[dpdk-stable] patch 'net/bnxt: fix Tx batching' has been queued to LTS release 18.11.3

[Kevin Traynor <ktraynor@redhat.com>]

Hi,

FYI, your patch has been queued to LTS release 18.11.3

Note it hasn't been pushed to http://dpdk.org/browse/dpdk-stable yet.
It will be pushed if I get no objections before 06/26/19. So please
shout if anyone has objections.

Also note that after the patch there's a diff of the upstream commit vs the
patch applied to the branch. This will indicate if there was any rebasing
needed to apply to the stable branch. If there were code changes for rebasing
(ie: not only metadata diffs), please double check that the rebase was
correctly done.

Queued patches are on a temporary branch at:
https://github.com/kevintraynor/dpdk-stable-queue

This queued commit can be viewed at:
https://github.com/kevintraynor/dpdk-stable-queue/commit/274bac969190de1b2922af85682ab64c92d2ec6e

Thanks.

Kevin Traynor

---
From 274bac969190de1b2922af85682ab64c92d2ec6e Mon Sep 17 00:00:00 2001
From: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Date: Wed, 15 May 2019 11:08:15 -0700
Subject: [PATCH] net/bnxt: fix Tx batching

[ upstream commit adcee0fc845f653ca288807467f29c7c7c3e532a ]

This patch addresses the following issues with Tx batching:

1. Tx stall observed in some conditions:

The batching code doesn't request for a completion when only a partial
chain of packets is transmitted due to mbuf allocation errors. Because
of this, Tx consumer index is not updated correctly and it eventually
leads to qfull condition. Fix this by requesting a completion for the
last packet in the partial chain that is transmitted successfully.

2. Tx stall seen with Jumbo frames:

With jumbo frames, number of TxBDs is > 1. While setting up these
additional BDs in bnxt_start_xmit(), the flags field is being set using
the OR-assignment operator. We end up using a stale value of the flags
field (from a previous use of that descriptor). This results in an
invalid completion and eventually leads to tx stall. Fix this to just
assign the flags field with the right value.

Fixes: 5735eb241947 ("net/bnxt: support Tx batching")

Signed-off-by: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
---
 drivers/net/bnxt/bnxt_txq.h |  1 -
 drivers/net/bnxt/bnxt_txr.c | 57 ++++++++++++++++++++++---------------
 2 files changed, 34 insertions(+), 24 deletions(-)

diff --git a/drivers/net/bnxt/bnxt_txq.h b/drivers/net/bnxt/bnxt_txq.h
index f2c712a75..720ca90cf 100644
--- a/drivers/net/bnxt/bnxt_txq.h
+++ b/drivers/net/bnxt/bnxt_txq.h
@@ -25,5 +25,4 @@ struct bnxt_tx_queue {
 	uint32_t		ctx_curr; /* Hardware context states */
 	uint8_t			tx_deferred_start; /* not in global dev start */
-	uint8_t			cmpl_next; /* Next BD to trigger a compl */
 
 	struct bnxt		*bp;
diff --git a/drivers/net/bnxt/bnxt_txr.c b/drivers/net/bnxt/bnxt_txr.c
index da5742ccc..52a926c02 100644
--- a/drivers/net/bnxt/bnxt_txr.c
+++ b/drivers/net/bnxt/bnxt_txr.c
@@ -104,12 +104,17 @@ int bnxt_init_tx_ring_struct(struct bnxt_tx_queue *txq, unsigned int socket_id)
 }
 
-static inline uint32_t bnxt_tx_avail(struct bnxt_tx_ring_info *txr)
+static inline uint32_t bnxt_tx_bds_in_hw(struct bnxt_tx_queue *txq)
+{
+	return ((txq->tx_ring->tx_prod - txq->tx_ring->tx_cons) &
+		txq->tx_ring->tx_ring_struct->ring_mask);
+}
+
+static inline uint32_t bnxt_tx_avail(struct bnxt_tx_queue *txq)
 {
 	/* Tell compiler to fetch tx indices from memory. */
 	rte_compiler_barrier();
 
