Skip to contentSkip to conversation

bnxt patchset

PATCH Superseded Ajit Khaparde 2 participants 36 messages Oct 2, 2019 Network drivers

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

v2 cover bnxt patchset

bnxt patchset against dpdk-next-net.
This patchset has been rebased and is dependent on "bnxt patchset to support device error recovery"
https://patchwork.dpdk.org/project/dpdk/list/?series=6660

Please apply.

--
v1->v2: Rebased to the newer dpdk-next-net code base.

Ajit Khaparde (2):
net/bnxt: subscribe to link speed change notification net/bnxt: fix async link handling and update

Kalesh AP (9):
net/bnxt: fix rxq stop/start handling
net/bnxt: fix L2 context calculation for Thor net/bnxt: fix use of deferred start of transmit queues net/bnxt: remove unnecessary return check net/bnxt: remove unused macro
net/bnxt: remove unnecessary interrupt disable net/bnxt: fix accessing variable before null check net/bnxt: fix mbuf flags for PTP packets in vector mode Rx net/bnxt: use macro for getting default vnic

Rahul Gupta (1):
net/bnxt: fix Rx queue count devop

Santoshkumar Karanappa Rastapur (3):
net/bnxt: fix incorrect flow steering net/bnxt: fix traffic failure with higher mbuf size net/bnxt: fix 40G link failure for Thor

Somnath Kotur (1):
net/bnxt: support for QinQ insertion and stripping

Stephen Hemminger (1):
net/bnxt: fix a segfault in secondary process

 drivers/net/bnxt/bnxt.h              |  10 +-
 drivers/net/bnxt/bnxt_cpr.c          |   2 +-
 drivers/net/bnxt/bnxt_ethdev.c       | 171 +++++++++++++++++----------
 drivers/net/bnxt/bnxt_flow.c         |   4 +
 drivers/net/bnxt/bnxt_hwrm.c         | 156 +++++++++++++-----------
 drivers/net/bnxt/bnxt_hwrm.h         |   2 +
 drivers/net/bnxt/bnxt_irq.c          |   2 +-
 drivers/net/bnxt/bnxt_irq.h          |   1 +
 drivers/net/bnxt/bnxt_ring.c         |   8 +-
 drivers/net/bnxt/bnxt_rxq.c          |  91 ++++++++------
 drivers/net/bnxt/bnxt_rxq.h          |   2 +-
 drivers/net/bnxt/bnxt_rxr.c          |  14 +--
 drivers/net/bnxt/bnxt_rxtx_vec_sse.c |  15 ++-
 drivers/net/bnxt/bnxt_txq.c          |   5 +
 drivers/net/bnxt/bnxt_txq.h          |   1 +
 drivers/net/bnxt/bnxt_txr.c          |  19 ++-
 16 files changed, 300 insertions(+), 203 deletions(-)
Signature · 1 line
2.20.1 (Apple Git-117)
#1 lore raw

PATCH v2 1/17 net/bnxt: fix incorrect flow steering

From: Santoshkumar Karanappa Rastapur <[email protected]>

When user creates a flow similar to an existing flow with just the destination queue change, we delete the old filter and allocate a new one with this destination queue change. We were also allocating a new L2 filter matching the same destination mac resulting in 2 L2 filters for the same destination mac. This was causing any flow matching the destination mac to be steered to this queue instead of the default queue. Fixed it by deleting this stale L2 filter.

---
 drivers/net/bnxt/bnxt_flow.c | 4 ++++
 1 file changed, 4 insertions(+)
drivers/net/bnxt/bnxt_flow.c +4 -0
index deb9733e99..be9b6fad39 100644
@@ -1028,6 +1028,10 @@ bnxt_match_filter(struct bnxt *bp, struct bnxt_filter_info *nf)
 				    sizeof(nf->dst_ipaddr_mask))) {
 				if (mf->dst_id == nf->dst_id)
 					return -EEXIST;
+				/* Clear the new L2 filter that was created
+				 * earlier in bnxt_validate_and_parse_flow.
+				 */
+				bnxt_hwrm_clear_l2_filter(bp, nf);
 				/*
 				 * Same Flow, Different queue
 				 * Clear the old ntuple filter
Signature · 1 line
2.20.1 (Apple Git-117)
#2 lore raw

PATCH v2 2/17 net/bnxt: fix traffic failure with higher mbuf size

In reply to Ajit Khaparde

From: Santoshkumar Karanappa Rastapur <[email protected]>

Maximum frame length supported by hw is 9600 bytes. When user launches testpmd with --mbuf-size=9729, we are posting buffer descriptors of size 9601 to the ring. This was causing packets getting discarded.

---
 drivers/net/bnxt/bnxt.h        |  5 ++++-
 drivers/net/bnxt/bnxt_ethdev.c | 29 +++++++++--------------------
 drivers/net/bnxt/bnxt_hwrm.c   |  3 +++
 drivers/net/bnxt/bnxt_ring.c   |  5 -----
 drivers/net/bnxt/bnxt_rxq.c    |  2 +-
 drivers/net/bnxt/bnxt_rxq.h    |  1 -
 drivers/net/bnxt/bnxt_rxr.c    |  7 ++-----
 7 files changed, 19 insertions(+), 33 deletions(-)
drivers/net/bnxt/bnxt.h +4 -1
index 818a49f461..68ac30bb58 100644
@@ -23,12 +23,15 @@
 
 #define BNXT_MAX_MTU		9574
 #define VLAN_TAG_SIZE		4
+#define BNXT_NUM_VLANS		2
+#define BNXT_MAX_PKT_LEN	(BNXT_MAX_MTU + RTE_ETHER_HDR_LEN +\
+				 RTE_ETHER_CRC_LEN +\
+				 (BNXT_NUM_VLANS * VLAN_TAG_SIZE))
 #define BNXT_VF_RSV_NUM_RSS_CTX	1
 #define BNXT_VF_RSV_NUM_L2_CTX	4
 /* TODO: For now, do not support VMDq/RFS on VFs. */
 #define BNXT_VF_RSV_NUM_VNIC	1
 #define BNXT_MAX_LED		4
-#define BNXT_NUM_VLANS		2
 #define BNXT_MIN_RING_DESC	16
 #define BNXT_MAX_TX_RING_DESC	4096
 #define BNXT_MAX_RX_RING_DESC	8192
drivers/net/bnxt/bnxt_ethdev.c +9 -20
index 0083ba6e83..4095e9c720 100644
@@ -545,10 +545,13 @@ static int bnxt_dev_info_get_op(struct rte_eth_dev *eth_dev,
 	dev_info->hash_key_size = 40;
 	max_vnics = bp->max_vnics;
 
+	/* MTU specifics */
+	dev_info->min_mtu = RTE_ETHER_MIN_MTU;
+	dev_info->max_mtu = BNXT_MAX_MTU;
+
 	/* Fast path specifics */
 	dev_info->min_rx_bufsize = 1;
-	dev_info->max_rx_pktlen = BNXT_MAX_MTU + RTE_ETHER_HDR_LEN +
-		RTE_ETHER_CRC_LEN + VLAN_TAG_SIZE * 2;
+	dev_info->max_rx_pktlen = BNXT_MAX_PKT_LEN;
 
 	dev_info->rx_offload_capa = BNXT_DEV_RX_OFFLOAD_SUPPORT;
 	if (bp->flags & BNXT_FLAG_PTP_SUPPORTED)
@@ -1967,7 +1970,6 @@ bnxt_txq_info_get_op(struct rte_eth_dev *dev, uint16_t queue_id,
 static int bnxt_mtu_set_op(struct rte_eth_dev *eth_dev, uint16_t new_mtu)
 {
 	struct bnxt *bp = eth_dev->data->dev_private;
-	struct rte_eth_dev_info dev_info;
 	uint32_t new_pkt_size;
 	uint32_t rc = 0;
 	uint32_t i;
@@ -1979,18 +1981,6 @@ static int bnxt_mtu_set_op(struct rte_eth_dev *eth_dev, uint16_t new_mtu)
 	new_pkt_size = new_mtu + RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN +
 		       VLAN_TAG_SIZE * BNXT_NUM_VLANS;
 
-	rc = bnxt_dev_info_get_op(eth_dev, &dev_info);
-	if (rc != 0) {
-		PMD_DRV_LOG(ERR, "Error during getting ethernet device info\n");
-		return rc;
-	}
-
-	if (new_mtu < RTE_ETHER_MIN_MTU || new_mtu > BNXT_MAX_MTU) {
-		PMD_DRV_LOG(ERR, "MTU requested must be within (%d, %d)\n",
-			RTE_ETHER_MIN_MTU, BNXT_MAX_MTU);
-		return -EINVAL;
-	}
-
 #ifdef RTE_ARCH_X86
 	/*
 	 * If vector-mode tx/rx is active, disallow any MTU change that would
@@ -2020,15 +2010,12 @@ static int bnxt_mtu_set_op(struct rte_eth_dev *eth_dev, uint16_t new_mtu)
 
 	eth_dev->data->dev_conf.rxmode.max_rx_pkt_len = new_pkt_size;
 
-	eth_dev->data->mtu = new_mtu;
-	PMD_DRV_LOG(INFO, "New MTU is %d\n", eth_dev->data->mtu);
-
 	for (i = 0; i < bp->nr_vnics; i++) {
 		struct bnxt_vnic_info *vnic = &bp->vnic_info[i];
 		uint16_t size = 0;
 
-		vnic->mru = bp->eth_dev->data->mtu + RTE_ETHER_HDR_LEN +
-					RTE_ETHER_CRC_LEN + VLAN_TAG_SIZE * 2;
+		vnic->mru = new_mtu + RTE_ETHER_HDR_LEN +
+				RTE_ETHER_CRC_LEN + VLAN_TAG_SIZE * 2;
 		rc = bnxt_hwrm_vnic_cfg(bp, vnic);
 		if (rc)
 			break;
@@ -2043,6 +2030,8 @@ static int bnxt_mtu_set_op(struct rte_eth_dev *eth_dev, uint16_t new_mtu)
 		}
 	}
 
+	PMD_DRV_LOG(INFO, "New MTU is %d\n", new_mtu);
+
 	return rc;
 }
 
drivers/net/bnxt/bnxt_hwrm.c +3 -0
index 174dc75d54..09330e2382 100644
@@ -1233,6 +1233,7 @@ int bnxt_hwrm_ring_alloc(struct bnxt *bp,
 			mb_pool = bp->rx_queues[0]->mb_pool;
 			rx_buf_size = rte_pktmbuf_data_room_size(mb_pool) -
 				      RTE_PKTMBUF_HEADROOM;
+			rx_buf_size = RTE_MIN(BNXT_MAX_PKT_LEN, rx_buf_size);
 			req.rx_buf_size = rte_cpu_to_le_16(rx_buf_size);
 			enables |=
 				HWRM_RING_ALLOC_INPUT_ENABLES_RX_BUF_SIZE_VALID;
@@ -1263,6 +1264,7 @@ int bnxt_hwrm_ring_alloc(struct bnxt *bp,
 		mb_pool = bp->rx_queues[0]->mb_pool;
 		rx_buf_size = rte_pktmbuf_data_room_size(mb_pool) -
 			      RTE_PKTMBUF_HEADROOM;
+		rx_buf_size = RTE_MIN(BNXT_MAX_PKT_LEN, rx_buf_size);
 		req.rx_buf_size = rte_cpu_to_le_16(rx_buf_size);
 
 		req.stat_ctx_id = rte_cpu_to_le_32(stats_ctx_id);
@@ -1879,6 +1881,7 @@ int bnxt_hwrm_vnic_plcmode_cfg(struct bnxt *bp,
 
 	size = rte_pktmbuf_data_room_size(bp->rx_queues[0]->mb_pool);
 	size -= RTE_PKTMBUF_HEADROOM;
+	size = RTE_MIN(BNXT_MAX_PKT_LEN, size);
 
 	req.jumbo_thresh = rte_cpu_to_le_16(size);
 	req.vnic_id = rte_cpu_to_le_16(vnic->fw_vnic_id);
drivers/net/bnxt/bnxt_ring.c +0 -5
index ec17783cf8..73eee43148 100644
@@ -561,9 +561,6 @@ int bnxt_alloc_hwrm_rx_ring(struct bnxt *bp, int queue_index)
 	if (rc)
 		goto err_out;
 
-	rxq->rx_buf_use_size = BNXT_MAX_MTU + RTE_ETHER_HDR_LEN +
-		RTE_ETHER_CRC_LEN + (2 * VLAN_TAG_SIZE);
-
 	if (bp->eth_dev->data->rx_queue_state[queue_index] ==
 	    RTE_ETH_QUEUE_STATE_STARTED) {
 		if (bnxt_init_one_rx_ring(rxq)) {
@@ -640,8 +637,6 @@ int bnxt_alloc_hwrm_rings(struct bnxt *bp)
 		if (bnxt_alloc_rx_agg_ring(bp, i))
 			goto err_out;
 
-		rxq->rx_buf_use_size = BNXT_MAX_MTU + RTE_ETHER_HDR_LEN +
-					RTE_ETHER_CRC_LEN + (2 * VLAN_TAG_SIZE);
 		if (bnxt_init_one_rx_ring(rxq)) {
 			PMD_DRV_LOG(ERR, "bnxt_init_one_rx_ring failed!\n");
 			bnxt_rx_queue_release_op(rxq);
drivers/net/bnxt/bnxt_rxq.c +1 -1
index d5fc5268db..192ffd8780 100644
@@ -331,13 +331,13 @@ int bnxt_rx_queue_setup_op(struct rte_eth_dev *eth_dev,
 	rxq->nb_rx_desc = nb_desc;
 	rxq->rx_free_thresh = rx_conf->rx_free_thresh;
 
-	PMD_DRV_LOG(DEBUG, "RX Buf size is %d\n", rxq->rx_buf_use_size);
 	PMD_DRV_LOG(DEBUG, "RX Buf MTU %d\n", eth_dev->data->mtu);
 
 	rc = bnxt_init_rx_ring_struct(rxq, socket_id);
 	if (rc)
 		goto out;
 
+	PMD_DRV_LOG(DEBUG, "RX Buf size is %d\n", rxq->rx_buf_size);
 	rxq->queue_id = queue_idx;
 	rxq->port_id = eth_dev->data->port_id;
 	if (rx_offloads & DEV_RX_OFFLOAD_KEEP_CRC)
drivers/net/bnxt/bnxt_rxq.h +0 -1
index b5e42d01c0..c60b04426c 100644
@@ -36,7 +36,6 @@ struct bnxt_rx_queue {
 	struct bnxt_vnic_info	*vnic;
 
 	uint32_t			rx_buf_size;
-	uint32_t			rx_buf_use_size;  /* useable size */
 	struct bnxt_rx_ring_info	*rx_ring;
 	struct bnxt_cp_ring_info	*cp_ring;
 	struct bnxt_cp_ring_info	*nq_ring;
drivers/net/bnxt/bnxt_rxr.c +2 -5
index 28487fb38e..f60880c63e 100644
@@ -713,9 +713,7 @@ int bnxt_init_rx_ring_struct(struct bnxt_rx_queue *rxq, unsigned int socket_id)
 	struct bnxt_rx_ring_info *rxr;
 	struct bnxt_ring *ring;
 
-	rxq->rx_buf_use_size = BNXT_MAX_MTU + RTE_ETHER_HDR_LEN +
-		RTE_ETHER_CRC_LEN + (2 * VLAN_TAG_SIZE);
-	rxq->rx_buf_size = rxq->rx_buf_use_size + sizeof(struct rte_mbuf);
+	rxq->rx_buf_size = BNXT_MAX_PKT_LEN + sizeof(struct rte_mbuf);
 
 	rxr = rte_zmalloc_socket("bnxt_rx_ring",
 				 sizeof(struct bnxt_rx_ring_info),
@@ -826,8 +824,7 @@ int bnxt_init_one_rx_ring(struct bnxt_rx_queue *rxq)
 	uint16_t size;
 
 	size = rte_pktmbuf_data_room_size(rxq->mb_pool) - RTE_PKTMBUF_HEADROOM;
-	if (rxq->rx_buf_use_size <= size)
-		size = rxq->rx_buf_use_size;
+	size = RTE_MIN(BNXT_MAX_PKT_LEN, size);
 
 	type = RX_PROD_PKT_BD_TYPE_RX_PROD_PKT | RX_PROD_PKT_BD_FLAGS_EOP_PAD;
 
Signature · 1 line
2.20.1 (Apple Git-117)
#3 lore raw

PATCH v2 3/17 net/bnxt: fix rxq stop/start handling

In reply to Ajit Khaparde

From: Kalesh AP <[email protected]>

Driver should not change "deferred_start" state of the rx queues. It should get the state in queue_setup_op() and use that value.

Since the deferred start state was being used in the packet receive functions to determine whether a stopped rx ring should be polled, introduced a per-rxq flag to track queue stopped/started state.

