Skip to contentSkip to conversation

add two ports hairpin mode support in mlx5 PMD

PATCH Superseded Bing Zhao 3 participants 28 messages Oct 8, 2020 – Oct 26, 2020 Network drivers

lore
Share & tools
lore

Bing Zhao

author
#0 lore raw

cover add two ports hairpin mode support in mlx5 PMD

This patch set will add the support for hairpin between two ports in mlx5 driver.

Depends-on: series-12779 ("introduce support for hairpin between two ports")

Bing Zhao (4):
net/mlx5: remove hairpin queue peer port checking net/mlx5: add support for two ports hairpin mode net/mlx5: conditional hairpin auto bind
doc: update hairpin support for mlx5 driver

 doc/guides/rel_notes/release_20_11.rst |   5 +
 drivers/net/mlx5/linux/mlx5_os.c       |  10 +
 drivers/net/mlx5/mlx5.h                |  19 ++
 drivers/net/mlx5/mlx5_rxq.c            |   4 +-
 drivers/net/mlx5/mlx5_rxtx.h           |   2 +
 drivers/net/mlx5/mlx5_trigger.c        | 502 ++++++++++++++++++++++++++++++++-
 drivers/net/mlx5/mlx5_txq.c            |   4 +-
 7 files changed, 536 insertions(+), 10 deletions(-)
Signature · 1 line
1.8.3.1
author
#1 lore raw

PATCH 1/4 net/mlx5: remove hairpin queue peer port checking

In the current implementation of single port mode hairpin, the peer queue should belong to the same port of the current port. When two ports hairpin mode is introduced, the checking should be removed to make the hairpin queue setup execute successfully.

---
 drivers/net/mlx5/mlx5_rxq.c | 4 +---
 drivers/net/mlx5/mlx5_txq.c | 4 +---
 2 files changed, 2 insertions(+), 6 deletions(-)
drivers/net/mlx5/mlx5_rxq.c +1 -3
index f1d8373..66abce7 100644
@@ -776,9 +776,7 @@
 	res = mlx5_rx_queue_pre_setup(dev, idx, &desc);
 	if (res)
 		return res;
-	if (hairpin_conf->peer_count != 1 ||
-	    hairpin_conf->peers[0].port != dev->data->port_id ||
-	    hairpin_conf->peers[0].queue >= priv->txqs_n) {
+	if (hairpin_conf->peer_count != 1) {
 		DRV_LOG(ERR, "port %u unable to setup hairpin queue index %u "
 			" invalid hairpind configuration", dev->data->port_id,
 			idx);
drivers/net/mlx5/mlx5_txq.c +1 -3
index af84f5f..17a9f5a 100644
@@ -421,9 +421,7 @@
 	res = mlx5_tx_queue_pre_setup(dev, idx, &desc);
 	if (res)
 		return res;
-	if (hairpin_conf->peer_count != 1 ||
-	    hairpin_conf->peers[0].port != dev->data->port_id ||
-	    hairpin_conf->peers[0].queue >= priv->rxqs_n) {
+	if (hairpin_conf->peer_count != 1) {
 		DRV_LOG(ERR, "port %u unable to setup hairpin queue index %u "
 			" invalid hairpind configuration", dev->data->port_id,
 			idx);
Signature · 1 line
1.8.3.1
author
#2 lore raw

PATCH 2/4 net/mlx5: add support for two ports hairpin mode

In reply to Bing Zhao

In order to support hairpin between two ports, mlx5 PMD needs to implement the functions and provide them as the function pointers.

The bind and unbind functions are executed per port pairs. All the hairpin queues between the two ports should have the same attributes during queues setup. Different configurations among queue pairs from the same ports are not supported. It is allowed that two ports only have one direction hairpin.

In order to set up the connection between two queues, peer RX queue HW information must be fetched via the internal RTE API and the queue information could be used to modify the SQ object. Then the RQ object will be modified with the TX queue HW information. The reverse operation is not supported right now.

When disconnecting the queues pair, SQ and RQ object should be reset without any peer HW information. The unbinding operation will try to disconnect all TX queues from the port from the RX queues of the peer port.

TX explicit mode attribute will be saved and used when creating a hairpin flow.

---
 drivers/net/mlx5/linux/mlx5_os.c |  10 +
 drivers/net/mlx5/mlx5.h          |  19 ++
 drivers/net/mlx5/mlx5_rxtx.h     |   2 +
 drivers/net/mlx5/mlx5_trigger.c  | 470 ++++++++++++++++++++++++++++++++++++++-
 4 files changed, 499 insertions(+), 2 deletions(-)
drivers/net/mlx5/linux/mlx5_os.c +10 -0
index 487714f..ee8e1bb 100644
@@ -2530,6 +2530,11 @@
 	.get_module_eeprom = mlx5_get_module_eeprom,
 	.hairpin_cap_get = mlx5_hairpin_cap_get,
 	.mtr_ops_get = mlx5_flow_meter_ops_get,
+	.hairpin_bind = mlx5_hairpin_bind,
+	.hairpin_unbind = mlx5_hairpin_unbind,
+	.hairpin_queue_peer_update = mlx5_hairpin_queue_peer_update,
+	.hairpin_queue_peer_bind = mlx5_hairpin_queue_peer_bind,
+	.hairpin_queue_peer_unbind = mlx5_hairpin_queue_peer_unbind,
 };
 
 /* Available operations from secondary process. */
@@ -2608,4 +2613,9 @@
 	.get_module_eeprom = mlx5_get_module_eeprom,
 	.hairpin_cap_get = mlx5_hairpin_cap_get,
 	.mtr_ops_get = mlx5_flow_meter_ops_get,
+	.hairpin_bind = mlx5_hairpin_bind,
+	.hairpin_unbind = mlx5_hairpin_unbind,
+	.hairpin_queue_peer_update = mlx5_hairpin_queue_peer_update,
+	.hairpin_queue_peer_bind = mlx5_hairpin_queue_peer_bind,
+	.hairpin_queue_peer_unbind = mlx5_hairpin_queue_peer_unbind,
 };
drivers/net/mlx5/mlx5.h +19 -0
index 87d3c15..80d0859 100644
@@ -878,6 +878,14 @@ struct mlx5_priv {
 #define PORT_ID(priv) ((priv)->dev_data->port_id)
 #define ETH_DEV(priv) (&rte_eth_devices[PORT_ID(priv)])
 
+struct rte_hairpin_peer_info {
+	uint32_t qp_id;
+	uint32_t vhca_id;
+	uint16_t peer_q;
+	uint16_t tx_explicit;
+	uint16_t manual_bind;
+};
+
 /* mlx5.c */
 
 int mlx5_getenv_int(const char *);
@@ -1028,6 +1036,17 @@ void mlx5_vlan_vmwa_acquire(struct rte_eth_dev *dev,
 int mlx5_traffic_enable(struct rte_eth_dev *dev);
 void mlx5_traffic_disable(struct rte_eth_dev *dev);
 int mlx5_traffic_restart(struct rte_eth_dev *dev);
+int mlx5_hairpin_queue_peer_update(struct rte_eth_dev *dev, uint16_t peer_queue,
+				   struct rte_hairpin_peer_info *current_info,
+				   struct rte_hairpin_peer_info *peer_info,
+				   uint32_t direction);
+int mlx5_hairpin_queue_peer_bind(struct rte_eth_dev *dev, uint16_t cur_queue,
+				 struct rte_hairpin_peer_info *peer_info,
+				 uint32_t direction);
+int mlx5_hairpin_queue_peer_unbind(struct rte_eth_dev *dev, uint16_t cur_queue,
+				   uint32_t direction);
+int mlx5_hairpin_bind(struct rte_eth_dev *dev, uint16_t rx_port);
+int mlx5_hairpin_unbind(struct rte_eth_dev *dev, uint16_t rx_port);
 
 /* mlx5_flow.c */
 
drivers/net/mlx5/mlx5_rxtx.h +2 -0
index 674296e..ac612ca 100644
@@ -184,6 +184,7 @@ struct mlx5_rxq_ctrl {
 	void *wq_umem; /* WQ buffer registration info. */
 	void *cq_umem; /* CQ buffer registration info. */
 	struct rte_eth_hairpin_conf hairpin_conf; /* Hairpin configuration. */
+	uint32_t hairpin_status;
 };
 
 /* TX queue send local data. */
@@ -279,6 +280,7 @@ struct mlx5_txq_ctrl {
 	off_t uar_mmap_offset; /* UAR mmap offset for non-primary process. */
 	void *bf_reg; /* BlueFlame register from Verbs. */
 	uint16_t dump_file_n; /* Number of dump files. */
+	uint32_t hairpin_status;
 	struct rte_eth_hairpin_conf hairpin_conf; /* Hairpin configuration. */
 	struct mlx5_txq_data txq; /* Data path structure. */
 	/* Must be the last field in the structure, contains elts[]. */
drivers/net/mlx5/mlx5_trigger.c +468 -2

488 lines not on this page · raw message

Signature · 1 line
1.8.3.1
author
#3 lore raw

PATCH 3/4 net/mlx5: conditional hairpin auto bind

In reply to Bing Zhao

In single port hairpin mode, after the queues are configured during start up. The binding process will be enabled automatically in the port start phase and the default control flow for egress will be created.

When switching to two ports hairpin mode, the auto binding process should be skipped if there is no TX queue with the peer RX queue on the same device, and it should be skipped also if the queues are configured with manual bind attribute.

If the explicit TX flow rule mode is configured or hairpin is between two ports, the default control flows for TX queues should not be created.