-	return txr->tx_ring_struct->ring_size -
-		((txr->tx_prod - txr->tx_cons) &
-			txr->tx_ring_struct->ring_mask) - 1;
+	return ((txq->tx_ring->tx_ring_struct->ring_size -
+		 bnxt_tx_bds_in_hw(txq)) - 1);
 }
 
@@ -117,5 +122,6 @@ static uint16_t bnxt_start_xmit(struct rte_mbuf *tx_pkt,
 				struct bnxt_tx_queue *txq,
 				uint16_t *coal_pkts,
-				uint16_t *cmpl_next)
+				uint16_t *cmpl_next,
+				struct tx_bd_long **last_txbd)
 {
 	struct bnxt_tx_ring_info *txr = txq->tx_ring;
@@ -124,4 +130,5 @@ static uint16_t bnxt_start_xmit(struct rte_mbuf *tx_pkt,
 	uint32_t vlan_tag_flags, cfa_action;
 	bool long_bd = false;
+	unsigned short nr_bds = 0;
 	struct rte_mbuf *m_seg;
 	struct bnxt_sw_tx_bd *tx_buf;
@@ -140,13 +147,12 @@ static uint16_t bnxt_start_xmit(struct rte_mbuf *tx_pkt,
 		long_bd = true;
 
-	tx_buf = &txr->tx_buf_ring[txr->tx_prod];
-	tx_buf->mbuf = tx_pkt;
-	tx_buf->nr_bds = long_bd + tx_pkt->nb_segs;
+	nr_bds = long_bd + tx_pkt->nb_segs;
+	if (unlikely(bnxt_tx_avail(txq) < nr_bds))
+		return -ENOMEM;
 
 	/* Check if number of Tx descriptors is above HW limit */
-	if (unlikely(tx_buf->nr_bds > BNXT_MAX_TSO_SEGS)) {
+	if (unlikely(nr_bds > BNXT_MAX_TSO_SEGS)) {
 		PMD_DRV_LOG(ERR,
-			    "Num descriptors %d exceeds HW limit\n",
-			    tx_buf->nr_bds);
+			    "Num descriptors %d exceeds HW limit\n", nr_bds);
 		return -ENOSPC;
 	}
@@ -171,10 +177,11 @@ static uint16_t bnxt_start_xmit(struct rte_mbuf *tx_pkt,
 	RTE_VERIFY(tx_pkt->data_len);
 
-	if (unlikely(bnxt_tx_avail(txr) < tx_buf->nr_bds))
-		return -ENOMEM;
+	tx_buf = &txr->tx_buf_ring[txr->tx_prod];
+	tx_buf->mbuf = tx_pkt;
+	tx_buf->nr_bds = nr_bds;
 
 	txbd = &txr->tx_desc_ring[txr->tx_prod];
 	txbd->opaque = *coal_pkts;
-	txbd->flags_type = tx_buf->nr_bds << TX_BD_LONG_FLAGS_BD_CNT_SFT;
+	txbd->flags_type = nr_bds << TX_BD_LONG_FLAGS_BD_CNT_SFT;
 	txbd->flags_type |= TX_BD_SHORT_FLAGS_COAL_NOW;
 	if (!*cmpl_next) {
@@ -190,4 +197,5 @@ static uint16_t bnxt_start_xmit(struct rte_mbuf *tx_pkt,
 		txbd->flags_type |= lhint_arr[tx_pkt->pkt_len >> 9];
 	txbd->address = rte_cpu_to_le_64(rte_mbuf_data_iova(tx_buf->mbuf));
+	*last_txbd = txbd;
 
 	if (long_bd) {
@@ -321,5 +329,5 @@ static uint16_t bnxt_start_xmit(struct rte_mbuf *tx_pkt,
 		txbd = &txr->tx_desc_ring[txr->tx_prod];
 		txbd->address = rte_cpu_to_le_64(rte_mbuf_data_iova(m_seg));
-		txbd->flags_type |= TX_BD_SHORT_TYPE_TX_BD_SHORT;
+		txbd->flags_type = TX_BD_SHORT_TYPE_TX_BD_SHORT;
 		txbd->len = m_seg->data_len;
 
