[dpdk-dev] [PATCH v2 0/5] net/mlx5: support new completion field for Multi-Packet RQ

[dpdk-dev] [PATCH v2 0/5] net/mlx5: support new completion field for Multi-Packet RQ

[Yongseok Koh]

For the RFC,
    https://mails.dpdk.org/archives/dev/2018-June/103664.html

Yongseok Koh (5):
  net/mlx5: change return value of Rx completion poll
  net/mlx5: add new fields in Rx completion entry
  net/mlx5: add warning message for Multi-Packet RQ
  net/mlx5: use stride index in Rx completion entry
  net/mlx5: increase number of strides

 doc/guides/nics/mlx5.rst     |  4 +++
 drivers/net/mlx5/Makefile    |  2 +-
 drivers/net/mlx5/mlx5.c      |  6 +++--
 drivers/net/mlx5/mlx5_defs.h |  2 +-
 drivers/net/mlx5/mlx5_prm.h  |  9 +++++--
 drivers/net/mlx5/mlx5_rxq.c  |  8 +++++-
 drivers/net/mlx5/mlx5_rxtx.c | 62 +++++++++++++++++++++++++++-----------------
 drivers/net/mlx5/mlx5_rxtx.h |  4 +--
 8 files changed, 64 insertions(+), 33 deletions(-)

-- 
2.11.0

[dpdk-dev] [PATCH v2 1/5] net/mlx5: change return value of Rx completion poll

[Yongseok Koh]

mlx5_rx_poll_len() returns Rx hash result extracted from either mini CQE or
regular CQE. As mini CQE may not have the hash result if configured
otherwise, it shouldn't assume the first DWORD of mini CQE is always hash
result. mlx5_rx_poll_len() is changed to return pointer to the mini CQE if
compressed.

Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
Acked-by: Shahaf Shuler <shahafs@mellanox.com>
---
 drivers/net/mlx5/mlx5_rxtx.c | 33 +++++++++++++++++++++------------
 1 file changed, 21 insertions(+), 12 deletions(-)

diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c
index 527859461..a4e15c519 100644
--- a/drivers/net/mlx5/mlx5_rxtx.c
+++ b/drivers/net/mlx5/mlx5_rxtx.c
@@ -38,7 +38,7 @@ rxq_cq_to_pkt_type(struct mlx5_rxq_data *rxq, volatile struct mlx5_cqe *cqe);
 
 static __rte_always_inline int
 mlx5_rx_poll_len(struct mlx5_rxq_data *rxq, volatile struct mlx5_cqe *cqe,
-		 uint16_t cqe_cnt, uint32_t *rss_hash);
+		 uint16_t cqe_cnt, volatile struct mlx5_mini_cqe8 **mcqe);
 
 static __rte_always_inline uint32_t
 rxq_cq_to_ol_flags(volatile struct mlx5_cqe *cqe);
@@ -1722,8 +1722,9 @@ rxq_cq_to_pkt_type(struct mlx5_rxq_data *rxq, volatile struct mlx5_cqe *cqe)
  *   Pointer to RX queue.
  * @param cqe
  *   CQE to process.
- * @param[out] rss_hash
- *   Packet RSS Hash result.
+ * @param[out] mcqe
+ *   Store pointer to mini-CQE if compressed. Otherwise, the pointer is not
+ *   written.
  *
  * @return
  *   Packet size in bytes (0 if there is none), -1 in case of completion
@@ -1731,7 +1732,7 @@ rxq_cq_to_pkt_type(struct mlx5_rxq_data *rxq, volatile struct mlx5_cqe *cqe)
  */
 static inline int
 mlx5_rx_poll_len(struct mlx5_rxq_data *rxq, volatile struct mlx5_cqe *cqe,
-		 uint16_t cqe_cnt, uint32_t *rss_hash)
+		 uint16_t cqe_cnt, volatile struct mlx5_mini_cqe8 **mcqe)
 {
 	struct rxq_zip *zip = &rxq->zip;
 	uint16_t cqe_n = cqe_cnt + 1;
@@ -1745,7 +1746,7 @@ mlx5_rx_poll_len(struct mlx5_rxq_data *rxq, volatile struct mlx5_cqe *cqe,
 			(uintptr_t)(&(*rxq->cqes)[zip->ca & cqe_cnt].pkt_info);
 