---
 drivers/net/mlx5/mlx5_trigger.c | 32 ++++++++++++++++++++++++++++++--
 1 file changed, 30 insertions(+), 2 deletions(-)
drivers/net/mlx5/mlx5_trigger.c +30 -2
index f326b57..77d84dd 100644
@@ -214,6 +214,8 @@
 	struct mlx5_devx_obj *rq;
 	unsigned int i;
 	int ret = 0;
+	bool need_auto = false;
+	uint16_t self_port = dev->data->port_id;
 
 	for (i = 0; i != priv->txqs_n; ++i) {
 		txq_ctrl = mlx5_txq_get(dev, i);
@@ -223,6 +225,25 @@
 			mlx5_txq_release(dev, i);
 			continue;
 		}
+		if (txq_ctrl->hairpin_conf.peers[0].port != self_port)
+			continue;
+		if (txq_ctrl->hairpin_conf.manual_bind) {
+			mlx5_txq_release(dev, i);
+			return 0;
+		}
+		need_auto = true;
+		mlx5_txq_release(dev, i);
+	}
+	if (!need_auto)
+		return 0;
+	for (i = 0; i != priv->txqs_n; ++i) {
+		txq_ctrl = mlx5_txq_get(dev, i);
+		if (!txq_ctrl)
+			continue;
+		if (txq_ctrl->type != MLX5_TXQ_TYPE_HAIRPIN) {
+			mlx5_txq_release(dev, i);
+			continue;
+		}
 		if (!txq_ctrl->obj) {
 			rte_errno = ENOMEM;
 			DRV_LOG(ERR, "port %u no txq object found: %d",
@@ -798,9 +819,13 @@
 			dev->data->port_id, strerror(rte_errno));
 		goto error;
 	}
+	/*
+	 * Such step will be skipped if there is no hairpin TX queue configured
+	 * with RX peer queue from the same device.
+	 */
 	ret = mlx5_hairpin_auto_bind(dev);
 	if (ret) {
-		DRV_LOG(ERR, "port %u hairpin binding failed: %s",
+		DRV_LOG(ERR, "port %u hairpin auto binding failed: %s",
 			dev->data->port_id, strerror(rte_errno));
 		goto error;
 	}
@@ -949,7 +974,10 @@
 		struct mlx5_txq_ctrl *txq_ctrl = mlx5_txq_get(dev, i);
 		if (!txq_ctrl)
 			continue;
-		if (txq_ctrl->type == MLX5_TXQ_TYPE_HAIRPIN) {
+		if (txq_ctrl->type == MLX5_TXQ_TYPE_HAIRPIN &&
+		    !txq_ctrl->hairpin_conf.manual_bind &&
+		    txq_ctrl->hairpin_conf.peers[0].port ==
+		    priv->dev_data->port_id) {
 			ret = mlx5_ctrl_flow_source_queue(dev, i);
 			if (ret) {
 				mlx5_txq_release(dev, i);
Signature · 1 line
1.8.3.1
author
#4 lore raw

PATCH 4/4 doc: update hairpin support for mlx5 driver

In reply to Bing Zhao

Hairpin between two ports will be supported by mlx5 PMD.

---
 doc/guides/rel_notes/release_20_11.rst | 5 +++++
 1 file changed, 5 insertions(+)
doc/guides/rel_notes/release_20_11.rst +5 -0
index 05ceea0..454472b 100644
@@ -70,6 +70,11 @@ New Features
   * Added support for non-zero priorities for group 0 flows
   * Added support for VXLAN decap combined with VLAN pop
 
+* **Updated Nvidia mlx5 driver.**
+
+  * Added support for hairpin between two ports and hairpin explicit
+    TX flow rules insertion.
+
 * **Updated Solarflare network PMD.**
 
   Updated the Solarflare ``sfc_efx`` driver with changes including:
Signature · 1 line
1.8.3.1
author
#5 lore raw

v2 cover add two ports hairpin mode support in mlx5 PMD

In reply to Bing Zhao

This patch set will add the support for hairpin between two ports in mlx5 PMD.

v2:
* Update the code and reorganize the patch set

Bing Zhao (6):
net/mlx5: change hairpin queue peer checking net/mlx5: add support for two ports hairpin mode net/mlx5: add support to get hairpin peer ports net/mlx5: conditional hairpin auto bind net/mlx5: change hairpin ingress flow validation net/mlx5: not split hairpin flow in explicit mode

 drivers/net/mlx5/linux/mlx5_os.c |  12 +
 drivers/net/mlx5/mlx5.h          |  21 ++
 drivers/net/mlx5/mlx5_flow.c     |   7 +
 drivers/net/mlx5/mlx5_flow_dv.c  |  15 +-
 drivers/net/mlx5/mlx5_rxq.c      |  50 ++-
 drivers/net/mlx5/mlx5_rxtx.h     |   4 +
 drivers/net/mlx5/mlx5_trigger.c  | 733 ++++++++++++++++++++++++++++++++++++++-
 drivers/net/mlx5/mlx5_txq.c      |  23 +-
 8 files changed, 846 insertions(+), 19 deletions(-)
Signature · 1 line
1.8.3.1
author
#6 lore raw

PATCH v2 1/6 net/mlx5: change hairpin queue peer checking

In the current implementation of single port mode hairpin, the peer queue should belong to the same port of the current queue. When the two ports hairpin mode is introduced, such checking should be removed to make the hairpin queue setup execute successfully since it is not a valid condition anymore.

In the meanwhile, different devices could have different queue configurations. The queues number of peer port is unknown to the current device. The checking should be removed also.

If the Tx and Rx port IDs of a hairpin peer are different, only the manual binding and explicit Tx flows are supported. Or else, the four combinations of modes could be supported. The mode attributes consistency checking will be done when connecting the queue with its peer queue.

---
 drivers/net/mlx5/mlx5_rxq.c | 23 +++++++++++++++++------
 drivers/net/mlx5/mlx5_txq.c | 23 +++++++++++++++++------
 2 files changed, 34 insertions(+), 12 deletions(-)
drivers/net/mlx5/mlx5_rxq.c +17 -6
index e1783ba..78e15e7 100644
@@ -777,15 +777,26 @@
 	res = mlx5_rx_queue_pre_setup(dev, idx, &desc);
 	if (res)
 		return res;
-	if (hairpin_conf->peer_count != 1 ||
-	    hairpin_conf->peers[0].port != dev->data->port_id ||
-	    hairpin_conf->peers[0].queue >= priv->txqs_n) {
-		DRV_LOG(ERR, "port %u unable to setup hairpin queue index %u "
-			" invalid hairpind configuration", dev->data->port_id,
-			idx);
+	if (hairpin_conf->peer_count != 1) {
 		rte_errno = EINVAL;
+		DRV_LOG(ERR, "port %u unable to setup Rx hairpin queue index %u"
+			" peer count is %u", dev->data->port_id,
+			idx, hairpin_conf->peer_count);
 		return -rte_errno;
 	}
+	if (hairpin_conf->peers[0].port != dev->data->port_id) {
+		if (hairpin_conf->manual_bind == 0 ||
+		    hairpin_conf->tx_explicit == 0) {
+			rte_errno = EINVAL;
+			DRV_LOG(ERR, "port %u unable to setup Rx hairpin queue"
+				" index %u peer port %u with attributes %u %u",
+				dev->data->port_id, idx,
+				hairpin_conf->peers[0].port,
+				hairpin_conf->manual_bind,
+				hairpin_conf->tx_explicit);
+			return -rte_errno;
+		}
+	}
 	rxq_ctrl = mlx5_rxq_hairpin_new(dev, idx, desc, hairpin_conf);
 	if (!rxq_ctrl) {
 		DRV_LOG(ERR, "port %u unable to allocate queue index %u",
drivers/net/mlx5/mlx5_txq.c +17 -6
index 9c2dd2a..850a85c 100644
@@ -421,15 +421,26 @@
 	res = mlx5_tx_queue_pre_setup(dev, idx, &desc);
 	if (res)
 		return res;
-	if (hairpin_conf->peer_count != 1 ||
-	    hairpin_conf->peers[0].port != dev->data->port_id ||
-	    hairpin_conf->peers[0].queue >= priv->rxqs_n) {
-		DRV_LOG(ERR, "port %u unable to setup hairpin queue index %u "
-			" invalid hairpind configuration", dev->data->port_id,
-			idx);
+	if (hairpin_conf->peer_count != 1) {
 		rte_errno = EINVAL;
+		DRV_LOG(ERR, "port %u unable to setup Tx hairpin queue index %u"
+			" peer count is %u", dev->data->port_id,
+			idx, hairpin_conf->peer_count);
 		return -rte_errno;
 	}
+	if (hairpin_conf->peers[0].port != dev->data->port_id) {
+		if (hairpin_conf->manual_bind == 0 ||
+		    hairpin_conf->tx_explicit == 0) {
+			rte_errno = EINVAL;
+			DRV_LOG(ERR, "port %u unable to setup Tx hairpin queue"
+				" index %u peer port %u with attributes %u %u",
+				dev->data->port_id, idx,
+				hairpin_conf->peers[0].port,
+				hairpin_conf->manual_bind,
+				hairpin_conf->tx_explicit);
+			return -rte_errno;
+		}
+	}
 	txq_ctrl = mlx5_txq_hairpin_new(dev, idx, desc,	hairpin_conf);
 	if (!txq_ctrl) {
 		DRV_LOG(ERR, "port %u unable to allocate queue index %u",
Signature · 1 line
1.8.3.1
Acked-by on v2 1/6
#7 lore raw

Re: [dpdk-dev] [PATCH v2 1/6] net/mlx5: change hairpin queue peer checking

Quoted 19 lines

-----Original Message-----
From: dev <[email protected]> On Behalf Of Bing Zhao
Sent: Thursday, October 22, 2020 17:07
To: [email protected]; [email protected]
Cc: [email protected]; Ori Kam <[email protected]>; Raslan Darawsheh <[email protected]>
Subject: [dpdk-dev] [PATCH v2 1/6] net/mlx5: change hairpin queue peer checking

In the current implementation of single port mode hairpin, the peer queue should belong to the same port of the current queue. When the two ports hairpin mode is introduced, such checking should be removed to make the hairpin queue setup execute successfully since it is not a valid condition anymore.

In the meanwhile, different devices could have different queue configurations. The queues number of peer port is unknown to the current device. The checking should be removed also.

Quoted 6 lines

If the Tx and Rx port IDs of a hairpin peer are different, only the manual binding and explicit Tx flows are supported. Or else, the four combinations of modes could be supported. The mode attributes consistency checking will be done when connecting the queue with its peer queue.

Quoted 4 lines
---
 drivers/net/mlx5/mlx5_rxq.c | 23 +++++++++++++++++------
drivers/net/mlx5/mlx5_txq.c | 23 +++++++++++++++++------
 2 files changed, 34 insertions(+), 12 deletions(-)
author
#8 lore raw

PATCH v2 2/6 net/mlx5: add support for two ports hairpin mode

In reply to Bing Zhao

In order to support hairpin between two ports, mlx5 PMD needs to implement the functions and provide them as the function pointers.

The bind and unbind functions are executed per port pairs. All the hairpin queues between the two ports should have the same attributes during queues setup. Different configurations among queue pairs from the same ports are not supported. It is allowed that two ports only have one direction hairpin.