@@ -371,6 +379,5 @@ static int bnxt_handle_tx_cp(struct bnxt_tx_queue *txq)
 	uint32_t opaque = 0;
 
-	if (((txq->tx_ring->tx_prod - txq->tx_ring->tx_cons) &
-		txq->tx_ring->tx_ring_struct->ring_mask) < txq->tx_free_thresh)
+	if (bnxt_tx_bds_in_hw(txq) < txq->tx_free_thresh)
 		return 0;
 
@@ -411,5 +418,6 @@ uint16_t bnxt_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
 	uint16_t nb_tx_pkts = 0;
 	uint16_t coal_pkts = 0;
-	uint16_t cmpl_next = txq->cmpl_next;
+	uint16_t cmpl_next = 0;
+	struct tx_bd_long *last_txbd = NULL;
 
 	/* Handle TX completions */
@@ -422,18 +430,21 @@ uint16_t bnxt_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
 	}
 
-	txq->cmpl_next = 0;
 	/* Handle TX burst request */
 	for (nb_tx_pkts = 0; nb_tx_pkts < nb_pkts; nb_tx_pkts++) {
 		int rc;
 
-		/* Request a completion on first and last packet */
+		/* Request a completion on the last packet */
 		cmpl_next |= (nb_pkts == nb_tx_pkts + 1);
 		coal_pkts++;
 		rc = bnxt_start_xmit(tx_pkts[nb_tx_pkts], txq,
-				&coal_pkts, &cmpl_next);
+				     &coal_pkts, &cmpl_next, &last_txbd);
 
 		if (unlikely(rc)) {
-			/* Request a completion in next cycle */
-			txq->cmpl_next = 1;
+			/* Request a completion on the last successfully
+			 * enqueued packet
+			 */
+			if (last_txbd)
+				last_txbd->flags_type &=
+					~TX_BD_LONG_FLAGS_NO_CMPL;
 			break;
 		}
-- 
2.20.1

---
  Diff of the applied patch vs upstream commit (please double-check if non-empty:
---
--- -	2019-06-21 17:22:12.377433281 +0100
+++ 0012-net-bnxt-fix-Tx-batching.patch	2019-06-21 17:22:11.720519194 +0100
@@ -1 +1 @@
-From adcee0fc845f653ca288807467f29c7c7c3e532a Mon Sep 17 00:00:00 2001
+From 274bac969190de1b2922af85682ab64c92d2ec6e Mon Sep 17 00:00:00 2001
@@ -5,0 +6,2 @@
+[ upstream commit adcee0fc845f653ca288807467f29c7c7c3e532a ]
+
@@ -26 +27,0 @@
-Cc: stable@dpdk.org
@@ -46 +47 @@
-index 9684fb177..186934136 100644
+index da5742ccc..52a926c02 100644
@@ -125 +126 @@
-@@ -322,5 +330,5 @@ static uint16_t bnxt_start_xmit(struct rte_mbuf *tx_pkt,
+@@ -321,5 +329,5 @@ static uint16_t bnxt_start_xmit(struct rte_mbuf *tx_pkt,
@@ -132 +133 @@
-@@ -372,6 +380,5 @@ static int bnxt_handle_tx_cp(struct bnxt_tx_queue *txq)
+@@ -371,6 +379,5 @@ static int bnxt_handle_tx_cp(struct bnxt_tx_queue *txq)
@@ -140 +141 @@
-@@ -412,5 +419,6 @@ uint16_t bnxt_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
+@@ -411,5 +418,6 @@ uint16_t bnxt_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
@@ -148 +149 @@
-@@ -423,18 +431,21 @@ uint16_t bnxt_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
+@@ -422,18 +430,21 @@ uint16_t bnxt_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,