DPDK patches and discussions
 help / color / mirror / Atom feed
From: Balazs Nemeth <bnemeth@redhat.com>
To: bnemeth@redhat.com, dev@dpdk.org
Cc: dsinghrawat@marvell.com, rmody@marvell.com,
	jerinjacobk@gmail.com, irusskikh@marvell.com
Subject: [dpdk-dev] [PATCH v2 7/7] net/qede: remove unnecessary field in Rx entry and simplify
Date: Fri, 26 Mar 2021 12:01:30 +0100	[thread overview]
Message-ID: <32e010b5391a6e36dac608b1ac02133abb032951.1616754844.git.bnemeth@redhat.com> (raw)
In-Reply-To: <cover.1616754844.git.bnemeth@redhat.com>

The member page_offset is always zero. Having this in the qede_rx_entry
makes it larger than it needs to be and this has cache performance
implications so remove that field. In addition, since qede_rx_entry only
has an rte_mbuf*, remove the definition of qede_rx_entry.

Signed-off-by: Balazs Nemeth <bnemeth@redhat.com>
Reviewed-by: Igor Russkikh <irusskikh@marvell.com>
---
 drivers/net/qede/qede_rxtx.c | 56 ++++++++++++++++++------------------
 drivers/net/qede/qede_rxtx.h | 11 +------
 2 files changed, 29 insertions(+), 38 deletions(-)

diff --git a/drivers/net/qede/qede_rxtx.c b/drivers/net/qede/qede_rxtx.c
index ed3617160..298f4e3e4 100644
--- a/drivers/net/qede/qede_rxtx.c
+++ b/drivers/net/qede/qede_rxtx.c
@@ -24,8 +24,7 @@ static inline int qede_alloc_rx_buffer(struct qede_rx_queue *rxq)
 			   rte_mempool_in_use_count(rxq->mb_pool));
 		return -ENOMEM;
 	}
-	rxq->sw_rx_ring[idx].mbuf = new_mb;
-	rxq->sw_rx_ring[idx].page_offset = 0;
+	rxq->sw_rx_ring[idx] = new_mb;
 	mapping = rte_mbuf_data_iova_default(new_mb);
 	/* Advance PROD and get BD pointer */
 	rx_bd = (struct eth_rx_bd *)ecore_chain_produce(&rxq->rx_bd_ring);
@@ -39,17 +38,24 @@ static inline int qede_alloc_rx_buffer(struct qede_rx_queue *rxq)
 
 static inline int qede_alloc_rx_bulk_mbufs(struct qede_rx_queue *rxq, int count)
 {
-	void *obj_p[QEDE_MAX_BULK_ALLOC_COUNT] __rte_cache_aligned;
 	struct rte_mbuf *mbuf = NULL;
 	struct eth_rx_bd *rx_bd;
 	dma_addr_t mapping;
 	int i, ret = 0;
 	uint16_t idx;
+	uint16_t mask = NUM_RX_BDS(rxq);
 
 	if (count > QEDE_MAX_BULK_ALLOC_COUNT)
 		count = QEDE_MAX_BULK_ALLOC_COUNT;
 
-	ret = rte_mempool_get_bulk(rxq->mb_pool, obj_p, count);
+	idx = rxq->sw_rx_prod & NUM_RX_BDS(rxq);
+
+	if (count > mask - idx + 1)
+		count = mask - idx + 1;
+
+	ret = rte_mempool_get_bulk(rxq->mb_pool, (void **)&rxq->sw_rx_ring[idx],
+				   count);
+
 	if (unlikely(ret)) {
 		PMD_RX_LOG(ERR, rxq,
 			   "Failed to allocate %d rx buffers "
@@ -63,20 +69,17 @@ static inline int qede_alloc_rx_bulk_mbufs(struct qede_rx_queue *rxq, int count)
 	}
 
 	for (i = 0; i < count; i++) {
-		mbuf = obj_p[i];
-		if (likely(i < count - 1))
-			rte_prefetch0(obj_p[i + 1]);
+		rte_prefetch0(rxq->sw_rx_ring[(idx + 1) & NUM_RX_BDS(rxq)]);
+		mbuf = rxq->sw_rx_ring[idx & NUM_RX_BDS(rxq)];
 
