Skip to contentSkip to conversation

bnxt patchset for thor and bnxt vector PMD

PATCH New Lance Richardson 1 participant 8 messages Aug 30, 2019 Network drivers

lore
Share & tools
lore
#0 lore raw

cover bnxt patchset for thor and bnxt vector PMD

Fixes and enhancements for adapters based on the BCM57500 controller and the bnxt vector PMD.

Patch set is against dpdk-next-net.

**Note** this patch series is dependent on the "bnxt patchset to support device error recovery" series.

Lance Richardson (7):
net/bnxt: fix thor tqm entry allocation net/bnxt: fix ring alignment for thor-based adapters net/bnxt: use common receive/transmit NQ ring net/bnxt: use correct default Rx queue for thor net/bnxt: add support for LRO for thor adapters net/bnxt: fix scatter receive offload capability net/bnxt: improve CPR handling in vector PMD

 doc/guides/rel_notes/release_19_11.rst |   6 ++
 drivers/net/bnxt/bnxt.h                |  17 ++++
 drivers/net/bnxt/bnxt_ethdev.c         |  19 +++-
 drivers/net/bnxt/bnxt_hwrm.c           |  43 ++++++---
 drivers/net/bnxt/bnxt_hwrm.h           |   1 +
 drivers/net/bnxt/bnxt_ring.c           | 124 +++++++++++++++++--------
 drivers/net/bnxt/bnxt_ring.h           |   3 +-
 drivers/net/bnxt/bnxt_rxq.c            |   8 +-
 drivers/net/bnxt/bnxt_rxq.h            |   1 -
 drivers/net/bnxt/bnxt_rxr.c            |  99 +++++++++++---------
 drivers/net/bnxt/bnxt_rxr.h            |  41 ++++++--
 drivers/net/bnxt/bnxt_rxtx_vec_sse.c   |  26 +-----
 drivers/net/bnxt/bnxt_txq.c            |   4 +-
 drivers/net/bnxt/bnxt_txq.h            |   1 -
 drivers/net/bnxt/bnxt_txr.c            |  25 -----
 15 files changed, 256 insertions(+), 162 deletions(-)
Signature · 1 line
2.17.1
#1 lore raw

PATCH 1/7 net/bnxt: fix thor tqm entry allocation

The current TQM backing store size isn't sufficient to allow 512 transmit rings. Fix by correcting TQM SP queue size calculation.

Fixes f8168ca0e690 ("net/bnxt: support thor controller")
---
 drivers/net/bnxt/bnxt_ethdev.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/net/bnxt/bnxt_ethdev.c +3 -1
index b94c9a122..e4c7b7c2a 100644
@@ -4057,7 +4057,9 @@ int bnxt_alloc_ctx_mem(struct bnxt *bp)
 	if (rc)
 		return rc;
 
-	entries = ctx->qp_max_l2_entries;
+	entries = ctx->qp_max_l2_entries +
+		  ctx->vnic_max_vnic_entries +
+		  ctx->tqm_min_entries_per_ring;
 	entries = bnxt_roundup(entries, ctx->tqm_entries_multiple);
 	entries = clamp_t(uint32_t, entries, ctx->tqm_min_entries_per_ring,
 			  ctx->tqm_max_entries_per_ring);
Signature · 1 line
2.17.1
#2 lore raw

PATCH 2/7 net/bnxt: fix ring alignment for thor-based adapters

In reply to Lance Richardson

When using transmit/receive queue sizes smaller than 256, alignment requirements are not being met for Thor-based adapters. Fix by forcing memory addresses used for transmit/receive/aggregation ring allocations to be on 4K boundaries.

Fixes f8168ca0e690 ("net/bnxt: support thor controller")
---
 drivers/net/bnxt/bnxt_ring.c | 3 +++
 1 file changed, 3 insertions(+)
drivers/net/bnxt/bnxt_ring.c +3 -0
index ec17783cf..bc8b92b04 100644
@@ -162,18 +162,21 @@ int bnxt_alloc_rings(struct bnxt *bp, uint16_t qidx,
 	int nq_ring_len = BNXT_CHIP_THOR(bp) ? cp_ring_len : 0;
 