Fixes 9b63c6fd70e3 ("net/bnxt: support Rx/Tx queue start/stop")
---
 drivers/net/bnxt/bnxt_ethdev.c       |  2 +-
 drivers/net/bnxt/bnxt_ring.c         |  3 +--
 drivers/net/bnxt/bnxt_rxq.c          | 36 +++++++++++++++++++---------
 drivers/net/bnxt/bnxt_rxq.h          |  1 +
 drivers/net/bnxt/bnxt_rxr.c          |  7 +++---
 drivers/net/bnxt/bnxt_rxtx_vec_sse.c |  2 +-
 6 files changed, 32 insertions(+), 19 deletions(-)
drivers/net/bnxt/bnxt_ethdev.c +1 -1
index 4095e9c720..d1c5acb3e2 100644
@@ -1945,7 +1945,7 @@ bnxt_rxq_info_get_op(struct rte_eth_dev *dev, uint16_t queue_id,
 
 	qinfo->conf.rx_free_thresh = rxq->rx_free_thresh;
 	qinfo->conf.rx_drop_en = 0;
-	qinfo->conf.rx_deferred_start = 0;
+	qinfo->conf.rx_deferred_start = rxq->rx_deferred_start;
 }
 
 static void
drivers/net/bnxt/bnxt_ring.c +1 -2
index 73eee43148..68a690de5a 100644
@@ -561,8 +561,7 @@ int bnxt_alloc_hwrm_rx_ring(struct bnxt *bp, int queue_index)
 	if (rc)
 		goto err_out;
 
-	if (bp->eth_dev->data->rx_queue_state[queue_index] ==
-	    RTE_ETH_QUEUE_STATE_STARTED) {
+	if (!rxq->rx_started) {
 		if (bnxt_init_one_rx_ring(rxq)) {
 			RTE_LOG(ERR, PMD,
 				"bnxt_init_one_rx_ring failed!\n");
drivers/net/bnxt/bnxt_rxq.c +25 -11
index 192ffd8780..4b506f8653 100644
@@ -357,9 +357,19 @@ int bnxt_rx_queue_setup_op(struct rte_eth_dev *eth_dev,
 	}
 	rte_atomic64_init(&rxq->rx_mbuf_alloc_fail);
 
-	rxq->rx_deferred_start = rx_conf->rx_deferred_start;
-	queue_state = rxq->rx_deferred_start ? RTE_ETH_QUEUE_STATE_STOPPED :
-						RTE_ETH_QUEUE_STATE_STARTED;
+	/* rxq 0 must not be stopped when used as async CPR */
+	if (!BNXT_NUM_ASYNC_CPR(bp) && queue_idx == 0)
+		rxq->rx_deferred_start = false;
+	else
+		rxq->rx_deferred_start = rx_conf->rx_deferred_start;
+
+	if (rxq->rx_deferred_start) {
+		queue_state = RTE_ETH_QUEUE_STATE_STOPPED;
+		rxq->rx_started = false;
+	} else {
+		queue_state = RTE_ETH_QUEUE_STATE_STARTED;
+		rxq->rx_started = true;
+	}
 	eth_dev->data->rx_queue_state[queue_idx] = queue_state;
 	rte_spinlock_init(&rxq->lock);
 
@@ -432,6 +442,11 @@ int bnxt_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 		return -EINVAL;
 	}
 
+	/* Set the queue state to started here.
+	 * We check the status of the queue while posting buffer.
+	 * If queue is it started, we do not post buffers for Rx.
+	 */
+	rxq->rx_started = true;
 	bnxt_free_hwrm_rx_ring(bp, rx_queue_id);
 	rc = bnxt_alloc_hwrm_rx_ring(bp, rx_queue_id);
 	if (rc)
@@ -448,20 +463,19 @@ int bnxt_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 
 			vnic->fw_grp_ids[rx_queue_id] =
 					bp->grp_info[rx_queue_id].fw_grp_id;
+			PMD_DRV_LOG(DEBUG,
+				    "vnic = %p fw_grp_id = %d\n",
+				    vnic, bp->grp_info[rx_queue_id].fw_grp_id);
 		}
 
-		PMD_DRV_LOG(DEBUG,
-			    "vnic = %p fw_grp_id = %d\n",
-			    vnic, bp->grp_info[rx_queue_id].fw_grp_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;
-		rxq->rx_deferred_start = false;
-	}
+	else
+		rxq->rx_started = false;
 
 	PMD_DRV_LOG(INFO,
 		    "queue %d, rx_deferred_start %d, state %d!\n",
@@ -500,7 +514,7 @@ int bnxt_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 	}
 
 	dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
-	rxq->rx_deferred_start = true;
+	rxq->rx_started = false;
 	PMD_DRV_LOG(DEBUG, "Rx queue stopped\n");
 
 	if (dev_conf->rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG) {
drivers/net/bnxt/bnxt_rxq.h +1 -0
index c60b04426c..3693d89a60 100644
@@ -30,6 +30,7 @@ struct bnxt_rx_queue {
 	uint16_t		port_id; /* Device port identifier */
 	uint8_t			crc_len; /* 0 if CRC stripped, 4 otherwise */
 	uint8_t			rx_deferred_start; /* not in global dev start */
+	uint8_t			rx_started; /* RX queue is started */
 
 	struct bnxt		*bp;
 	int			index;
drivers/net/bnxt/bnxt_rxr.c +3 -4
index f60880c63e..b3cc0d8a04 100644
@@ -574,10 +574,9 @@ uint16_t bnxt_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
 	if (unlikely(is_bnxt_in_error(rxq->bp)))
 		return 0;
 
-	/* If Rx Q was stopped return. RxQ0 cannot be stopped. */
-	if (unlikely(((rxq->rx_deferred_start ||
-		       !rte_spinlock_trylock(&rxq->lock)) &&
-		      rxq->queue_id)))
+	/* If Rx Q was stopped return */
+	if (unlikely(!rxq->rx_started ||
+		     !rte_spinlock_trylock(&rxq->lock)))
 		return 0;
 
 	/* Handle RX burst request */
drivers/net/bnxt/bnxt_rxtx_vec_sse.c +1 -1
index 2e6e83c949..5ae812f3b9 100644
@@ -224,7 +224,7 @@ bnxt_recv_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts,
 			     0xFF, 0xFF, 0xFF, 0xFF); /* pkt_type (zeroes) */
 
 	/* If Rx Q was stopped return */
-	if (rxq->rx_deferred_start)
+	if (unlikely(!rxq->rx_started))
 		return 0;
 
 	if (rxq->rxrearm_nb >= RTE_BNXT_RXQ_REARM_THRESH)
Signature · 1 line
2.20.1 (Apple Git-117)
#4 lore raw

PATCH v2 4/17 net/bnxt: subscribe to link speed change notification

In reply to Ajit Khaparde

We are not subscribing to link speed change notification. This patch addresses that.

---
 drivers/net/bnxt/bnxt_hwrm.c | 1 +
 drivers/net/bnxt/bnxt_hwrm.h | 2 ++
 2 files changed, 3 insertions(+)
drivers/net/bnxt/bnxt_hwrm.c +1 -0
index 09330e2382..b2033ea416 100644
@@ -738,6 +738,7 @@ int bnxt_hwrm_func_driver_register(struct bnxt *bp)
 		rte_cpu_to_le_32(ASYNC_CMPL_EVENT_ID_LINK_STATUS_CHANGE |
 				 ASYNC_CMPL_EVENT_ID_PORT_CONN_NOT_ALLOWED |
 				 ASYNC_CMPL_EVENT_ID_LINK_SPEED_CFG_CHANGE |
+				 ASYNC_CMPL_EVENT_ID_LINK_SPEED_CHANGE |
 				 ASYNC_CMPL_EVENT_ID_RESET_NOTIFY);
 	if (bp->flags & BNXT_FLAG_FW_CAP_ERROR_RECOVERY)
 		req.async_event_fwd[0] |=
drivers/net/bnxt/bnxt_hwrm.h +2 -0
index 0d386952b6..5190f43290 100644
@@ -21,6 +21,8 @@ struct bnxt_cp_ring_info;
 	(1 << HWRM_ASYNC_EVENT_CMPL_EVENT_ID_PORT_CONN_NOT_ALLOWED)
 #define ASYNC_CMPL_EVENT_ID_LINK_SPEED_CFG_CHANGE	\
 	(1 << HWRM_ASYNC_EVENT_CMPL_EVENT_ID_LINK_SPEED_CFG_CHANGE)
+#define ASYNC_CMPL_EVENT_ID_LINK_SPEED_CHANGE \
+	(1 << HWRM_ASYNC_EVENT_CMPL_EVENT_ID_LINK_SPEED_CHANGE)
 #define ASYNC_CMPL_EVENT_ID_RESET_NOTIFY \
 	(1 << HWRM_ASYNC_EVENT_CMPL_EVENT_ID_RESET_NOTIFY)
 #define ASYNC_CMPL_EVENT_ID_ERROR_RECOVERY \
Signature · 1 line
2.20.1 (Apple Git-117)
#5 lore raw

PATCH v2 5/17 net/bnxt: fix async link handling and update

In reply to Ajit Khaparde

When updating the link because of an async link notification there is no need to set wait_for_completion. At this point the link related information should be available without need to poll. Use rte_eth_linkstatus_set instead of memcpy to ensure atomicity while updating the link status.
We force the physical link down as a part of device stop.
But we are not waiting there enough and handling the async notification before exiting. It just sits in the default CQ till we do a device start. Fix it by calling the CQ handler in device stop.

Fixes 7bc8e9a227cc ("net/bnxt: support async link notification")
---
 drivers/net/bnxt/bnxt_cpr.c    |  2 +-
 drivers/net/bnxt/bnxt_ethdev.c | 11 +++++++----
 drivers/net/bnxt/bnxt_irq.c    |  2 +-
 drivers/net/bnxt/bnxt_irq.h    |  1 +
 4 files changed, 10 insertions(+), 6 deletions(-)
drivers/net/bnxt/bnxt_cpr.c +1 -1
index 4817672ef0..f58372516a 100644
@@ -66,7 +66,7 @@ void bnxt_handle_async_event(struct bnxt *bp,
 	case HWRM_ASYNC_EVENT_CMPL_EVENT_ID_LINK_SPEED_CHANGE:
 	case HWRM_ASYNC_EVENT_CMPL_EVENT_ID_LINK_SPEED_CFG_CHANGE:
 		/* FALLTHROUGH */
-		bnxt_link_update_op(bp->eth_dev, 1);
+		bnxt_link_update_op(bp->eth_dev, 0);
 		break;
 	case HWRM_ASYNC_EVENT_CMPL_EVENT_ID_PF_DRVR_UNLOAD:
 		PMD_DRV_LOG(INFO, "Async event: PF driver unloaded\n");
drivers/net/bnxt/bnxt_ethdev.c +7 -4
index d1c5acb3e2..4d15ba5ceb 100644
@@ -834,6 +834,7 @@ static int bnxt_dev_start_op(struct rte_eth_dev *eth_dev)
 			bp->rx_cp_nr_rings, RTE_ETHDEV_QUEUE_STAT_CNTRS);
 	}
 
+	bnxt_enable_int(bp);
 	rc = bnxt_hwrm_if_change(bp, 1);
 	if (!rc) {
 		if (bp->flags & BNXT_FLAG_IF_CHANGE_HOT_FW_RESET_DONE) {
@@ -862,7 +863,6 @@ static int bnxt_dev_start_op(struct rte_eth_dev *eth_dev)
 	eth_dev->rx_pkt_burst = bnxt_receive_function(eth_dev);
 	eth_dev->tx_pkt_burst = bnxt_transmit_function(eth_dev);
 
-	bnxt_enable_int(bp);
 	bp->flags |= BNXT_FLAG_INIT_DONE;
 	eth_dev->data->dev_started = 1;
 	bp->dev_stopped = 0;
@@ -926,7 +926,9 @@ static void bnxt_dev_stop_op(struct rte_eth_dev *eth_dev)
 		/* TBD: STOP HW queues DMA */
 		eth_dev->data->dev_link.link_status = 0;
 	}
-	bnxt_set_hwrm_link_config(bp, false);
+	bnxt_dev_set_link_down_op(eth_dev);
+	/* Wait for link to be reset and the async notification to process. */
+	rte_delay_ms(BNXT_LINK_WAIT_INTERVAL * 2);
 
 	/* Clean queue intr-vector mapping */
 	rte_intr_efd_disable(intr_handle);
@@ -938,6 +940,8 @@ static void bnxt_dev_stop_op(struct rte_eth_dev *eth_dev)
 	bnxt_hwrm_port_clr_stats(bp);
 	bnxt_free_tx_mbufs(bp);
 	bnxt_free_rx_mbufs(bp);
+	/* Process any remaining notifications in default completion queue */
+	bnxt_int_handler(eth_dev);
 	bnxt_shutdown_nic(bp);
 	bnxt_hwrm_if_change(bp, 0);
 	bp->dev_stopped = 1;
@@ -1084,8 +1088,7 @@ int bnxt_link_update_op(struct rte_eth_dev *eth_dev, int wait_to_complete)
 	/* Timed out or success */
 	if (new.link_status != eth_dev->data->dev_link.link_status ||
 	new.link_speed != eth_dev->data->dev_link.link_speed) {
-		memcpy(&eth_dev->data->dev_link, &new,
-			sizeof(struct rte_eth_link));
+		rte_eth_linkstatus_set(eth_dev, &new);
 
 		_rte_eth_dev_callback_process(eth_dev,
 					      RTE_ETH_EVENT_INTR_LSC,
drivers/net/bnxt/bnxt_irq.c +1 -1
index a22700a0da..729d68d704 100644
@@ -18,7 +18,7 @@
  * Interrupts
  */
 
-static void bnxt_int_handler(void *param)
+void bnxt_int_handler(void *param)
 {
 	struct rte_eth_dev *eth_dev = (struct rte_eth_dev *)param;
 	struct bnxt *bp = eth_dev->data->dev_private;
drivers/net/bnxt/bnxt_irq.h +1 -0
index 460a97a09c..1b56e08068 100644
@@ -22,5 +22,6 @@ void bnxt_disable_int(struct bnxt *bp);
 void bnxt_enable_int(struct bnxt *bp);
 int bnxt_setup_int(struct bnxt *bp);
 int bnxt_request_int(struct bnxt *bp);
+void bnxt_int_handler(void *param);
 
 #endif
Signature · 1 line
2.20.1 (Apple Git-117)
#6 lore raw

PATCH v2 6/17 net/bnxt: fix 40G link failure for Thor

In reply to Ajit Khaparde

From: Santoshkumar Karanappa Rastapur <[email protected]>

Link autonegotiation for 40G is not supported by Thor.
Hence speed needs to be forced and autoneg disabled to configure 40G speed.

Fixes f8168ca0e690 ("net/bnxt: support thor controller")
---
 drivers/net/bnxt/bnxt_hwrm.c | 22 +++++++++++++++++++---
 1 file changed, 19 insertions(+), 3 deletions(-)
drivers/net/bnxt/bnxt_hwrm.c +19 -3
index b2033ea416..6e95acb89e 100644
@@ -2679,11 +2679,27 @@ int bnxt_set_hwrm_link_config(struct bnxt *bp, bool link_up)
 		goto port_phy_cfg;
 
 	autoneg = bnxt_check_eth_link_autoneg(dev_conf->link_speeds);
+	if (BNXT_CHIP_THOR(bp) &&
+	    dev_conf->link_speeds == ETH_LINK_SPEED_40G) {
+		/* 40G is not supported as part of media auto detect.
+		 * The speed should be forced and autoneg disabled
+		 * to configure 40G speed.
+		 */
+		PMD_DRV_LOG(INFO, "Disabling autoneg for 40G\n");
+		autoneg = 0;
+	}
+
 	speed = bnxt_parse_eth_link_speed(dev_conf->link_speeds);
 	link_req.phy_flags = HWRM_PORT_PHY_CFG_INPUT_FLAGS_RESET_PHY;
-	/* Autoneg can be done only when the FW allows */
-	if (autoneg == 1 && !(bp->link_info.auto_link_speed ||
-				bp->link_info.force_link_speed)) {
+	/* Autoneg can be done only when the FW allows.
+	 * When user configures fixed speed of 40G and later changes to
+	 * any other speed, auto_link_speed/force_link_speed is still set
+	 * to 40G until link comes up at new speed.
+	 */
+	if (autoneg == 1 &&
+	    !(!BNXT_CHIP_THOR(bp) &&
+	      (bp->link_info.auto_link_speed ||
+	       bp->link_info.force_link_speed))) {
 		link_req.phy_flags |=
 				HWRM_PORT_PHY_CFG_INPUT_FLAGS_RESTART_AUTONEG;
 		link_req.auto_link_speed_mask =
Signature · 1 line
2.20.1 (Apple Git-117)
#7 lore raw

PATCH v2 7/17 net/bnxt: fix L2 context calculation for Thor

In reply to Ajit Khaparde

From: Kalesh AP <[email protected]>

On Wh+, number of L2 context supported by FW is the sum of number of EM flow count and number of L2 context count returned in HWRM_FUNC_QCFG. This is not true for Thor. Restrict this only for Whitney for now.