-		idx = rxq->sw_rx_prod & NUM_RX_BDS(rxq);
-		rxq->sw_rx_ring[idx].mbuf = mbuf;
-		rxq->sw_rx_ring[idx].page_offset = 0;
 		mapping = rte_mbuf_data_iova_default(mbuf);
 		rx_bd = (struct eth_rx_bd *)
 			ecore_chain_produce(&rxq->rx_bd_ring);
 		rx_bd->addr.hi = rte_cpu_to_le_32(U64_HI(mapping));
 		rx_bd->addr.lo = rte_cpu_to_le_32(U64_LO(mapping));
-		rxq->sw_rx_prod++;
+		idx++;
 	}
+	rxq->sw_rx_prod = idx;
 
 	return 0;
 }
@@ -309,9 +312,9 @@ static void qede_rx_queue_release_mbufs(struct qede_rx_queue *rxq)
 
 	if (rxq->sw_rx_ring) {
 		for (i = 0; i < rxq->nb_rx_desc; i++) {
-			if (rxq->sw_rx_ring[i].mbuf) {
-				rte_pktmbuf_free(rxq->sw_rx_ring[i].mbuf);
-				rxq->sw_rx_ring[i].mbuf = NULL;
+			if (rxq->sw_rx_ring[i]) {
+				rte_pktmbuf_free(rxq->sw_rx_ring[i]);
+				rxq->sw_rx_ring[i] = NULL;
 			}
 		}
 	}
@@ -1318,18 +1321,15 @@ static inline void qede_rx_bd_ring_consume(struct qede_rx_queue *rxq)
 
 static inline void
 qede_reuse_page(__rte_unused struct qede_dev *qdev,
-		struct qede_rx_queue *rxq, struct qede_rx_entry *curr_cons)
+		struct qede_rx_queue *rxq, struct rte_mbuf *curr_cons)
 {
 	struct eth_rx_bd *rx_bd_prod = ecore_chain_produce(&rxq->rx_bd_ring);
 	uint16_t idx = rxq->sw_rx_prod & NUM_RX_BDS(rxq);
-	struct qede_rx_entry *curr_prod;
 	dma_addr_t new_mapping;
 
-	curr_prod = &rxq->sw_rx_ring[idx];
-	*curr_prod = *curr_cons;
+	rxq->sw_rx_ring[idx] = curr_cons;
 
-	new_mapping = rte_mbuf_data_iova_default(curr_prod->mbuf) +
-		      curr_prod->page_offset;
+	new_mapping = rte_mbuf_data_iova_default(curr_cons);
 
 	rx_bd_prod->addr.hi = rte_cpu_to_le_32(U64_HI(new_mapping));
 	rx_bd_prod->addr.lo = rte_cpu_to_le_32(U64_LO(new_mapping));
@@ -1341,10 +1341,10 @@ static inline void
 qede_recycle_rx_bd_ring(struct qede_rx_queue *rxq,
 			struct qede_dev *qdev, uint8_t count)
 {
-	struct qede_rx_entry *curr_cons;
+	struct rte_mbuf *curr_cons;
 
 	for (; count > 0; count--) {
-		curr_cons = &rxq->sw_rx_ring[rxq->sw_rx_cons & NUM_RX_BDS(rxq)];
+		curr_cons = rxq->sw_rx_ring[rxq->sw_rx_cons & NUM_RX_BDS(rxq)];
 		qede_reuse_page(qdev, rxq, curr_cons);
 		qede_rx_bd_ring_consume(rxq);
 	}
@@ -1366,7 +1366,7 @@ qede_rx_process_tpa_cmn_cont_end_cqe(__rte_unused struct qede_dev *qdev,
 	if (rte_le_to_cpu_16(len)) {
 		tpa_info = &rxq->tpa_info[agg_index];
 		cons_idx = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
-		curr_frag = rxq->sw_rx_ring[cons_idx].mbuf;
+		curr_frag = rxq->sw_rx_ring[cons_idx];
 		assert(curr_frag);
 		curr_frag->nb_segs = 1;
 		curr_frag->pkt_len = rte_le_to_cpu_16(len);
@@ -1498,7 +1498,7 @@ qede_process_sg_pkts(void *p_rxq,  struct rte_mbuf *rx_mb,
 			return -EINVAL;
 		}
 		sw_rx_index = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
-		seg2 = rxq->sw_rx_ring[sw_rx_index].mbuf;
+		seg2 = rxq->sw_rx_ring[sw_rx_index];
 		qede_rx_bd_ring_consume(rxq);
 		pkt_len -= cur_size;
 		seg2->data_len = cur_size;
@@ -1617,7 +1617,7 @@ qede_recv_pkts_regular(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
 
 		/* Get the data from the SW ring */
 		sw_rx_index = rxq->sw_rx_cons & num_rx_bds;
-		rx_mb = rxq->sw_rx_ring[sw_rx_index].mbuf;
+		rx_mb = rxq->sw_rx_ring[sw_rx_index];
 		assert(rx_mb != NULL);
 
 		parse_flag = rte_le_to_cpu_16(fp_cqe->pars_flags.flags);
@@ -1716,7 +1716,7 @@ qede_recv_pkts_regular(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
 