 	int tx_ring_start = nq_ring_start + nq_ring_len;
+	tx_ring_start = RTE_ALIGN(tx_ring_start, 4096);
 	int tx_ring_len = tx_ring_info ?
 	    RTE_CACHE_LINE_ROUNDUP(tx_ring_info->tx_ring_struct->ring_size *
 				   sizeof(struct tx_bd_long)) : 0;
 	tx_ring_len = RTE_ALIGN(tx_ring_len, 4096);
 
 	int rx_ring_start = tx_ring_start + tx_ring_len;
+	rx_ring_start = RTE_ALIGN(rx_ring_start, 4096);
 	int rx_ring_len =  rx_ring_info ?
 		RTE_CACHE_LINE_ROUNDUP(rx_ring_info->rx_ring_struct->ring_size *
 		sizeof(struct rx_prod_pkt_bd)) : 0;
 	rx_ring_len = RTE_ALIGN(rx_ring_len, 4096);
 
 	int ag_ring_start = rx_ring_start + rx_ring_len;
+	ag_ring_start = RTE_ALIGN(ag_ring_start, 4096);
 	int ag_ring_len = rx_ring_len * AGG_RING_SIZE_FACTOR;
 	ag_ring_len = RTE_ALIGN(ag_ring_len, 4096);
 
Signature · 1 line
2.17.1
#3 lore raw

PATCH 3/7 net/bnxt: use common receive/transmit NQ ring

In reply to Lance Richardson

Thor queue scaling is currently limited by the number of NQs that can be allocated. Fix by using a common NQ for all receive/transmit rings instead of allocating a separate NQ for each ring.

Fixes f8168ca0e690 ("net/bnxt: support thor controller")
---
 drivers/net/bnxt/bnxt.h        |   1 +
 drivers/net/bnxt/bnxt_ethdev.c |   5 ++
 drivers/net/bnxt/bnxt_hwrm.c   |   7 +--
 drivers/net/bnxt/bnxt_ring.c   | 107 ++++++++++++++++++++++-----------
 drivers/net/bnxt/bnxt_ring.h   |   2 +
 drivers/net/bnxt/bnxt_rxq.c    |   4 +-
 drivers/net/bnxt/bnxt_rxq.h    |   1 -
 drivers/net/bnxt/bnxt_rxr.c    |  27 ---------
 drivers/net/bnxt/bnxt_txq.c    |   4 +-
 drivers/net/bnxt/bnxt_txq.h    |   1 -
 drivers/net/bnxt/bnxt_txr.c    |  25 --------
 11 files changed, 84 insertions(+), 100 deletions(-)
drivers/net/bnxt/bnxt.h +1 -0
index ac602fe52..3e508ca1f 100644
@@ -459,6 +459,7 @@ struct bnxt {
 
 	/* Default completion ring */
 	struct bnxt_cp_ring_info	*async_cp_ring;
+	struct bnxt_cp_ring_info	*rxtx_nq_ring;
 	uint32_t		max_ring_grps;
 	struct bnxt_ring_grp_info	*grp_info;
 
drivers/net/bnxt/bnxt_ethdev.c +5 -0
index e4c7b7c2a..c2ab8df7b 100644
@@ -220,6 +220,7 @@ static void bnxt_free_mem(struct bnxt *bp, bool reconfig)
 		bnxt_free_rx_rings(bp);
 	}
 	bnxt_free_async_cp_ring(bp);
+	bnxt_free_rxtx_nq_ring(bp);
 }
 
 static int bnxt_alloc_mem(struct bnxt *bp, bool reconfig)
@@ -250,6 +251,10 @@ static int bnxt_alloc_mem(struct bnxt *bp, bool reconfig)
 	if (rc)
 		goto alloc_mem_err;
 
+	rc = bnxt_alloc_rxtx_nq_ring(bp);
+	if (rc)
+		goto alloc_mem_err;
+
 	return 0;
 
 alloc_mem_err:
drivers/net/bnxt/bnxt_hwrm.c +1 -6
index bd2cc01e1..4b1230453 100644
@@ -2175,11 +2175,8 @@ void bnxt_free_hwrm_rx_ring(struct bnxt *bp, int queue_index)
 			bp->grp_info[queue_index].ag_fw_ring_id =
 							INVALID_HW_RING_ID;
 	}
-	if (cpr->cp_ring_struct->fw_ring_id != INVALID_HW_RING_ID) {
+	if (cpr->cp_ring_struct->fw_ring_id != INVALID_HW_RING_ID)
 		bnxt_free_cp_ring(bp, cpr);
-		if (rxq->nq_ring)
-			bnxt_free_nq_ring(bp, rxq->nq_ring);
-	}
 
