Skip to contentSkip to conversation

net/bnxt: rxq stop/start fixes

PATCH In review Lance Richardson 2 participants 12 messages Nov 4, 2019 Network drivers

lore
Share & tools
lore
#0 lore raw

cover net/bnxt: rxq stop/start fixes

This patch set fixes several problems involving stopping/starting receive queues in the bnxt PMD.

Lance Richardson (4):
net/bnxt: fix RSS table update for start/stop rxq net/bnxt: keep consistent rxq start/stop state net/bnxt: release hwrm lock before returning net/bnxt: fix rxq start/stop for Thor based NICs

 drivers/net/bnxt/bnxt_hwrm.c | 31 +++++++++++++++++----
 drivers/net/bnxt/bnxt_rxq.c  | 52 +++++++++++++++++++++++++++++-------
 drivers/net/bnxt/bnxt_vnic.c |  1 +
 3 files changed, 69 insertions(+), 15 deletions(-)
Signature · 1 line
2.17.1
#1 lore raw

PATCH 1/4 net/bnxt: fix RSS table update for start/stop rxq

A previous commit made updating of the RSS table when a receive queue is stopped/started conditional on vnic->rx_queue_cnt being nonzero. This count is only nonzero for dynamically created VNICs, so the RSS table was not being updated in the normal path. Fix by restoring the original logic.

Also ensure that vnic->rx_queue_cnt is initialized to zero when reinitializing the VNIC array.

Fixes 36024b2e7fe5 ("net/bnxt: allow dynamic creation of VNIC")
---
 drivers/net/bnxt/bnxt_rxq.c  | 6 ++----
 drivers/net/bnxt/bnxt_vnic.c | 1 +
 2 files changed, 3 insertions(+), 4 deletions(-)
drivers/net/bnxt/bnxt_rxq.c +2 -4
index 6420281d3..ab889c10e 100644
@@ -466,8 +466,7 @@ int bnxt_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 		}
 
 		PMD_DRV_LOG(DEBUG, "Rx Queue Count %d\n", vnic->rx_queue_cnt);
-		if (vnic->rx_queue_cnt > 1)
-			rc = bnxt_vnic_rss_configure(bp, vnic);
+		rc = bnxt_vnic_rss_configure(bp, vnic);
 	}
 
 	if (rc == 0)
@@ -522,8 +521,7 @@ int bnxt_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 			vnic->fw_grp_ids[rx_queue_id] = INVALID_HW_RING_ID;
 
 		PMD_DRV_LOG(DEBUG, "Rx Queue Count %d\n", vnic->rx_queue_cnt);
-		if (vnic->rx_queue_cnt > 1)
-			rc = bnxt_vnic_rss_configure(bp, vnic);
+		rc = bnxt_vnic_rss_configure(bp, vnic);
 	}
 
 	if (rc == 0)
drivers/net/bnxt/bnxt_vnic.c +1 -0
index 412a53dc0..52a4badfc 100644
@@ -51,6 +51,7 @@ static void bnxt_init_vnics(struct bnxt *bp)
 		vnic->lb_rule = (uint16_t)HWRM_NA_SIGNATURE;
 		vnic->hash_mode =
 			HWRM_VNIC_RSS_CFG_INPUT_HASH_MODE_FLAGS_DEFAULT;
+		vnic->rx_queue_cnt = 0;
 
 		STAILQ_INIT(&vnic->filter);
 		STAILQ_INIT(&vnic->flow_list);
Signature · 1 line
2.17.1
#2 lore raw

PATCH 2/4 net/bnxt: keep consistent rxq start/stop state

In reply to Lance Richardson

Receive queue state needs to reflect "started" state when rebuilding the RSS table for Thor-based NICs. Move state update so that receive queues being started are included in the RSS table.