In order to set up the connection between two queues, peer Rx queue HW information must be fetched via the internal RTE API and the queue information could be used to modify the SQ object. Then the RQ object will be modified with the Tx queue HW information. The reverse operation is not supported right now.

When disconnecting the queues pair, SQ and RQ object should be reset without any peer HW information. The unbinding operation will try to disconnect all Tx queues from the port from the Rx queues of the peer port.

Tx explicit mode attribute will be saved and used when creating a hairpin flow.

---
 drivers/net/mlx5/linux/mlx5_os.c |  10 +
 drivers/net/mlx5/mlx5.h          |  19 ++
 drivers/net/mlx5/mlx5_rxtx.h     |   2 +
 drivers/net/mlx5/mlx5_trigger.c  | 611 ++++++++++++++++++++++++++++++++++++++-
 4 files changed, 640 insertions(+), 2 deletions(-)
drivers/net/mlx5/linux/mlx5_os.c +10 -0
index 40f9446..83a8b56 100644
@@ -2552,6 +2552,11 @@
 	.get_module_eeprom = mlx5_get_module_eeprom,
 	.hairpin_cap_get = mlx5_hairpin_cap_get,
 	.mtr_ops_get = mlx5_flow_meter_ops_get,
+	.hairpin_bind = mlx5_hairpin_bind,
+	.hairpin_unbind = mlx5_hairpin_unbind,
+	.hairpin_queue_peer_update = mlx5_hairpin_queue_peer_update,
+	.hairpin_queue_peer_bind = mlx5_hairpin_queue_peer_bind,
+	.hairpin_queue_peer_unbind = mlx5_hairpin_queue_peer_unbind,
 };
 
 /* Available operations from secondary process. */
@@ -2630,4 +2635,9 @@
 	.get_module_eeprom = mlx5_get_module_eeprom,
 	.hairpin_cap_get = mlx5_hairpin_cap_get,
 	.mtr_ops_get = mlx5_flow_meter_ops_get,
+	.hairpin_bind = mlx5_hairpin_bind,
+	.hairpin_unbind = mlx5_hairpin_unbind,
+	.hairpin_queue_peer_update = mlx5_hairpin_queue_peer_update,
+	.hairpin_queue_peer_bind = mlx5_hairpin_queue_peer_bind,
+	.hairpin_queue_peer_unbind = mlx5_hairpin_queue_peer_unbind,
 };
drivers/net/mlx5/mlx5.h +19 -0
index c9d5d71..38d0977 100644
@@ -891,6 +891,14 @@ struct mlx5_priv {
 #define PORT_ID(priv) ((priv)->dev_data->port_id)
 #define ETH_DEV(priv) (&rte_eth_devices[PORT_ID(priv)])
 
+struct rte_hairpin_peer_info {
+	uint32_t qp_id;
+	uint32_t vhca_id;
+	uint16_t peer_q;
+	uint16_t tx_explicit;
+	uint16_t manual_bind;
+};
+
 /* mlx5.c */
 
 int mlx5_getenv_int(const char *);
@@ -1041,6 +1049,17 @@ void mlx5_vlan_vmwa_acquire(struct rte_eth_dev *dev,
 int mlx5_traffic_enable(struct rte_eth_dev *dev);
 void mlx5_traffic_disable(struct rte_eth_dev *dev);
 int mlx5_traffic_restart(struct rte_eth_dev *dev);
+int mlx5_hairpin_queue_peer_update(struct rte_eth_dev *dev, uint16_t peer_queue,
+				   struct rte_hairpin_peer_info *current_info,
+				   struct rte_hairpin_peer_info *peer_info,
+				   uint32_t direction);
+int mlx5_hairpin_queue_peer_bind(struct rte_eth_dev *dev, uint16_t cur_queue,
+				 struct rte_hairpin_peer_info *peer_info,
+				 uint32_t direction);
+int mlx5_hairpin_queue_peer_unbind(struct rte_eth_dev *dev, uint16_t cur_queue,
+				   uint32_t direction);
+int mlx5_hairpin_bind(struct rte_eth_dev *dev, uint16_t rx_port);
+int mlx5_hairpin_unbind(struct rte_eth_dev *dev, uint16_t rx_port);
 
 /* mlx5_flow.c */
 
drivers/net/mlx5/mlx5_rxtx.h +2 -0
index b243b6f..b50b643 100644
@@ -184,6 +184,7 @@ struct mlx5_rxq_ctrl {
 	void *wq_umem; /* WQ buffer registration info. */
 	void *cq_umem; /* CQ buffer registration info. */
 	struct rte_eth_hairpin_conf hairpin_conf; /* Hairpin configuration. */
+	uint32_t hairpin_status; /* Hairpin binding status. */
 };
 
 /* TX queue send local data. */
@@ -280,6 +281,7 @@ struct mlx5_txq_ctrl {
 	void *bf_reg; /* BlueFlame register from Verbs. */
 	uint16_t dump_file_n; /* Number of dump files. */
 	struct rte_eth_hairpin_conf hairpin_conf; /* Hairpin configuration. */
+	uint32_t hairpin_status; /* Hairpin binding status. */
 	struct mlx5_txq_data txq; /* Data path structure. */
 	/* Must be the last field in the structure, contains elts[]. */
 };
drivers/net/mlx5/mlx5_trigger.c +609 -2

629 lines not on this page · raw message

Signature · 1 line
1.8.3.1
Acked-by on v2 2/6
#9 lore raw

Re: [dpdk-dev] [PATCH v2 2/6] net/mlx5: add support for two ports hairpin mode

Quoted 22 lines

-----Original Message-----
From: Bing Zhao <[email protected]>
Sent: Thursday, October 22, 2020 17:07
To: [email protected]; [email protected]
Cc: [email protected]; Ori Kam <[email protected]>; Raslan Darawsheh <[email protected]>
Subject: [PATCH v2 2/6] net/mlx5: add support for two ports hairpin mode

In order to support hairpin between two ports, mlx5 PMD needs to implement the functions and provide them as the function pointers.

The bind and unbind functions are executed per port pairs. All the hairpin queues between the two ports should have the same attributes during queues setup. Different configurations among queue pairs from the same ports are not supported. It is allowed that two ports only have one direction hairpin.

In order to set up the connection between two queues, peer Rx queue HW information must be fetched via the internal RTE API and the queue information could be used to modify the SQ object. Then the RQ object will be modified with the Tx queue HW information. The reverse operation is not supported right now.

Quoted 7 lines

When disconnecting the queues pair, SQ and RQ object should be reset without any peer HW information. The unbinding operation will try to disconnect all Tx queues from the port from the Rx queues of the peer port.

Tx explicit mode attribute will be saved and used when creating a hairpin flow.

Quoted 8 lines
---
 drivers/net/mlx5/linux/mlx5_os.c |  10 +
 drivers/net/mlx5/mlx5.h          |  19 ++
 drivers/net/mlx5/mlx5_rxtx.h     |   2 +
 drivers/net/mlx5/mlx5_trigger.c  | 611
++++++++++++++++++++++++++++++++++++++-
 4 files changed, 640 insertions(+), 2 deletions(-)
author
#10 lore raw

PATCH v2 3/6 net/mlx5: add support to get hairpin peer ports

In reply to Bing Zhao

In real-life business, one device could be attached and detached dynamically. The hairpin configuration of this port to/from all the other ports should be enabled and disabled accordingly.

The RTE ethdev lib and PMD should provide this ability to get the peer ports list in case that the application doesn't save it. It is recommended that the size of the array to save the port IDs is as large as the "RTE_MAX_ETHPORTS" to have the maximal capacity.

The order of the peer port IDs may be different from that during hairpin queues set in the initialization stage. The peer port ID could be the same as the current device port ID when the hairpin peer ports contain itself - the single port hairpin.

The application should check the ports' status and decide if the peer port should be bound / unbound when starting / stopping the current device.

---
 drivers/net/mlx5/linux/mlx5_os.c |  2 +
 drivers/net/mlx5/mlx5.h          |  2 +
 drivers/net/mlx5/mlx5_trigger.c  | 89 ++++++++++++++++++++++++++++++++++++++++
 3 files changed, 93 insertions(+)
drivers/net/mlx5/linux/mlx5_os.c +2 -0
index 83a8b56..17d3767 100644
@@ -2554,6 +2554,7 @@
 	.mtr_ops_get = mlx5_flow_meter_ops_get,
 	.hairpin_bind = mlx5_hairpin_bind,
 	.hairpin_unbind = mlx5_hairpin_unbind,
+	.hairpin_get_peer_ports = mlx5_hairpin_get_peer_ports,
 	.hairpin_queue_peer_update = mlx5_hairpin_queue_peer_update,
 	.hairpin_queue_peer_bind = mlx5_hairpin_queue_peer_bind,
 	.hairpin_queue_peer_unbind = mlx5_hairpin_queue_peer_unbind,
@@ -2637,6 +2638,7 @@
 	.mtr_ops_get = mlx5_flow_meter_ops_get,
 	.hairpin_bind = mlx5_hairpin_bind,
 	.hairpin_unbind = mlx5_hairpin_unbind,
+	.hairpin_get_peer_ports = mlx5_hairpin_get_peer_ports,
 	.hairpin_queue_peer_update = mlx5_hairpin_queue_peer_update,
 	.hairpin_queue_peer_bind = mlx5_hairpin_queue_peer_bind,
 	.hairpin_queue_peer_unbind = mlx5_hairpin_queue_peer_unbind,
drivers/net/mlx5/mlx5.h +2 -0
index 38d0977..70a0d3d 100644
@@ -1060,6 +1060,8 @@ int mlx5_hairpin_queue_peer_unbind(struct rte_eth_dev *dev, uint16_t cur_queue,
 				   uint32_t direction);
 int mlx5_hairpin_bind(struct rte_eth_dev *dev, uint16_t rx_port);
 int mlx5_hairpin_unbind(struct rte_eth_dev *dev, uint16_t rx_port);
+int mlx5_hairpin_get_peer_ports(struct rte_eth_dev *dev, uint16_t *peer_ports,
+				size_t len, uint32_t direction);
 
 /* mlx5_flow.c */
 
drivers/net/mlx5/mlx5_trigger.c +89 -0

95 lines not on this page · raw message

Signature · 1 line
1.8.3.1
Acked-by on v2 3/6
#11 lore raw

Re: [dpdk-dev] [PATCH v2 3/6] net/mlx5: add support to get hairpin peer ports

Quoted 22 lines

-----Original Message-----
From: Bing Zhao <[email protected]>
Sent: Thursday, October 22, 2020 17:07
To: [email protected]; [email protected]
Cc: [email protected]; Ori Kam <[email protected]>; Raslan Darawsheh <[email protected]>
Subject: [PATCH v2 3/6] net/mlx5: add support to get hairpin peer ports

In real-life business, one device could be attached and detached dynamically. The hairpin configuration of this port to/from all the other ports should be enabled and disabled accordingly.