Fixes ff9c0ca47e81176 ("net/bnxt: save the number of EM flow count")
---
 drivers/net/bnxt/bnxt_hwrm.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)
drivers/net/bnxt/bnxt_hwrm.c +6 -5
index 6e95acb89e..d0f4054412 100644
@@ -611,8 +611,9 @@ static int __bnxt_hwrm_func_qcaps(struct bnxt *bp)
 	bp->max_rx_rings = rte_le_to_cpu_16(resp->max_rx_rings);
 	bp->first_vf_id = rte_le_to_cpu_16(resp->first_vf_id);
 	bp->max_rx_em_flows = rte_le_to_cpu_16(resp->max_rx_em_flows);
-	bp->max_l2_ctx =
-		rte_le_to_cpu_16(resp->max_l2_ctxs) + bp->max_rx_em_flows;
+	bp->max_l2_ctx = rte_le_to_cpu_16(resp->max_l2_ctxs);
+	if (!BNXT_CHIP_THOR(bp))
+		bp->max_l2_ctx += bp->max_rx_em_flows;
 	/* TODO: For now, do not support VMDq/RFS on VFs. */
 	if (BNXT_PF(bp)) {
 		if (bp->pf.max_vfs)
@@ -858,9 +859,9 @@ int bnxt_hwrm_func_resc_qcaps(struct bnxt *bp)
 		/* func_resource_qcaps does not return max_rx_em_flows.
 		 * So use the value provided by func_qcaps.
 		 */
-		bp->max_l2_ctx =
-			rte_le_to_cpu_16(resp->max_l2_ctxs) +
-			bp->max_rx_em_flows;
+		bp->max_l2_ctx = rte_le_to_cpu_16(resp->max_l2_ctxs);
+		if (!BNXT_CHIP_THOR(bp))
+			bp->max_l2_ctx += bp->max_rx_em_flows;
 		bp->max_vnics = rte_le_to_cpu_16(resp->max_vnics);
 		bp->max_stat_ctx = rte_le_to_cpu_16(resp->max_stat_ctx);
 	}
Signature · 1 line
2.20.1 (Apple Git-117)
#8 lore raw

PATCH v2 8/17 net/bnxt: fix use of deferred start of transmit queues

In reply to Ajit Khaparde

From: Kalesh AP <[email protected]>

Driver should not change "deferred_start" state of the tx queues. It should get the state in queue_setup_op() and use that value.

Since the deferred start state was being used in the packet transmit functions to determine whether the queue has been stopped already, introduced a per-txq flag to track queue stopped/started state.

Fixes 9b63c6fd70e3 ("net/bnxt: support Rx/Tx queue start/stop")
---
 drivers/net/bnxt/bnxt_rxtx_vec_sse.c | 2 +-
 drivers/net/bnxt/bnxt_txq.c          | 5 +++++
 drivers/net/bnxt/bnxt_txq.h          | 1 +
 drivers/net/bnxt/bnxt_txr.c          | 6 +++---
 4 files changed, 10 insertions(+), 4 deletions(-)
drivers/net/bnxt/bnxt_rxtx_vec_sse.c +1 -1
index 5ae812f3b9..029053e305 100644
@@ -482,7 +482,7 @@ bnxt_xmit_pkts_vec(void *tx_queue, struct rte_mbuf **tx_pkts,
 	struct bnxt_tx_queue *txq = tx_queue;
 
 	/* Tx queue was stopped; wait for it to be restarted */
-	if (unlikely(txq->tx_deferred_start)) {
+	if (unlikely(!txq->tx_started)) {
 		PMD_DRV_LOG(DEBUG, "Tx q stopped;return\n");
 		return 0;
 	}
drivers/net/bnxt/bnxt_txq.c +5 -0
index 0901324793..ea20d737fe 100644
@@ -131,6 +131,7 @@ int bnxt_tx_queue_setup_op(struct rte_eth_dev *eth_dev,
 	txq->bp = bp;
 	txq->nb_tx_desc = nb_desc;
 	txq->tx_free_thresh = tx_conf->tx_free_thresh;
+	txq->tx_deferred_start = tx_conf->tx_deferred_start;
 
 	rc = bnxt_init_tx_ring_struct(txq, socket_id);
 	if (rc)
@@ -157,6 +158,10 @@ int bnxt_tx_queue_setup_op(struct rte_eth_dev *eth_dev,
 
 	eth_dev->data->tx_queues[queue_idx] = txq;
 
+	if (txq->tx_deferred_start)
+		txq->tx_started = false;
+	else
+		txq->tx_started = true;
 out:
 	return rc;
 }
drivers/net/bnxt/bnxt_txq.h +1 -0
index 9190e3f731..7a442516d2 100644
@@ -24,6 +24,7 @@ struct bnxt_tx_queue {
 	uint8_t			wthresh; /* Write-back threshold reg */
 	uint32_t		ctx_curr; /* Hardware context states */
 	uint8_t			tx_deferred_start; /* not in global dev start */
+	uint8_t			tx_started; /* TX queue is started */
 
 	struct bnxt		*bp;
 	int			index;
drivers/net/bnxt/bnxt_txr.c +3 -3
index 172b480b2e..99d2009055 100644
@@ -469,7 +469,7 @@ uint16_t bnxt_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
 	bnxt_handle_tx_cp(txq);
 
 	/* Tx queue was stopped; wait for it to be restarted */
-	if (txq->tx_deferred_start) {
+	if (unlikely(!txq->tx_started)) {
 		PMD_DRV_LOG(DEBUG, "Tx q stopped;return\n");
 		return 0;
 	}
@@ -518,7 +518,7 @@ int bnxt_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 		return rc;
 
 	dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
-	txq->tx_deferred_start = false;
+	txq->tx_started = true;
 	PMD_DRV_LOG(DEBUG, "Tx queue started\n");
 
 	return 0;
@@ -538,7 +538,7 @@ int bnxt_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 	bnxt_handle_tx_cp(txq);
 
 	dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
-	txq->tx_deferred_start = true;
+	txq->tx_started = false;
 	PMD_DRV_LOG(DEBUG, "Tx queue stopped\n");
 
 	return 0;
Signature · 1 line
2.20.1 (Apple Git-117)
#9 lore raw

PATCH v2 9/17 net/bnxt: remove unnecessary return check

In reply to Ajit Khaparde

From: Kalesh AP <[email protected]>

HWRM_CHECK_RESULT_SILENT() returns if the HWRM command fails. There is no need to check the return value after this call.

---
 drivers/net/bnxt/bnxt_hwrm.c | 119 +++++++++++++++++------------------
 1 file changed, 56 insertions(+), 63 deletions(-)
drivers/net/bnxt/bnxt_hwrm.c +56 -63
index d0f4054412..1065f21de7 100644
@@ -4379,7 +4379,10 @@ int bnxt_hwrm_func_backing_store_qcaps(struct bnxt *bp)
 	struct hwrm_func_backing_store_qcaps_input req = {0};
 	struct hwrm_func_backing_store_qcaps_output *resp =
 		bp->hwrm_cmd_resp_addr;
-	int rc;
+	struct bnxt_ctx_pg_info *ctx_pg;
+	struct bnxt_ctx_mem_info *ctx;
+	int total_alloc_len;
+	int rc, i;
 
 	if (!BNXT_CHIP_THOR(bp) ||
 	    bp->hwrm_spec_code < HWRM_VERSION_1_9_2 ||
@@ -4391,70 +4394,60 @@ int bnxt_hwrm_func_backing_store_qcaps(struct bnxt *bp)
 	rc = bnxt_hwrm_send_message(bp, &req, sizeof(req), BNXT_USE_CHIMP_MB);
 	HWRM_CHECK_RESULT_SILENT();
 
-	if (!rc) {
-		struct bnxt_ctx_pg_info *ctx_pg;
-		struct bnxt_ctx_mem_info *ctx;
-		int total_alloc_len;
-		int i;
-
-		total_alloc_len = sizeof(*ctx);
-		ctx = rte_malloc("bnxt_ctx_mem", total_alloc_len,
-				 RTE_CACHE_LINE_SIZE);
-		if (!ctx) {
-			rc = -ENOMEM;
-			goto ctx_err;
-		}
-		memset(ctx, 0, total_alloc_len);
+	total_alloc_len = sizeof(*ctx);
+	ctx = rte_zmalloc("bnxt_ctx_mem", total_alloc_len,
+			  RTE_CACHE_LINE_SIZE);
+	if (!ctx) {
+		rc = -ENOMEM;
+		goto ctx_err;
+	}
 
-		ctx_pg = rte_malloc("bnxt_ctx_pg_mem",
-				    sizeof(*ctx_pg) * BNXT_MAX_Q,
-				    RTE_CACHE_LINE_SIZE);
-		if (!ctx_pg) {
-			rc = -ENOMEM;
-			goto ctx_err;
-		}
-		for (i = 0; i < BNXT_MAX_Q; i++, ctx_pg++)
-			ctx->tqm_mem[i] = ctx_pg;
-
-		bp->ctx = ctx;
-		ctx->qp_max_entries = rte_le_to_cpu_32(resp->qp_max_entries);
-		ctx->qp_min_qp1_entries =
-			rte_le_to_cpu_16(resp->qp_min_qp1_entries);
-		ctx->qp_max_l2_entries =
-			rte_le_to_cpu_16(resp->qp_max_l2_entries);
-		ctx->qp_entry_size = rte_le_to_cpu_16(resp->qp_entry_size);
-		ctx->srq_max_l2_entries =
-			rte_le_to_cpu_16(resp->srq_max_l2_entries);
-		ctx->srq_max_entries = rte_le_to_cpu_32(resp->srq_max_entries);
-		ctx->srq_entry_size = rte_le_to_cpu_16(resp->srq_entry_size);
-		ctx->cq_max_l2_entries =
-			rte_le_to_cpu_16(resp->cq_max_l2_entries);
-		ctx->cq_max_entries = rte_le_to_cpu_32(resp->cq_max_entries);
-		ctx->cq_entry_size = rte_le_to_cpu_16(resp->cq_entry_size);
-		ctx->vnic_max_vnic_entries =
-			rte_le_to_cpu_16(resp->vnic_max_vnic_entries);
-		ctx->vnic_max_ring_table_entries =
-			rte_le_to_cpu_16(resp->vnic_max_ring_table_entries);
-		ctx->vnic_entry_size = rte_le_to_cpu_16(resp->vnic_entry_size);
-		ctx->stat_max_entries =
-			rte_le_to_cpu_32(resp->stat_max_entries);
-		ctx->stat_entry_size = rte_le_to_cpu_16(resp->stat_entry_size);
-		ctx->tqm_entry_size = rte_le_to_cpu_16(resp->tqm_entry_size);
-		ctx->tqm_min_entries_per_ring =
-			rte_le_to_cpu_32(resp->tqm_min_entries_per_ring);
-		ctx->tqm_max_entries_per_ring =
-			rte_le_to_cpu_32(resp->tqm_max_entries_per_ring);
-		ctx->tqm_entries_multiple = resp->tqm_entries_multiple;
-		if (!ctx->tqm_entries_multiple)
-			ctx->tqm_entries_multiple = 1;
-		ctx->mrav_max_entries =
-			rte_le_to_cpu_32(resp->mrav_max_entries);
-		ctx->mrav_entry_size = rte_le_to_cpu_16(resp->mrav_entry_size);
-		ctx->tim_entry_size = rte_le_to_cpu_16(resp->tim_entry_size);
-		ctx->tim_max_entries = rte_le_to_cpu_32(resp->tim_max_entries);
-	} else {
-		rc = 0;
+	ctx_pg = rte_malloc("bnxt_ctx_pg_mem",
+			    sizeof(*ctx_pg) * BNXT_MAX_Q,
+			    RTE_CACHE_LINE_SIZE);
+	if (!ctx_pg) {
+		rc = -ENOMEM;
+		goto ctx_err;
 	}
+	for (i = 0; i < BNXT_MAX_Q; i++, ctx_pg++)
+		ctx->tqm_mem[i] = ctx_pg;
+
+	bp->ctx = ctx;
+	ctx->qp_max_entries = rte_le_to_cpu_32(resp->qp_max_entries);
+	ctx->qp_min_qp1_entries =
+		rte_le_to_cpu_16(resp->qp_min_qp1_entries);
+	ctx->qp_max_l2_entries =
+		rte_le_to_cpu_16(resp->qp_max_l2_entries);
+	ctx->qp_entry_size = rte_le_to_cpu_16(resp->qp_entry_size);
+	ctx->srq_max_l2_entries =
+		rte_le_to_cpu_16(resp->srq_max_l2_entries);
+	ctx->srq_max_entries = rte_le_to_cpu_32(resp->srq_max_entries);
+	ctx->srq_entry_size = rte_le_to_cpu_16(resp->srq_entry_size);
+	ctx->cq_max_l2_entries =
+		rte_le_to_cpu_16(resp->cq_max_l2_entries);
+	ctx->cq_max_entries = rte_le_to_cpu_32(resp->cq_max_entries);
+	ctx->cq_entry_size = rte_le_to_cpu_16(resp->cq_entry_size);
+	ctx->vnic_max_vnic_entries =
+		rte_le_to_cpu_16(resp->vnic_max_vnic_entries);
+	ctx->vnic_max_ring_table_entries =
+		rte_le_to_cpu_16(resp->vnic_max_ring_table_entries);
+	ctx->vnic_entry_size = rte_le_to_cpu_16(resp->vnic_entry_size);
+	ctx->stat_max_entries =
+		rte_le_to_cpu_32(resp->stat_max_entries);
+	ctx->stat_entry_size = rte_le_to_cpu_16(resp->stat_entry_size);
+	ctx->tqm_entry_size = rte_le_to_cpu_16(resp->tqm_entry_size);
+	ctx->tqm_min_entries_per_ring =
+		rte_le_to_cpu_32(resp->tqm_min_entries_per_ring);
+	ctx->tqm_max_entries_per_ring =
+		rte_le_to_cpu_32(resp->tqm_max_entries_per_ring);
+	ctx->tqm_entries_multiple = resp->tqm_entries_multiple;
+	if (!ctx->tqm_entries_multiple)
+		ctx->tqm_entries_multiple = 1;
+	ctx->mrav_max_entries =
+		rte_le_to_cpu_32(resp->mrav_max_entries);
+	ctx->mrav_entry_size = rte_le_to_cpu_16(resp->mrav_entry_size);
+	ctx->tim_entry_size = rte_le_to_cpu_16(resp->tim_entry_size);
+	ctx->tim_max_entries = rte_le_to_cpu_32(resp->tim_max_entries);
 ctx_err:
 	HWRM_UNLOCK();
 	return rc;
Signature · 1 line
2.20.1 (Apple Git-117)
#10 lore raw

PATCH v2 10/17 net/bnxt: remove unused macro

In reply to Ajit Khaparde

From: Kalesh AP <[email protected]>

remove MAX_NUM_MAC_ADDR macro

---
 drivers/net/bnxt/bnxt.h | 1 -
 1 file changed, 1 deletion(-)
drivers/net/bnxt/bnxt.h +0 -1
index 68ac30bb58..089f86fec7 100644
@@ -488,7 +488,6 @@ struct bnxt {
 
 	struct bnxt_irq         *irq_tbl;
 
-#define MAX_NUM_MAC_ADDR	32
 	uint8_t			mac_addr[RTE_ETHER_ADDR_LEN];
 
 	uint16_t			hwrm_cmd_seq;
Signature · 1 line
2.20.1 (Apple Git-117)
#11 lore raw

PATCH v2 11/17 net/bnxt: remove unnecessary interrupt disable

In reply to Ajit Khaparde

From: Kalesh AP <[email protected]>

Since bnxt_enable_int() is called at start time, invoke bnxt_disable_int() during stop only. Remove it from device uninit function.

---
 drivers/net/bnxt/bnxt_ethdev.c | 1 -
 1 file changed, 1 deletion(-)
drivers/net/bnxt/bnxt_ethdev.c +0 -1
index 4d15ba5ceb..74ded7717a 100644
@@ -4576,7 +4576,6 @@ bnxt_uninit_resources(struct bnxt *bp, bool reconfig_dev)
 {
 	int rc;
 
-	bnxt_disable_int(bp);
 	bnxt_free_int(bp);
 	bnxt_free_mem(bp, reconfig_dev);
 	bnxt_hwrm_func_buf_unrgtr(bp);
Signature · 1 line
2.20.1 (Apple Git-117)
#12 lore raw

PATCH v2 12/17 net/bnxt: fix accessing variable before null check

In reply to Ajit Khaparde

From: Kalesh AP <[email protected]>

Check input argument "rxq" in bnxt_rx_queue_release_mbufs(), to be sure variable is not NULL before accessing it.