Fixes 38412304b50a ("net/bnxt: enable RSS for thor-based controllers")
---
 drivers/net/bnxt/bnxt_rxq.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
drivers/net/bnxt/bnxt_rxq.c +5 -3
index ab889c10e..e82eda9d7 100644
@@ -444,6 +444,8 @@ int bnxt_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 	 * If queue is it started, we do not post buffers for Rx.
 	 */
 	rxq->rx_started = true;
+	dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
+
 	bnxt_free_hwrm_rx_ring(bp, rx_queue_id);
 	rc = bnxt_alloc_hwrm_rx_ring(bp, rx_queue_id);
 	if (rc)
@@ -469,11 +471,11 @@ int bnxt_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 		rc = bnxt_vnic_rss_configure(bp, vnic);
 	}
 
-	if (rc == 0)
+	if (rc != 0) {
 		dev->data->rx_queue_state[rx_queue_id] =
-				RTE_ETH_QUEUE_STATE_STARTED;
-	else
+				RTE_ETH_QUEUE_STATE_STOPPED;
 		rxq->rx_started = false;
+	}
 
 	PMD_DRV_LOG(INFO,
 		    "queue %d, rx_deferred_start %d, state %d!\n",
Signature · 1 line
2.17.1
#3 lore raw

PATCH 3/4 net/bnxt: release hwrm lock before returning

In reply to Lance Richardson

The function bnxt_vnic_rss_configure_thor() returns early when all receive queues are stopped without releasing the hwrm spinlock, which causes subsequent HWRM operations to hang. Fix by ensuring that the lock is released before returning from this function.

Fixes 38412304b50a ("net/bnxt: enable RSS for thor-based controllers")
---
 drivers/net/bnxt/bnxt_hwrm.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/net/bnxt/bnxt_hwrm.c +3 -1
index ab6cb1dc3..bb4dcf4b2 100644
@@ -4337,8 +4337,10 @@ bnxt_vnic_rss_configure_thor(struct bnxt *bp, struct bnxt_vnic_info *vnic)
 			}
 
 			/* Return if no rings are active. */
-			if (cnt == max_rings)
+			if (cnt == max_rings) {
+				HWRM_UNLOCK();
 				return 0;
+			}
 
 			/* Add rx/cp ring pair to RSS table. */
 			rxr = rxqs[k]->rx_ring;
Signature · 1 line
2.17.1
#4 lore raw

PATCH 4/4 net/bnxt: fix rxq start/stop for Thor based NICs

In reply to Lance Richardson

Controller-specific handling is required for Thor-based NICs when stopping or starting a receive queue, otherwise packet reception may not be reliably resumed when a stopped receive queue is restarted:
- The VNIC default receive ring needs to be recomputed when a receive queue is stopped or started.
- When stopping the last (or only) receive queue for a given VNIC, ensure that no packets can reach the default receive ring by temporarily setting the VNIC MRU to zero.

Fixes f8168ca0e690 ("net/bnxt: support thor controller")
---
 drivers/net/bnxt/bnxt_hwrm.c | 27 +++++++++++++++++++++----
 drivers/net/bnxt/bnxt_rxq.c  | 38 +++++++++++++++++++++++++++++++++---
 2 files changed, 58 insertions(+), 7 deletions(-)
drivers/net/bnxt/bnxt_hwrm.c +23 -4
index bb4dcf4b2..2cba007ea 100644
@@ -1696,10 +1696,29 @@ int bnxt_hwrm_vnic_cfg(struct bnxt *bp, struct bnxt_vnic_info *vnic)
 	HWRM_PREP(req, VNIC_CFG, BNXT_USE_CHIMP_MB);
 
 	if (BNXT_CHIP_THOR(bp)) {
-		struct bnxt_rx_queue *rxq =
-			bp->eth_dev->data->rx_queues[vnic->start_grp_id];
-		struct bnxt_rx_ring_info *rxr = rxq->rx_ring;
-		struct bnxt_cp_ring_info *cpr = rxq->cp_ring;
+		int dflt_rxq = vnic->start_grp_id;
+		struct bnxt_rx_ring_info *rxr;
+		struct bnxt_cp_ring_info *cpr;
+		struct bnxt_rx_queue *rxq;
+		int i;
+
+		/*
+		 * The first active receive ring is used as the VNIC
+		 * default receive ring. If there are no active receive
+		 * rings (all corresponding receive queues are stopped),
+		 * the first receive ring is used.
+		 */
+		for (i = vnic->start_grp_id; i < vnic->end_grp_id; i++) {
+			rxq = bp->eth_dev->data->rx_queues[i];
+			if (rxq->rx_started) {
+				dflt_rxq = i;
+				break;
+			}
+		}
+
+		rxq = bp->eth_dev->data->rx_queues[dflt_rxq];
+		rxr = rxq->rx_ring;
+		cpr = rxq->cp_ring;
 
 		req.default_rx_ring_id =
 			rte_cpu_to_le_16(rxr->rx_ring_struct->fw_ring_id);
drivers/net/bnxt/bnxt_rxq.c +35 -3
index e82eda9d7..e7c012f68 100644
@@ -451,6 +451,10 @@ int bnxt_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 	if (rc)
 		return rc;
 