 	if (BNXT_HAS_RING_GRPS(bp))
 		bp->grp_info[queue_index].cp_fw_ring_id = INVALID_HW_RING_ID;
@@ -2211,8 +2208,6 @@ int bnxt_free_all_hwrm_rings(struct bnxt *bp)
 		if (cpr->cp_ring_struct->fw_ring_id != INVALID_HW_RING_ID) {
 			bnxt_free_cp_ring(bp, cpr);
 			cpr->cp_ring_struct->fw_ring_id = INVALID_HW_RING_ID;
-			if (txq->nq_ring)
-				bnxt_free_nq_ring(bp, txq->nq_ring);
 		}
 	}
 
drivers/net/bnxt/bnxt_ring.c +71 -36

178 lines not on this page · raw message

drivers/net/bnxt/bnxt_ring.h +2 -0
index a31d59ea3..200d85532 100644
@@ -79,6 +79,8 @@ int bnxt_alloc_hwrm_rings(struct bnxt *bp);
 int bnxt_alloc_async_cp_ring(struct bnxt *bp);
 void bnxt_free_async_cp_ring(struct bnxt *bp);
 int bnxt_alloc_async_ring_struct(struct bnxt *bp);
+int bnxt_alloc_rxtx_nq_ring(struct bnxt *bp);
+void bnxt_free_rxtx_nq_ring(struct bnxt *bp);
 
 static inline void bnxt_db_write(struct bnxt_db_info *db, uint32_t idx)
 {
drivers/net/bnxt/bnxt_rxq.c +2 -2
index d5fc5268d..eb45436c3 100644
@@ -347,8 +347,8 @@ int bnxt_rx_queue_setup_op(struct rte_eth_dev *eth_dev,
 
 	eth_dev->data->rx_queues[queue_idx] = rxq;
 	/* Allocate RX ring hardware descriptors */
-	if (bnxt_alloc_rings(bp, queue_idx, NULL, rxq, rxq->cp_ring,
-			rxq->nq_ring, "rxr")) {
+	if (bnxt_alloc_rings(bp, queue_idx, NULL, rxq, rxq->cp_ring, NULL,
+			     "rxr")) {
 		PMD_DRV_LOG(ERR,
 			"ring_dma_zone_reserve for rx_ring failed!\n");
 		bnxt_rx_queue_release_op(rxq);
drivers/net/bnxt/bnxt_rxq.h +0 -1
index b5e42d01c..7c6b4dec9 100644
@@ -39,7 +39,6 @@ struct bnxt_rx_queue {
 	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;
 	rte_atomic64_t		rx_mbuf_alloc_fail;
 	const struct rte_memzone *mz;
 };
drivers/net/bnxt/bnxt_rxr.c +0 -27
index 12313dd53..95750367d 100644
@@ -677,7 +677,6 @@ void bnxt_free_rx_rings(struct bnxt *bp)
 int bnxt_init_rx_ring_struct(struct bnxt_rx_queue *rxq, unsigned int socket_id)
 {
 	struct bnxt_cp_ring_info *cpr;
-	struct bnxt_cp_ring_info *nqr;
 	struct bnxt_rx_ring_info *rxr;
 	struct bnxt_ring *ring;
 
@@ -726,32 +725,6 @@ int bnxt_init_rx_ring_struct(struct bnxt_rx_queue *rxq, unsigned int socket_id)
 	ring->vmem_size = 0;
 	ring->vmem = NULL;
 