Fixes 9b63c6fd70e3 ("net/bnxt: support Rx/Tx queue start/stop")
---
 drivers/net/bnxt/bnxt_rxq.c | 53 +++++++++++++++++++------------------
 1 file changed, 27 insertions(+), 26 deletions(-)
drivers/net/bnxt/bnxt_rxq.c +27 -26
index 4b506f8653..dac8ccb167 100644
@@ -207,39 +207,40 @@ void bnxt_rx_queue_release_mbufs(struct bnxt_rx_queue *rxq)
 	struct bnxt_tpa_info *tpa_info;
 	uint16_t i;
 
+	if (!rxq)
+		return;
+
 	rte_spinlock_lock(&rxq->lock);
 
-	if (rxq) {
-		sw_ring = rxq->rx_ring->rx_buf_ring;
-		if (sw_ring) {
-			for (i = 0;
-			     i < rxq->rx_ring->rx_ring_struct->ring_size; i++) {
-				if (sw_ring[i].mbuf) {
-					rte_pktmbuf_free_seg(sw_ring[i].mbuf);
-					sw_ring[i].mbuf = NULL;
-				}
+	sw_ring = rxq->rx_ring->rx_buf_ring;
+	if (sw_ring) {
+		for (i = 0;
+		     i < rxq->rx_ring->rx_ring_struct->ring_size; i++) {
+			if (sw_ring[i].mbuf) {
+				rte_pktmbuf_free_seg(sw_ring[i].mbuf);
+				sw_ring[i].mbuf = NULL;
 			}
 		}
-		/* Free up mbufs in Agg ring */
-		sw_ring = rxq->rx_ring->ag_buf_ring;
-		if (sw_ring) {
-			for (i = 0;
-			     i < rxq->rx_ring->ag_ring_struct->ring_size; i++) {
-				if (sw_ring[i].mbuf) {
-					rte_pktmbuf_free_seg(sw_ring[i].mbuf);
-					sw_ring[i].mbuf = NULL;
-				}
+	}
+	/* Free up mbufs in Agg ring */
+	sw_ring = rxq->rx_ring->ag_buf_ring;
+	if (sw_ring) {
+		for (i = 0;
+		     i < rxq->rx_ring->ag_ring_struct->ring_size; i++) {
+			if (sw_ring[i].mbuf) {
+				rte_pktmbuf_free_seg(sw_ring[i].mbuf);
+				sw_ring[i].mbuf = NULL;
 			}
 		}
+	}
 
-		/* Free up mbufs in TPA */
-		tpa_info = rxq->rx_ring->tpa_info;
-		if (tpa_info) {
-			for (i = 0; i < BNXT_TPA_MAX; i++) {
-				if (tpa_info[i].mbuf) {
-					rte_pktmbuf_free_seg(tpa_info[i].mbuf);
-					tpa_info[i].mbuf = NULL;
-				}
+	/* Free up mbufs in TPA */
+	tpa_info = rxq->rx_ring->tpa_info;
+	if (tpa_info) {
+		for (i = 0; i < BNXT_TPA_MAX; i++) {
+			if (tpa_info[i].mbuf) {
+				rte_pktmbuf_free_seg(tpa_info[i].mbuf);
+				tpa_info[i].mbuf = NULL;
 			}
 		}
 	}
Signature · 1 line
2.20.1 (Apple Git-117)
#13 lore raw

PATCH v2 13/17 net/bnxt: fix Rx queue count devop

In reply to Ajit Khaparde

From: Rahul Gupta <[email protected]>

Fix Computing of number of used descriptors in an Rx queue.

Fixes 1b7ceba3e375 ("net/bnxt: support Rx queue count")
---
 drivers/net/bnxt/bnxt_ethdev.c | 31 +++++++------------------------
 1 file changed, 7 insertions(+), 24 deletions(-)
drivers/net/bnxt/bnxt_ethdev.c +7 -24
index 74ded7717a..0159be3462 100644
@@ -2096,9 +2096,6 @@ bnxt_rx_queue_count_op(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 	struct bnxt_cp_ring_info *cpr;
 	struct bnxt_rx_queue *rxq;
 	struct rx_pkt_cmpl *rxcmp;
-	uint16_t cmp_type;
-	uint8_t cmp = 1;
-	bool valid;
 	int rc;
 
 	rc = is_bnxt_in_error(bp);
@@ -2107,33 +2104,19 @@ bnxt_rx_queue_count_op(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 
 	rxq = dev->data->rx_queues[rx_queue_id];
 	cpr = rxq->cp_ring;
-	valid = cpr->valid;
+	raw_cons = cpr->cp_raw_cons;
 
-	while (raw_cons < rxq->nb_rx_desc) {
+	while (1) {
 		cons = RING_CMP(cpr->cp_ring_struct, raw_cons);
+		rte_prefetch0(&cpr->cp_desc_ring[cons]);
 		rxcmp = (struct rx_pkt_cmpl *)&cpr->cp_desc_ring[cons];
 
-		if (!CMPL_VALID(rxcmp, valid))
-			goto nothing_to_do;
-		valid = FLIP_VALID(cons, cpr->cp_ring_struct->ring_mask, valid);
-		cmp_type = CMP_TYPE(rxcmp);
-		if (cmp_type == RX_TPA_END_CMPL_TYPE_RX_TPA_END) {
-			cmp = (rte_le_to_cpu_32(
-					((struct rx_tpa_end_cmpl *)
-					 (rxcmp))->agg_bufs_v1) &
-			       RX_TPA_END_CMPL_AGG_BUFS_MASK) >>
-				RX_TPA_END_CMPL_AGG_BUFS_SFT;
-			desc++;
-		} else if (cmp_type == 0x11) {
-			desc++;
-			cmp = (rxcmp->agg_bufs_v1 &
-				   RX_PKT_CMPL_AGG_BUFS_MASK) >>
-				RX_PKT_CMPL_AGG_BUFS_SFT;
+		if (!CMP_VALID(rxcmp, raw_cons, cpr->cp_ring_struct)) {
+			break;
 		} else {
-			cmp = 1;
+			raw_cons++;
+			desc++;
 		}
-nothing_to_do:
-		raw_cons += cmp ? cmp : 2;
 	}
 
 	return desc;
Signature · 1 line
2.20.1 (Apple Git-117)
#14 lore raw

PATCH v2 14/17 net/bnxt: fix mbuf flags for PTP packets in vector mode Rx

In reply to Ajit Khaparde

From: Kalesh AP <[email protected]>

PKT_RX_IEEE1588_PTP and PKT_RX_IEEE1588_TMST flags need to be set in the offload flags for the received pkt in case of PTP offload.

Fixes bc4a000f2f53c ("net/bnxt: implement SSE vector mode")
---
 drivers/net/bnxt/bnxt_rxtx_vec_sse.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)
drivers/net/bnxt/bnxt_rxtx_vec_sse.c +10 -1
index 029053e305..b853a8810a 100644
@@ -255,6 +255,7 @@ bnxt_recv_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts,
 			uint16_t cp_cons;
 			struct rte_mbuf *mbuf;
 			__m128i mm_rxcmp, pkt_mb;
+			uint16_t flags_type;
 
 			tmp_raw_cons = NEXT_RAW_CMP(raw_cons);
 			cp_cons = RING_CMP(cpr->cp_ring_struct, tmp_raw_cons);
@@ -288,9 +289,17 @@ bnxt_recv_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts,
 
 			rte_compiler_barrier();
 
-			if (rxcmp->flags_type & RX_PKT_CMPL_FLAGS_RSS_VALID)
+			flags_type = rte_le_to_cpu_16(rxcmp->flags_type);
+
+			if (flags_type & RX_PKT_CMPL_FLAGS_RSS_VALID)
 				mbuf->ol_flags |= PKT_RX_RSS_HASH;
 
+#ifdef RTE_LIBRTE_IEEE1588
+			if ((flags_type & RX_PKT_CMPL_FLAGS_MASK) ==
+			    RX_PKT_CMPL_FLAGS_ITYPE_PTP_W_TIMESTAMP)
+				mbuf->ol_flags |= PKT_RX_IEEE1588_PTP |
+					PKT_RX_IEEE1588_TMST;
+#endif
 			if (rxcmp1->flags2 &
 			    RX_PKT_CMPL_FLAGS2_META_FORMAT_VLAN) {
 				mbuf->vlan_tci = rxcmp1->metadata &
Signature · 1 line
2.20.1 (Apple Git-117)
#15 lore raw

PATCH v2 15/17 net/bnxt: fix a segfault in secondary process

In reply to Ajit Khaparde

From: Stephen Hemminger <[email protected]>

The secondary process should not modify device state when init is called. The pci device information pointed to by dev_private pointer is not correct in secondary process.

Fixes b7778e8a1c00 ("net/bnxt: refactor to properly allocate resources for PF/VF")
---
 drivers/net/bnxt/bnxt_ethdev.c | 16 +++++++++-------
 1 file changed, 9 insertions(+), 7 deletions(-)
drivers/net/bnxt/bnxt_ethdev.c +9 -7
index 0159be3462..5985963bf7 100644
@@ -521,6 +521,7 @@ static int bnxt_init_nic(struct bnxt *bp)
 static int bnxt_dev_info_get_op(struct rte_eth_dev *eth_dev,
 				struct rte_eth_dev_info *dev_info)
 {
+	struct rte_pci_device *pdev = RTE_DEV_TO_PCI(eth_dev->device);
 	struct bnxt *bp = eth_dev->data->dev_private;
 	uint16_t max_vnics, i, j, vpool, vrxq;
 	unsigned int max_rx_rings;
@@ -536,7 +537,8 @@ static int bnxt_dev_info_get_op(struct rte_eth_dev *eth_dev,
 
 	/* PF/VF specifics */
 	if (BNXT_PF(bp))
-		dev_info->max_vfs = bp->pdev->max_vfs;
+		dev_info->max_vfs = pdev->max_vfs;
+
 	max_rx_rings = RTE_MIN(bp->max_rx_rings, bp->max_stat_ctx);
 	/* For the sake of symmetry, max_rx_queues = max_tx_queues */
 	dev_info->max_rx_queues = max_rx_rings;
@@ -4488,12 +4490,6 @@ bnxt_dev_init(struct rte_eth_dev *eth_dev)
 	if (version_printed++ == 0)
 		PMD_DRV_LOG(INFO, "%s\n", bnxt_version);
 
-	rte_eth_copy_pci_info(eth_dev, pci_dev);
-
-	bp = eth_dev->data->dev_private;
-
-	bp->dev_stopped = 1;
-
 	eth_dev->dev_ops = &bnxt_dev_ops;
 	eth_dev->rx_pkt_burst = &bnxt_recv_pkts;
 	eth_dev->tx_pkt_burst = &bnxt_xmit_pkts;
@@ -4505,6 +4501,12 @@ bnxt_dev_init(struct rte_eth_dev *eth_dev)
 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 		return 0;
 
+	rte_eth_copy_pci_info(eth_dev, pci_dev);
+
+	bp = eth_dev->data->dev_private;
+
+	bp->dev_stopped = 1;
+
 	if (bnxt_vf_pciid(pci_dev->id.device_id))
 		bp->flags |= BNXT_FLAG_VF;
 
Signature · 1 line
2.20.1 (Apple Git-117)
#16 lore raw

PATCH v2 16/17 net/bnxt: support for QinQ insertion and stripping

In reply to Ajit Khaparde

From: Somnath Kotur <[email protected]>

Driver will accelerate only outer/S-VLAN insertion by turning on the appropriate bits in the Tx Buffer Descriptor when the packet arrives for transmission.
The TPID to be used for this S-VLAN is conveyed by the vlan_tpid_set dev_op which will terminate in the driver.
In the Rx path, driver will continue providing the stripped vlan tag in the mbuf's vlan tci field. This would be the outermost vlan tag in a double-tagged packet or the vlan tag for a single vlan tagged pkt.

---
 drivers/net/bnxt/bnxt.h        |  4 ++
 drivers/net/bnxt/bnxt_ethdev.c | 71 ++++++++++++++++++++++++++++++++--
 drivers/net/bnxt/bnxt_txr.c    | 13 ++++++-
 3 files changed, 83 insertions(+), 5 deletions(-)
drivers/net/bnxt/bnxt.h +4 -0
index 089f86fec7..f8a550ba54 100644
@@ -521,6 +521,10 @@ struct bnxt {
 	uint16_t		max_stat_ctx;
 	uint16_t		first_vf_id;
 	uint16_t		vlan;
+#define BNXT_OUTER_TPID_MASK	0x0000ffff
+#define BNXT_OUTER_TPID_BD_MASK	0xffff0000
+#define BNXT_OUTER_TPID_BD_SHFT	16
+	uint32_t		outer_tpid_bd;
 	struct bnxt_pf_info	pf;
 	uint8_t			port_partition_type;
 	uint8_t			dev_stopped;
drivers/net/bnxt/bnxt_ethdev.c +68 -3

101 lines not on this page · raw message

drivers/net/bnxt/bnxt_txr.c +11 -2
index 99d2009055..0ed6581bed 100644
@@ -134,6 +134,7 @@ static uint16_t bnxt_start_xmit(struct rte_mbuf *tx_pkt,
 				struct tx_bd_long **last_txbd)
 {
 	struct bnxt_tx_ring_info *txr = txq->tx_ring;
+	uint32_t outer_tpid_bd = 0;
 	struct tx_bd_long *txbd;
 	struct tx_bd_long_hi *txbd1 = NULL;
 	uint32_t vlan_tag_flags, cfa_action;
@@ -155,7 +156,8 @@ static uint16_t bnxt_start_xmit(struct rte_mbuf *tx_pkt,
 				PKT_TX_UDP_CKSUM | PKT_TX_IP_CKSUM |
 				PKT_TX_VLAN_PKT | PKT_TX_OUTER_IP_CKSUM |
 				PKT_TX_TUNNEL_GRE | PKT_TX_TUNNEL_VXLAN |
-				PKT_TX_TUNNEL_GENEVE | PKT_TX_IEEE1588_TMST))
+				PKT_TX_TUNNEL_GENEVE | PKT_TX_IEEE1588_TMST |
+				PKT_TX_QINQ_PKT))
 		long_bd = true;
 
 	nr_bds = long_bd + tx_pkt->nb_segs;
@@ -209,7 +211,14 @@ static uint16_t bnxt_start_xmit(struct rte_mbuf *tx_pkt,
 		txbd->flags_type |= TX_BD_LONG_TYPE_TX_BD_LONG;
 		vlan_tag_flags = 0;
 		cfa_action = 0;
-		if (tx_buf->mbuf->ol_flags & PKT_TX_VLAN_PKT) {
+		/* HW can accelerate only outer vlan in QinQ mode */
+		if (tx_buf->mbuf->ol_flags & PKT_TX_QINQ_PKT) {
+			vlan_tag_flags = TX_BD_LONG_CFA_META_KEY_VLAN_TAG |
+				tx_buf->mbuf->vlan_tci_outer;
+			outer_tpid_bd = txq->bp->outer_tpid_bd &
+				BNXT_OUTER_TPID_BD_MASK;
+			vlan_tag_flags |= outer_tpid_bd;
+		} else if (tx_buf->mbuf->ol_flags & PKT_TX_VLAN_PKT) {
 			/* shurd: Should this mask at
 			 * TX_BD_LONG_CFA_META_VLAN_VID_MASK?
 			 */
Signature · 1 line
2.20.1 (Apple Git-117)
#17 lore raw

PATCH v2 17/17 net/bnxt: use macro for getting default vnic

In reply to Ajit Khaparde

From: Kalesh AP <[email protected]>

Minor code cleanup. BNXT_GET_DEFAULT_VNIC macro returns the default VNIC of the function and fixed code to use the macro.

---
 drivers/net/bnxt/bnxt_ethdev.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)
drivers/net/bnxt/bnxt_ethdev.c +5 -5
index b3a37e1554..bc9a910ee3 100644
@@ -1118,7 +1118,7 @@ static int bnxt_promiscuous_enable_op(struct rte_eth_dev *eth_dev)
 	if (bp->vnic_info == NULL)
 		return 0;
 
-	vnic = &bp->vnic_info[0];
+	vnic = BNXT_GET_DEFAULT_VNIC(bp);
 
 	old_flags = vnic->flags;
 	vnic->flags |= BNXT_VNIC_INFO_PROMISC;
@@ -1143,7 +1143,7 @@ static int bnxt_promiscuous_disable_op(struct rte_eth_dev *eth_dev)
 	if (bp->vnic_info == NULL)
 		return 0;
 
-	vnic = &bp->vnic_info[0];
+	vnic = BNXT_GET_DEFAULT_VNIC(bp);
 
 	old_flags = vnic->flags;
 	vnic->flags &= ~BNXT_VNIC_INFO_PROMISC;
@@ -1168,7 +1168,7 @@ static int bnxt_allmulticast_enable_op(struct rte_eth_dev *eth_dev)
 	if (bp->vnic_info == NULL)
 		return 0;
 
-	vnic = &bp->vnic_info[0];
+	vnic = BNXT_GET_DEFAULT_VNIC(bp);
 
 	old_flags = vnic->flags;
 	vnic->flags |= BNXT_VNIC_INFO_ALLMULTI;
@@ -1193,7 +1193,7 @@ static int bnxt_allmulticast_disable_op(struct rte_eth_dev *eth_dev)
 	if (bp->vnic_info == NULL)
 		return 0;
 
-	vnic = &bp->vnic_info[0];
+	vnic = BNXT_GET_DEFAULT_VNIC(bp);
 
 	old_flags = vnic->flags;
 	vnic->flags &= ~BNXT_VNIC_INFO_ALLMULTI;
@@ -1960,7 +1960,7 @@ bnxt_dev_set_mc_addr_list_op(struct rte_eth_dev *eth_dev,
 	if (rc)
 		return rc;
 
-	vnic = &bp->vnic_info[0];
+	vnic = BNXT_GET_DEFAULT_VNIC(bp);
 
 	if (nb_mc_addr > BNXT_MAX_MC_ADDRS) {
 		vnic->flags |= BNXT_VNIC_INFO_ALLMULTI;
Signature · 1 line
2.20.1 (Apple Git-117)
#18 lore raw

v3 cover bnxt patchset

In reply to Ajit Khaparde

bnxt patchset against dpdk-next-net.
This patchset has been rebased and is dependent on "bnxt patchset to support device error recovery" commit-id 5f30ec0329bf81d8a4a3404faaf450dbceb9463e

Please apply.

v1->v2: 5f30ec0329bf81d8a4a3404faaf450dbceb9463e v2->v3: Removed an unnecessary patch from v2. No other change.