 		len = rte_be_to_cpu_32((*mc)[zip->ai & 7].byte_cnt);
-		*rss_hash = rte_be_to_cpu_32((*mc)[zip->ai & 7].rx_hash_result);
+		*mcqe = &(*mc)[zip->ai & 7];
 		if ((++zip->ai & 7) == 0) {
 			/* Invalidate consumed CQEs */
 			idx = zip->ca;
@@ -1810,7 +1811,7 @@ mlx5_rx_poll_len(struct mlx5_rxq_data *rxq, volatile struct mlx5_cqe *cqe,
 			zip->cq_ci = rxq->cq_ci + zip->cqe_cnt;
 			/* Get packet size to return. */
 			len = rte_be_to_cpu_32((*mc)[0].byte_cnt);
-			*rss_hash = rte_be_to_cpu_32((*mc)[0].rx_hash_result);
+			*mcqe = &(*mc)[0];
 			zip->ai = 1;
 			/* Prefetch all the entries to be invalidated */
 			idx = zip->ca;
@@ -1821,7 +1822,6 @@ mlx5_rx_poll_len(struct mlx5_rxq_data *rxq, volatile struct mlx5_cqe *cqe,
 			}
 		} else {
 			len = rte_be_to_cpu_32(cqe->byte_cnt);
-			*rss_hash = rte_be_to_cpu_32(cqe->rx_hash_res);
 		}
 		/* Error while receiving packet. */
 		if (unlikely(MLX5_CQE_OPCODE(op_own) == MLX5_CQE_RESP_ERR))
@@ -1934,7 +1934,8 @@ mlx5_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		volatile struct mlx5_wqe_data_seg *wqe =
 			&((volatile struct mlx5_wqe_data_seg *)rxq->wqes)[idx];
 		struct rte_mbuf *rep = (*rxq->elts)[idx];
-		uint32_t rss_hash_res = 0;
+		volatile struct mlx5_mini_cqe8 *mcqe = NULL;
+		uint32_t rss_hash_res;
 
 		if (pkt)
 			NEXT(seg) = rep;
@@ -1964,8 +1965,7 @@ mlx5_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		}
 		if (!pkt) {
 			cqe = &(*rxq->cqes)[rxq->cq_ci & cqe_cnt];
-			len = mlx5_rx_poll_len(rxq, cqe, cqe_cnt,
-					       &rss_hash_res);
+			len = mlx5_rx_poll_len(rxq, cqe, cqe_cnt, &mcqe);
 			if (!len) {
 				rte_mbuf_raw_free(rep);
 				break;
@@ -1979,6 +1979,10 @@ mlx5_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 			pkt = seg;
 			assert(len >= (rxq->crc_present << 2));
 			pkt->ol_flags = 0;
+			/* If compressed, take hash result from mini-CQE. */
+			rss_hash_res = rte_be_to_cpu_32(mcqe == NULL ?
+							cqe->rx_hash_res :
+							mcqe->rx_hash_result);
 			rxq_cq_to_mbuf(rxq, pkt, cqe, rss_hash_res);
 			if (rxq->crc_present)
 				len -= ETHER_CRC_LEN;
@@ -2115,7 +2119,8 @@ mlx5_rx_burst_mprq(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		uint16_t consumed_strd;
 		uint32_t offset;
 		uint32_t byte_cnt;
-		uint32_t rss_hash_res = 0;
+		volatile struct mlx5_mini_cqe8 *mcqe = NULL;
+		uint32_t rss_hash_res;
 
 		if (strd_idx == strd_n) {
 			/* Replace WQE only if the buffer is still in use. */
@@ -2142,7 +2147,7 @@ mlx5_rx_burst_mprq(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 			buf = (*rxq->mprq_bufs)[rq_ci & wq_mask];
 		}
 		cqe = &(*rxq->cqes)[rxq->cq_ci & cq_mask];
-		ret = mlx5_rx_poll_len(rxq, cqe, cq_mask, &rss_hash_res);
+		ret = mlx5_rx_poll_len(rxq, cqe, cq_mask, &mcqe);
 		if (!ret)
 			break;
 		if (unlikely(ret == -1)) {
@@ -2237,6 +2242,10 @@ mlx5_rx_burst_mprq(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 				continue;
 			}
 		}
+		/* If compressed, take hash result from mini-CQE. */
+		rss_hash_res = rte_be_to_cpu_32(mcqe == NULL ?
+						cqe->rx_hash_res :
+						mcqe->rx_hash_result);
 		rxq_cq_to_mbuf(rxq, pkt, cqe, rss_hash_res);
 		PKT_LEN(pkt) = len;
 		DATA_LEN(pkt) = len;
-- 
2.11.0

[dpdk-dev] [PATCH v2 2/5] net/mlx5: add new fields in Rx completion entry

[Yongseok Koh]

Stride index is added to mlx5_mini_cqe8 structure and WQE ID is added to
mlx5_cqe structure.

Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
Acked-by: Shahaf Shuler <shahafs@mellanox.com>
---
 drivers/net/mlx5/mlx5_prm.h  | 9 +++++++--
 drivers/net/mlx5/mlx5_rxtx.h | 2 +-
 2 files changed, 8 insertions(+), 3 deletions(-)

diff --git a/drivers/net/mlx5/mlx5_prm.h b/drivers/net/mlx5/mlx5_prm.h
index 0cf370cd7..f9fae1e50 100644
--- a/drivers/net/mlx5/mlx5_prm.h
+++ b/drivers/net/mlx5/mlx5_prm.h
@@ -240,7 +240,9 @@ struct mlx5_cqe {
 	uint8_t padding[64];
 #endif
 	uint8_t pkt_info;
-	uint8_t rsvd0[11];
+	uint8_t rsvd0;
+	uint16_t wqe_id;
+	uint8_t rsvd3[8];
 	uint32_t rx_hash_res;
 	uint8_t rx_hash_type;
 	uint8_t rsvd1[11];
@@ -285,7 +287,10 @@ struct mlx5_cqe {
 struct mlx5_mini_cqe8 {
 	union {
 		uint32_t rx_hash_result;
-		uint32_t checksum;
+		struct {
+			uint16_t checksum;
+			uint16_t stride_idx;
+		};
 		struct {
 			uint16_t wqe_counter;
 			uint8_t  s_wqe_opcode;
diff --git a/drivers/net/mlx5/mlx5_rxtx.h b/drivers/net/mlx5/mlx5_rxtx.h
index f53bb43c3..04a432adf 100644
--- a/drivers/net/mlx5/mlx5_rxtx.h
+++ b/drivers/net/mlx5/mlx5_rxtx.h
@@ -362,7 +362,7 @@ static inline int
 check_cqe_seen(volatile struct mlx5_cqe *cqe)
 {
 	static const uint8_t magic[] = "seen";
-	volatile uint8_t (*buf)[sizeof(cqe->rsvd0)] = &cqe->rsvd0;
+	volatile uint8_t (*buf)[sizeof(cqe->rsvd1)] = &cqe->rsvd1;
 	int ret = 1;
 	unsigned int i;
 
-- 
2.11.0

[dpdk-dev] [PATCH v2 3/5] net/mlx5: add warning message for Multi-Packet RQ

[Yongseok Koh]

If Multi-Packet RQ is enabled but not supported by device or
kernel/library, print out a warning message.

Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
---
 drivers/net/mlx5/mlx5.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index 49b400503..f0e6ed726 100644
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -1024,8 +1024,7 @@ mlx5_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
 			DRV_LOG(WARNING, "Rx CQE compression isn't supported");
 			config.cqe_comp = 0;
 		}
-		config.mprq.enabled = config.mprq.enabled && mprq;
-		if (config.mprq.enabled) {
+		if (config.mprq.enabled && mprq) {
 			if (config.mprq.stride_num_n > mprq_max_stride_num_n ||
 			    config.mprq.stride_num_n < mprq_min_stride_num_n) {
 				config.mprq.stride_num_n =
@@ -1039,6 +1038,9 @@ mlx5_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
 			}
 			config.mprq.min_stride_size_n = mprq_min_stride_size_n;
 			config.mprq.max_stride_size_n = mprq_max_stride_size_n;
+		} else if (config.mprq.enabled && !mprq) {
+			DRV_LOG(WARNING, "Multi-Packet RQ isn't supported");
+			config.mprq.enabled = 0;
 		}
 		eth_dev = rte_eth_dev_allocate(name);
 		if (eth_dev == NULL) {
-- 
2.11.0

[dpdk-dev] [PATCH v2 4/5] net/mlx5: use stride index in Rx completion entry

[Yongseok Koh]

Multi-Packet Receive Queue is to receive multiple packets on a single large
buffer. The number of consumed strides in CQE is accumulated to keep track
of the current stride index. However, it is safer to directly use stride
index in CQE to avoid out-of-order situation which can possibly be caused
by introducing LRO in the future.