-	if (BNXT_HAS_NQ(rxq->bp)) {
-		nqr = rte_zmalloc_socket("bnxt_rx_ring_cq",
-					 sizeof(struct bnxt_cp_ring_info),
-					 RTE_CACHE_LINE_SIZE, socket_id);
-		if (nqr == NULL)
-			return -ENOMEM;
-
-		rxq->nq_ring = nqr;
-
-		ring = rte_zmalloc_socket("bnxt_rx_ring_struct",
-					  sizeof(struct bnxt_ring),
-					  RTE_CACHE_LINE_SIZE, socket_id);
-		if (ring == NULL)
-			return -ENOMEM;
-
-		nqr->cp_ring_struct = ring;
-		ring->ring_size =
-			rte_align32pow2(rxr->rx_ring_struct->ring_size *
-					(2 + AGG_RING_SIZE_FACTOR));
-		ring->ring_mask = ring->ring_size - 1;
-		ring->bd = (void *)nqr->cp_desc_ring;
-		ring->bd_dma = nqr->cp_desc_mapping;
-		ring->vmem_size = 0;
-		ring->vmem = NULL;
-	}
-
 	/* Allocate Aggregator rings */
 	ring = rte_zmalloc_socket("bnxt_rx_ring_struct",
 				   sizeof(struct bnxt_ring),
drivers/net/bnxt/bnxt_txq.c +2 -2
index 090132479..8f43c1755 100644
@@ -140,8 +140,8 @@ int bnxt_tx_queue_setup_op(struct rte_eth_dev *eth_dev,
 	txq->port_id = eth_dev->data->port_id;
 
 	/* Allocate TX ring hardware descriptors */
-	if (bnxt_alloc_rings(bp, queue_idx, txq, NULL, txq->cp_ring,
-			txq->nq_ring, "txr")) {
+	if (bnxt_alloc_rings(bp, queue_idx, txq, NULL, txq->cp_ring, NULL,
+			     "txr")) {
 		PMD_DRV_LOG(ERR, "ring_dma_zone_reserve for tx_ring failed!");
 		bnxt_tx_queue_release_op(txq);
 		rc = -ENOMEM;
drivers/net/bnxt/bnxt_txq.h +0 -1
index 9190e3f73..a0d4678d9 100644
@@ -32,7 +32,6 @@ struct bnxt_tx_queue {
 
 	unsigned int		cp_nr_rings;
 	struct bnxt_cp_ring_info	*cp_ring;
-	struct bnxt_cp_ring_info        *nq_ring;
 	const struct rte_memzone *mz;
 	struct rte_mbuf **free;
 };
drivers/net/bnxt/bnxt_txr.c +0 -25
index 35e7166be..b24dcfc2e 100644
@@ -57,7 +57,6 @@ int bnxt_init_one_tx_ring(struct bnxt_tx_queue *txq)
 int bnxt_init_tx_ring_struct(struct bnxt_tx_queue *txq, unsigned int socket_id)
 {
 	struct bnxt_cp_ring_info *cpr;
-	struct bnxt_cp_ring_info *nqr;
 	struct bnxt_tx_ring_info *txr;
 	struct bnxt_ring *ring;
 
@@ -101,30 +100,6 @@ int bnxt_init_tx_ring_struct(struct bnxt_tx_queue *txq, unsigned int socket_id)
 	ring->vmem_size = 0;
 	ring->vmem = NULL;
 
-	if (BNXT_HAS_NQ(txq->bp)) {
-		nqr = rte_zmalloc_socket("bnxt_tx_ring_nq",
-					 sizeof(struct bnxt_cp_ring_info),
-					 RTE_CACHE_LINE_SIZE, socket_id);
-		if (nqr == NULL)
-			return -ENOMEM;
-
-		txq->nq_ring = nqr;
-
-		ring = rte_zmalloc_socket("bnxt_tx_ring_struct",
-					  sizeof(struct bnxt_ring),
-					  RTE_CACHE_LINE_SIZE, socket_id);
-		if (ring == NULL)
-			return -ENOMEM;
-
-		nqr->cp_ring_struct = ring;
-		ring->ring_size = txr->tx_ring_struct->ring_size;
-		ring->ring_mask = ring->ring_size - 1;
-		ring->bd = (void *)nqr->cp_desc_ring;
-		ring->bd_dma = nqr->cp_desc_mapping;
-		ring->vmem_size = 0;
-		ring->vmem = NULL;
-	}
-
 	return 0;
 }
 
Signature · 1 line
2.17.1
#4 lore raw

PATCH 4/7 net/bnxt: use correct default Rx queue for thor

In reply to Lance Richardson

Use first receive queue assigned to VNIC as the default receive queue when configuring Thor VNICs. This is necessary e.g. in order for flow redirection to a specific receive queue to work correctly.