Ajit Khaparde (2):
net/bnxt: subscribe to link speed change notification net/bnxt: fix async link handling and update

Kalesh AP (8):
net/bnxt: fix rxq stop/start handling net/bnxt: fix L2 context calculation for Thor net/bnxt: fix use of deferred start of transmit queues net/bnxt: remove unnecessary return check net/bnxt: remove unused macro
net/bnxt: remove unnecessary interrupt disable net/bnxt: fix accessing variable before null check net/bnxt: use macro for getting default vnic

Rahul Gupta (1):
net/bnxt: fix Rx queue count devop

Santoshkumar Karanappa Rastapur (3):
net/bnxt: fix incorrect flow steering net/bnxt: fix traffic failure with higher mbuf size net/bnxt: fix 40G link failure for Thor

Somnath Kotur (1):
net/bnxt: support for QinQ insertion and stripping

Stephen Hemminger (1):
net/bnxt: fix a segfault in secondary process

 drivers/net/bnxt/bnxt.h              |  10 +-
 drivers/net/bnxt/bnxt_cpr.c          |   2 +-
 drivers/net/bnxt/bnxt_ethdev.c       | 171 +++++++++++++++++----------
 drivers/net/bnxt/bnxt_flow.c         |   4 +
 drivers/net/bnxt/bnxt_hwrm.c         | 156 +++++++++++++-----------
 drivers/net/bnxt/bnxt_hwrm.h         |   2 +
 drivers/net/bnxt/bnxt_irq.c          |   2 +-
 drivers/net/bnxt/bnxt_irq.h          |   1 +
 drivers/net/bnxt/bnxt_ring.c         |   8 +-
 drivers/net/bnxt/bnxt_rxq.c          |  91 ++++++++------
 drivers/net/bnxt/bnxt_rxq.h          |   2 +-
 drivers/net/bnxt/bnxt_rxr.c          |  14 +--
 drivers/net/bnxt/bnxt_rxtx_vec_sse.c |   4 +-
 drivers/net/bnxt/bnxt_txq.c          |   5 +
 drivers/net/bnxt/bnxt_txq.h          |   1 +
 drivers/net/bnxt/bnxt_txr.c          |  19 ++-
 16 files changed, 290 insertions(+), 202 deletions(-)
Signature · 1 line
2.20.1 (Apple Git-117)
#19 lore raw

PATCH v3 1/16 net/bnxt: fix incorrect flow steering

From: Santoshkumar Karanappa Rastapur <[email protected]>

When user creates a flow similar to an existing flow with just the destination queue change, we delete the old filter and allocate a new one with this destination queue change. We were also allocating a new L2 filter matching the same destination mac resulting in 2 L2 filters for the same destination mac. This was causing any flow matching the destination mac to be steered to this queue instead of the default queue. Fixed it by deleting this stale L2 filter.

---
 drivers/net/bnxt/bnxt_flow.c | 4 ++++
 1 file changed, 4 insertions(+)
drivers/net/bnxt/bnxt_flow.c +4 -0
index deb9733e99..be9b6fad39 100644
@@ -1028,6 +1028,10 @@ bnxt_match_filter(struct bnxt *bp, struct bnxt_filter_info *nf)
 				    sizeof(nf->dst_ipaddr_mask))) {
 				if (mf->dst_id == nf->dst_id)
 					return -EEXIST;
+				/* Clear the new L2 filter that was created
+				 * earlier in bnxt_validate_and_parse_flow.
+				 */
+				bnxt_hwrm_clear_l2_filter(bp, nf);
 				/*
 				 * Same Flow, Different queue
 				 * Clear the old ntuple filter
Signature · 1 line
2.20.1 (Apple Git-117)
#20 lore raw

PATCH v3 2/16 net/bnxt: fix traffic failure with higher mbuf size

In reply to Ajit Khaparde

From: Santoshkumar Karanappa Rastapur <[email protected]>

Maximum frame length supported by hw is 9600 bytes. When user launches testpmd with --mbuf-size=9729, we are posting buffer descriptors of size 9601 to the ring. This was causing packets getting discarded.

---
 drivers/net/bnxt/bnxt.h        |  5 ++++-
 drivers/net/bnxt/bnxt_ethdev.c | 29 +++++++++--------------------
 drivers/net/bnxt/bnxt_hwrm.c   |  3 +++
 drivers/net/bnxt/bnxt_ring.c   |  5 -----
 drivers/net/bnxt/bnxt_rxq.c    |  2 +-
 drivers/net/bnxt/bnxt_rxq.h    |  1 -
 drivers/net/bnxt/bnxt_rxr.c    |  7 ++-----
 7 files changed, 19 insertions(+), 33 deletions(-)
drivers/net/bnxt/bnxt.h +4 -1
index 818a49f461..68ac30bb58 100644
@@ -23,12 +23,15 @@
 
 #define BNXT_MAX_MTU		9574
 #define VLAN_TAG_SIZE		4
+#define BNXT_NUM_VLANS		2
+#define BNXT_MAX_PKT_LEN	(BNXT_MAX_MTU + RTE_ETHER_HDR_LEN +\
+				 RTE_ETHER_CRC_LEN +\
+				 (BNXT_NUM_VLANS * VLAN_TAG_SIZE))
 #define BNXT_VF_RSV_NUM_RSS_CTX	1
 #define BNXT_VF_RSV_NUM_L2_CTX	4
 /* TODO: For now, do not support VMDq/RFS on VFs. */
 #define BNXT_VF_RSV_NUM_VNIC	1
 #define BNXT_MAX_LED		4
-#define BNXT_NUM_VLANS		2
 #define BNXT_MIN_RING_DESC	16
 #define BNXT_MAX_TX_RING_DESC	4096
 #define BNXT_MAX_RX_RING_DESC	8192
drivers/net/bnxt/bnxt_ethdev.c +9 -20
index 0083ba6e83..4095e9c720 100644
@@ -545,10 +545,13 @@ static int bnxt_dev_info_get_op(struct rte_eth_dev *eth_dev,
 	dev_info->hash_key_size = 40;
 	max_vnics = bp->max_vnics;
 
+	/* MTU specifics */
+	dev_info->min_mtu = RTE_ETHER_MIN_MTU;
+	dev_info->max_mtu = BNXT_MAX_MTU;
+
 	/* Fast path specifics */
 	dev_info->min_rx_bufsize = 1;
-	dev_info->max_rx_pktlen = BNXT_MAX_MTU + RTE_ETHER_HDR_LEN +
-		RTE_ETHER_CRC_LEN + VLAN_TAG_SIZE * 2;
+	dev_info->max_rx_pktlen = BNXT_MAX_PKT_LEN;
 
 	dev_info->rx_offload_capa = BNXT_DEV_RX_OFFLOAD_SUPPORT;
 	if (bp->flags & BNXT_FLAG_PTP_SUPPORTED)
@@ -1967,7 +1970,6 @@ bnxt_txq_info_get_op(struct rte_eth_dev *dev, uint16_t queue_id,
 static int bnxt_mtu_set_op(struct rte_eth_dev *eth_dev, uint16_t new_mtu)
 {
 	struct bnxt *bp = eth_dev->data->dev_private;
-	struct rte_eth_dev_info dev_info;
 	uint32_t new_pkt_size;
 	uint32_t rc = 0;
 	uint32_t i;
@@ -1979,18 +1981,6 @@ static int bnxt_mtu_set_op(struct rte_eth_dev *eth_dev, uint16_t new_mtu)
 	new_pkt_size = new_mtu + RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN +
 		       VLAN_TAG_SIZE * BNXT_NUM_VLANS;
 
-	rc = bnxt_dev_info_get_op(eth_dev, &dev_info);
-	if (rc != 0) {
-		PMD_DRV_LOG(ERR, "Error during getting ethernet device info\n");
-		return rc;
-	}
-
-	if (new_mtu < RTE_ETHER_MIN_MTU || new_mtu > BNXT_MAX_MTU) {
-		PMD_DRV_LOG(ERR, "MTU requested must be within (%d, %d)\n",
-			RTE_ETHER_MIN_MTU, BNXT_MAX_MTU);
-		return -EINVAL;
-	}
-
 #ifdef RTE_ARCH_X86
 	/*
 	 * If vector-mode tx/rx is active, disallow any MTU change that would
@@ -2020,15 +2010,12 @@ static int bnxt_mtu_set_op(struct rte_eth_dev *eth_dev, uint16_t new_mtu)
 
 	eth_dev->data->dev_conf.rxmode.max_rx_pkt_len = new_pkt_size;
 
-	eth_dev->data->mtu = new_mtu;
-	PMD_DRV_LOG(INFO, "New MTU is %d\n", eth_dev->data->mtu);
-
 	for (i = 0; i < bp->nr_vnics; i++) {
 		struct bnxt_vnic_info *vnic = &bp->vnic_info[i];
 		uint16_t size = 0;
 
-		vnic->mru = bp->eth_dev->data->mtu + RTE_ETHER_HDR_LEN +
-					RTE_ETHER_CRC_LEN + VLAN_TAG_SIZE * 2;
+		vnic->mru = new_mtu + RTE_ETHER_HDR_LEN +
+				RTE_ETHER_CRC_LEN + VLAN_TAG_SIZE * 2;
 		rc = bnxt_hwrm_vnic_cfg(bp, vnic);
 		if (rc)
 			break;
@@ -2043,6 +2030,8 @@ static int bnxt_mtu_set_op(struct rte_eth_dev *eth_dev, uint16_t new_mtu)
 		}
 	}
 
+	PMD_DRV_LOG(INFO, "New MTU is %d\n", new_mtu);
+
 	return rc;
 }
 
drivers/net/bnxt/bnxt_hwrm.c +3 -0
index 174dc75d54..09330e2382 100644
@@ -1233,6 +1233,7 @@ int bnxt_hwrm_ring_alloc(struct bnxt *bp,
 			mb_pool = bp->rx_queues[0]->mb_pool;
 			rx_buf_size = rte_pktmbuf_data_room_size(mb_pool) -
 				      RTE_PKTMBUF_HEADROOM;
+			rx_buf_size = RTE_MIN(BNXT_MAX_PKT_LEN, rx_buf_size);
 			req.rx_buf_size = rte_cpu_to_le_16(rx_buf_size);
 			enables |=
 				HWRM_RING_ALLOC_INPUT_ENABLES_RX_BUF_SIZE_VALID;
@@ -1263,6 +1264,7 @@ int bnxt_hwrm_ring_alloc(struct bnxt *bp,
 		mb_pool = bp->rx_queues[0]->mb_pool;
 		rx_buf_size = rte_pktmbuf_data_room_size(mb_pool) -
 			      RTE_PKTMBUF_HEADROOM;
+		rx_buf_size = RTE_MIN(BNXT_MAX_PKT_LEN, rx_buf_size);
 		req.rx_buf_size = rte_cpu_to_le_16(rx_buf_size);
 
 		req.stat_ctx_id = rte_cpu_to_le_32(stats_ctx_id);
@@ -1879,6 +1881,7 @@ int bnxt_hwrm_vnic_plcmode_cfg(struct bnxt *bp,
 
 	size = rte_pktmbuf_data_room_size(bp->rx_queues[0]->mb_pool);
 	size -= RTE_PKTMBUF_HEADROOM;
+	size = RTE_MIN(BNXT_MAX_PKT_LEN, size);
 
 	req.jumbo_thresh = rte_cpu_to_le_16(size);
 	req.vnic_id = rte_cpu_to_le_16(vnic->fw_vnic_id);
drivers/net/bnxt/bnxt_ring.c +0 -5
index ec17783cf8..73eee43148 100644
@@ -561,9 +561,6 @@ int bnxt_alloc_hwrm_rx_ring(struct bnxt *bp, int queue_index)
 	if (rc)
 		goto err_out;
 
-	rxq->rx_buf_use_size = BNXT_MAX_MTU + RTE_ETHER_HDR_LEN +
-		RTE_ETHER_CRC_LEN + (2 * VLAN_TAG_SIZE);
-
 	if (bp->eth_dev->data->rx_queue_state[queue_index] ==
 	    RTE_ETH_QUEUE_STATE_STARTED) {
 		if (bnxt_init_one_rx_ring(rxq)) {
@@ -640,8 +637,6 @@ int bnxt_alloc_hwrm_rings(struct bnxt *bp)
 		if (bnxt_alloc_rx_agg_ring(bp, i))
 			goto err_out;
 
-		rxq->rx_buf_use_size = BNXT_MAX_MTU + RTE_ETHER_HDR_LEN +
-					RTE_ETHER_CRC_LEN + (2 * VLAN_TAG_SIZE);
 		if (bnxt_init_one_rx_ring(rxq)) {
 			PMD_DRV_LOG(ERR, "bnxt_init_one_rx_ring failed!\n");
 			bnxt_rx_queue_release_op(rxq);
drivers/net/bnxt/bnxt_rxq.c +1 -1
index d5fc5268db..192ffd8780 100644
@@ -331,13 +331,13 @@ int bnxt_rx_queue_setup_op(struct rte_eth_dev *eth_dev,
 	rxq->nb_rx_desc = nb_desc;
 	rxq->rx_free_thresh = rx_conf->rx_free_thresh;
 
-	PMD_DRV_LOG(DEBUG, "RX Buf size is %d\n", rxq->rx_buf_use_size);
 	PMD_DRV_LOG(DEBUG, "RX Buf MTU %d\n", eth_dev->data->mtu);
 
 	rc = bnxt_init_rx_ring_struct(rxq, socket_id);
 	if (rc)
 		goto out;
 
+	PMD_DRV_LOG(DEBUG, "RX Buf size is %d\n", rxq->rx_buf_size);
 	rxq->queue_id = queue_idx;
 	rxq->port_id = eth_dev->data->port_id;
 	if (rx_offloads & DEV_RX_OFFLOAD_KEEP_CRC)
drivers/net/bnxt/bnxt_rxq.h +0 -1
index b5e42d01c0..c60b04426c 100644
@@ -36,7 +36,6 @@ struct bnxt_rx_queue {
 	struct bnxt_vnic_info	*vnic;
 
 	uint32_t			rx_buf_size;
-	uint32_t			rx_buf_use_size;  /* useable size */
 	struct bnxt_rx_ring_info	*rx_ring;
 	struct bnxt_cp_ring_info	*cp_ring;
 	struct bnxt_cp_ring_info	*nq_ring;
drivers/net/bnxt/bnxt_rxr.c +2 -5
index 28487fb38e..f60880c63e 100644
@@ -713,9 +713,7 @@ int bnxt_init_rx_ring_struct(struct bnxt_rx_queue *rxq, unsigned int socket_id)
 	struct bnxt_rx_ring_info *rxr;
 	struct bnxt_ring *ring;
 
-	rxq->rx_buf_use_size = BNXT_MAX_MTU + RTE_ETHER_HDR_LEN +
-		RTE_ETHER_CRC_LEN + (2 * VLAN_TAG_SIZE);
-	rxq->rx_buf_size = rxq->rx_buf_use_size + sizeof(struct rte_mbuf);
+	rxq->rx_buf_size = BNXT_MAX_PKT_LEN + sizeof(struct rte_mbuf);
 
 	rxr = rte_zmalloc_socket("bnxt_rx_ring",
 				 sizeof(struct bnxt_rx_ring_info),
@@ -826,8 +824,7 @@ int bnxt_init_one_rx_ring(struct bnxt_rx_queue *rxq)
 	uint16_t size;
 
 	size = rte_pktmbuf_data_room_size(rxq->mb_pool) - RTE_PKTMBUF_HEADROOM;
-	if (rxq->rx_buf_use_size <= size)
-		size = rxq->rx_buf_use_size;
+	size = RTE_MIN(BNXT_MAX_PKT_LEN, size);
 
 	type = RX_PROD_PKT_BD_TYPE_RX_PROD_PKT | RX_PROD_PKT_BD_FLAGS_EOP_PAD;
 
Signature · 1 line
2.20.1 (Apple Git-117)
#21 lore raw

PATCH v3 3/16 net/bnxt: fix rxq stop/start handling

In reply to Ajit Khaparde

From: Kalesh AP <[email protected]>

Driver should not change "deferred_start" state of the rx queues. It should get the state in queue_setup_op() and use that value.

Since the deferred start state was being used in the packet receive functions to determine whether a stopped rx ring should be polled, introduced a per-rxq flag to track queue stopped/started state.