+	if (BNXT_CHIP_THOR(bp)) {
+		/* Reconfigure default receive ring and MRU. */
+		bnxt_hwrm_vnic_cfg(bp, rxq->vnic);
+	}
 	PMD_DRV_LOG(INFO, "Rx queue started %d\n", rx_queue_id);
 
 	if (dev_conf->rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG) {
@@ -491,7 +495,8 @@ int bnxt_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 	struct rte_eth_conf *dev_conf = &bp->eth_dev->data->dev_conf;
 	struct bnxt_vnic_info *vnic = NULL;
 	struct bnxt_rx_queue *rxq = NULL;
-	int rc = 0;
+	int active_queue_cnt = 0;
+	int i, rc = 0;
 
 	rc = is_bnxt_in_error(bp);
 	if (rc)
@@ -507,8 +512,9 @@ int bnxt_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 	}
 
 	rxq = bp->rx_queues[rx_queue_id];
+	vnic = rxq->vnic;
 
-	if (rxq == NULL) {
+	if (!rxq || !vnic) {
 		PMD_DRV_LOG(ERR, "Invalid Rx queue %d\n", rx_queue_id);
 		return -EINVAL;
 	}
@@ -518,7 +524,6 @@ int bnxt_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 	PMD_DRV_LOG(DEBUG, "Rx queue stopped\n");
 
 	if (dev_conf->rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG) {
-		vnic = rxq->vnic;
 		if (BNXT_HAS_RING_GRPS(bp))
 			vnic->fw_grp_ids[rx_queue_id] = INVALID_HW_RING_ID;
 
@@ -526,6 +531,33 @@ int bnxt_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 		rc = bnxt_vnic_rss_configure(bp, vnic);
 	}
 
+	if (BNXT_CHIP_THOR(bp)) {
+		/* Compute current number of active receive queues. */
+		for (i = vnic->start_grp_id; i < vnic->end_grp_id; i++)
+			if (bp->rx_queues[i]->rx_started)
+				active_queue_cnt++;
+
+		/*
+		 * For Thor, we need to ensure that the VNIC default receive
+		 * ring corresponds to an active receive queue. When no queue
+		 * is active, we need to temporarily set the MRU to zero so
+		 * that packets are dropped early in the receive pipeline in
+		 * order to prevent the VNIC default receive ring from being
+		 * accessed.
+		 */
+		if (active_queue_cnt == 0) {
+			uint16_t saved_mru = vnic->mru;
+
+			vnic->mru = 0;
+			/* Reconfigure default receive ring and MRU. */
+			bnxt_hwrm_vnic_cfg(bp, vnic);
+			vnic->mru = saved_mru;
+		} else {
+			/* Reconfigure default receive ring. */
+			bnxt_hwrm_vnic_cfg(bp, vnic);
+		}
+	}
+
 	if (rc == 0)
 		bnxt_rx_queue_release_mbufs(rxq);
 
Signature · 1 line
2.17.1
#5 lore raw
In reply to Lance Richardson

This patch set depends on the "bnxt patchset with bug fixes" series:

    http://patchwork.dpdk.org/project/dpdk/list/?series=7216

It is based on current dpdk-next-net-brcm.

Lance

Quoted 19 lines · Lance Richardson

This patch set fixes several problems involving stopping/starting receive queues in the bnxt PMD.

Lance Richardson (4):
net/bnxt: fix RSS table update for start/stop rxq net/bnxt: keep consistent rxq start/stop state net/bnxt: release hwrm lock before returning net/bnxt: fix rxq start/stop for Thor based NICs

 drivers/net/bnxt/bnxt_hwrm.c | 31 +++++++++++++++++----
 drivers/net/bnxt/bnxt_rxq.c  | 52 +++++++++++++++++++++++++++++-------
 drivers/net/bnxt/bnxt_vnic.c |  1 +
 3 files changed, 69 insertions(+), 15 deletions(-)

--
2.17.1

#6 lore raw

v2 cover net/bnxt: rxq stop/start fixes

This patch set fixes several problems involving stopping/starting receive queues in the bnxt PMD.