The RTE ethdev lib and PMD should provide this ability to get the peer ports list in case that the application doesn't save it. It is recommended that the size of the array to save the port IDs is as large as the "RTE_MAX_ETHPORTS" to have the maximal capacity.

The order of the peer port IDs may be different from that during hairpin queues set in the initialization stage. The peer port ID could be the same as the current device port ID when the hairpin peer ports contain itself - the single port hairpin.

Quoted 4 lines

The application should check the ports' status and decide if the peer port should be bound / unbound when starting / stopping the current device.

Quoted 7 lines
---
 drivers/net/mlx5/linux/mlx5_os.c |  2 +
 drivers/net/mlx5/mlx5.h          |  2 +
 drivers/net/mlx5/mlx5_trigger.c  | 89
++++++++++++++++++++++++++++++++++++++++
 3 files changed, 93 insertions(+)
author
#12 lore raw

PATCH v2 4/6 net/mlx5: conditional hairpin auto bind

In reply to Bing Zhao

In single port hairpin mode, after the queues are configured during start up. The binding process will be enabled automatically in the port start phase and the default control flow for egress will be created.

When switching to two ports hairpin mode, the auto binding process should be skipped if there is no TX queue with the peer RX queue on the same device, and it should be skipped also if the queues are configured with manual bind attribute.

If the explicit TX flow rule mode is configured or hairpin is between two ports, the default control flows for TX queues should not be created.

---
 drivers/net/mlx5/mlx5_trigger.c | 33 +++++++++++++++++++++++++++++++--
 1 file changed, 31 insertions(+), 2 deletions(-)
drivers/net/mlx5/mlx5_trigger.c +31 -2
index 497f731..27bd3e9 100644
@@ -214,6 +214,8 @@
 	struct mlx5_devx_obj *rq;
 	unsigned int i;
 	int ret = 0;
+	bool need_auto = false;
+	uint16_t self_port = dev->data->port_id;
 
 	for (i = 0; i != priv->txqs_n; ++i) {
 		txq_ctrl = mlx5_txq_get(dev, i);
@@ -223,6 +225,25 @@
 			mlx5_txq_release(dev, i);
 			continue;
 		}
+		if (txq_ctrl->hairpin_conf.peers[0].port != self_port)
+			continue;
+		if (txq_ctrl->hairpin_conf.manual_bind) {
+			mlx5_txq_release(dev, i);
+			return 0;
+		}
+		need_auto = true;
+		mlx5_txq_release(dev, i);
+	}
+	if (!need_auto)
+		return 0;
+	for (i = 0; i != priv->txqs_n; ++i) {
+		txq_ctrl = mlx5_txq_get(dev, i);
+		if (!txq_ctrl)
+			continue;
+		if (txq_ctrl->type != MLX5_TXQ_TYPE_HAIRPIN) {
+			mlx5_txq_release(dev, i);
+			continue;
+		}
 		if (!txq_ctrl->obj) {
 			rte_errno = ENOMEM;
 			DRV_LOG(ERR, "port %u no txq object found: %d",
@@ -1028,9 +1049,13 @@
 			dev->data->port_id, strerror(rte_errno));
 		goto error;
 	}
+	/*
+	 * Such step will be skipped if there is no hairpin TX queue configured
+	 * with RX peer queue from the same device.
+	 */
 	ret = mlx5_hairpin_auto_bind(dev);
 	if (ret) {
-		DRV_LOG(ERR, "port %u hairpin binding failed: %s",
+		DRV_LOG(ERR, "port %u hairpin auto binding failed: %s",
 			dev->data->port_id, strerror(rte_errno));
 		goto error;
 	}
@@ -1181,7 +1206,11 @@
 		struct mlx5_txq_ctrl *txq_ctrl = mlx5_txq_get(dev, i);
 		if (!txq_ctrl)
 			continue;
-		if (txq_ctrl->type == MLX5_TXQ_TYPE_HAIRPIN) {
+		/* Only Tx implicit mode requires the default Tx flow. */
+		if (txq_ctrl->type == MLX5_TXQ_TYPE_HAIRPIN &&
+		    txq_ctrl->hairpin_conf.tx_explicit == 0 &&
+		    txq_ctrl->hairpin_conf.peers[0].port ==
+		    priv->dev_data->port_id) {
 			ret = mlx5_ctrl_flow_source_queue(dev, i);
 			if (ret) {
 				mlx5_txq_release(dev, i);
Signature · 1 line
1.8.3.1
Acked-by on v2 4/6
#13 lore raw

Re: [dpdk-dev] [PATCH v2 4/6] net/mlx5: conditional hairpin auto bind

Quoted 17 lines

-----Original Message-----
From: Bing Zhao <[email protected]>
Sent: Thursday, October 22, 2020 17:07
To: [email protected]; [email protected]
Cc: [email protected]; Ori Kam <[email protected]>; Raslan Darawsheh <[email protected]>
Subject: [PATCH v2 4/6] net/mlx5: conditional hairpin auto bind

In single port hairpin mode, after the queues are configured during start up. The binding process will be enabled automatically in the port start phase and the default control flow for egress will be created.

When switching to two ports hairpin mode, the auto binding process should be skipped if there is no TX queue with the peer RX queue on the same device, and it should be skipped also if the queues are configured with manual bind attribute.

Quoted 4 lines

If the explicit TX flow rule mode is configured or hairpin is between two ports, the default control flows for TX queues should not be created.

Quoted 4 lines
---
 drivers/net/mlx5/mlx5_trigger.c | 33 +++++++++++++++++++++++++++++++--
 1 file changed, 31 insertions(+), 2 deletions(-)
author
#14 lore raw

PATCH v2 5/6 net/mlx5: change hairpin ingress flow validation

In reply to Bing Zhao

In the current implementation of the single port hairpin, there is a implicit splitting process for actions. When inserting a hairpin flow, all the actions will be included with the ingress attribute. The flow engine will check and decide which actions should be moved into the TX flow part, e.g., encapsulation, VLAN push.

In some NICs, some actions can only be done in one direction. Since the hairpin flow will be split into two parts, such validation will be skipped.

With the hairpin explicit TX flow mode, no splitting is needed any more. The hairpin flow may have no big difference from a standard flow (except the queue). The application should take full charge of the actions and the flow engine should validate the hairpin flow in the same way as other flows.

In the meanwhile, a new internal API is added to get the hairpin configuration. This will bypass the useless atomic operation to save the CPU cycles.

---
 drivers/net/mlx5/mlx5_flow_dv.c | 15 ++++++++++++---
 drivers/net/mlx5/mlx5_rxq.c     | 27 +++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_rxtx.h    |  2 ++
 3 files changed, 41 insertions(+), 3 deletions(-)
drivers/net/mlx5/mlx5_flow_dv.c +12 -3
index 15cd34e..d5be6f0 100644
@@ -6058,11 +6058,17 @@ struct field_modify_info modify_tcp[] = {
 						  actions,
 						  "no fate action is found");
 	}
-	/* Continue validation for Xcap and VLAN actions.*/
+	/*
+	 * Continue validation for Xcap and VLAN actions.
+	 * If hairpin is working in explicit TX rule mode, there is no actions
+	 * splitting and the validation of hairpin ingress flow should be the
+	 * same as other standard flows.
+	 */
 	if ((action_flags & (MLX5_FLOW_XCAP_ACTIONS |
 			     MLX5_FLOW_VLAN_ACTIONS)) &&
 	    (queue_index == 0xFFFF ||
-	     mlx5_rxq_get_type(dev, queue_index) != MLX5_RXQ_TYPE_HAIRPIN)) {
+	     mlx5_rxq_get_type(dev, queue_index) != MLX5_RXQ_TYPE_HAIRPIN ||
+	     !!mlx5_rxq_get_hairpin_conf(dev, queue_index)->tx_explicit)) {
 		if ((action_flags & MLX5_FLOW_XCAP_ACTIONS) ==
 		    MLX5_FLOW_XCAP_ACTIONS)
 			return rte_flow_error_set(error, ENOTSUP,
@@ -6091,7 +6097,10 @@ struct field_modify_info modify_tcp[] = {
 						 "multiple VLAN actions");
 		}
 	}
-	/* Hairpin flow will add one more TAG action. */
+	/*
+	 * Hairpin flow will add one more TAG action in TX implicit mode.
+	 * In TX explicit mode, there will be no hairpin flow ID.
+	 */
 	if (hairpin > 0)
 		rw_act_num += MLX5_ACT_NUM_SET_TAG;
 	/* extra metadata enabled: one more TAG action will be add. */
drivers/net/mlx5/mlx5_rxq.c +27 -0
index 78e15e7..d328d4a 100644
@@ -1720,6 +1720,33 @@ enum mlx5_rxq_type
 	return MLX5_RXQ_TYPE_UNDEFINED;
 }
 
+/*
+ * Get a Rx hairpin queue configuration.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @param idx
+ *   Rx queue index.
+ *
+ * @return
+ *   Pointer to the configuration if a hairpin RX queue, otherwise NULL.
+ */
+const struct rte_eth_hairpin_conf *
+mlx5_rxq_get_hairpin_conf(struct rte_eth_dev *dev, uint16_t idx)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_rxq_ctrl *rxq_ctrl = NULL;
+
+	if (idx < priv->rxqs_n && (*priv->rxqs)[idx]) {
+		rxq_ctrl = container_of((*priv->rxqs)[idx],
+					struct mlx5_rxq_ctrl,
+					rxq);
+		if (rxq_ctrl->type == MLX5_RXQ_TYPE_HAIRPIN)
+			return &rxq_ctrl->hairpin_conf;
+	}
+	return NULL;
+}
+
 /**
  * Get an indirection table.
  *
drivers/net/mlx5/mlx5_rxtx.h +2 -0
index b50b643..d91ed0f 100644
@@ -344,6 +344,8 @@ uint32_t mlx5_hrxq_get(struct rte_eth_dev *dev,
 int mlx5_hrxq_release(struct rte_eth_dev *dev, uint32_t hxrq_idx);
 int mlx5_hrxq_verify(struct rte_eth_dev *dev);
 enum mlx5_rxq_type mlx5_rxq_get_type(struct rte_eth_dev *dev, uint16_t idx);
+const struct rte_eth_hairpin_conf *mlx5_rxq_get_hairpin_conf
+	(struct rte_eth_dev *dev, uint16_t idx);
 struct mlx5_hrxq *mlx5_drop_action_create(struct rte_eth_dev *dev);
 void mlx5_drop_action_destroy(struct rte_eth_dev *dev);
 uint64_t mlx5_get_rx_port_offloads(void);
Signature · 1 line
1.8.3.1
Acked-by on v2 5/6
#15 lore raw

Re: [dpdk-dev] [PATCH v2 5/6] net/mlx5: change hairpin ingress flow validation

Quoted 22 lines

-----Original Message-----
From: Bing Zhao <[email protected]>
Sent: Thursday, October 22, 2020 17:07
To: [email protected]; [email protected]
Cc: [email protected]; Ori Kam <[email protected]>; Raslan Darawsheh <[email protected]>
Subject: [PATCH v2 5/6] net/mlx5: change hairpin ingress flow validation

In the current implementation of the single port hairpin, there is a implicit splitting process for actions. When inserting a hairpin flow, all the actions will be included with the ingress attribute.
The flow engine will check and decide which actions should be moved into the TX flow part, e.g., encapsulation, VLAN push.