Fixes 9b63c6fd70e3 ("net/bnxt: support Rx/Tx queue start/stop")
---
 drivers/net/bnxt/bnxt_ethdev.c       |  2 +-
 drivers/net/bnxt/bnxt_ring.c         |  3 +--
 drivers/net/bnxt/bnxt_rxq.c          | 36 +++++++++++++++++++---------
 drivers/net/bnxt/bnxt_rxq.h          |  1 +
 drivers/net/bnxt/bnxt_rxr.c          |  7 +++---
 drivers/net/bnxt/bnxt_rxtx_vec_sse.c |  2 +-
 6 files changed, 32 insertions(+), 19 deletions(-)
drivers/net/bnxt/bnxt_ethdev.c +1 -1
index 4095e9c720..d1c5acb3e2 100644
@@ -1945,7 +1945,7 @@ bnxt_rxq_info_get_op(struct rte_eth_dev *dev, uint16_t queue_id,
 
 	qinfo->conf.rx_free_thresh = rxq->rx_free_thresh;
 	qinfo->conf.rx_drop_en = 0;
-	qinfo->conf.rx_deferred_start = 0;
+	qinfo->conf.rx_deferred_start = rxq->rx_deferred_start;
 }
 
 static void
drivers/net/bnxt/bnxt_ring.c +1 -2
index 73eee43148..68a690de5a 100644
@@ -561,8 +561,7 @@ int bnxt_alloc_hwrm_rx_ring(struct bnxt *bp, int queue_index)
 	if (rc)
 		goto err_out;
 
-	if (bp->eth_dev->data->rx_queue_state[queue_index] ==
-	    RTE_ETH_QUEUE_STATE_STARTED) {
+	if (!rxq->rx_started) {
 		if (bnxt_init_one_rx_ring(rxq)) {
 			RTE_LOG(ERR, PMD,
 				"bnxt_init_one_rx_ring failed!\n");
drivers/net/bnxt/bnxt_rxq.c +25 -11
index 192ffd8780..4b506f8653 100644
@@ -357,9 +357,19 @@ int bnxt_rx_queue_setup_op(struct rte_eth_dev *eth_dev,
 	}
 	rte_atomic64_init(&rxq->rx_mbuf_alloc_fail);
 
-	rxq->rx_deferred_start = rx_conf->rx_deferred_start;
-	queue_state = rxq->rx_deferred_start ? RTE_ETH_QUEUE_STATE_STOPPED :
-						RTE_ETH_QUEUE_STATE_STARTED;
+	/* rxq 0 must not be stopped when used as async CPR */
+	if (!BNXT_NUM_ASYNC_CPR(bp) && queue_idx == 0)
+		rxq->rx_deferred_start = false;
+	else
+		rxq->rx_deferred_start = rx_conf->rx_deferred_start;
+
+	if (rxq->rx_deferred_start) {
+		queue_state = RTE_ETH_QUEUE_STATE_STOPPED;
+		rxq->rx_started = false;
+	} else {
+		queue_state = RTE_ETH_QUEUE_STATE_STARTED;
+		rxq->rx_started = true;
+	}
 	eth_dev->data->rx_queue_state[queue_idx] = queue_state;
 	rte_spinlock_init(&rxq->lock);
 
@@ -432,6 +442,11 @@ int bnxt_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 		return -EINVAL;
 	}
 
+	/* Set the queue state to started here.
+	 * We check the status of the queue while posting buffer.
+	 * If queue is it started, we do not post buffers for Rx.
+	 */
+	rxq->rx_started = true;
 	bnxt_free_hwrm_rx_ring(bp, rx_queue_id);
 	rc = bnxt_alloc_hwrm_rx_ring(bp, rx_queue_id);
 	if (rc)
@@ -448,20 +463,19 @@ int bnxt_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 
 			vnic->fw_grp_ids[rx_queue_id] =
 					bp->grp_info[rx_queue_id].fw_grp_id;
+			PMD_DRV_LOG(DEBUG,
+				    "vnic = %p fw_grp_id = %d\n",
+				    vnic, bp->grp_info[rx_queue_id].fw_grp_id);
 		}
 
-		PMD_DRV_LOG(DEBUG,
-			    "vnic = %p fw_grp_id = %d\n",
-			    vnic, bp->grp_info[rx_queue_id].fw_grp_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;
-		rxq->rx_deferred_start = false;
-	}
+	else
+		rxq->rx_started = false;
 
 	PMD_DRV_LOG(INFO,
 		    "queue %d, rx_deferred_start %d, state %d!\n",
@@ -500,7 +514,7 @@ int bnxt_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 	}
 
 	dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
-	rxq->rx_deferred_start = true;
+	rxq->rx_started = false;
 	PMD_DRV_LOG(DEBUG, "Rx queue stopped\n");
 
 	if (dev_conf->rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG) {
drivers/net/bnxt/bnxt_rxq.h +1 -0
index c60b04426c..3693d89a60 100644
@@ -30,6 +30,7 @@ struct bnxt_rx_queue {
 	uint16_t		port_id; /* Device port identifier */
 	uint8_t			crc_len; /* 0 if CRC stripped, 4 otherwise */
 	uint8_t			rx_deferred_start; /* not in global dev start */
+	uint8_t			rx_started; /* RX queue is started */
 
 	struct bnxt		*bp;
 	int			index;
drivers/net/bnxt/bnxt_rxr.c +3 -4
index f60880c63e..b3cc0d8a04 100644
@@ -574,10 +574,9 @@ uint16_t bnxt_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
 	if (unlikely(is_bnxt_in_error(rxq->bp)))
 		return 0;
 
-	/* If Rx Q was stopped return. RxQ0 cannot be stopped. */
-	if (unlikely(((rxq->rx_deferred_start ||
-		       !rte_spinlock_trylock(&rxq->lock)) &&
-		      rxq->queue_id)))
+	/* If Rx Q was stopped return */
+	if (unlikely(!rxq->rx_started ||
+		     !rte_spinlock_trylock(&rxq->lock)))
 		return 0;
 
 	/* Handle RX burst request */
drivers/net/bnxt/bnxt_rxtx_vec_sse.c +1 -1
index 2e6e83c949..5ae812f3b9 100644
@@ -224,7 +224,7 @@ bnxt_recv_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts,
 			     0xFF, 0xFF, 0xFF, 0xFF); /* pkt_type (zeroes) */
 
 	/* If Rx Q was stopped return */
-	if (rxq->rx_deferred_start)
+	if (unlikely(!rxq->rx_started))
 		return 0;
 
 	if (rxq->rxrearm_nb >= RTE_BNXT_RXQ_REARM_THRESH)
Signature · 1 line
2.20.1 (Apple Git-117)
#22 lore raw

PATCH v3 4/16 net/bnxt: subscribe to link speed change notification

In reply to Ajit Khaparde

We are not subscribing to link speed change notification. This patch addresses that.

---
 drivers/net/bnxt/bnxt_hwrm.c | 1 +
 drivers/net/bnxt/bnxt_hwrm.h | 2 ++
 2 files changed, 3 insertions(+)
drivers/net/bnxt/bnxt_hwrm.c +1 -0
index 09330e2382..b2033ea416 100644
@@ -738,6 +738,7 @@ int bnxt_hwrm_func_driver_register(struct bnxt *bp)
 		rte_cpu_to_le_32(ASYNC_CMPL_EVENT_ID_LINK_STATUS_CHANGE |
 				 ASYNC_CMPL_EVENT_ID_PORT_CONN_NOT_ALLOWED |
 				 ASYNC_CMPL_EVENT_ID_LINK_SPEED_CFG_CHANGE |
+				 ASYNC_CMPL_EVENT_ID_LINK_SPEED_CHANGE |
 				 ASYNC_CMPL_EVENT_ID_RESET_NOTIFY);
 	if (bp->flags & BNXT_FLAG_FW_CAP_ERROR_RECOVERY)
 		req.async_event_fwd[0] |=
drivers/net/bnxt/bnxt_hwrm.h +2 -0
index 0d386952b6..5190f43290 100644
@@ -21,6 +21,8 @@ struct bnxt_cp_ring_info;
 	(1 << HWRM_ASYNC_EVENT_CMPL_EVENT_ID_PORT_CONN_NOT_ALLOWED)
 #define ASYNC_CMPL_EVENT_ID_LINK_SPEED_CFG_CHANGE	\
 	(1 << HWRM_ASYNC_EVENT_CMPL_EVENT_ID_LINK_SPEED_CFG_CHANGE)
+#define ASYNC_CMPL_EVENT_ID_LINK_SPEED_CHANGE \
+	(1 << HWRM_ASYNC_EVENT_CMPL_EVENT_ID_LINK_SPEED_CHANGE)
 #define ASYNC_CMPL_EVENT_ID_RESET_NOTIFY \
 	(1 << HWRM_ASYNC_EVENT_CMPL_EVENT_ID_RESET_NOTIFY)
 #define ASYNC_CMPL_EVENT_ID_ERROR_RECOVERY \
Signature · 1 line
2.20.1 (Apple Git-117)
#23 lore raw

PATCH v3 5/16 net/bnxt: fix async link handling and update

In reply to Ajit Khaparde

When updating the link because of an async link notification there is no need to set wait_for_completion. At this point the link related information should be available without need to poll. Use rte_eth_linkstatus_set instead of memcpy to ensure atomicity while updating the link status.
We force the physical link down as a part of device stop.
But we are not waiting there enough and handling the async notification before exiting. It just sits in the default CQ till we do a device start. Fix it by calling the CQ handler in device stop.

Fixes 7bc8e9a227cc ("net/bnxt: support async link notification")
---
 drivers/net/bnxt/bnxt_cpr.c    |  2 +-
 drivers/net/bnxt/bnxt_ethdev.c | 11 +++++++----
 drivers/net/bnxt/bnxt_irq.c    |  2 +-
 drivers/net/bnxt/bnxt_irq.h    |  1 +
 4 files changed, 10 insertions(+), 6 deletions(-)
drivers/net/bnxt/bnxt_cpr.c +1 -1
index 4817672ef0..f58372516a 100644
@@ -66,7 +66,7 @@ void bnxt_handle_async_event(struct bnxt *bp,
 	case HWRM_ASYNC_EVENT_CMPL_EVENT_ID_LINK_SPEED_CHANGE:
 	case HWRM_ASYNC_EVENT_CMPL_EVENT_ID_LINK_SPEED_CFG_CHANGE:
 		/* FALLTHROUGH */
-		bnxt_link_update_op(bp->eth_dev, 1);
+		bnxt_link_update_op(bp->eth_dev, 0);
 		break;
 	case HWRM_ASYNC_EVENT_CMPL_EVENT_ID_PF_DRVR_UNLOAD:
 		PMD_DRV_LOG(INFO, "Async event: PF driver unloaded\n");
drivers/net/bnxt/bnxt_ethdev.c +7 -4
index d1c5acb3e2..4d15ba5ceb 100644
@@ -834,6 +834,7 @@ static int bnxt_dev_start_op(struct rte_eth_dev *eth_dev)
 			bp->rx_cp_nr_rings, RTE_ETHDEV_QUEUE_STAT_CNTRS);
 	}
 
+	bnxt_enable_int(bp);
 	rc = bnxt_hwrm_if_change(bp, 1);
 	if (!rc) {
 		if (bp->flags & BNXT_FLAG_IF_CHANGE_HOT_FW_RESET_DONE) {
@@ -862,7 +863,6 @@ static int bnxt_dev_start_op(struct rte_eth_dev *eth_dev)
 	eth_dev->rx_pkt_burst = bnxt_receive_function(eth_dev);
 	eth_dev->tx_pkt_burst = bnxt_transmit_function(eth_dev);
 
-	bnxt_enable_int(bp);
 	bp->flags |= BNXT_FLAG_INIT_DONE;
 	eth_dev->data->dev_started = 1;
 	bp->dev_stopped = 0;
@@ -926,7 +926,9 @@ static void bnxt_dev_stop_op(struct rte_eth_dev *eth_dev)
 		/* TBD: STOP HW queues DMA */
 		eth_dev->data->dev_link.link_status = 0;
 	}
-	bnxt_set_hwrm_link_config(bp, false);
+	bnxt_dev_set_link_down_op(eth_dev);
+	/* Wait for link to be reset and the async notification to process. */
+	rte_delay_ms(BNXT_LINK_WAIT_INTERVAL * 2);
 
 	/* Clean queue intr-vector mapping */
 	rte_intr_efd_disable(intr_handle);
@@ -938,6 +940,8 @@ static void bnxt_dev_stop_op(struct rte_eth_dev *eth_dev)
 	bnxt_hwrm_port_clr_stats(bp);
 	bnxt_free_tx_mbufs(bp);
 	bnxt_free_rx_mbufs(bp);
+	/* Process any remaining notifications in default completion queue */
+	bnxt_int_handler(eth_dev);
 	bnxt_shutdown_nic(bp);
 	bnxt_hwrm_if_change(bp, 0);
 	bp->dev_stopped = 1;
@@ -1084,8 +1088,7 @@ int bnxt_link_update_op(struct rte_eth_dev *eth_dev, int wait_to_complete)
 	/* Timed out or success */
 	if (new.link_status != eth_dev->data->dev_link.link_status ||
 	new.link_speed != eth_dev->data->dev_link.link_speed) {
-		memcpy(&eth_dev->data->dev_link, &new,
-			sizeof(struct rte_eth_link));
+		rte_eth_linkstatus_set(eth_dev, &new);
 
 		_rte_eth_dev_callback_process(eth_dev,
 					      RTE_ETH_EVENT_INTR_LSC,
drivers/net/bnxt/bnxt_irq.c +1 -1
index a22700a0da..729d68d704 100644
@@ -18,7 +18,7 @@
  * Interrupts
  */
 
-static void bnxt_int_handler(void *param)
+void bnxt_int_handler(void *param)
 {
 	struct rte_eth_dev *eth_dev = (struct rte_eth_dev *)param;
 	struct bnxt *bp = eth_dev->data->dev_private;
drivers/net/bnxt/bnxt_irq.h +1 -0
index 460a97a09c..1b56e08068 100644
@@ -22,5 +22,6 @@ void bnxt_disable_int(struct bnxt *bp);
 void bnxt_enable_int(struct bnxt *bp);
 int bnxt_setup_int(struct bnxt *bp);
 int bnxt_request_int(struct bnxt *bp);
+void bnxt_int_handler(void *param);
 
 #endif
Signature · 1 line
2.20.1 (Apple Git-117)
#24 lore raw

PATCH v3 6/16 net/bnxt: fix 40G link failure for Thor

In reply to Ajit Khaparde

From: Santoshkumar Karanappa Rastapur <[email protected]>

Link autonegotiation for 40G is not supported by Thor.
Hence speed needs to be forced and autoneg disabled to configure 40G speed.

Fixes f8168ca0e690 ("net/bnxt: support thor controller")
---
 drivers/net/bnxt/bnxt_hwrm.c | 22 +++++++++++++++++++---
 1 file changed, 19 insertions(+), 3 deletions(-)
drivers/net/bnxt/bnxt_hwrm.c +19 -3
index b2033ea416..6e95acb89e 100644
@@ -2679,11 +2679,27 @@ int bnxt_set_hwrm_link_config(struct bnxt *bp, bool link_up)
 		goto port_phy_cfg;
 
 	autoneg = bnxt_check_eth_link_autoneg(dev_conf->link_speeds);
+	if (BNXT_CHIP_THOR(bp) &&
+	    dev_conf->link_speeds == ETH_LINK_SPEED_40G) {
+		/* 40G is not supported as part of media auto detect.
+		 * The speed should be forced and autoneg disabled
+		 * to configure 40G speed.
+		 */
+		PMD_DRV_LOG(INFO, "Disabling autoneg for 40G\n");
+		autoneg = 0;
+	}
+
 	speed = bnxt_parse_eth_link_speed(dev_conf->link_speeds);
 	link_req.phy_flags = HWRM_PORT_PHY_CFG_INPUT_FLAGS_RESET_PHY;
-	/* Autoneg can be done only when the FW allows */
-	if (autoneg == 1 && !(bp->link_info.auto_link_speed ||
-				bp->link_info.force_link_speed)) {
+	/* Autoneg can be done only when the FW allows.
+	 * When user configures fixed speed of 40G and later changes to
+	 * any other speed, auto_link_speed/force_link_speed is still set
+	 * to 40G until link comes up at new speed.
+	 */
+	if (autoneg == 1 &&
+	    !(!BNXT_CHIP_THOR(bp) &&
+	      (bp->link_info.auto_link_speed ||
+	       bp->link_info.force_link_speed))) {
 		link_req.phy_flags |=
 				HWRM_PORT_PHY_CFG_INPUT_FLAGS_RESTART_AUTONEG;
 		link_req.auto_link_speed_mask =
Signature · 1 line
2.20.1 (Apple Git-117)
#25 lore raw

PATCH v3 7/16 net/bnxt: fix L2 context calculation for Thor

In reply to Ajit Khaparde

From: Kalesh AP <[email protected]>

On Wh+, number of L2 context supported by FW is the sum of number of EM flow count and number of L2 context count returned in HWRM_FUNC_QCFG. This is not true for Thor. Restrict this only for Whitney for now.