 		/* Prefetch next mbuf while processing current one. */
 		preload_idx = rxq->sw_rx_cons & num_rx_bds;
-		rte_prefetch0(rxq->sw_rx_ring[preload_idx].mbuf);
+		rte_prefetch0(rxq->sw_rx_ring[preload_idx]);
 
 		/* Update rest of the MBUF fields */
 		rx_mb->data_off = offset + RTE_PKTMBUF_HEADROOM;
@@ -1874,7 +1874,7 @@ qede_recv_pkts(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
 
 		/* Get the data from the SW ring */
 		sw_rx_index = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
-		rx_mb = rxq->sw_rx_ring[sw_rx_index].mbuf;
+		rx_mb = rxq->sw_rx_ring[sw_rx_index];
 		assert(rx_mb != NULL);
 
 		/* Handle regular CQE or TPA start CQE */
@@ -2005,7 +2005,7 @@ qede_recv_pkts(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
 
 		/* Prefetch next mbuf while processing current one. */
 		preload_idx = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
-		rte_prefetch0(rxq->sw_rx_ring[preload_idx].mbuf);
+		rte_prefetch0(rxq->sw_rx_ring[preload_idx]);
 
 		/* Update rest of the MBUF fields */
 		rx_mb->data_off = offset + RTE_PKTMBUF_HEADROOM;
diff --git a/drivers/net/qede/qede_rxtx.h b/drivers/net/qede/qede_rxtx.h
index 335016847..c9334448c 100644
--- a/drivers/net/qede/qede_rxtx.h
+++ b/drivers/net/qede/qede_rxtx.h
@@ -159,15 +159,6 @@
 #define QEDE_TX_OFFLOAD_NOTSUP_MASK \
 	(PKT_TX_OFFLOAD_MASK ^ QEDE_TX_OFFLOAD_MASK)
 
-/*
- * RX BD descriptor ring
- */
-struct qede_rx_entry {
-	struct rte_mbuf *mbuf;
-	uint32_t page_offset;
-	/* allows expansion .. */
-};
-
 /* TPA related structures */
 struct qede_agg_info {
 	struct rte_mbuf *tpa_head; /* Pointer to first TPA segment */
@@ -185,7 +176,7 @@ struct qede_rx_queue {
 	struct ecore_chain rx_comp_ring;
 	uint16_t *hw_cons_ptr;
 	void OSAL_IOMEM *hw_rxq_prod_addr;
-	struct qede_rx_entry *sw_rx_ring;
+	struct rte_mbuf **sw_rx_ring;
 	struct ecore_sb_info *sb_info;
 	uint16_t sw_rx_cons;
 	uint16_t sw_rx_prod;
-- 
2.30.2


  parent reply	other threads:[~2021-03-26 11:03 UTC|newest]

Thread overview: 9+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-03-26 11:01 [dpdk-dev] [PATCH v2 0/7] Optimize qede use of Rx/Tx entries Balazs Nemeth
2021-03-26 11:01 ` [dpdk-dev] [PATCH v2 1/7] net/qede: remove flags from Tx entry Balazs Nemeth
2021-03-26 11:01 ` [dpdk-dev] [PATCH v2 2/7] net/qede: get consumer index once Balazs Nemeth
2021-03-26 11:01 ` [dpdk-dev] [PATCH v2 3/7] net/qede: assume mbuf to free is never null Balazs Nemeth
2021-03-26 11:01 ` [dpdk-dev] [PATCH v2 4/7] net/qede: free packets in bulk instead of one by one Balazs Nemeth
2021-03-26 11:01 ` [dpdk-dev] [PATCH v2 5/7] net/qede: prefetch hardware consumer Balazs Nemeth
2021-03-26 11:01 ` [dpdk-dev] [PATCH v2 6/7] net/qede: prefetch next packet to free Balazs Nemeth
2021-03-26 11:01 ` Balazs Nemeth [this message]
2021-03-27 14:02 ` [dpdk-dev] [PATCH v2 0/7] Optimize qede use of Rx/Tx entries Jerin Jacob

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=32e010b5391a6e36dac608b1ac02133abb032951.1616754844.git.bnemeth@redhat.com \
    --to=bnemeth@redhat.com \
    --cc=dev@dpdk.org \
    --cc=dsinghrawat@marvell.com \
    --cc=irusskikh@marvell.com \
    --cc=jerinjacobk@gmail.com \
    --cc=rmody@marvell.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).