In some NICs, some actions can only be done in one direction. Since the hairpin flow will be split into two parts, such validation will be skipped.

With the hairpin explicit TX flow mode, no splitting is needed any more. The hairpin flow may have no big difference from a standard flow (except the queue). The application should take full charge of the actions and the flow engine should validate the hairpin flow in the same way as other flows.

Quoted 4 lines

In the meanwhile, a new internal API is added to get the hairpin configuration. This will bypass the useless atomic operation to save the CPU cycles.

Quoted 96 lines
---
 drivers/net/mlx5/mlx5_flow_dv.c | 15 ++++++++++++---
 drivers/net/mlx5/mlx5_rxq.c     | 27 +++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_rxtx.h    |  2 ++
 3 files changed, 41 insertions(+), 3 deletions(-)
(quoted diff) +0 -0

b/drivers/net/mlx5/mlx5_flow_dv.c index 15cd34e..d5be6f0 100644

drivers/net/mlx5/mlx5_flow_dv.c +7 -2
@@ -6058,11 +6058,17 @@ struct field_modify_info modify_tcp[] = {
 						  actions,
 						  "no fate action is found");
 	}
-	/* Continue validation for Xcap and VLAN actions.*/
+	/*
+	 * Continue validation for Xcap and VLAN actions.
+	 * If hairpin is working in explicit TX rule mode, there is no actions
+	 * splitting and the validation of hairpin ingress flow should be the
+	 * same as other standard flows.
+	 */
 	if ((action_flags & (MLX5_FLOW_XCAP_ACTIONS |
 			     MLX5_FLOW_VLAN_ACTIONS)) &&
 	    (queue_index == 0xFFFF ||
-	     mlx5_rxq_get_type(dev, queue_index) !=
 MLX5_RXQ_TYPE_HAIRPIN)) {
+	     mlx5_rxq_get_type(dev, queue_index) !=

MLX5_RXQ_TYPE_HAIRPIN ||

drivers/net/mlx5/mlx5_flow_dv.c +1 -0
+	     !!mlx5_rxq_get_hairpin_conf(dev, queue_index)->tx_explicit)) {
 		if ((action_flags & MLX5_FLOW_XCAP_ACTIONS) ==
 		    MLX5_FLOW_XCAP_ACTIONS)
 			return rte_flow_error_set(error, ENOTSUP, @@ -

6091,7 +6097,10 @@ struct field_modify_info modify_tcp[] = {

drivers/net/mlx5/mlx5_flow_dv.c +4 -1
 						 "multiple VLAN actions");
 		}
 	}
-	/* Hairpin flow will add one more TAG action. */
+	/*
+	 * Hairpin flow will add one more TAG action in TX implicit mode.
+	 * In TX explicit mode, there will be no hairpin flow ID.
+	 */
 	if (hairpin > 0)
 		rw_act_num += MLX5_ACT_NUM_SET_TAG;
 	/* extra metadata enabled: one more TAG action will be add. */ diff --

git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c index 78e15e7..d328d4a 100644

drivers/net/mlx5/mlx5_rxq.c +26 -0
@@ -1720,6 +1720,33 @@ enum mlx5_rxq_type
 	return MLX5_RXQ_TYPE_UNDEFINED;
 }
 
+/*
+ * Get a Rx hairpin queue configuration.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @param idx
+ *   Rx queue index.
+ *
+ * @return
+ *   Pointer to the configuration if a hairpin RX queue, otherwise NULL.
+ */
+const struct rte_eth_hairpin_conf *
+mlx5_rxq_get_hairpin_conf(struct rte_eth_dev *dev, uint16_t idx) {
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_rxq_ctrl *rxq_ctrl = NULL;
+
+	if (idx < priv->rxqs_n && (*priv->rxqs)[idx]) {
+		rxq_ctrl = container_of((*priv->rxqs)[idx],
+					struct mlx5_rxq_ctrl,
+					rxq);
+		if (rxq_ctrl->type == MLX5_RXQ_TYPE_HAIRPIN)
+			return &rxq_ctrl->hairpin_conf;
+	}
+	return NULL;
+}
+
 /**
  * Get an indirection table.
  *
drivers/net/mlx5/mlx5_rxtx.h index renamed from drivers/net/mlx5/mlx5_rxtx.h +0 -0

b50b643..d91ed0f 100644

drivers/net/mlx5/mlx5_rxtx.h +0 -0
@@ -344,6 +344,8 @@ uint32_t mlx5_hrxq_get(struct rte_eth_dev *dev,  int

mlx5_hrxq_release(struct rte_eth_dev *dev, uint32_t hxrq_idx); int mlx5_hrxq_verify(struct rte_eth_dev *dev); enum mlx5_rxq_type mlx5_rxq_get_type(struct rte_eth_dev *dev, uint16_t idx);

drivers/net/mlx5/mlx5_rxtx.h +2 -0
+const struct rte_eth_hairpin_conf *mlx5_rxq_get_hairpin_conf
+	(struct rte_eth_dev *dev, uint16_t idx);
 struct mlx5_hrxq *mlx5_drop_action_create(struct rte_eth_dev *dev);  void

mlx5_drop_action_destroy(struct rte_eth_dev *dev); uint64_t mlx5_get_rx_port_offloads(void);
--
1.8.3.1

author
#16 lore raw

PATCH v2 6/6 net/mlx5: not split hairpin flow in explicit mode

In reply to Bing Zhao

In the current implementation, the hairpin flow will be split into two flows implicitly if there is some action that only belongs to the TX part. A TX device flow will be inserted by the mlx5 PMD itself.

In hairpin between two ports, the explicit TX flow mode will be the only one to be supported. It is not the appropriate behavior to insert a TX flow into another device implicitly. The application could create any flow as it likes and has full control of the user flows. Hairpin flows will have no difference from standard flows and the application can decide how to chain RX and TX flows together.

Even in the single port hairpin, this explicit TX flow mode could also be supported.

When checking if the hairpin needs to be split, just return if the hairpin queue is with "tx_explicit" attribute. Then in the following steps for validation and translation, the code path will be the same as that for standard flows.

---
 drivers/net/mlx5/mlx5_flow.c | 7 +++++++
 1 file changed, 7 insertions(+)
drivers/net/mlx5/mlx5_flow.c +7 -0
index d7243a8..8a114a6 100644
@@ -3261,6 +3261,7 @@ struct mlx5_flow_tunnel_info {
 	const struct rte_flow_action_queue *queue;
 	const struct rte_flow_action_rss *rss;
 	const struct rte_flow_action_raw_encap *raw_encap;
+	const struct rte_eth_hairpin_conf *conf;
 
 	if (!attr->ingress)
 		return 0;
@@ -3273,6 +3274,9 @@ struct mlx5_flow_tunnel_info {
 			if (mlx5_rxq_get_type(dev, queue->index) !=
 			    MLX5_RXQ_TYPE_HAIRPIN)
 				return 0;
+			conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
+			if (!!conf->tx_explicit)
+				return 0;
 			queue_action = 1;
 			action_n++;
 			break;
@@ -3283,6 +3287,9 @@ struct mlx5_flow_tunnel_info {
 			if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
 			    MLX5_RXQ_TYPE_HAIRPIN)
 				return 0;
+			conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
+			if (conf != NULL && !!conf->tx_explicit)
+				return 0;
 			queue_action = 1;
 			action_n++;
 			break;
Signature · 1 line
1.8.3.1
Acked-by on v2 6/6
#17 lore raw

Re: [dpdk-dev] [PATCH v2 6/6] net/mlx5: not split hairpin flow in explicit mode

Quoted 19 lines

-----Original Message-----
From: Bing Zhao <[email protected]>
Sent: Thursday, October 22, 2020 17:07
To: [email protected]; [email protected]
Cc: [email protected]; Ori Kam <[email protected]>; Raslan Darawsheh <[email protected]>
Subject: [PATCH v2 6/6] net/mlx5: not split hairpin flow in explicit mode

In the current implementation, the hairpin flow will be split into two flows implicitly if there is some action that only belongs to the TX part. A TX device flow will be inserted by the mlx5 PMD itself.

In hairpin between two ports, the explicit TX flow mode will be the only one to be supported. It is not the appropriate behavior to insert a TX flow into another device implicitly. The application could create any flow as it likes and has full control of the user flows. Hairpin flows will have no difference from standard flows and the application can decide how to chain RX and TX flows together.

Quoted 8 lines

Even in the single port hairpin, this explicit TX flow mode could also be supported.

When checking if the hairpin needs to be split, just return if the hairpin queue is with "tx_explicit" attribute. Then in the following steps for validation and translation, the code path will be the same as that for standard flows.

