@ -615,6 +616,11 @@ Limitations
 
 - The NIC egress flow rules on representor port are not supported.
 
+- Match on aggregated affinity:
+
+  - Supports NIC ingress flow in group 0.
+  - Supports E-Switch flow in group 0 and depends on
+    device-managed flow steering (DMFS) mode.
 
 Statistics
 ----------
drivers/net/mlx5/linux/mlx5_os.c +2 -0
index 60462da39d..1c26b30702 100644
@@ -1394,6 +1394,8 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
 			sh->lag_rx_port_affinity_en = 1;
 			DRV_LOG(DEBUG, "LAG Rx Port Affinity enabled");
 		}
+		priv->num_lag_ports = hca_attr->num_lag_ports;
+		DRV_LOG(DEBUG, "The number of lag ports is %d", priv->num_lag_ports);
 	}
 	/* Process parameters and store port configuration on priv structure. */
 	err = mlx5_port_args_config(priv, mkvlist, &priv->config);
drivers/net/mlx5/mlx5.h +1 -0
index 32797008c1..581179cecd 100644
@@ -1681,6 +1681,7 @@ struct mlx5_priv {
 	unsigned int mtr_reg_share:1; /* Whether support meter REG_C share. */
 	unsigned int lb_used:1; /* Loopback queue is referred to. */
 	uint32_t mark_enabled:1; /* If mark action is enabled on rxqs. */
+	uint32_t num_lag_ports:4; /* Number of ports can be bonded. */
 	uint16_t domain_id; /* Switch domain identifier. */
 	uint16_t vport_id; /* Associated VF vport index (if any). */
 	uint32_t vport_meta_tag; /* Used for vport index match ove VF LAG. */
drivers/net/mlx5/mlx5_flow.h +3 -0
index 4bef2296b8..3a72f4eb4d 100644
@@ -223,6 +223,9 @@ enum mlx5_feature_name {
 #define MLX5_FLOW_ITEM_OUTER_IPV6_ROUTING_EXT (UINT64_C(1) << 45)
 #define MLX5_FLOW_ITEM_INNER_IPV6_ROUTING_EXT (UINT64_C(1) << 46)
 
+/* Aggregated affinity item */
+#define MLX5_FLOW_ITEM_AGGR_AFFINITY (UINT64_C(1) << 49)
+
 /* Outer Masks. */
 #define MLX5_FLOW_LAYER_OUTER_L3 \
 	(MLX5_FLOW_LAYER_OUTER_L3_IPV4 | MLX5_FLOW_LAYER_OUTER_L3_IPV6)
drivers/net/mlx5/mlx5_flow_dv.c +98 -2

136 lines not on this page · raw message

drivers/net/mlx5/mlx5_flow_hw.c +14 -0
index a9c7045a3e..07766dd8f1 100644
@@ -4723,6 +4723,20 @@ flow_hw_pattern_validate(struct rte_eth_dev *dev,
 							  "Unsupported meter color register");
 			break;
 		}
+		case RTE_FLOW_ITEM_TYPE_AGGR_AFFINITY:
+		{
+			if (!priv->sh->lag_rx_port_affinity_en)
+				return rte_flow_error_set(error, EINVAL,
+							  RTE_FLOW_ERROR_TYPE_ITEM, NULL,
+							  "Unsupported aggregated affinity with Older FW");
+			if ((attr->transfer && priv->fdb_def_rule) || attr->egress)
+				return rte_flow_error_set(error, EINVAL,
+							  RTE_FLOW_ERROR_TYPE_ITEM, NULL,
+							  "Aggregated affinity item not supported"
+							  " with egress or transfer"
+							  " attribute");
+			break;
+		}
 		case RTE_FLOW_ITEM_TYPE_VOID:
 		case RTE_FLOW_ITEM_TYPE_ETH:
 		case RTE_FLOW_ITEM_TYPE_VLAN:
Signature · 1 line
2.18.1
#7 lore raw

PATCH v2 3/3 net/mlx5: enhance the Tx queue affinity

In reply to Jiawei Wang

The rte_eth_dev_map_aggr_tx_affinity() was introduced in ethdev lib, it was used to set the affinity value per Tx queue.

This patch adds the MLX5 PMD support for two device ops:
- map_aggr_tx_affinity
- count_aggr_ports

After maps a Tx queue with an aggregated port by call map_aggr_tx_affinity() and starts sending traffic, the MLX5 PMD updates TIS creation with tx_aggr_affinity value of Tx queue. TIS index 1 goes to first physical port, TIS index 2 goes to second physical port, and so on, TIS index 0 is reserved for default HW hash mode.