Fixes f8168ca0e690 ("net/bnxt: support thor controller")
---
 drivers/net/bnxt/bnxt_hwrm.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/net/bnxt/bnxt_hwrm.c +2 -1
index 4b1230453..4d8866ac1 100644
@@ -1580,7 +1580,8 @@ int bnxt_hwrm_vnic_cfg(struct bnxt *bp, struct bnxt_vnic_info *vnic)
 	HWRM_PREP(req, VNIC_CFG, BNXT_USE_CHIMP_MB);
 
 	if (BNXT_CHIP_THOR(bp)) {
-		struct bnxt_rx_queue *rxq = bp->eth_dev->data->rx_queues[0];
+		struct bnxt_rx_queue *rxq =
+			bp->eth_dev->data->rx_queues[vnic->start_grp_id];
 		struct bnxt_rx_ring_info *rxr = rxq->rx_ring;
 		struct bnxt_cp_ring_info *cpr = rxq->cp_ring;
 
Signature · 1 line
2.17.1
#5 lore raw

PATCH 5/7 net/bnxt: add support for LRO for thor adapters

In reply to Lance Richardson

Add support for LRO for adapters based on Thor (BCM57500).

---
 doc/guides/rel_notes/release_19_11.rst |  6 +++
 drivers/net/bnxt/bnxt.h                | 16 ++++++
 drivers/net/bnxt/bnxt_ethdev.c         |  4 ++
 drivers/net/bnxt/bnxt_hwrm.c           | 33 ++++++++++--
 drivers/net/bnxt/bnxt_hwrm.h           |  1 +
 drivers/net/bnxt/bnxt_ring.c           | 14 +++--
 drivers/net/bnxt/bnxt_ring.h           |  1 -
 drivers/net/bnxt/bnxt_rxq.c            |  4 +-
 drivers/net/bnxt/bnxt_rxr.c            | 72 +++++++++++++++++++-------
 drivers/net/bnxt/bnxt_rxr.h            | 41 ++++++++++++---
 10 files changed, 155 insertions(+), 37 deletions(-)
doc/guides/rel_notes/release_19_11.rst +6 -0
index 27cfbd9e3..a044db46f 100644
@@ -56,6 +56,12 @@ New Features
      Also, make sure to start the actual text at the margin.
      =========================================================
 
+* **Updated the Broadcom bnxt PMD.**
+
+  Updated the Broadcom bnxt PMD. The major enhancements include:
+
+  * Added LRO support for BCM57500 adapters.
+
 
 Removed Items
 -------------
drivers/net/bnxt/bnxt.h +16 -0
index 3e508ca1f..6da5126a8 100644
@@ -34,6 +34,21 @@
 #define BNXT_MAX_RX_RING_DESC	8192
 #define BNXT_DB_SIZE		0x80
 
+#define TPA_MAX_AGGS		64
+#define TPA_MAX_AGGS_TH		1024
+
+#define TPA_MAX_NUM_SEGS	32
+#define TPA_MAX_SEGS_TH		8 /* 32 segments in 4-segment units */
+#define TPA_MAX_SEGS		5 /* 32 segments in log2 units */
+
+#define BNXT_TPA_MAX_AGGS(bp) \
+	(BNXT_CHIP_THOR(bp) ? TPA_MAX_AGGS_TH : \
+			     TPA_MAX_AGGS)
+
+#define BNXT_TPA_MAX_SEGS(bp) \
+	(BNXT_CHIP_THOR(bp) ? TPA_MAX_SEGS_TH : \
+			      TPA_MAX_SEGS)
+
 #ifdef RTE_ARCH_ARM64
 #define BNXT_NUM_ASYNC_CPR(bp) (BNXT_STINGRAY(bp) ? 0 : 1)
 #else
@@ -506,6 +521,7 @@ struct bnxt {
 	uint16_t		max_rx_em_flows;
 	uint16_t		max_vnics;
 	uint16_t		max_stat_ctx;
+	uint16_t		max_tpa_v2;
 	uint16_t		first_vf_id;
 	uint16_t		vlan;
 	struct bnxt_pf_info	pf;
drivers/net/bnxt/bnxt_ethdev.c +4 -0
index c2ab8df7b..227960d4e 100644
@@ -4303,6 +4303,10 @@ static int bnxt_init_fw(struct bnxt *bp)
 	if (rc)
 		return rc;
 