--
v1->v2: Updated commit logs for consistency.

Lance Richardson (4):
net/bnxt: fix RSS table update for start/stop rxq net/bnxt: keep consistent rxq start/stop state net/bnxt: release hwrm lock before returning net/bnxt: fix rxq start/stop for Thor based NICs

 drivers/net/bnxt/bnxt_hwrm.c | 31 +++++++++++++++++----
 drivers/net/bnxt/bnxt_rxq.c  | 52 +++++++++++++++++++++++++++++-------
 drivers/net/bnxt/bnxt_vnic.c |  1 +
 3 files changed, 69 insertions(+), 15 deletions(-)
Signature · 1 line
2.21.0 (Apple Git-122)
#7 lore raw

PATCH v2 1/4 net/bnxt: fix RSS table update for start/stop rxq

From: Lance Richardson <[email protected]>

A previous commit made updating of the RSS table when a receive queue is stopped/started conditional on vnic->rx_queue_cnt being nonzero. This count is only nonzero for dynamically created VNICs, so the RSS table was not being updated in the normal path. Fix by restoring the original logic.

Also ensure that vnic->rx_queue_cnt is initialized to zero when reinitializing the VNIC array.

Fixes 36024b2e7fe5 ("net/bnxt: allow dynamic creation of VNIC")
---
 drivers/net/bnxt/bnxt_rxq.c  | 6 ++----
 drivers/net/bnxt/bnxt_vnic.c | 1 +
 2 files changed, 3 insertions(+), 4 deletions(-)
drivers/net/bnxt/bnxt_rxq.c +2 -4
index 6420281d3..ab889c10e 100644
@@ -466,8 +466,7 @@ int bnxt_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 		}
 
 		PMD_DRV_LOG(DEBUG, "Rx Queue Count %d\n", vnic->rx_queue_cnt);
-		if (vnic->rx_queue_cnt > 1)
-			rc = bnxt_vnic_rss_configure(bp, vnic);
+		rc = bnxt_vnic_rss_configure(bp, vnic);
 	}
 
 	if (rc == 0)
@@ -522,8 +521,7 @@ int bnxt_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 			vnic->fw_grp_ids[rx_queue_id] = INVALID_HW_RING_ID;
 
 		PMD_DRV_LOG(DEBUG, "Rx Queue Count %d\n", vnic->rx_queue_cnt);
-		if (vnic->rx_queue_cnt > 1)
-			rc = bnxt_vnic_rss_configure(bp, vnic);
+		rc = bnxt_vnic_rss_configure(bp, vnic);
 	}
 
 	if (rc == 0)
drivers/net/bnxt/bnxt_vnic.c +1 -0
index 412a53dc0..52a4badfc 100644
@@ -51,6 +51,7 @@ static void bnxt_init_vnics(struct bnxt *bp)
 		vnic->lb_rule = (uint16_t)HWRM_NA_SIGNATURE;
 		vnic->hash_mode =
 			HWRM_VNIC_RSS_CFG_INPUT_HASH_MODE_FLAGS_DEFAULT;
+		vnic->rx_queue_cnt = 0;
 
 		STAILQ_INIT(&vnic->filter);
 		STAILQ_INIT(&vnic->flow_list);
Signature · 1 line
2.21.0 (Apple Git-122)
#8 lore raw

PATCH v2 2/4 net/bnxt: keep consistent rxq start/stop state

In reply to Ajit Khaparde

From: Lance Richardson <[email protected]>

Receive queue state needs to reflect "started" state when rebuilding the RSS table for Thor-based NICs. Move state update so that receive queues being started are included in the RSS table.