---
 drivers/common/mlx5/mlx5_prm.h |  8 ------
 drivers/net/mlx5/mlx5.c        | 49 +++++++++++++++++-----------------
 drivers/net/mlx5/mlx5_devx.c   | 24 +++++++++--------
 drivers/net/mlx5/mlx5_tx.h     |  4 +++
 drivers/net/mlx5/mlx5_txq.c    | 38 ++++++++++++++++++++++++++
 5 files changed, 80 insertions(+), 43 deletions(-)
drivers/common/mlx5/mlx5_prm.h +0 -8
index 26a1f0717d..2f5aeecaa9 100644
@@ -2363,14 +2363,6 @@ struct mlx5_ifc_query_nic_vport_context_in_bits {
 	u8 reserved_at_68[0x18];
 };
 
-/*
- * lag_tx_port_affinity: 0 auto-selection, 1 PF1, 2 PF2 vice versa.
- * Each TIS binds to one PF by setting lag_tx_port_affinity (>0).
- * Once LAG enabled, we create multiple TISs and bind each one to
- * different PFs, then TIS[i] gets affinity i+1 and goes to PF i+1.
- */
-#define MLX5_IFC_LAG_MAP_TIS_AFFINITY(index, num) ((num) ? \
-						    (index) % (num) + 1 : 0)
 struct mlx5_ifc_tisc_bits {
 	u8 strict_lag_tx_port_affinity[0x1];
 	u8 reserved_at_1[0x3];
drivers/net/mlx5/mlx5.c +25 -24

87 lines not on this page · raw message

drivers/net/mlx5/mlx5_devx.c +13 -11
index f6e1943fd7..d02cedb202 100644
@@ -1190,17 +1190,19 @@ static uint32_t
 mlx5_get_txq_tis_num(struct rte_eth_dev *dev, uint16_t queue_idx)
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
-	int tis_idx;
-
-	if (priv->sh->bond.n_port && priv->sh->lag.affinity_mode ==
-			MLX5_LAG_MODE_TIS) {
-		tis_idx = (priv->lag_affinity_idx + queue_idx) %
-			priv->sh->bond.n_port;
-		DRV_LOG(INFO, "port %d txq %d gets affinity %d and maps to PF %d.",
-			dev->data->port_id, queue_idx, tis_idx + 1,
-			priv->sh->lag.tx_remap_affinity[tis_idx]);
-	} else {
-		tis_idx = 0;
+	struct mlx5_txq_data *txq_data = (*priv->txqs)[queue_idx];
+	int tis_idx = 0;
+
+	if (priv->sh->bond.n_port) {
+		if (txq_data->tx_aggr_affinity) {
+			tis_idx = txq_data->tx_aggr_affinity;
+		} else if (priv->sh->lag.affinity_mode == MLX5_LAG_MODE_TIS) {
+			tis_idx = (priv->lag_affinity_idx + queue_idx) %
+				priv->sh->bond.n_port + 1;
+			DRV_LOG(INFO, "port %d txq %d gets affinity %d and maps to PF %d.",
+				dev->data->port_id, queue_idx, tis_idx,
+				priv->sh->lag.tx_remap_affinity[tis_idx - 1]);
+		}
 	}
 	MLX5_ASSERT(priv->sh->tis[tis_idx]);
 	return priv->sh->tis[tis_idx]->id;
drivers/net/mlx5/mlx5_tx.h +4 -0
index a056be7ca8..d0c6303a2d 100644
@@ -144,6 +144,7 @@ struct mlx5_txq_data {
 	uint16_t inlen_send; /* Ordinary send data inline size. */
 	uint16_t inlen_empw; /* eMPW max packet size to inline. */
 	uint16_t inlen_mode; /* Minimal data length to inline. */
+	uint8_t tx_aggr_affinity; /* TxQ affinity configuration. */
 	uint32_t qp_num_8s; /* QP number shifted by 8. */
 	uint64_t offloads; /* Offloads for Tx Queue. */
 	struct mlx5_mr_ctrl mr_ctrl; /* MR control descriptor. */
@@ -218,6 +219,9 @@ void txq_alloc_elts(struct mlx5_txq_ctrl *txq_ctrl);
 void txq_free_elts(struct mlx5_txq_ctrl *txq_ctrl);
 uint64_t mlx5_get_tx_port_offloads(struct rte_eth_dev *dev);
 void mlx5_txq_dynf_timestamp_set(struct rte_eth_dev *dev);
+int mlx5_count_aggr_ports(struct rte_eth_dev *dev);
+int mlx5_map_aggr_tx_affinity(struct rte_eth_dev *dev, uint16_t tx_queue_id,
+			      uint8_t affinity);
 