+	rc = bnxt_hwrm_vnic_qcaps(bp);
+	if (rc)
+		return rc;
+
 	rc = bnxt_hwrm_func_qcfg(bp, &mtu);
 	if (rc)
 		return rc;
drivers/net/bnxt/bnxt_hwrm.c +28 -5
index 4d8866ac1..404e52491 100644
@@ -656,6 +656,27 @@ int bnxt_hwrm_func_qcaps(struct bnxt *bp)
 	return rc;
 }
 
+int bnxt_hwrm_vnic_qcaps(struct bnxt *bp)
+{
+	int rc = 0;
+	struct hwrm_vnic_qcaps_input req = {.req_type = 0 };
+	struct hwrm_vnic_qcaps_output *resp = bp->hwrm_cmd_resp_addr;
+
+	HWRM_PREP(req, VNIC_QCAPS, BNXT_USE_CHIMP_MB);
+
+	req.target_id = rte_cpu_to_le_16(0xffff);
+
+	rc = bnxt_hwrm_send_message(bp, &req, sizeof(req), BNXT_USE_CHIMP_MB);
+
+	HWRM_CHECK_RESULT();
+
+	bp->max_tpa_v2 = rte_le_to_cpu_16(resp->max_aggs_supported);
+
+	HWRM_UNLOCK();
+
+	return rc;
+}
+
 int bnxt_hwrm_func_reset(struct bnxt *bp)
 {
 	int rc = 0;
@@ -1878,8 +1899,11 @@ int bnxt_hwrm_vnic_tpa_cfg(struct bnxt *bp,
 	struct hwrm_vnic_tpa_cfg_input req = {.req_type = 0 };
 	struct hwrm_vnic_tpa_cfg_output *resp = bp->hwrm_cmd_resp_addr;
 
-	if (BNXT_CHIP_THOR(bp))
-		return 0;
+	if (BNXT_CHIP_THOR(bp) && !bp->max_tpa_v2) {
+		if (enable)
+			PMD_DRV_LOG(ERR, "No HW support for LRO\n");
+		return -ENOTSUP;
+	}
 
 	HWRM_PREP(req, VNIC_TPA_CFG, BNXT_USE_CHIMP_MB);
 
@@ -1895,9 +1919,8 @@ int bnxt_hwrm_vnic_tpa_cfg(struct bnxt *bp,
 				HWRM_VNIC_TPA_CFG_INPUT_FLAGS_GRO |
 				HWRM_VNIC_TPA_CFG_INPUT_FLAGS_AGG_WITH_ECN |
 			HWRM_VNIC_TPA_CFG_INPUT_FLAGS_AGG_WITH_SAME_GRE_SEQ);
-		req.max_agg_segs = rte_cpu_to_le_16(5);
-		req.max_aggs =
-			rte_cpu_to_le_16(HWRM_VNIC_TPA_CFG_INPUT_MAX_AGGS_MAX);
+		req.max_agg_segs = rte_cpu_to_le_16(BNXT_TPA_MAX_AGGS(bp));
+		req.max_aggs = rte_cpu_to_le_16(BNXT_TPA_MAX_SEGS(bp));
 		req.min_agg_len = rte_cpu_to_le_32(512);
 	}
 	req.vnic_id = rte_cpu_to_le_16(vnic->fw_vnic_id);
drivers/net/bnxt/bnxt_hwrm.h +1 -0
index db25ad591..a26698ce3 100644
@@ -105,6 +105,7 @@ int bnxt_hwrm_vnic_alloc(struct bnxt *bp, struct bnxt_vnic_info *vnic);
 int bnxt_hwrm_vnic_cfg(struct bnxt *bp, struct bnxt_vnic_info *vnic);
 int bnxt_hwrm_vnic_qcfg(struct bnxt *bp, struct bnxt_vnic_info *vnic,
 				int16_t fw_vf_id);
+int bnxt_hwrm_vnic_qcaps(struct bnxt *bp);
 int bnxt_hwrm_vnic_ctx_alloc(struct bnxt *bp, struct bnxt_vnic_info *vnic,
 			     uint16_t ctx_idx);
 int bnxt_hwrm_vnic_ctx_free(struct bnxt *bp, struct bnxt_vnic_info *vnic,
drivers/net/bnxt/bnxt_ring.c +9 -5
index 85a10c584..77a3c8ba1 100644
@@ -187,13 +187,17 @@ int bnxt_alloc_rings(struct bnxt *bp, uint16_t qidx,
 			AGG_RING_SIZE_FACTOR)) : 0;
 