Quoted 4 lines
---
 drivers/net/mlx5/mlx5_flow.c | 7 +++++++
 1 file changed, 7 insertions(+)
author
#18 lore raw

v3 cover add two ports hairpin mode support in mlx5 PMD

In reply to Bing Zhao

This patch set will add the support for hairpin between two ports in mlx5 PMD.

v2:
* Update the code and reorganize the patch set
v3:
* Doc update
* fix code bugs and code style update

Bing Zhao (7):
net/mlx5: change hairpin queue peer checking net/mlx5: add support for two ports hairpin mode net/mlx5: add support to get hairpin peer ports net/mlx5: conditional hairpin auto bind net/mlx5: change hairpin ingress flow validation net/mlx5: not split hairpin flow in explicit mode
doc: update mlx5 hairpin support and limitations

 doc/guides/nics/mlx5.rst               |   5 +
 doc/guides/rel_notes/release_20_11.rst |   1 +
 drivers/net/mlx5/linux/mlx5_os.c       |  12 +
 drivers/net/mlx5/mlx5.h                |  21 +
 drivers/net/mlx5/mlx5_flow.c           |   9 +-
 drivers/net/mlx5/mlx5_flow_dv.c        |  17 +-
 drivers/net/mlx5/mlx5_rxq.c            |  59 ++-
 drivers/net/mlx5/mlx5_rxtx.h           |   4 +
 drivers/net/mlx5/mlx5_trigger.c        | 757 ++++++++++++++++++++++++++++++++-
 drivers/net/mlx5/mlx5_txq.c            |  32 +-
 10 files changed, 894 insertions(+), 23 deletions(-)
Signature · 1 line
1.8.3.1
author
#19 lore raw

PATCH v3 1/7 net/mlx5: change hairpin queue peer checking

In the current implementation of single port mode hairpin, the peer queue should belong to the same port of the current queue. When the two ports hairpin mode is introduced, such checking should be removed to make the hairpin queue setup execute successfully since it is not an invalid condition, if the Tx port and Rx port are not the same.

In the meanwhile, different devices could have different queue configurations. The queues number of peer port is unknown to the current device. The checking should be removed also.

If the Tx and Rx port IDs of a hairpin peer are different, only the manual binding and explicit Tx flows are supported. Or else, the four combinations of modes could be supported. The mode attributes consistency checking will be done when connecting the queue with its peer queue.

---
v3: fix attributes checking

---
 drivers/net/mlx5/mlx5_rxq.c | 32 ++++++++++++++++++++++++++------
 drivers/net/mlx5/mlx5_txq.c | 32 ++++++++++++++++++++++++++------
 2 files changed, 52 insertions(+), 12 deletions(-)
drivers/net/mlx5/mlx5_rxq.c +26 -6
index 1cc477a..034f43e 100644
@@ -818,15 +818,35 @@
 	res = mlx5_rx_queue_pre_setup(dev, idx, &desc);
 	if (res)
 		return res;
-	if (hairpin_conf->peer_count != 1 ||
-	    hairpin_conf->peers[0].port != dev->data->port_id ||
-	    hairpin_conf->peers[0].queue >= priv->txqs_n) {
-		DRV_LOG(ERR, "port %u unable to setup hairpin queue index %u "
-			" invalid hairpind configuration", dev->data->port_id,
-			idx);
+	if (hairpin_conf->peer_count != 1) {
 		rte_errno = EINVAL;
+		DRV_LOG(ERR, "port %u unable to setup Rx hairpin queue index %u"
+			" peer count is %u", dev->data->port_id,
+			idx, hairpin_conf->peer_count);
 		return -rte_errno;
 	}
+	if (hairpin_conf->peers[0].port == dev->data->port_id) {
+		if (hairpin_conf->peers[0].queue >= priv->txqs_n) {
+			rte_errno = EINVAL;
+			DRV_LOG(ERR, "port %u unable to setup Rx hairpin queue"
+				" index %u, Tx %u is larger than %u",
+				dev->data->port_id, idx,
+				hairpin_conf->peers[0].queue, priv->txqs_n);
+			return -rte_errno;
+		}
+	} else {
+		if (hairpin_conf->manual_bind == 0 ||
+		    hairpin_conf->tx_explicit == 0) {
+			rte_errno = EINVAL;
+			DRV_LOG(ERR, "port %u unable to setup Rx hairpin queue"
+				" index %u peer port %u with attributes %u %u",
+				dev->data->port_id, idx,
+				hairpin_conf->peers[0].port,
+				hairpin_conf->manual_bind,
+				hairpin_conf->tx_explicit);
+			return -rte_errno;
+		}
+	}
 	rxq_ctrl = mlx5_rxq_hairpin_new(dev, idx, desc, hairpin_conf);
 	if (!rxq_ctrl) {
 		DRV_LOG(ERR, "port %u unable to allocate queue index %u",
drivers/net/mlx5/mlx5_txq.c +26 -6
index 9c2dd2a..dca9c05 100644
@@ -421,15 +421,35 @@
 	res = mlx5_tx_queue_pre_setup(dev, idx, &desc);
 	if (res)
 		return res;
-	if (hairpin_conf->peer_count != 1 ||
-	    hairpin_conf->peers[0].port != dev->data->port_id ||
-	    hairpin_conf->peers[0].queue >= priv->rxqs_n) {
-		DRV_LOG(ERR, "port %u unable to setup hairpin queue index %u "
-			" invalid hairpind configuration", dev->data->port_id,
-			idx);
+	if (hairpin_conf->peer_count != 1) {
 		rte_errno = EINVAL;
+		DRV_LOG(ERR, "port %u unable to setup Tx hairpin queue index %u"
+			" peer count is %u", dev->data->port_id,
+			idx, hairpin_conf->peer_count);
 		return -rte_errno;
 	}
+	if (hairpin_conf->peers[0].port == dev->data->port_id) {
+		if (hairpin_conf->peers[0].queue >= priv->rxqs_n) {
+			rte_errno = EINVAL;
+			DRV_LOG(ERR, "port %u unable to setup Tx hairpin queue"
+				" index %u, Rx %u is larger than %u",
+				dev->data->port_id, idx,
+				hairpin_conf->peers[0].queue, priv->txqs_n);
+			return -rte_errno;
+		}
+	} else {
+		if (hairpin_conf->manual_bind == 0 ||
+		    hairpin_conf->tx_explicit == 0) {
+			rte_errno = EINVAL;
+			DRV_LOG(ERR, "port %u unable to setup Tx hairpin queue"
+				" index %u peer port %u with attributes %u %u",
+				dev->data->port_id, idx,
+				hairpin_conf->peers[0].port,
+				hairpin_conf->manual_bind,
+				hairpin_conf->tx_explicit);
+			return -rte_errno;
+		}
+	}
 	txq_ctrl = mlx5_txq_hairpin_new(dev, idx, desc,	hairpin_conf);
 	if (!txq_ctrl) {
 		DRV_LOG(ERR, "port %u unable to allocate queue index %u",
Signature · 1 line
1.8.3.1
author
#20 lore raw

PATCH v3 2/7 net/mlx5: add support for two ports hairpin mode

In reply to Bing Zhao

In order to support hairpin between two ports, mlx5 PMD needs to implement the functions and provide them as the function pointers.

The bind and unbind functions are executed per port pairs. All the hairpin queues between the two ports should have the same attributes during queues setup. Different configurations among queue pairs from the same ports are not supported. It is allowed that two ports only have one direction hairpin.

In order to set up the connection between two queues, peer Rx queue HW information must be fetched via the internal RTE API and the queue information could be used to modify the SQ object. Then the RQ object will be modified with the Tx queue HW information. The reverse operation is not supported right now.

When disconnecting the queues pair, SQ and RQ object should be reset without any peer HW information. The unbinding operation will try to disconnect all Tx queues from the port from the Rx queues of the peer port.

Tx explicit mode attribute will be saved and used when creating a hairpin flow.

---
v3:
* fix code review comments and bugs for unbinding
* code style update
* Checking for mlx5 driver type

---
 drivers/net/mlx5/linux/mlx5_os.c |  10 +
 drivers/net/mlx5/mlx5.h          |  19 ++
 drivers/net/mlx5/mlx5_rxtx.h     |   2 +
 drivers/net/mlx5/mlx5_trigger.c  | 629 ++++++++++++++++++++++++++++++++++++++-
 4 files changed, 658 insertions(+), 2 deletions(-)
drivers/net/mlx5/linux/mlx5_os.c +10 -0
index ed3f020..b791859 100644
@@ -2570,6 +2570,11 @@
 	.get_module_eeprom = mlx5_get_module_eeprom,
 	.hairpin_cap_get = mlx5_hairpin_cap_get,
 	.mtr_ops_get = mlx5_flow_meter_ops_get,
+	.hairpin_bind = mlx5_hairpin_bind,
+	.hairpin_unbind = mlx5_hairpin_unbind,
+	.hairpin_queue_peer_update = mlx5_hairpin_queue_peer_update,
+	.hairpin_queue_peer_bind = mlx5_hairpin_queue_peer_bind,
+	.hairpin_queue_peer_unbind = mlx5_hairpin_queue_peer_unbind,
 };
 
 /* Available operations from secondary process. */
@@ -2648,4 +2653,9 @@
 	.get_module_eeprom = mlx5_get_module_eeprom,
 	.hairpin_cap_get = mlx5_hairpin_cap_get,
 	.mtr_ops_get = mlx5_flow_meter_ops_get,
+	.hairpin_bind = mlx5_hairpin_bind,
+	.hairpin_unbind = mlx5_hairpin_unbind,
+	.hairpin_queue_peer_update = mlx5_hairpin_queue_peer_update,
+	.hairpin_queue_peer_bind = mlx5_hairpin_queue_peer_bind,
+	.hairpin_queue_peer_unbind = mlx5_hairpin_queue_peer_unbind,
 };
drivers/net/mlx5/mlx5.h +19 -0
index 258be03..010152c 100644
@@ -904,6 +904,14 @@ struct mlx5_priv {
 #define PORT_ID(priv) ((priv)->dev_data->port_id)
 #define ETH_DEV(priv) (&rte_eth_devices[PORT_ID(priv)])
 
+struct rte_hairpin_peer_info {
+	uint32_t qp_id;
+	uint32_t vhca_id;
+	uint16_t peer_q;
+	uint16_t tx_explicit;
+	uint16_t manual_bind;
+};
+
 /* mlx5.c */
 
 int mlx5_getenv_int(const char *);
@@ -1054,6 +1062,17 @@ void mlx5_vlan_vmwa_acquire(struct rte_eth_dev *dev,
 int mlx5_traffic_enable(struct rte_eth_dev *dev);
 void mlx5_traffic_disable(struct rte_eth_dev *dev);
 int mlx5_traffic_restart(struct rte_eth_dev *dev);
+int mlx5_hairpin_queue_peer_update(struct rte_eth_dev *dev, uint16_t peer_queue,
+				   struct rte_hairpin_peer_info *current_info,
+				   struct rte_hairpin_peer_info *peer_info,
+				   uint32_t direction);
+int mlx5_hairpin_queue_peer_bind(struct rte_eth_dev *dev, uint16_t cur_queue,
+				 struct rte_hairpin_peer_info *peer_info,
+				 uint32_t direction);
+int mlx5_hairpin_queue_peer_unbind(struct rte_eth_dev *dev, uint16_t cur_queue,
+				   uint32_t direction);
+int mlx5_hairpin_bind(struct rte_eth_dev *dev, uint16_t rx_port);
+int mlx5_hairpin_unbind(struct rte_eth_dev *dev, uint16_t rx_port);
 
