[dpdk-dev] [PATCH v1 3/7] net/mlx4: save lkey in big-endian format

[Ophir Munk]

mlx4 NIC is using lkey in big endian format. Save lkey in this format
in order to avoid conversions during Tx fast path

Signed-off-by: Ophir Munk <ophirmu at mellanox.com>
---
 drivers/net/mlx4/mlx4_rxtx.c | 14 ++++++--------
 drivers/net/mlx4/mlx4_rxtx.h | 10 ++++++----
 2 files changed, 12 insertions(+), 12 deletions(-)

diff --git a/drivers/net/mlx4/mlx4_rxtx.c b/drivers/net/mlx4/mlx4_rxtx.c
index ae37f9b..4e06cca 100644
--- a/drivers/net/mlx4/mlx4_rxtx.c
+++ b/drivers/net/mlx4/mlx4_rxtx.c
@@ -260,10 +260,10 @@ uint32_t mlx4_txq_add_mr(struct txq *txq, struct rte_mempool *mp, uint32_t i)
 	/* Store the new entry. */
 	txq->mp2mr[i].mp = mp;
 	txq->mp2mr[i].mr = mr;
-	txq->mp2mr[i].lkey = mr->lkey;
+	txq->mp2mr[i].be_lkey = rte_cpu_to_be_32(mr->lkey);
 	DEBUG("%p: new MR lkey for MP \"%s\" (%p): 0x%08" PRIu32,
-	      (void *)txq, mp->name, (void *)mp, txq->mp2mr[i].lkey);
-	return txq->mp2mr[i].lkey;
+	      (void *)txq, mp->name, (void *)mp, mr->lkey);
+	return txq->mp2mr[i].be_lkey;
 }
 
 /**
@@ -289,7 +289,6 @@ mlx4_post_send(struct txq *txq, struct rte_mbuf *pkt)
 		uint16_t flags16[2];
 	} srcrb;
 	uint32_t head_idx = sq->head & sq->txbb_cnt_mask;
-	uint32_t lkey;
 	uintptr_t addr;
 	uint32_t owner_opcode = MLX4_OPCODE_SEND;
 	uint32_t byte_count;
@@ -323,10 +322,10 @@ mlx4_post_send(struct txq *txq, struct rte_mbuf *pkt)
 		if (unlikely(dseg >= (struct mlx4_wqe_data_seg *)sq->eob))
 			dseg = (struct mlx4_wqe_data_seg *)sq->buf;
 		dseg->addr = rte_cpu_to_be_64(addr);
-		/* Memory region key for this memory pool. */
-		lkey = mlx4_txq_mp2mr(txq, mlx4_txq_mb2mp(buf));
+		/* Memory region key (big endian) for this memory pool. */
+		dseg->lkey = mlx4_txq_mp2mr(txq, mlx4_txq_mb2mp(buf));
 #ifndef NDEBUG
-		if (unlikely(lkey == (uint32_t)-1)) {
+		if (unlikely(dseg->lkey == rte_cpu_to_be_32((uint32_t)-1))) {
 			/* MR does not exist. */
 			DEBUG("%p: unable to get MP <-> MR association",
 			      (void *)txq);
@@ -341,7 +340,6 @@ mlx4_post_send(struct txq *txq, struct rte_mbuf *pkt)
 			return -EFAULT;
 		}
 #endif /* NDEBUG */
-		dseg->lkey = rte_cpu_to_be_32(lkey);
 		if (likely(buf->data_len)) {
 			byte_count = rte_cpu_to_be_32(buf->data_len);
 		} else {
diff --git a/drivers/net/mlx4/mlx4_rxtx.h b/drivers/net/mlx4/mlx4_rxtx.h
index 719ef45..8d39afb 100644
--- a/drivers/net/mlx4/mlx4_rxtx.h
+++ b/drivers/net/mlx4/mlx4_rxtx.h
@@ -135,7 +135,7 @@ struct txq {
 	struct {
 		const struct rte_mempool *mp; /**< Cached memory pool. */
 		struct ibv_mr *mr; /**< Memory region (for mp). */
-		uint32_t lkey; /**< mr->lkey copy. */
+		uint32_t be_lkey; /**< mr->lkey copy. (big endian) */
 	} mp2mr[MLX4_PMD_TX_MP_CACHE]; /**< MP to MR translation table. */
 	struct priv *priv; /**< Back pointer to private data. */
 	unsigned int socket; /**< CPU socket ID for allocations. */
@@ -220,9 +220,11 @@ mlx4_txq_mp2mr(struct txq *txq, struct rte_mempool *mp)
 			break;
 		}
 		if (txq->mp2mr[i].mp == mp) {
-			assert(txq->mp2mr[i].lkey != (uint32_t)-1);
-			assert(txq->mp2mr[i].mr->lkey == txq->mp2mr[i].lkey);
-			return txq->mp2mr[i].lkey;
+			assert(txq->mp2mr[i].be_lkey !=
+				rte_cpu_to_be_32((uint32_t)-1));
+			assert(txq->mp2mr[i].mr->lkey ==
+				rte_be_to_cpu_32(txq->mp2mr[i].be_lkey));
+			return txq->mp2mr[i].be_lkey;
 		}
 	}
 	return mlx4_txq_add_mr(txq, mp, i);
-- 
2.7.4