If Rx CQE compression is enabled, HW can be configured to store the stride
index in a mini-CQE but this will need newer version of library/driver.
Therefore, since this change, MPRQ is only supported with the newer
library/driver and Rx hash result is not supported if MPRQ is enabled along
with Rx CQE compression.

Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
Acked-by: Shahaf Shuler <shahafs@mellanox.com>
---
 doc/guides/nics/mlx5.rst     |  4 ++++
 drivers/net/mlx5/Makefile    |  2 +-
 drivers/net/mlx5/mlx5_rxq.c  |  8 +++++++-
 drivers/net/mlx5/mlx5_rxtx.c | 41 +++++++++++++++++++++++------------------
 drivers/net/mlx5/mlx5_rxtx.h |  2 +-
 5 files changed, 36 insertions(+), 21 deletions(-)

diff --git a/doc/guides/nics/mlx5.rst b/doc/guides/nics/mlx5.rst
index 79c982e29..7dd9c1c5e 100644
--- a/doc/guides/nics/mlx5.rst
+++ b/doc/guides/nics/mlx5.rst
@@ -132,6 +132,10 @@ Limitations
   the external buffers will be freed by PMD and the application which still
   holds the external buffers may be corrupted.
 
+- If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
+  enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
+  supported. Some Rx packets may not have PKT_RX_RSS_HASH.
+
 Statistics
 ----------
 
diff --git a/drivers/net/mlx5/Makefile b/drivers/net/mlx5/Makefile
index 8a5229e61..955861a41 100644
--- a/drivers/net/mlx5/Makefile
+++ b/drivers/net/mlx5/Makefile
@@ -100,7 +100,7 @@ mlx5_autoconf.h.new: $(RTE_SDK)/buildtools/auto-config-h.sh
 	$Q sh -- '$<' '$@' \
 		HAVE_IBV_DEVICE_STRIDING_RQ_SUPPORT \
 		infiniband/mlx5dv.h \
-		enum MLX5DV_CONTEXT_MASK_STRIDING_RQ \
+		enum MLX5DV_CQE_RES_FORMAT_CSUM_STRIDX \
 		$(AUTOCONF_OUTPUT)
 	$Q sh -- '$<' '$@' \
 		HAVE_IBV_DEVICE_TUNNEL_SUPPORT \
diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
index 17db7c160..08dd5596b 100644
--- a/drivers/net/mlx5/mlx5_rxq.c
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -818,7 +818,13 @@ mlx5_rxq_ibv_new(struct rte_eth_dev *dev, uint16_t idx)
 	if (config->cqe_comp && !rxq_data->hw_timestamp) {
 		attr.cq.mlx5.comp_mask |=
 			MLX5DV_CQ_INIT_ATTR_MASK_COMPRESSED_CQE;
+#ifdef HAVE_IBV_DEVICE_STRIDING_RQ_SUPPORT
+		attr.cq.mlx5.cqe_comp_res_format =
+			mprq_en ? MLX5DV_CQE_RES_FORMAT_CSUM_STRIDX :
+				  MLX5DV_CQE_RES_FORMAT_HASH;
+#else
 		attr.cq.mlx5.cqe_comp_res_format = MLX5DV_CQE_RES_FORMAT_HASH;
+#endif
 		/*
 		 * For vectorized Rx, it must not be doubled in order to
 		 * make cq_ci and rq_ci aligned.
@@ -976,7 +982,7 @@ mlx5_rxq_ibv_new(struct rte_eth_dev *dev, uint16_t idx)
 	rxq_data->rq_db = rwq.dbrec;
 	rxq_data->cqe_n = log2above(cq_info.cqe_cnt);
 	rxq_data->cq_ci = 0;
-	rxq_data->strd_ci = 0;
+	rxq_data->consumed_strd = 0;
 	rxq_data->rq_pi = 0;
 	rxq_data->zip = (struct rxq_zip){
 		.ai = 0,
diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c
index a4e15c519..a7ed8d8e4 100644
--- a/drivers/net/mlx5/mlx5_rxtx.c
+++ b/drivers/net/mlx5/mlx5_rxtx.c
@@ -2108,7 +2108,7 @@ mlx5_rx_burst_mprq(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 	volatile struct mlx5_cqe *cqe = &(*rxq->cqes)[rxq->cq_ci & cq_mask];
 	unsigned int i = 0;
 	uint16_t rq_ci = rxq->rq_ci;
-	uint16_t strd_idx = rxq->strd_ci;
+	uint16_t consumed_strd = rxq->consumed_strd;
 	struct mlx5_mprq_buf *buf = (*rxq->mprq_bufs)[rq_ci & wq_mask];
 