Fixes 38412304b50a ("net/bnxt: enable RSS for thor-based controllers")
---
 drivers/net/bnxt/bnxt_rxq.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
drivers/net/bnxt/bnxt_rxq.c +5 -3
index ab889c10e..e82eda9d7 100644
@@ -444,6 +444,8 @@ int bnxt_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 	 * If queue is it started, we do not post buffers for Rx.
 	 */
 	rxq->rx_started = true;
+	dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
+
 	bnxt_free_hwrm_rx_ring(bp, rx_queue_id);
 	rc = bnxt_alloc_hwrm_rx_ring(bp, rx_queue_id);
 	if (rc)
@@ -469,11 +471,11 @@ int bnxt_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 		rc = bnxt_vnic_rss_configure(bp, vnic);
 	}
 
-	if (rc == 0)
+	if (rc != 0) {
 		dev->data->rx_queue_state[rx_queue_id] =
-				RTE_ETH_QUEUE_STATE_STARTED;
-	else
+				RTE_ETH_QUEUE_STATE_STOPPED;
 		rxq->rx_started = false;
+	}
 
 	PMD_DRV_LOG(INFO,
 		    "queue %d, rx_deferred_start %d, state %d!\n",
Signature · 1 line
2.21.0 (Apple Git-122)
#9 lore raw

PATCH v2 3/4 net/bnxt: release hwrm lock before returning

In reply to Ajit Khaparde

From: Lance Richardson <[email protected]>

The function bnxt_vnic_rss_configure_thor() returns early when all receive queues are stopped without releasing the hwrm spinlock, which causes subsequent HWRM operations to hang. Fix by ensuring that the lock is released before returning from this function.

Fixes 38412304b50a ("net/bnxt: enable RSS for thor-based controllers")
---
 drivers/net/bnxt/bnxt_hwrm.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/net/bnxt/bnxt_hwrm.c +3 -1
index ab6cb1dc3..bb4dcf4b2 100644
@@ -4337,8 +4337,10 @@ bnxt_vnic_rss_configure_thor(struct bnxt *bp, struct bnxt_vnic_info *vnic)
 			}
 
 			/* Return if no rings are active. */
-			if (cnt == max_rings)
+			if (cnt == max_rings) {
+				HWRM_UNLOCK();
 				return 0;
+			}
 
 			/* Add rx/cp ring pair to RSS table. */
 			rxr = rxqs[k]->rx_ring;
Signature · 1 line
2.21.0 (Apple Git-122)
#10 lore raw

PATCH v2 4/4 net/bnxt: fix rxq start/stop for Thor based NICs

In reply to Ajit Khaparde

From: Lance Richardson <[email protected]>

Controller-specific handling is required for Thor-based NICs when stopping or starting a receive queue, otherwise packet reception may not be reliably resumed when a stopped receive queue is restarted:
- The VNIC default receive ring needs to be recomputed when a receive queue is stopped or started.
- When stopping the last (or only) receive queue for a given VNIC, ensure that no packets can reach the default receive ring by temporarily setting the VNIC MRU to zero.

Fixes f8168ca0e690 ("net/bnxt: support thor controller")
---
 drivers/net/bnxt/bnxt_hwrm.c | 27 +++++++++++++++++++++----
 drivers/net/bnxt/bnxt_rxq.c  | 38 +++++++++++++++++++++++++++++++++---
 2 files changed, 58 insertions(+), 7 deletions(-)
drivers/net/bnxt/bnxt_hwrm.c +23 -4
index bb4dcf4b2..2cba007ea 100644
@@ -1696,10 +1696,29 @@ int bnxt_hwrm_vnic_cfg(struct bnxt *bp, struct bnxt_vnic_info *vnic)
 	HWRM_PREP(req, VNIC_CFG, BNXT_USE_CHIMP_MB);
 
 	if (BNXT_CHIP_THOR(bp)) {
-		struct bnxt_rx_queue *rxq =
-			bp->eth_dev->data->rx_queues[vnic->start_grp_id];
-		struct bnxt_rx_ring_info *rxr = rxq->rx_ring;
-		struct bnxt_cp_ring_info *cpr = rxq->cp_ring;
+		int dflt_rxq = vnic->start_grp_id;
+		struct bnxt_rx_ring_info *rxr;
+		struct bnxt_cp_ring_info *cpr;
+		struct bnxt_rx_queue *rxq;
+		int i;
+
+		/*
+		 * The first active receive ring is used as the VNIC
+		 * default receive ring. If there are no active receive
+		 * rings (all corresponding receive queues are stopped),
+		 * the first receive ring is used.
+		 */
+		for (i = vnic->start_grp_id; i < vnic->end_grp_id; i++) {
+			rxq = bp->eth_dev->data->rx_queues[i];
+			if (rxq->rx_started) {
+				dflt_rxq = i;
+				break;
+			}
+		}
+
+		rxq = bp->eth_dev->data->rx_queues[dflt_rxq];
+		rxr = rxq->rx_ring;
+		cpr = rxq->cp_ring;
 
 		req.default_rx_ring_id =
 			rte_cpu_to_le_16(rxr->rx_ring_struct->fw_ring_id);
drivers/net/bnxt/bnxt_rxq.c +35 -3
index e82eda9d7..e7c012f68 100644
@@ -451,6 +451,10 @@ int bnxt_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 	if (rc)
 		return rc;
 