 	int tpa_info_start = ag_bitmap_start + ag_bitmap_len;
-	int tpa_info_len = rx_ring_info ?
-		RTE_CACHE_LINE_ROUNDUP(BNXT_TPA_MAX *
-				       sizeof(struct bnxt_tpa_info)) : 0;
+	int tpa_info_len = 0;
+
+	if (rx_ring_info && (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)) {
+		int tpa_max = BNXT_TPA_MAX_AGGS(bp);
+
+		tpa_info_len = tpa_max * sizeof(struct bnxt_tpa_info);
+		tpa_info_len = RTE_CACHE_LINE_ROUNDUP(tpa_info_len);
+	}
 
 	int total_alloc_len = tpa_info_start;
-	if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
-		total_alloc_len += tpa_info_len;
+	total_alloc_len += tpa_info_len;
 
 	snprintf(mz_name, RTE_MEMZONE_NAMESIZE,
 		 "bnxt_%04x:%02x:%02x:%02x-%04x_%s", pdev->addr.domain,
drivers/net/bnxt/bnxt_ring.h +0 -1
index 200d85532..833118391 100644
@@ -27,7 +27,6 @@
 #define DEFAULT_RX_RING_SIZE	256
 #define DEFAULT_TX_RING_SIZE	256
 
-#define BNXT_TPA_MAX		64
 #define AGG_RING_SIZE_FACTOR	2
 #define AGG_RING_MULTIPLIER	2
 
drivers/net/bnxt/bnxt_rxq.c +3 -1
index eb45436c3..d2f899106 100644
@@ -235,7 +235,9 @@ void bnxt_rx_queue_release_mbufs(struct bnxt_rx_queue *rxq)
 		/* Free up mbufs in TPA */
 		tpa_info = rxq->rx_ring->tpa_info;
 		if (tpa_info) {
-			for (i = 0; i < BNXT_TPA_MAX; i++) {
+			int max_aggs = BNXT_TPA_MAX_AGGS(rxq->bp);
+
+			for (i = 0; i < max_aggs; i++) {
 				if (tpa_info[i].mbuf) {
 					rte_pktmbuf_free_seg(tpa_info[i].mbuf);
 					tpa_info[i].mbuf = NULL;
drivers/net/bnxt/bnxt_rxr.c +54 -18

149 lines not on this page · raw message

drivers/net/bnxt/bnxt_rxr.h +34 -7

60 lines not on this page · raw message

Signature · 1 line
2.17.1
#6 lore raw

PATCH 6/7 net/bnxt: fix scatter receive offload capability

In reply to Lance Richardson

Scattered receive is supported but not included in receive offload capabilities. Fix by adding it and including in scattered receive calculation.

Fixes 9c1507d96ab8 ("net/bnxt: switch to the new offload API")
---
 drivers/net/bnxt/bnxt_ethdev.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
drivers/net/bnxt/bnxt_ethdev.c +5 -1
index 227960d4e..7e756ea45 100644
@@ -161,7 +161,8 @@ static const struct rte_pci_id bnxt_pci_id_map[] = {
 				     DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM | \
 				     DEV_RX_OFFLOAD_JUMBO_FRAME | \
 				     DEV_RX_OFFLOAD_KEEP_CRC | \
-				     DEV_RX_OFFLOAD_TCP_LRO)
+				     DEV_RX_OFFLOAD_TCP_LRO | \
+				     DEV_RX_OFFLOAD_SCATTER)
 
 static int bnxt_vlan_offload_set_op(struct rte_eth_dev *dev, int mask);
 static void bnxt_print_link_info(struct rte_eth_dev *eth_dev);
@@ -725,6 +726,9 @@ static int bnxt_scattered_rx(struct rte_eth_dev *eth_dev)
 	uint16_t buf_size;
 	int i;
 
+	if (eth_dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_SCATTER)
+		return 1;
+
 	for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 		struct bnxt_rx_queue *rxq = eth_dev->data->rx_queues[i];
 
Signature · 1 line
2.17.1
#7 lore raw

PATCH 7/7 net/bnxt: improve CPR handling in vector PMD

In reply to Lance Richardson

Reduce overhead of CPR descriptor validity checking in vector receive and transmit functions.