 	while (i < pkts_n) {
@@ -2116,13 +2116,14 @@ mlx5_rx_burst_mprq(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		void *addr;
 		int ret;
 		unsigned int len;
-		uint16_t consumed_strd;
+		uint16_t strd_cnt;
+		uint16_t strd_idx;
 		uint32_t offset;
 		uint32_t byte_cnt;
 		volatile struct mlx5_mini_cqe8 *mcqe = NULL;
-		uint32_t rss_hash_res;
+		uint32_t rss_hash_res = 0;
 
-		if (strd_idx == strd_n) {
+		if (consumed_strd == strd_n) {
 			/* Replace WQE only if the buffer is still in use. */
 			if (rte_atomic16_read(&buf->refcnt) > 1) {
 				mprq_buf_replace(rxq, rq_ci & wq_mask);
@@ -2142,7 +2143,7 @@ mlx5_rx_burst_mprq(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 					rxq->mprq_repl = rep;
 			}
 			/* Advance to the next WQE. */
-			strd_idx = 0;
+			consumed_strd = 0;
 			++rq_ci;
 			buf = (*rxq->mprq_bufs)[rq_ci & wq_mask];
 		}
@@ -2156,14 +2157,21 @@ mlx5_rx_burst_mprq(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 			continue;
 		}
 		byte_cnt = ret;
-		consumed_strd = (byte_cnt & MLX5_MPRQ_STRIDE_NUM_MASK) >>
-				MLX5_MPRQ_STRIDE_NUM_SHIFT;
-		assert(consumed_strd);
-		/* Calculate offset before adding up stride index. */
-		offset = strd_idx * strd_sz + strd_shift;
-		strd_idx += consumed_strd;
+		strd_cnt = (byte_cnt & MLX5_MPRQ_STRIDE_NUM_MASK) >>
+			   MLX5_MPRQ_STRIDE_NUM_SHIFT;
+		assert(strd_cnt);
+		consumed_strd += strd_cnt;
 		if (byte_cnt & MLX5_MPRQ_FILLER_MASK)
 			continue;
+		if (mcqe == NULL) {
+			rss_hash_res = rte_be_to_cpu_32(cqe->rx_hash_res);
+			strd_idx = rte_be_to_cpu_16(cqe->wqe_counter);
+		} else {
+			/* mini-CQE for MPRQ doesn't have hash result. */
+			strd_idx = rte_be_to_cpu_16(mcqe->stride_idx);
+		}
+		assert(strd_idx < strd_n);
+		assert(!((rte_be_to_cpu_16(cqe->wqe_id) ^ rq_ci) & wq_mask));
 		/*
 		 * Currently configured to receive a packet per a stride. But if
 		 * MTU is adjusted through kernel interface, device could
@@ -2171,7 +2179,7 @@ mlx5_rx_burst_mprq(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		 * case, the packet should be dropped because it is bigger than
 		 * the max_rx_pkt_len.
 		 */
-		if (unlikely(consumed_strd > 1)) {
+		if (unlikely(strd_cnt > 1)) {
 			++rxq->stats.idropped;
 			continue;
 		}
@@ -2184,6 +2192,7 @@ mlx5_rx_burst_mprq(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		assert((int)len >= (rxq->crc_present << 2));
 		if (rxq->crc_present)
 			len -= ETHER_CRC_LEN;
+		offset = strd_idx * strd_sz + strd_shift;
 		addr = RTE_PTR_ADD(mlx5_mprq_buf_addr(buf), offset);
 		/* Initialize the offload flag. */
 		pkt->ol_flags = 0;
@@ -2206,7 +2215,7 @@ mlx5_rx_burst_mprq(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		} else {
 			rte_iova_t buf_iova;
 			struct rte_mbuf_ext_shared_info *shinfo;
-			uint16_t buf_len = consumed_strd * strd_sz;
+			uint16_t buf_len = strd_cnt * strd_sz;
 