 /* mlx5_flow.c */
 
drivers/net/mlx5/mlx5_rxtx.h +2 -0
index f204f7e..cdc18e3 100644
@@ -199,6 +199,7 @@ struct mlx5_rxq_ctrl {
 	void *wq_umem; /* WQ buffer registration info. */
 	void *cq_umem; /* CQ buffer registration info. */
 	struct rte_eth_hairpin_conf hairpin_conf; /* Hairpin configuration. */
+	uint32_t hairpin_status; /* Hairpin binding status. */
 };
 
 /* TX queue send local data. */
@@ -295,6 +296,7 @@ struct mlx5_txq_ctrl {
 	void *bf_reg; /* BlueFlame register from Verbs. */
 	uint16_t dump_file_n; /* Number of dump files. */
 	struct rte_eth_hairpin_conf hairpin_conf; /* Hairpin configuration. */
+	uint32_t hairpin_status; /* Hairpin binding status. */
 	struct mlx5_txq_data txq; /* Data path structure. */
 	/* Must be the last field in the structure, contains elts[]. */
 };
drivers/net/mlx5/mlx5_trigger.c +627 -2

647 lines not on this page · raw message

Signature · 1 line
1.8.3.1
author
#21 lore raw

PATCH v3 3/7 net/mlx5: add support to get hairpin peer ports

In reply to Bing Zhao

In real-life business, one device could be attached and detached dynamically. The hairpin configuration of this port to/from all the other ports should be enabled and disabled accordingly.

The RTE ethdev lib and PMD should provide this ability to get the peer ports list in case that the application doesn't save it. It is recommended that the size of the array to save the port IDs is as large as the "RTE_MAX_ETHPORTS" to have the maximal capacity.

The order of the peer port IDs may be different from that during hairpin queues set in the initialization stage. The peer port ID could be the same as the current device port ID when the hairpin peer ports contain itself - the single port hairpin.

The application should check the ports' status and decide if the peer port should be bound / unbound when starting / stopping the current device.

---
 drivers/net/mlx5/linux/mlx5_os.c |  2 +
 drivers/net/mlx5/mlx5.h          |  2 +
 drivers/net/mlx5/mlx5_trigger.c  | 89 ++++++++++++++++++++++++++++++++++++++++
 3 files changed, 93 insertions(+)
drivers/net/mlx5/linux/mlx5_os.c +2 -0
index b791859..c890998 100644
@@ -2572,6 +2572,7 @@
 	.mtr_ops_get = mlx5_flow_meter_ops_get,
 	.hairpin_bind = mlx5_hairpin_bind,
 	.hairpin_unbind = mlx5_hairpin_unbind,
+	.hairpin_get_peer_ports = mlx5_hairpin_get_peer_ports,
 	.hairpin_queue_peer_update = mlx5_hairpin_queue_peer_update,
 	.hairpin_queue_peer_bind = mlx5_hairpin_queue_peer_bind,
 	.hairpin_queue_peer_unbind = mlx5_hairpin_queue_peer_unbind,
@@ -2655,6 +2656,7 @@
 	.mtr_ops_get = mlx5_flow_meter_ops_get,
 	.hairpin_bind = mlx5_hairpin_bind,
 	.hairpin_unbind = mlx5_hairpin_unbind,
+	.hairpin_get_peer_ports = mlx5_hairpin_get_peer_ports,
 	.hairpin_queue_peer_update = mlx5_hairpin_queue_peer_update,
 	.hairpin_queue_peer_bind = mlx5_hairpin_queue_peer_bind,
 	.hairpin_queue_peer_unbind = mlx5_hairpin_queue_peer_unbind,
drivers/net/mlx5/mlx5.h +2 -0
index 010152c..c537af9 100644
@@ -1073,6 +1073,8 @@ int mlx5_hairpin_queue_peer_unbind(struct rte_eth_dev *dev, uint16_t cur_queue,
 				   uint32_t direction);
 int mlx5_hairpin_bind(struct rte_eth_dev *dev, uint16_t rx_port);
 int mlx5_hairpin_unbind(struct rte_eth_dev *dev, uint16_t rx_port);
+int mlx5_hairpin_get_peer_ports(struct rte_eth_dev *dev, uint16_t *peer_ports,
+				size_t len, uint32_t direction);
 
 /* mlx5_flow.c */
 
drivers/net/mlx5/mlx5_trigger.c +89 -0

95 lines not on this page · raw message

Signature · 1 line
1.8.3.1
author
#22 lore raw

PATCH v3 4/7 net/mlx5: conditional hairpin auto bind

In reply to Bing Zhao

In single port hairpin mode, after the queues are configured during start up. The binding process will be enabled automatically in the port start phase and the default control flow for egress will be created.

When switching to two ports hairpin mode, the auto binding process should be skipped if there is no TX queue with the peer RX queue on the same device, and it should be skipped also if the queues are configured with manual bind attribute.

If the explicit TX flow rule mode is configured or hairpin is between two ports, the default control flows for TX queues should not be created.

---
 drivers/net/mlx5/mlx5_trigger.c | 39 +++++++++++++++++++++++++++++++++++++--
 1 file changed, 37 insertions(+), 2 deletions(-)
drivers/net/mlx5/mlx5_trigger.c +37 -2
index 3f56592..52691b6 100644
@@ -218,6 +218,8 @@
 	struct mlx5_devx_obj *rq;
 	unsigned int i;
 	int ret = 0;
+	bool need_auto = false;
+	uint16_t self_port = dev->data->port_id;
 
 	for (i = 0; i != priv->txqs_n; ++i) {
 		txq_ctrl = mlx5_txq_get(dev, i);
@@ -227,6 +229,28 @@
 			mlx5_txq_release(dev, i);
 			continue;
 		}
+		if (txq_ctrl->hairpin_conf.peers[0].port != self_port)
+			continue;
+		if (txq_ctrl->hairpin_conf.manual_bind) {
+			mlx5_txq_release(dev, i);
+			return 0;
+		}
+		need_auto = true;
+		mlx5_txq_release(dev, i);
+	}
+	if (!need_auto)
+		return 0;
+	for (i = 0; i != priv->txqs_n; ++i) {
+		txq_ctrl = mlx5_txq_get(dev, i);
+		if (!txq_ctrl)
+			continue;
+		if (txq_ctrl->type != MLX5_TXQ_TYPE_HAIRPIN) {
+			mlx5_txq_release(dev, i);
+			continue;
+		}
+		/* Skip hairpin queues with other peer ports. */
+		if (txq_ctrl->hairpin_conf.peers[0].port != self_port)
+			continue;
 		if (!txq_ctrl->obj) {
 			rte_errno = ENOMEM;
 			DRV_LOG(ERR, "port %u no txq object found: %d",
@@ -275,6 +299,9 @@
 		ret = mlx5_devx_cmd_modify_rq(rq, &rq_attr);
 		if (ret)
 			goto error;
+		/* Qs with auto-bind will be destroyed directly. */
+		rxq_ctrl->hairpin_status = 1;
+		txq_ctrl->hairpin_status = 1;
 		mlx5_txq_release(dev, i);
 		mlx5_rxq_release(dev, txq_ctrl->hairpin_conf.peers[0].queue);
 	}
@@ -1050,9 +1077,13 @@
 			dev->data->port_id, strerror(rte_errno));
 		goto error;
 	}
+	/*
+	 * Such step will be skipped if there is no hairpin TX queue configured
+	 * with RX peer queue from the same device.
+	 */
 	ret = mlx5_hairpin_auto_bind(dev);
 	if (ret) {
-		DRV_LOG(ERR, "port %u hairpin binding failed: %s",
+		DRV_LOG(ERR, "port %u hairpin auto binding failed: %s",
 			dev->data->port_id, strerror(rte_errno));
 		goto error;
 	}
@@ -1203,7 +1234,11 @@
 		struct mlx5_txq_ctrl *txq_ctrl = mlx5_txq_get(dev, i);
 		if (!txq_ctrl)
 			continue;
-		if (txq_ctrl->type == MLX5_TXQ_TYPE_HAIRPIN) {
+		/* Only Tx implicit mode requires the default Tx flow. */
+		if (txq_ctrl->type == MLX5_TXQ_TYPE_HAIRPIN &&
+		    txq_ctrl->hairpin_conf.tx_explicit == 0 &&
+		    txq_ctrl->hairpin_conf.peers[0].port ==
+		    priv->dev_data->port_id) {
 			ret = mlx5_ctrl_flow_source_queue(dev, i);
 			if (ret) {
 				mlx5_txq_release(dev, i);
Signature · 1 line
1.8.3.1
author
#23 lore raw

PATCH v3 5/7 net/mlx5: change hairpin ingress flow validation

In reply to Bing Zhao

In the current implementation of the single port hairpin, there is an implicit splitting process for actions. When inserting a hairpin flow, all the actions will be included with the ingress attribute. The flow engine will check and decide which actions should be moved into the TX flow part, e.g., encapsulation, VLAN push.

In some NICs, some actions can only be done in one direction. Since the hairpin flow will be split into two parts, such validation will be skipped.

With the hairpin explicit TX flow mode, no splitting is needed any more. The hairpin flow may have no big difference from a standard flow (except the queue). The application should take full charge of the actions and the flow engine should validate the hairpin flow in the same way as other flows.

In the meanwhile, a new internal API is added to get the hairpin configuration. This will bypass the useless atomic operation to save the CPU cycles.