Preserve raw cpr consumer index in vector transmit completion function.

Remove an unneeded prefetch (per benchmarking) from vector transmit completion function.

Fixes bc4a000f2f53 ("net/bnxt: implement SSE vector mode")
---
 drivers/net/bnxt/bnxt_rxtx_vec_sse.c | 26 ++++----------------------
 1 file changed, 4 insertions(+), 22 deletions(-)
drivers/net/bnxt/bnxt_rxtx_vec_sse.c +4 -22
index 2e6e83c94..980fddb1f 100644
@@ -245,10 +245,6 @@ bnxt_recv_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts,
 		if (!CMP_VALID(rxcmp, raw_cons, cpr->cp_ring_struct))
 			break;
 
-		cpr->valid = FLIP_VALID(cons,
-					cpr->cp_ring_struct->ring_mask,
-					cpr->valid);
-
 		if (likely(CMP_TYPE(rxcmp) == RX_PKT_CMPL_TYPE_RX_L2)) {
 			struct rx_pkt_cmpl_hi *rxcmp1;
 			uint32_t tmp_raw_cons;
@@ -272,10 +268,6 @@ bnxt_recv_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts,
 			rte_prefetch0(mbuf);
 			rxr->rx_buf_ring[cons].mbuf = NULL;
 
-			cpr->valid = FLIP_VALID(cp_cons,
-						cpr->cp_ring_struct->ring_mask,
-						cpr->valid);
-
 			/* Set constant fields from mbuf initializer. */
 			_mm_store_si128((__m128i *)&mbuf->rearm_data,
 					mbuf_init);
@@ -318,22 +310,13 @@ bnxt_recv_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts,
 
 	rxq->rxrearm_nb += nb_rx_pkts;
 	cpr->cp_raw_cons = raw_cons;
+	cpr->valid = !!(cpr->cp_raw_cons & cpr->cp_ring_struct->ring_size);
 	if (nb_rx_pkts || evt)
 		bnxt_db_cq(cpr);
 
 	return nb_rx_pkts;
 }
 
-static inline void bnxt_next_cmpl(struct bnxt_cp_ring_info *cpr, uint32_t *idx,
-				  bool *v, uint32_t inc)
-{
-	*idx += inc;
-	if (unlikely(*idx == cpr->cp_ring_struct->ring_size)) {
-		*v = !*v;
-		*idx = 0;
-	}
-}
-
 static void
 bnxt_tx_cmp_vec(struct bnxt_tx_queue *txq, int nr_pkts)
 {
@@ -379,10 +362,8 @@ bnxt_handle_tx_cp_vec(struct bnxt_tx_queue *txq)
 		cons = RING_CMPL(ring_mask, raw_cons);
 		txcmp = (struct tx_cmpl *)&cp_desc_ring[cons];
 
-		if (!CMPL_VALID(txcmp, cpr->valid))
+		if (!CMP_VALID(txcmp, raw_cons, cp_ring_struct))
 			break;
-		bnxt_next_cmpl(cpr, &cons, &cpr->valid, 1);
-		rte_prefetch0(&cp_desc_ring[cons]);
 
 		if (likely(CMP_TYPE(txcmp) == TX_CMPL_TYPE_TX_L2))
 			nb_tx_pkts += txcmp->opaque;
@@ -390,9 +371,10 @@ bnxt_handle_tx_cp_vec(struct bnxt_tx_queue *txq)
 			RTE_LOG_DP(ERR, PMD,
 				   "Unhandled CMP type %02x\n",
 				   CMP_TYPE(txcmp));
-		raw_cons = cons;
+		raw_cons = NEXT_RAW_CMP(raw_cons);
 	} while (nb_tx_pkts < ring_mask);
 
+	cpr->valid = !!(raw_cons & cp_ring_struct->ring_size);
 	if (nb_tx_pkts) {
 		bnxt_tx_cmp_vec(txq, nb_tx_pkts);
 		cpr->cp_raw_cons = raw_cons;
Signature · 1 line
2.17.1