+	if (BNXT_CHIP_THOR(bp)) {
+		/* Reconfigure default receive ring and MRU. */
+		bnxt_hwrm_vnic_cfg(bp, rxq->vnic);
+	}
 	PMD_DRV_LOG(INFO, "Rx queue started %d\n", rx_queue_id);
 
 	if (dev_conf->rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG) {
@@ -491,7 +495,8 @@ int bnxt_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 	struct rte_eth_conf *dev_conf = &bp->eth_dev->data->dev_conf;
 	struct bnxt_vnic_info *vnic = NULL;
 	struct bnxt_rx_queue *rxq = NULL;
-	int rc = 0;
+	int active_queue_cnt = 0;
+	int i, rc = 0;
 
 	rc = is_bnxt_in_error(bp);
 	if (rc)
@@ -507,8 +512,9 @@ int bnxt_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 	}
 
 	rxq = bp->rx_queues[rx_queue_id];
+	vnic = rxq->vnic;
 
-	if (rxq == NULL) {
+	if (!rxq || !vnic) {
 		PMD_DRV_LOG(ERR, "Invalid Rx queue %d\n", rx_queue_id);
 		return -EINVAL;
 	}
@@ -518,7 +524,6 @@ int bnxt_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 	PMD_DRV_LOG(DEBUG, "Rx queue stopped\n");
 
 	if (dev_conf->rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG) {
-		vnic = rxq->vnic;
 		if (BNXT_HAS_RING_GRPS(bp))
 			vnic->fw_grp_ids[rx_queue_id] = INVALID_HW_RING_ID;
 
@@ -526,6 +531,33 @@ int bnxt_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 		rc = bnxt_vnic_rss_configure(bp, vnic);
 	}
 
+	if (BNXT_CHIP_THOR(bp)) {
+		/* Compute current number of active receive queues. */
+		for (i = vnic->start_grp_id; i < vnic->end_grp_id; i++)
+			if (bp->rx_queues[i]->rx_started)
+				active_queue_cnt++;
+
+		/*
+		 * For Thor, we need to ensure that the VNIC default receive
+		 * ring corresponds to an active receive queue. When no queue
+		 * is active, we need to temporarily set the MRU to zero so
+		 * that packets are dropped early in the receive pipeline in
+		 * order to prevent the VNIC default receive ring from being
+		 * accessed.
+		 */
+		if (active_queue_cnt == 0) {
+			uint16_t saved_mru = vnic->mru;
+
+			vnic->mru = 0;
+			/* Reconfigure default receive ring and MRU. */
+			bnxt_hwrm_vnic_cfg(bp, vnic);
+			vnic->mru = saved_mru;
+		} else {
+			/* Reconfigure default receive ring. */
+			bnxt_hwrm_vnic_cfg(bp, vnic);
+		}
+	}
+
 	if (rc == 0)
 		bnxt_rx_queue_release_mbufs(rxq);
 
Signature · 1 line
2.21.0 (Apple Git-122)
#11 lore raw
In reply to Ajit Khaparde
Quoted 12 lines · Ajit Khaparde

This patch set fixes several problems involving stopping/starting receive queues in the bnxt PMD.

--
v1->v2: Updated commit logs for consistency.

Lance Richardson (4):
net/bnxt: fix RSS table update for start/stop rxq net/bnxt: keep consistent rxq start/stop state net/bnxt: release hwrm lock before returning net/bnxt: fix rxq start/stop for Thor based NICs

Patchset applied to dpdk-next-net-brcm.

Quoted 10 lines
 drivers/net/bnxt/bnxt_hwrm.c | 31 +++++++++++++++++----
 drivers/net/bnxt/bnxt_rxq.c  | 52 +++++++++++++++++++++++++++++-------
 drivers/net/bnxt/bnxt_vnic.c |  1 +
 3 files changed, 69 insertions(+), 15 deletions(-)

--
2.21.0 (Apple Git-122)