 			/* Increment the refcnt of the whole chunk. */
 			rte_atomic16_add_return(&buf->refcnt, 1);
@@ -2242,10 +2251,6 @@ mlx5_rx_burst_mprq(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 				continue;
 			}
 		}
-		/* If compressed, take hash result from mini-CQE. */
-		rss_hash_res = rte_be_to_cpu_32(mcqe == NULL ?
-						cqe->rx_hash_res :
-						mcqe->rx_hash_result);
 		rxq_cq_to_mbuf(rxq, pkt, cqe, rss_hash_res);
 		PKT_LEN(pkt) = len;
 		DATA_LEN(pkt) = len;
@@ -2259,7 +2264,7 @@ mlx5_rx_burst_mprq(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
 		++i;
 	}
 	/* Update the consumer indexes. */
-	rxq->strd_ci = strd_idx;
+	rxq->consumed_strd = consumed_strd;
 	rte_cio_wmb();
 	*rxq->cq_db = rte_cpu_to_be_32(rxq->cq_ci);
 	if (rq_ci != rxq->rq_ci) {
diff --git a/drivers/net/mlx5/mlx5_rxtx.h b/drivers/net/mlx5/mlx5_rxtx.h
index 04a432adf..0007be08b 100644
--- a/drivers/net/mlx5/mlx5_rxtx.h
+++ b/drivers/net/mlx5/mlx5_rxtx.h
@@ -93,7 +93,7 @@ struct mlx5_rxq_data {
 	volatile uint32_t *cq_db;
 	uint16_t port_id;
 	uint16_t rq_ci;
-	uint16_t strd_ci; /* Stride index in a WQE for Multi-Packet RQ. */
+	uint16_t consumed_strd; /* Number of consumed strides in WQE. */
 	uint16_t rq_pi;
 	uint16_t cq_ci;
 	struct mlx5_mr_ctrl mr_ctrl; /* MR control descriptor. */
-- 
2.11.0

[dpdk-dev] [PATCH v2 5/5] net/mlx5: increase number of strides

[Yongseok Koh]

If WQE ID is used in CQE for Multi-Packet RQ, the ratio of CQE compression
drops a little bit.  In order to reach to 100Gbps with 64B traffic, it is
needed to further save PCIe bandwidth by increasing the number of strides
in a WQE. It is now 64 by default but adjustable by a PMD parameter -
mprq_log_stride_num.

Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
Acked-by: Shahaf Shuler <shahafs@mellanox.com>
---
 drivers/net/mlx5/mlx5_defs.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/net/mlx5/mlx5_defs.h b/drivers/net/mlx5/mlx5_defs.h
index 51124cdcc..79bb23d2a 100644
--- a/drivers/net/mlx5/mlx5_defs.h
+++ b/drivers/net/mlx5/mlx5_defs.h
@@ -96,7 +96,7 @@
 #define MLX5_UAR_OFFSET (1ULL << 32)
 
 /* Log 2 of the default number of strides per WQE for Multi-Packet RQ. */
-#define MLX5_MPRQ_STRIDE_NUM_N 4U
+#define MLX5_MPRQ_STRIDE_NUM_N 6U
 
 /* Two-byte shift is disabled for Multi-Packet RQ. */
 #define MLX5_MPRQ_TWO_BYTE_SHIFT 0
-- 
2.11.0

Re: [dpdk-dev] [PATCH v2 0/5] net/mlx5: support new completion field for Multi-Packet RQ

[Shahaf Shuler]

Tuesday, June 26, 2018 3:39 PM, Yongseok Koh:
> Subject: [PATCH v2 0/5] net/mlx5: support new completion field for Multi-
> Packet RQ
> 
> For the RFC,
>     https://mails.dpdk.org/archives/dev/2018-June/103664.html
> 
> Yongseok Koh (5):
>   net/mlx5: change return value of Rx completion poll
>   net/mlx5: add new fields in Rx completion entry
>   net/mlx5: add warning message for Multi-Packet RQ
>   net/mlx5: use stride index in Rx completion entry
>   net/mlx5: increase number of strides
> 
>  doc/guides/nics/mlx5.rst     |  4 +++
>  drivers/net/mlx5/Makefile    |  2 +-
>  drivers/net/mlx5/mlx5.c      |  6 +++--
>  drivers/net/mlx5/mlx5_defs.h |  2 +-
>  drivers/net/mlx5/mlx5_prm.h  |  9 +++++--  drivers/net/mlx5/mlx5_rxq.c  |
> 8 +++++-  drivers/net/mlx5/mlx5_rxtx.c | 62
> +++++++++++++++++++++++++++-----------------
>  drivers/net/mlx5/mlx5_rxtx.h |  4 +--
>  8 files changed, 64 insertions(+), 33 deletions(-)

Series applied to next-net-mlx, thanks.