 /* mlx5_tx.c */
 
drivers/net/mlx5/mlx5_txq.c +38 -0
index 419e913559..1e0e61a620 100644
@@ -1365,3 +1365,41 @@ mlx5_txq_dynf_timestamp_set(struct rte_eth_dev *dev)
 				     ts_mask : 0;
 	}
 }
+
+int mlx5_count_aggr_ports(struct rte_eth_dev *dev)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+
+	return priv->sh->bond.n_port;
+}
+
+int mlx5_map_aggr_tx_affinity(struct rte_eth_dev *dev, uint16_t tx_queue_id,
+			      uint8_t affinity)
+{
+	struct mlx5_txq_ctrl *txq_ctrl;
+	struct mlx5_txq_data *txq;
+	struct mlx5_priv *priv;
+
+	priv = dev->data->dev_private;
+	txq = (*priv->txqs)[tx_queue_id];
+	if (!txq)
+		return -1;
+	txq_ctrl = container_of(txq, struct mlx5_txq_ctrl, txq);
+	if (tx_queue_id >= priv->txqs_n) {
+		DRV_LOG(ERR, "port %u Tx queue index out of range (%u >= %u)",
+			dev->data->port_id, tx_queue_id, priv->txqs_n);
+		rte_errno = EOVERFLOW;
+		return -rte_errno;
+	}
+	if (affinity > priv->num_lag_ports) {
+		DRV_LOG(ERR, "port %u unable to setup Tx queue index %u"
+			" affinity is %u exceeds the maximum %u", dev->data->port_id,
+			tx_queue_id, affinity, priv->num_lag_ports);
+		rte_errno = EINVAL;
+		return -rte_errno;
+	}
+	DRV_LOG(DEBUG, "port %u configuring queue %u for aggregated affinity %u",
+		dev->data->port_id, tx_queue_id, affinity);
+	txq_ctrl->txq.tx_aggr_affinity = affinity;
+	return 0;
+}
Signature · 1 line
2.18.1
#8 lore raw

RE: [PATCH v2 0/3] Add Tx queue mapping of aggregated ports in MLX5 PMD

In reply to Jiawei Wang

Hi,

Quoted 44 lines

-----Original Message-----
From: Jiawei(Jonny) Wang <[email protected]>
Sent: Wednesday, February 22, 2023 2:26 PM
To: Slava Ovsiienko <[email protected]>; Ori Kam <[email protected]>
Cc: [email protected]; Raslan Darawsheh <[email protected]>
Subject: [PATCH v2 0/3] Add Tx queue mapping of aggregated ports in MLX5 PMD

This patch implements RTE_FLOW_ITEM_TYPE_AGGR_AFFINITY support in MLX5 PMD.

This patch adds the MLX5 PMD support for two device ops:
- map_aggr_tx_affinity
- count_aggr_ports

This patch series relies on the preceding RTE API:
http://patches.dpdk.org/project/dpdk/list/?series=27064

v2:
* update the PMD code based on the new RTE API.

Jiawei Wang (3):
net/mlx5: add lag Rx port affinity in PRM
net/mlx5: add aggregated affinity item support
net/mlx5: enhance the Tx queue affinity

 doc/guides/nics/features/default.ini |   1 +
 doc/guides/nics/features/mlx5.ini    |   1 +
 doc/guides/nics/mlx5.rst             |   6 ++
 drivers/common/mlx5/mlx5_devx_cmds.c |   3 +
 drivers/common/mlx5/mlx5_devx_cmds.h |   1 +
 drivers/common/mlx5/mlx5_prm.h       |  15 ++--
 drivers/net/mlx5/linux/mlx5_os.c     |   6 ++
 drivers/net/mlx5/mlx5.c              |  49 ++++++-------
 drivers/net/mlx5/mlx5.h              |   3 +
 drivers/net/mlx5/mlx5_devx.c         |  24 ++++---
 drivers/net/mlx5/mlx5_flow.h         |   3 +
 drivers/net/mlx5/mlx5_flow_dv.c      | 100
++++++++++++++++++++++++++-
 drivers/net/mlx5/mlx5_flow_hw.c      |  14 ++++
 drivers/net/mlx5/mlx5_tx.h           |   4 ++
 drivers/net/mlx5/mlx5_txq.c          |  38 ++++++++++
 15 files changed, 221 insertions(+), 47 deletions(-)
Quoted 2 lines

--
2.18.1

Series applied to next-net-mlx,

Kindest regards,
Raslan Darawsheh