---
 drivers/net/mlx5/mlx5_flow_dv.c | 17 ++++++++++++++---
 drivers/net/mlx5/mlx5_rxq.c     | 27 +++++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_rxtx.h    |  2 ++
 3 files changed, 43 insertions(+), 3 deletions(-)
drivers/net/mlx5/mlx5_flow_dv.c +14 -3
index 504d842..62e1d19 100644
@@ -5310,6 +5310,7 @@ struct field_modify_info modify_tcp[] = {
 		.transfer = !!attr->transfer,
 		.fdb_def_rule = !!priv->fdb_def_rule,
 	};
+	const struct rte_eth_hairpin_conf *conf;
 
 	if (items == NULL)
 		return -1;
@@ -6155,11 +6156,18 @@ struct field_modify_info modify_tcp[] = {
 						  actions,
 						  "no fate action is found");
 	}
-	/* Continue validation for Xcap and VLAN actions.*/
+	/*
+	 * Continue validation for Xcap and VLAN actions.
+	 * If hairpin is working in explicit TX rule mode, there is no actions
+	 * splitting and the validation of hairpin ingress flow should be the
+	 * same as other standard flows.
+	 */
 	if ((action_flags & (MLX5_FLOW_XCAP_ACTIONS |
 			     MLX5_FLOW_VLAN_ACTIONS)) &&
 	    (queue_index == 0xFFFF ||
-	     mlx5_rxq_get_type(dev, queue_index) != MLX5_RXQ_TYPE_HAIRPIN)) {
+	     mlx5_rxq_get_type(dev, queue_index) != MLX5_RXQ_TYPE_HAIRPIN ||
+	     ((conf = mlx5_rxq_get_hairpin_conf(dev, queue_index)) != NULL &&
+	     conf->tx_explicit != 0))) {
 		if ((action_flags & MLX5_FLOW_XCAP_ACTIONS) ==
 		    MLX5_FLOW_XCAP_ACTIONS)
 			return rte_flow_error_set(error, ENOTSUP,
@@ -6188,7 +6196,10 @@ struct field_modify_info modify_tcp[] = {
 						 "multiple VLAN actions");
 		}
 	}
-	/* Hairpin flow will add one more TAG action. */
+	/*
+	 * Hairpin flow will add one more TAG action in TX implicit mode.
+	 * In TX explicit mode, there will be no hairpin flow ID.
+	 */
 	if (hairpin > 0)
 		rw_act_num += MLX5_ACT_NUM_SET_TAG;
 	/* extra metadata enabled: one more TAG action will be add. */
drivers/net/mlx5/mlx5_rxq.c +27 -0
index 034f43e..493c5f2 100644
@@ -1845,6 +1845,33 @@ enum mlx5_rxq_type
 	return MLX5_RXQ_TYPE_UNDEFINED;
 }
 
+/*
+ * Get a Rx hairpin queue configuration.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @param idx
+ *   Rx queue index.
+ *
+ * @return
+ *   Pointer to the configuration if a hairpin RX queue, otherwise NULL.
+ */
+const struct rte_eth_hairpin_conf *
+mlx5_rxq_get_hairpin_conf(struct rte_eth_dev *dev, uint16_t idx)
+{
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_rxq_ctrl *rxq_ctrl = NULL;
+
+	if (idx < priv->rxqs_n && (*priv->rxqs)[idx]) {
+		rxq_ctrl = container_of((*priv->rxqs)[idx],
+					struct mlx5_rxq_ctrl,
+					rxq);
+		if (rxq_ctrl->type == MLX5_RXQ_TYPE_HAIRPIN)
+			return &rxq_ctrl->hairpin_conf;
+	}
+	return NULL;
+}
+
 /**
  * Match queues listed in arguments to queues contained in indirection table
  * object.
drivers/net/mlx5/mlx5_rxtx.h +2 -0
index cdc18e3..1b5fba4 100644
@@ -360,6 +360,8 @@ uint32_t mlx5_hrxq_get(struct rte_eth_dev *dev,
 int mlx5_hrxq_release(struct rte_eth_dev *dev, uint32_t hxrq_idx);
 int mlx5_hrxq_verify(struct rte_eth_dev *dev);
 enum mlx5_rxq_type mlx5_rxq_get_type(struct rte_eth_dev *dev, uint16_t idx);
+const struct rte_eth_hairpin_conf *mlx5_rxq_get_hairpin_conf
+	(struct rte_eth_dev *dev, uint16_t idx);
 struct mlx5_hrxq *mlx5_drop_action_create(struct rte_eth_dev *dev);
 void mlx5_drop_action_destroy(struct rte_eth_dev *dev);
 uint64_t mlx5_get_rx_port_offloads(void);
Signature · 1 line
1.8.3.1
author
#24 lore raw

PATCH v3 6/7 net/mlx5: not split hairpin flow in explicit mode

In reply to Bing Zhao

In the current implementation, the hairpin flow will be split into two flows implicitly if there is some action that only belongs to the Tx part. A Tx device flow will be inserted by the mlx5 PMD itself.

In hairpin between two ports, the explicit Tx flow mode will be the only one to be supported. It is not the appropriate behavior to insert a Tx flow into another device implicitly. The application could create any flow as it likes and has full control of the user flows. Hairpin flows will have no difference from standard flows and the application can decide how to chain Rx and Tx flows together.

Even in the single port hairpin, this explicit Tx flow mode could also be supported.

When checking if the hairpin needs to be split, it will just return if the hairpin queue is with "tx_explicit" attribute. Then in the following steps for validation and translation, the code path will be the same as that for standard flows.

---
v3: remove unnecessary checking of hairpin queue type

---
 drivers/net/mlx5/mlx5_flow.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)
drivers/net/mlx5/mlx5_flow.c +5 -4
index 949b9ce..4756cf9 100644
@@ -3618,6 +3618,7 @@ struct rte_flow_shared_action *
 	const struct rte_flow_action_queue *queue;
 	const struct rte_flow_action_rss *rss;
 	const struct rte_flow_action_raw_encap *raw_encap;
+	const struct rte_eth_hairpin_conf *conf;
 
 	if (!attr->ingress)
 		return 0;
@@ -3627,8 +3628,8 @@ struct rte_flow_shared_action *
 			queue = actions->conf;
 			if (queue == NULL)
 				return 0;
-			if (mlx5_rxq_get_type(dev, queue->index) !=
-			    MLX5_RXQ_TYPE_HAIRPIN)
+			conf = mlx5_rxq_get_hairpin_conf(dev, queue->index);
+			if (conf != NULL && !!conf->tx_explicit)
 				return 0;
 			queue_action = 1;
 			action_n++;
@@ -3637,8 +3638,8 @@ struct rte_flow_shared_action *
 			rss = actions->conf;
 			if (rss == NULL || rss->queue_num == 0)
 				return 0;
-			if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
-			    MLX5_RXQ_TYPE_HAIRPIN)
+			conf = mlx5_rxq_get_hairpin_conf(dev, rss->queue[0]);
+			if (conf != NULL && !!conf->tx_explicit)
 				return 0;
 			queue_action = 1;
 			action_n++;
Signature · 1 line
1.8.3.1
author
#25 lore raw

PATCH v3 7/7 doc: update mlx5 hairpin support and limitations

In reply to Bing Zhao

Hairpin between two ports will be supported by mlx5 PMD.

The supported scenarios and limitations are listed in "mlx5.rst".

---
 doc/guides/nics/mlx5.rst               | 5 +++++
 doc/guides/rel_notes/release_20_11.rst | 1 +
 2 files changed, 6 insertions(+)
doc/guides/nics/mlx5.rst +5 -0
index 8dc7c62..ab5cc62 100644
@@ -326,6 +326,11 @@ Limitations
   The last extension header item 'next header' field can specify the following
   header protocol type.
 
+- Hairpin:
+
+  - Hairpin between two ports could only manual binding and explicit Tx flow mode. For single port hairpin, all the combinations of auto/manual binding and explicit/implicit Tx flow mode could be supported.
+  - Hairpin in switchdev SR-IOV mode is not supported till now.
+
 Statistics
 ----------
 
doc/guides/rel_notes/release_20_11.rst +1 -0
index f9ef4fe..1c5c2e0 100644
@@ -354,6 +354,7 @@ New Features
   * Updated the supported timeout for Age action to the maximal value supported
     by rte_flow API.
   * Added support of Age action query.
+  * Added support of multi-ports hairpin.
 
 * **Updated vhost sample application.**
 
Signature · 1 line
1.8.3.1
Acked-by on v3 7/7
#26 lore raw

Re: [dpdk-dev] [PATCH v3 7/7] doc: update mlx5 hairpin support and limitations

Quoted 8 lines

-----Original Message-----
From: Bing Zhao <[email protected]>
Sent: Monday, October 26, 2020 18:38
To: [email protected]; [email protected]
Cc: [email protected]; Ori Kam <[email protected]>; Raslan Darawsheh <[email protected]>
Subject: [PATCH v3 7/7] doc: update mlx5 hairpin support and limitations

Quoted 5 lines

Hairpin between two ports will be supported by mlx5 PMD.

The supported scenarios and limitations are listed in "mlx5.rst".

#27 lore raw
In reply to Bing Zhao

Hi,

Quoted 39 lines

-----Original Message-----
From: Bing Zhao <[email protected]>
Sent: Monday, October 26, 2020 6:38 PM
To: [email protected]; [email protected]
Cc: [email protected]; Ori Kam <[email protected]>; Raslan Darawsheh <[email protected]>
Subject: [PATCH v3 0/7] add two ports hairpin mode support in mlx5 PMD

This patch set will add the support for hairpin between two ports in mlx5 PMD.

v2:
* Update the code and reorganize the patch set
v3:
* Doc update
* fix code bugs and code style update

Bing Zhao (7):
net/mlx5: change hairpin queue peer checking net/mlx5: add support for two ports hairpin mode net/mlx5: add support to get hairpin peer ports net/mlx5: conditional hairpin auto bind net/mlx5: change hairpin ingress flow validation net/mlx5: not split hairpin flow in explicit mode
doc: update mlx5 hairpin support and limitations

 doc/guides/nics/mlx5.rst               |   5 +
 doc/guides/rel_notes/release_20_11.rst |   1 +
 drivers/net/mlx5/linux/mlx5_os.c       |  12 +
 drivers/net/mlx5/mlx5.h                |  21 +
 drivers/net/mlx5/mlx5_flow.c           |   9 +-
 drivers/net/mlx5/mlx5_flow_dv.c        |  17 +-
 drivers/net/mlx5/mlx5_rxq.c            |  59 ++-
 drivers/net/mlx5/mlx5_rxtx.h           |   4 +
 drivers/net/mlx5/mlx5_trigger.c        | 757
++++++++++++++++++++++++++++++++-
 drivers/net/mlx5/mlx5_txq.c            |  32 +-
 10 files changed, 894 insertions(+), 23 deletions(-)
Quoted 2 lines

--
1.8.3.1

Series applied to next-net-mlx,

Kindest regards,
Raslan Darawsheh