Fixes ff9c0ca47e81176 ("net/bnxt: save the number of EM flow count")
---
 drivers/net/bnxt/bnxt_hwrm.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)
drivers/net/bnxt/bnxt_hwrm.c +6 -5
index 6e95acb89e..d0f4054412 100644
@@ -611,8 +611,9 @@ static int __bnxt_hwrm_func_qcaps(struct bnxt *bp)
 	bp->max_rx_rings = rte_le_to_cpu_16(resp->max_rx_rings);
 	bp->first_vf_id = rte_le_to_cpu_16(resp->first_vf_id);
 	bp->max_rx_em_flows = rte_le_to_cpu_16(resp->max_rx_em_flows);
-	bp->max_l2_ctx =
-		rte_le_to_cpu_16(resp->max_l2_ctxs) + bp->max_rx_em_flows;
+	bp->max_l2_ctx = rte_le_to_cpu_16(resp->max_l2_ctxs);
+	if (!BNXT_CHIP_THOR(bp))
+		bp->max_l2_ctx += bp->max_rx_em_flows;
 	/* TODO: For now, do not support VMDq/RFS on VFs. */
 	if (BNXT_PF(bp)) {
 		if (bp->pf.max_vfs)
@@ -858,9 +859,9 @@ int bnxt_hwrm_func_resc_qcaps(struct bnxt *bp)
 		/* func_resource_qcaps does not return max_rx_em_flows.
 		 * So use the value provided by func_qcaps.
 		 */
-		bp->max_l2_ctx =
-			rte_le_to_cpu_16(resp->max_l2_ctxs) +
-			bp->max_rx_em_flows;
+		bp->max_l2_ctx = rte_le_to_cpu_16(resp->max_l2_ctxs);
+		if (!BNXT_CHIP_THOR(bp))
+			bp->max_l2_ctx += bp->max_rx_em_flows;
 		bp->max_vnics = rte_le_to_cpu_16(resp->max_vnics);
 		bp->max_stat_ctx = rte_le_to_cpu_16(resp->max_stat_ctx);
 	}
Signature · 1 line
2.20.1 (Apple Git-117)
#26 lore raw

PATCH v3 8/16 net/bnxt: fix use of deferred start of transmit queues

In reply to Ajit Khaparde

From: Kalesh AP <[email protected]>

Driver should not change "deferred_start" state of the tx queues. It should get the state in queue_setup_op() and use that value.

Since the deferred start state was being used in the packet transmit functions to determine whether the queue has been stopped already, introduced a per-txq flag to track queue stopped/started state.

Fixes 9b63c6fd70e3 ("net/bnxt: support Rx/Tx queue start/stop")
---
 drivers/net/bnxt/bnxt_rxtx_vec_sse.c | 2 +-
 drivers/net/bnxt/bnxt_txq.c          | 5 +++++
 drivers/net/bnxt/bnxt_txq.h          | 1 +
 drivers/net/bnxt/bnxt_txr.c          | 6 +++---
 4 files changed, 10 insertions(+), 4 deletions(-)
drivers/net/bnxt/bnxt_rxtx_vec_sse.c +1 -1
index 5ae812f3b9..029053e305 100644
@@ -482,7 +482,7 @@ bnxt_xmit_pkts_vec(void *tx_queue, struct rte_mbuf **tx_pkts,
 	struct bnxt_tx_queue *txq = tx_queue;
 
 	/* Tx queue was stopped; wait for it to be restarted */
-	if (unlikely(txq->tx_deferred_start)) {
+	if (unlikely(!txq->tx_started)) {
 		PMD_DRV_LOG(DEBUG, "Tx q stopped;return\n");
 		return 0;
 	}
drivers/net/bnxt/bnxt_txq.c +5 -0
index 0901324793..ea20d737fe 100644
@@ -131,6 +131,7 @@ int bnxt_tx_queue_setup_op(struct rte_eth_dev *eth_dev,
 	txq->bp = bp;
 	txq->nb_tx_desc = nb_desc;
 	txq->tx_free_thresh = tx_conf->tx_free_thresh;
+	txq->tx_deferred_start = tx_conf->tx_deferred_start;
 
 	rc = bnxt_init_tx_ring_struct(txq, socket_id);
 	if (rc)
@@ -157,6 +158,10 @@ int bnxt_tx_queue_setup_op(struct rte_eth_dev *eth_dev,
 
 	eth_dev->data->tx_queues[queue_idx] = txq;
 
+	if (txq->tx_deferred_start)
+		txq->tx_started = false;
+	else
+		txq->tx_started = true;
 out:
 	return rc;
 }
drivers/net/bnxt/bnxt_txq.h +1 -0
index 9190e3f731..7a442516d2 100644
@@ -24,6 +24,7 @@ struct bnxt_tx_queue {
 	uint8_t			wthresh; /* Write-back threshold reg */
 	uint32_t		ctx_curr; /* Hardware context states */
 	uint8_t			tx_deferred_start; /* not in global dev start */
+	uint8_t			tx_started; /* TX queue is started */
 
 	struct bnxt		*bp;
 	int			index;
drivers/net/bnxt/bnxt_txr.c +3 -3
index 172b480b2e..99d2009055 100644
@@ -469,7 +469,7 @@ uint16_t bnxt_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
 	bnxt_handle_tx_cp(txq);
 
 	/* Tx queue was stopped; wait for it to be restarted */
-	if (txq->tx_deferred_start) {
+	if (unlikely(!txq->tx_started)) {
 		PMD_DRV_LOG(DEBUG, "Tx q stopped;return\n");
 		return 0;
 	}
@@ -518,7 +518,7 @@ int bnxt_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 		return rc;
 
 	dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
-	txq->tx_deferred_start = false;
+	txq->tx_started = true;
 	PMD_DRV_LOG(DEBUG, "Tx queue started\n");
 
 	return 0;
@@ -538,7 +538,7 @@ int bnxt_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 	bnxt_handle_tx_cp(txq);
 
 	dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
-	txq->tx_deferred_start = true;
+	txq->tx_started = false;
 	PMD_DRV_LOG(DEBUG, "Tx queue stopped\n");
 
 	return 0;
Signature · 1 line
2.20.1 (Apple Git-117)
#27 lore raw

PATCH v3 9/16 net/bnxt: remove unnecessary return check

In reply to Ajit Khaparde

From: Kalesh AP <[email protected]>

HWRM_CHECK_RESULT_SILENT() returns if the HWRM command fails. There is no need to check the return value after this call.

---
 drivers/net/bnxt/bnxt_hwrm.c | 119 +++++++++++++++++------------------
 1 file changed, 56 insertions(+), 63 deletions(-)
drivers/net/bnxt/bnxt_hwrm.c +56 -63
index d0f4054412..1065f21de7 100644
@@ -4379,7 +4379,10 @@ int bnxt_hwrm_func_backing_store_qcaps(struct bnxt *bp)
 	struct hwrm_func_backing_store_qcaps_input req = {0};
 	struct hwrm_func_backing_store_qcaps_output *resp =
 		bp->hwrm_cmd_resp_addr;
-	int rc;
+	struct bnxt_ctx_pg_info *ctx_pg;
+	struct bnxt_ctx_mem_info *ctx;
+	int total_alloc_len;
+	int rc, i;
 
 	if (!BNXT_CHIP_THOR(bp) ||
 	    bp->hwrm_spec_code < HWRM_VERSION_1_9_2 ||
@@ -4391,70 +4394,60 @@ int bnxt_hwrm_func_backing_store_qcaps(struct bnxt *bp)
 	rc = bnxt_hwrm_send_message(bp, &req, sizeof(req), BNXT_USE_CHIMP_MB);
 	HWRM_CHECK_RESULT_SILENT();
 
-	if (!rc) {
-		struct bnxt_ctx_pg_info *ctx_pg;
-		struct bnxt_ctx_mem_info *ctx;
-		int total_alloc_len;
-		int i;
-
-		total_alloc_len = sizeof(*ctx);
-		ctx = rte_malloc("bnxt_ctx_mem", total_alloc_len,
-				 RTE_CACHE_LINE_SIZE);
-		if (!ctx) {
-			rc = -ENOMEM;
-			goto ctx_err;
-		}
-		memset(ctx, 0, total_alloc_len);
+	total_alloc_len = sizeof(*ctx);
+	ctx = rte_zmalloc("bnxt_ctx_mem", total_alloc_len,
+			  RTE_CACHE_LINE_SIZE);
+	if (!ctx) {
+		rc = -ENOMEM;
+		goto ctx_err;
+	}
 
-		ctx_pg = rte_malloc("bnxt_ctx_pg_mem",
-				    sizeof(*ctx_pg) * BNXT_MAX_Q,
-				    RTE_CACHE_LINE_SIZE);
-		if (!ctx_pg) {
-			rc = -ENOMEM;
-			goto ctx_err;
-		}
-		for (i = 0; i < BNXT_MAX_Q; i++, ctx_pg++)
-			ctx->tqm_mem[i] = ctx_pg;
-
-		bp->ctx = ctx;
-		ctx->qp_max_entries = rte_le_to_cpu_32(resp->qp_max_entries);
-		ctx->qp_min_qp1_entries =
-			rte_le_to_cpu_16(resp->qp_min_qp1_entries);
-		ctx->qp_max_l2_entries =
-			rte_le_to_cpu_16(resp->qp_max_l2_entries);
-		ctx->qp_entry_size = rte_le_to_cpu_16(resp->qp_entry_size);
-		ctx->srq_max_l2_entries =
-			rte_le_to_cpu_16(resp->srq_max_l2_entries);
-		ctx->srq_max_entries = rte_le_to_cpu_32(resp->srq_max_entries);
-		ctx->srq_entry_size = rte_le_to_cpu_16(resp->srq_entry_size);
-		ctx->cq_max_l2_entries =
-			rte_le_to_cpu_16(resp->cq_max_l2_entries);
-		ctx->cq_max_entries = rte_le_to_cpu_32(resp->cq_max_entries);
-		ctx->cq_entry_size = rte_le_to_cpu_16(resp->cq_entry_size);
-		ctx->vnic_max_vnic_entries =
-			rte_le_to_cpu_16(resp->vnic_max_vnic_entries);
-		ctx->vnic_max_ring_table_entries =
-			rte_le_to_cpu_16(resp->vnic_max_ring_table_entries);
-		ctx->vnic_entry_size = rte_le_to_cpu_16(resp->vnic_entry_size);
-		ctx->stat_max_entries =
-			rte_le_to_cpu_32(resp->stat_max_entries);
-		ctx->stat_entry_size = rte_le_to_cpu_16(resp->stat_entry_size);
-		ctx->tqm_entry_size = rte_le_to_cpu_16(resp->tqm_entry_size);
-		ctx->tqm_min_entries_per_ring =
-			rte_le_to_cpu_32(resp->tqm_min_entries_per_ring);
-		ctx->tqm_max_entries_per_ring =
-			rte_le_to_cpu_32(resp->tqm_max_entries_per_ring);
-		ctx->tqm_entries_multiple = resp->tqm_entries_multiple;
-		if (!ctx->tqm_entries_multiple)
-			ctx->tqm_entries_multiple = 1;
-		ctx->mrav_max_entries =
-			rte_le_to_cpu_32(resp->mrav_max_entries);
-		ctx->mrav_entry_size = rte_le_to_cpu_16(resp->mrav_entry_size);
-		ctx->tim_entry_size = rte_le_to_cpu_16(resp->tim_entry_size);
-		ctx->tim_max_entries = rte_le_to_cpu_32(resp->tim_max_entries);
-	} else {
-		rc = 0;
+	ctx_pg = rte_malloc("bnxt_ctx_pg_mem",
+			    sizeof(*ctx_pg) * BNXT_MAX_Q,
+			    RTE_CACHE_LINE_SIZE);
+	if (!ctx_pg) {
+		rc = -ENOMEM;
+		goto ctx_err;
 	}
+	for (i = 0; i < BNXT_MAX_Q; i++, ctx_pg++)
+		ctx->tqm_mem[i] = ctx_pg;
+
+	bp->ctx = ctx;
+	ctx->qp_max_entries = rte_le_to_cpu_32(resp->qp_max_entries);
+	ctx->qp_min_qp1_entries =
+		rte_le_to_cpu_16(resp->qp_min_qp1_entries);
+	ctx->qp_max_l2_entries =
+		rte_le_to_cpu_16(resp->qp_max_l2_entries);
+	ctx->qp_entry_size = rte_le_to_cpu_16(resp->qp_entry_size);
+	ctx->srq_max_l2_entries =
+		rte_le_to_cpu_16(resp->srq_max_l2_entries);
+	ctx->srq_max_entries = rte_le_to_cpu_32(resp->srq_max_entries);
+	ctx->srq_entry_size = rte_le_to_cpu_16(resp->srq_entry_size);
+	ctx->cq_max_l2_entries =
+		rte_le_to_cpu_16(resp->cq_max_l2_entries);
+	ctx->cq_max_entries = rte_le_to_cpu_32(resp->cq_max_entries);
+	ctx->cq_entry_size = rte_le_to_cpu_16(resp->cq_entry_size);
+	ctx->vnic_max_vnic_entries =
+		rte_le_to_cpu_16(resp->vnic_max_vnic_entries);
+	ctx->vnic_max_ring_table_entries =
+		rte_le_to_cpu_16(resp->vnic_max_ring_table_entries);
+	ctx->vnic_entry_size = rte_le_to_cpu_16(resp->vnic_entry_size);
+	ctx->stat_max_entries =
+		rte_le_to_cpu_32(resp->stat_max_entries);
+	ctx->stat_entry_size = rte_le_to_cpu_16(resp->stat_entry_size);
+	ctx->tqm_entry_size = rte_le_to_cpu_16(resp->tqm_entry_size);
+	ctx->tqm_min_entries_per_ring =
+		rte_le_to_cpu_32(resp->tqm_min_entries_per_ring);
+	ctx->tqm_max_entries_per_ring =
+		rte_le_to_cpu_32(resp->tqm_max_entries_per_ring);
+	ctx->tqm_entries_multiple = resp->tqm_entries_multiple;
+	if (!ctx->tqm_entries_multiple)
+		ctx->tqm_entries_multiple = 1;
+	ctx->mrav_max_entries =
+		rte_le_to_cpu_32(resp->mrav_max_entries);
+	ctx->mrav_entry_size = rte_le_to_cpu_16(resp->mrav_entry_size);
+	ctx->tim_entry_size = rte_le_to_cpu_16(resp->tim_entry_size);
+	ctx->tim_max_entries = rte_le_to_cpu_32(resp->tim_max_entries);
 ctx_err:
 	HWRM_UNLOCK();
 	return rc;
Signature · 1 line
2.20.1 (Apple Git-117)
#28 lore raw

PATCH v3 10/16 net/bnxt: remove unused macro

In reply to Ajit Khaparde

From: Kalesh AP <[email protected]>

remove MAX_NUM_MAC_ADDR macro

---
 drivers/net/bnxt/bnxt.h | 1 -
 1 file changed, 1 deletion(-)
drivers/net/bnxt/bnxt.h +0 -1
index 68ac30bb58..089f86fec7 100644
@@ -488,7 +488,6 @@ struct bnxt {
 
 	struct bnxt_irq         *irq_tbl;
 
-#define MAX_NUM_MAC_ADDR	32
 	uint8_t			mac_addr[RTE_ETHER_ADDR_LEN];
 
 	uint16_t			hwrm_cmd_seq;
Signature · 1 line
2.20.1 (Apple Git-117)
#29 lore raw

PATCH v3 11/16 net/bnxt: remove unnecessary interrupt disable

In reply to Ajit Khaparde

From: Kalesh AP <[email protected]>

Since bnxt_enable_int() is called at start time, invoke bnxt_disable_int() during stop only. Remove it from device uninit function.

---
 drivers/net/bnxt/bnxt_ethdev.c | 1 -
 1 file changed, 1 deletion(-)
drivers/net/bnxt/bnxt_ethdev.c +0 -1
index 4d15ba5ceb..74ded7717a 100644
@@ -4576,7 +4576,6 @@ bnxt_uninit_resources(struct bnxt *bp, bool reconfig_dev)
 {
 	int rc;
 
-	bnxt_disable_int(bp);
 	bnxt_free_int(bp);
 	bnxt_free_mem(bp, reconfig_dev);
 	bnxt_hwrm_func_buf_unrgtr(bp);
Signature · 1 line
2.20.1 (Apple Git-117)
#30 lore raw

PATCH v3 12/16 net/bnxt: fix accessing variable before null check

In reply to Ajit Khaparde

From: Kalesh AP <[email protected]>

Check input argument "rxq" in bnxt_rx_queue_release_mbufs(), to be sure variable is not NULL before accessing it.

Fixes 9b63c6fd70e3 ("net/bnxt: support Rx/Tx queue start/stop")
---
 drivers/net/bnxt/bnxt_rxq.c | 53 +++++++++++++++++++------------------
 1 file changed, 27 insertions(+), 26 deletions(-)
drivers/net/bnxt/bnxt_rxq.c +27 -26
index 4b506f8653..dac8ccb167 100644
@@ -207,39 +207,40 @@ void bnxt_rx_queue_release_mbufs(struct bnxt_rx_queue *rxq)
 	struct bnxt_tpa_info *tpa_info;
 	uint16_t i;
 
+	if (!rxq)
+		return;
+
 	rte_spinlock_lock(&rxq->lock);
 
-	if (rxq) {
-		sw_ring = rxq->rx_ring->rx_buf_ring;
-		if (sw_ring) {
-			for (i = 0;
-			     i < rxq->rx_ring->rx_ring_struct->ring_size; i++) {
-				if (sw_ring[i].mbuf) {
-					rte_pktmbuf_free_seg(sw_ring[i].mbuf);
-					sw_ring[i].mbuf = NULL;
-				}
+	sw_ring = rxq->rx_ring->rx_buf_ring;
+	if (sw_ring) {
+		for (i = 0;
+		     i < rxq->rx_ring->rx_ring_struct->ring_size; i++) {
+			if (sw_ring[i].mbuf) {
+				rte_pktmbuf_free_seg(sw_ring[i].mbuf);
+				sw_ring[i].mbuf = NULL;
 			}
 		}
-		/* Free up mbufs in Agg ring */
-		sw_ring = rxq->rx_ring->ag_buf_ring;
-		if (sw_ring) {
-			for (i = 0;
-			     i < rxq->rx_ring->ag_ring_struct->ring_size; i++) {
-				if (sw_ring[i].mbuf) {
-					rte_pktmbuf_free_seg(sw_ring[i].mbuf);
-					sw_ring[i].mbuf = NULL;
-				}
+	}
+	/* Free up mbufs in Agg ring */
+	sw_ring = rxq->rx_ring->ag_buf_ring;
+	if (sw_ring) {
+		for (i = 0;
+		     i < rxq->rx_ring->ag_ring_struct->ring_size; i++) {
+			if (sw_ring[i].mbuf) {
+				rte_pktmbuf_free_seg(sw_ring[i].mbuf);
+				sw_ring[i].mbuf = NULL;
 			}
 		}
+	}
 
-		/* Free up mbufs in TPA */
-		tpa_info = rxq->rx_ring->tpa_info;
-		if (tpa_info) {
-			for (i = 0; i < BNXT_TPA_MAX; i++) {
-				if (tpa_info[i].mbuf) {
-					rte_pktmbuf_free_seg(tpa_info[i].mbuf);
-					tpa_info[i].mbuf = NULL;
-				}
+	/* Free up mbufs in TPA */
+	tpa_info = rxq->rx_ring->tpa_info;
+	if (tpa_info) {
+		for (i = 0; i < BNXT_TPA_MAX; i++) {
+			if (tpa_info[i].mbuf) {
+				rte_pktmbuf_free_seg(tpa_info[i].mbuf);
+				tpa_info[i].mbuf = NULL;
 			}
 		}
 	}
Signature · 1 line
2.20.1 (Apple Git-117)
#31 lore raw

PATCH v3 13/16 net/bnxt: fix Rx queue count devop

In reply to Ajit Khaparde

From: Rahul Gupta <[email protected]>

Fix Computing of number of used descriptors in an Rx queue.

Fixes 1b7ceba3e375 ("net/bnxt: support Rx queue count")
---
 drivers/net/bnxt/bnxt_ethdev.c | 31 +++++++------------------------
 1 file changed, 7 insertions(+), 24 deletions(-)
drivers/net/bnxt/bnxt_ethdev.c +7 -24
index 74ded7717a..0159be3462 100644
@@ -2096,9 +2096,6 @@ bnxt_rx_queue_count_op(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 	struct bnxt_cp_ring_info *cpr;
 	struct bnxt_rx_queue *rxq;
 	struct rx_pkt_cmpl *rxcmp;
-	uint16_t cmp_type;
-	uint8_t cmp = 1;
-	bool valid;
 	int rc;
 
 	rc = is_bnxt_in_error(bp);
@@ -2107,33 +2104,19 @@ bnxt_rx_queue_count_op(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 
 	rxq = dev->data->rx_queues[rx_queue_id];
 	cpr = rxq->cp_ring;
-	valid = cpr->valid;
+	raw_cons = cpr->cp_raw_cons;
 
-	while (raw_cons < rxq->nb_rx_desc) {
+	while (1) {
 		cons = RING_CMP(cpr->cp_ring_struct, raw_cons);
+		rte_prefetch0(&cpr->cp_desc_ring[cons]);
 		rxcmp = (struct rx_pkt_cmpl *)&cpr->cp_desc_ring[cons];
 
-		if (!CMPL_VALID(rxcmp, valid))
-			goto nothing_to_do;
-		valid = FLIP_VALID(cons, cpr->cp_ring_struct->ring_mask, valid);
-		cmp_type = CMP_TYPE(rxcmp);
-		if (cmp_type == RX_TPA_END_CMPL_TYPE_RX_TPA_END) {
-			cmp = (rte_le_to_cpu_32(
-					((struct rx_tpa_end_cmpl *)
-					 (rxcmp))->agg_bufs_v1) &
-			       RX_TPA_END_CMPL_AGG_BUFS_MASK) >>
-				RX_TPA_END_CMPL_AGG_BUFS_SFT;
-			desc++;
-		} else if (cmp_type == 0x11) {
-			desc++;
-			cmp = (rxcmp->agg_bufs_v1 &
-				   RX_PKT_CMPL_AGG_BUFS_MASK) >>
-				RX_PKT_CMPL_AGG_BUFS_SFT;
+		if (!CMP_VALID(rxcmp, raw_cons, cpr->cp_ring_struct)) {
+			break;
 		} else {
-			cmp = 1;
+			raw_cons++;
+			desc++;
 		}
-nothing_to_do:
-		raw_cons += cmp ? cmp : 2;
 	}
 
 	return desc;
Signature · 1 line
2.20.1 (Apple Git-117)
#32 lore raw

PATCH v3 14/16 net/bnxt: fix a segfault in secondary process

In reply to Ajit Khaparde

From: Stephen Hemminger <[email protected]>

The secondary process should not modify device state when init is called. The pci device information pointed to by dev_private pointer is not correct in secondary process.

Fixes b7778e8a1c00 ("net/bnxt: refactor to properly allocate resources for PF/VF")
---
 drivers/net/bnxt/bnxt_ethdev.c | 16 +++++++++-------
 1 file changed, 9 insertions(+), 7 deletions(-)
drivers/net/bnxt/bnxt_ethdev.c +9 -7
index 0159be3462..5985963bf7 100644
@@ -521,6 +521,7 @@ static int bnxt_init_nic(struct bnxt *bp)
 static int bnxt_dev_info_get_op(struct rte_eth_dev *eth_dev,
 				struct rte_eth_dev_info *dev_info)
 {
+	struct rte_pci_device *pdev = RTE_DEV_TO_PCI(eth_dev->device);
 	struct bnxt *bp = eth_dev->data->dev_private;
 	uint16_t max_vnics, i, j, vpool, vrxq;
 	unsigned int max_rx_rings;
@@ -536,7 +537,8 @@ static int bnxt_dev_info_get_op(struct rte_eth_dev *eth_dev,
 
 	/* PF/VF specifics */
 	if (BNXT_PF(bp))
-		dev_info->max_vfs = bp->pdev->max_vfs;
+		dev_info->max_vfs = pdev->max_vfs;
+
 	max_rx_rings = RTE_MIN(bp->max_rx_rings, bp->max_stat_ctx);
 	/* For the sake of symmetry, max_rx_queues = max_tx_queues */
 	dev_info->max_rx_queues = max_rx_rings;
@@ -4488,12 +4490,6 @@ bnxt_dev_init(struct rte_eth_dev *eth_dev)
 	if (version_printed++ == 0)
 		PMD_DRV_LOG(INFO, "%s\n", bnxt_version);
 
-	rte_eth_copy_pci_info(eth_dev, pci_dev);
-
-	bp = eth_dev->data->dev_private;
-
-	bp->dev_stopped = 1;
-
 	eth_dev->dev_ops = &bnxt_dev_ops;
 	eth_dev->rx_pkt_burst = &bnxt_recv_pkts;
 	eth_dev->tx_pkt_burst = &bnxt_xmit_pkts;
@@ -4505,6 +4501,12 @@ bnxt_dev_init(struct rte_eth_dev *eth_dev)
 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 		return 0;
 
+	rte_eth_copy_pci_info(eth_dev, pci_dev);
+
+	bp = eth_dev->data->dev_private;
+
+	bp->dev_stopped = 1;
+
 	if (bnxt_vf_pciid(pci_dev->id.device_id))
 		bp->flags |= BNXT_FLAG_VF;
 
Signature · 1 line
2.20.1 (Apple Git-117)
#33 lore raw

PATCH v3 15/16 net/bnxt: support for QinQ insertion and stripping

In reply to Ajit Khaparde

From: Somnath Kotur <[email protected]>

Driver will accelerate only outer/S-VLAN insertion by turning on the appropriate bits in the Tx Buffer Descriptor when the packet arrives for transmission.
The TPID to be used for this S-VLAN is conveyed by the vlan_tpid_set dev_op which will terminate in the driver.
In the Rx path, driver will continue providing the stripped vlan tag in the mbuf's vlan tci field. This would be the outermost vlan tag in a double-tagged packet or the vlan tag for a single vlan tagged pkt.

---
 drivers/net/bnxt/bnxt.h        |  4 ++
 drivers/net/bnxt/bnxt_ethdev.c | 71 ++++++++++++++++++++++++++++++++--
 drivers/net/bnxt/bnxt_txr.c    | 13 ++++++-
 3 files changed, 83 insertions(+), 5 deletions(-)
drivers/net/bnxt/bnxt.h +4 -0
index 089f86fec7..f8a550ba54 100644
@@ -521,6 +521,10 @@ struct bnxt {
 	uint16_t		max_stat_ctx;
 	uint16_t		first_vf_id;
 	uint16_t		vlan;
+#define BNXT_OUTER_TPID_MASK	0x0000ffff
+#define BNXT_OUTER_TPID_BD_MASK	0xffff0000
+#define BNXT_OUTER_TPID_BD_SHFT	16
+	uint32_t		outer_tpid_bd;
 	struct bnxt_pf_info	pf;
 	uint8_t			port_partition_type;
 	uint8_t			dev_stopped;
drivers/net/bnxt/bnxt_ethdev.c +68 -3

101 lines not on this page · raw message

drivers/net/bnxt/bnxt_txr.c +11 -2
index 99d2009055..0ed6581bed 100644
@@ -134,6 +134,7 @@ static uint16_t bnxt_start_xmit(struct rte_mbuf *tx_pkt,
 				struct tx_bd_long **last_txbd)
 {
 	struct bnxt_tx_ring_info *txr = txq->tx_ring;
+	uint32_t outer_tpid_bd = 0;
 	struct tx_bd_long *txbd;
 	struct tx_bd_long_hi *txbd1 = NULL;
 	uint32_t vlan_tag_flags, cfa_action;
@@ -155,7 +156,8 @@ static uint16_t bnxt_start_xmit(struct rte_mbuf *tx_pkt,
 				PKT_TX_UDP_CKSUM | PKT_TX_IP_CKSUM |
 				PKT_TX_VLAN_PKT | PKT_TX_OUTER_IP_CKSUM |
 				PKT_TX_TUNNEL_GRE | PKT_TX_TUNNEL_VXLAN |
-				PKT_TX_TUNNEL_GENEVE | PKT_TX_IEEE1588_TMST))
+				PKT_TX_TUNNEL_GENEVE | PKT_TX_IEEE1588_TMST |
+				PKT_TX_QINQ_PKT))
 		long_bd = true;
 
 	nr_bds = long_bd + tx_pkt->nb_segs;
@@ -209,7 +211,14 @@ static uint16_t bnxt_start_xmit(struct rte_mbuf *tx_pkt,
 		txbd->flags_type |= TX_BD_LONG_TYPE_TX_BD_LONG;
 		vlan_tag_flags = 0;
 		cfa_action = 0;
-		if (tx_buf->mbuf->ol_flags & PKT_TX_VLAN_PKT) {
+		/* HW can accelerate only outer vlan in QinQ mode */
+		if (tx_buf->mbuf->ol_flags & PKT_TX_QINQ_PKT) {
+			vlan_tag_flags = TX_BD_LONG_CFA_META_KEY_VLAN_TAG |
+				tx_buf->mbuf->vlan_tci_outer;
+			outer_tpid_bd = txq->bp->outer_tpid_bd &
+				BNXT_OUTER_TPID_BD_MASK;
+			vlan_tag_flags |= outer_tpid_bd;
+		} else if (tx_buf->mbuf->ol_flags & PKT_TX_VLAN_PKT) {
 			/* shurd: Should this mask at
 			 * TX_BD_LONG_CFA_META_VLAN_VID_MASK?
 			 */
Signature · 1 line
2.20.1 (Apple Git-117)
#34 lore raw

PATCH v3 16/16 net/bnxt: use macro for getting default vnic

In reply to Ajit Khaparde

From: Kalesh AP <[email protected]>

Minor code cleanup. BNXT_GET_DEFAULT_VNIC macro returns the default VNIC of the function and fixed code to use the macro.

---
 drivers/net/bnxt/bnxt_ethdev.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)
drivers/net/bnxt/bnxt_ethdev.c +5 -5
index b3a37e1554..bc9a910ee3 100644
@@ -1118,7 +1118,7 @@ static int bnxt_promiscuous_enable_op(struct rte_eth_dev *eth_dev)
 	if (bp->vnic_info == NULL)
 		return 0;
 
-	vnic = &bp->vnic_info[0];
+	vnic = BNXT_GET_DEFAULT_VNIC(bp);
 
 	old_flags = vnic->flags;
 	vnic->flags |= BNXT_VNIC_INFO_PROMISC;
@@ -1143,7 +1143,7 @@ static int bnxt_promiscuous_disable_op(struct rte_eth_dev *eth_dev)
 	if (bp->vnic_info == NULL)
 		return 0;
 
-	vnic = &bp->vnic_info[0];
+	vnic = BNXT_GET_DEFAULT_VNIC(bp);
 
 	old_flags = vnic->flags;
 	vnic->flags &= ~BNXT_VNIC_INFO_PROMISC;
@@ -1168,7 +1168,7 @@ static int bnxt_allmulticast_enable_op(struct rte_eth_dev *eth_dev)
 	if (bp->vnic_info == NULL)
 		return 0;
 
-	vnic = &bp->vnic_info[0];
+	vnic = BNXT_GET_DEFAULT_VNIC(bp);
 
 	old_flags = vnic->flags;
 	vnic->flags |= BNXT_VNIC_INFO_ALLMULTI;
@@ -1193,7 +1193,7 @@ static int bnxt_allmulticast_disable_op(struct rte_eth_dev *eth_dev)
 	if (bp->vnic_info == NULL)
 		return 0;
 
-	vnic = &bp->vnic_info[0];
+	vnic = BNXT_GET_DEFAULT_VNIC(bp);
 
 	old_flags = vnic->flags;
 	vnic->flags &= ~BNXT_VNIC_INFO_ALLMULTI;
@@ -1960,7 +1960,7 @@ bnxt_dev_set_mc_addr_list_op(struct rte_eth_dev *eth_dev,
 	if (rc)
 		return rc;
 
-	vnic = &bp->vnic_info[0];
+	vnic = BNXT_GET_DEFAULT_VNIC(bp);
 
 	if (nb_mc_addr > BNXT_MAX_MC_ADDRS) {
 		vnic->flags |= BNXT_VNIC_INFO_ALLMULTI;
Signature · 1 line
2.20.1 (Apple Git-117)
#35 lore raw
In reply to Ajit Khaparde
Quoted 32 lines · Ajit Khaparde

bnxt patchset against dpdk-next-net.
This patchset has been rebased and is dependent on "bnxt patchset to support device error recovery" commit-id 5f30ec0329bf81d8a4a3404faaf450dbceb9463e

Please apply.

v1->v2: 5f30ec0329bf81d8a4a3404faaf450dbceb9463e v2->v3: Removed an unnecessary patch from v2. No other change.

Ajit Khaparde (2):
net/bnxt: subscribe to link speed change notification net/bnxt: fix async link handling and update

Kalesh AP (8):
net/bnxt: fix rxq stop/start handling net/bnxt: fix L2 context calculation for Thor net/bnxt: fix use of deferred start of transmit queues net/bnxt: remove unnecessary return check net/bnxt: remove unused macro
net/bnxt: remove unnecessary interrupt disable net/bnxt: fix accessing variable before null check net/bnxt: use macro for getting default vnic

Rahul Gupta (1):
net/bnxt: fix Rx queue count devop

Santoshkumar Karanappa Rastapur (3):
net/bnxt: fix incorrect flow steering net/bnxt: fix traffic failure with higher mbuf size net/bnxt: fix 40G link failure for Thor

Quoted 5 lines

Somnath Kotur (1):
net/bnxt: support for QinQ insertion and stripping

Stephen Hemminger (1):
net/bnxt: fix a segfault in secondary process

Series applied to dpdk-next-net/master, thanks.