[PATCH v4 12/19] net/thunderx: add single and multi segment tx functions

[Jerin Jacob <jerin.jacob@caviumnetworks.com>]

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Signed-off-by: Maciej Czekaj <maciej.czekaj@caviumnetworks.com>
Signed-off-by: Kamil Rytarowski <Kamil.Rytarowski@caviumnetworks.com>
Signed-off-by: Zyta Szpak <zyta.szpak@semihalf.com>
Signed-off-by: Slawomir Rosek <slawomir.rosek@semihalf.com>
Signed-off-by: Radoslaw Biernacki <rad@semihalf.com>
---
 drivers/net/thunderx/Makefile       |   2 +
 drivers/net/thunderx/nicvf_ethdev.c |   5 +-
 drivers/net/thunderx/nicvf_rxtx.c   | 255 ++++++++++++++++++++++++++++++++++++
 drivers/net/thunderx/nicvf_rxtx.h   |  93 +++++++++++++
 4 files changed, 354 insertions(+), 1 deletion(-)
 create mode 100644 drivers/net/thunderx/nicvf_rxtx.c
 create mode 100644 drivers/net/thunderx/nicvf_rxtx.h

diff --git a/drivers/net/thunderx/Makefile b/drivers/net/thunderx/Makefile
index eb9f100..9079b5b 100644
--- a/drivers/net/thunderx/Makefile
+++ b/drivers/net/thunderx/Makefile
@@ -51,10 +51,12 @@ VPATH += $(SRCDIR)/base
 #
 # all source are stored in SRCS-y
 #
+SRCS-$(CONFIG_RTE_LIBRTE_THUNDERX_NICVF_PMD) += nicvf_rxtx.c
 SRCS-$(CONFIG_RTE_LIBRTE_THUNDERX_NICVF_PMD) += nicvf_hw.c
 SRCS-$(CONFIG_RTE_LIBRTE_THUNDERX_NICVF_PMD) += nicvf_mbox.c
 SRCS-$(CONFIG_RTE_LIBRTE_THUNDERX_NICVF_PMD) += nicvf_ethdev.c
 
+CFLAGS_nicvf_rxtx.o += -fno-prefetch-loop-arrays -Ofast
 
 # this lib depends upon:
 DEPDIRS-$(CONFIG_RTE_LIBRTE_THUNDERX_NICVF_PMD) += lib/librte_eal lib/librte_ether
diff --git a/drivers/net/thunderx/nicvf_ethdev.c b/drivers/net/thunderx/nicvf_ethdev.c
index 19ad85a..15f5cfc 100644
--- a/drivers/net/thunderx/nicvf_ethdev.c
+++ b/drivers/net/thunderx/nicvf_ethdev.c
@@ -66,7 +66,7 @@
 #include "base/nicvf_plat.h"
 
 #include "nicvf_ethdev.h"
-
+#include "nicvf_rxtx.h"
 #include "nicvf_logs.h"
 
 static inline int
@@ -617,6 +617,9 @@ nicvf_dev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
 		(tx_conf->tx_free_thresh == NICVF_DEFAULT_TX_FREE_THRESH ?
 				NICVF_TX_FREE_MPOOL_THRESH :
 				tx_conf->tx_free_thresh);
+		txq->pool_free = nicvf_multi_pool_free_xmited_buffers;
+	} else {
+		txq->pool_free = nicvf_single_pool_free_xmited_buffers;
 	}
 
 	/* Allocate software ring */
diff --git a/drivers/net/thunderx/nicvf_rxtx.c b/drivers/net/thunderx/nicvf_rxtx.c
new file mode 100644
index 0000000..88a5152
--- /dev/null
+++ b/drivers/net/thunderx/nicvf_rxtx.c
@@ -0,0 +1,255 @@
+/*
+ *   BSD LICENSE
+ *
+ *   Copyright (C) Cavium networks Ltd. 2016.
+ *
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
+ *   are met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in
+ *       the documentation and/or other materials provided with the
+ *       distribution.
+ *     * Neither the name of Cavium networks nor the names of its
+ *       contributors may be used to endorse or promote products derived
+ *       from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <unistd.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+#include <rte_atomic.h>
+#include <rte_branch_prediction.h>
+#include <rte_byteorder.h>
+#include <rte_common.h>
+#include <rte_cycles.h>
+#include <rte_errno.h>
+#include <rte_ethdev.h>
+#include <rte_ether.h>
+#include <rte_log.h>
+#include <rte_mbuf.h>
+#include <rte_prefetch.h>
+
+#include "base/nicvf_plat.h"
+
+#include "nicvf_ethdev.h"
+#include "nicvf_rxtx.h"
+#include "nicvf_logs.h"
+
+static inline void __hot
+fill_sq_desc_header(union sq_entry_t *entry, struct rte_mbuf *pkt)
+{
+	/* Local variable sqe to avoid read from sq desc memory*/
+	union sq_entry_t sqe;
+	uint64_t ol_flags;
+
+	/* Fill SQ header descriptor */
+	sqe.buff[0] = 0;
+	sqe.hdr.subdesc_type = SQ_DESC_TYPE_HEADER;
+	/* Number of sub-descriptors following this one */
+	sqe.hdr.subdesc_cnt = pkt->nb_segs;
+	sqe.hdr.tot_len = pkt->pkt_len;
+
+	ol_flags = pkt->ol_flags & NICVF_TX_OFFLOAD_MASK;
+	if (unlikely(ol_flags)) {
+		/* L4 cksum */
+		if (ol_flags & PKT_TX_TCP_CKSUM)
+			sqe.hdr.csum_l4 = SEND_L4_CSUM_TCP;
+		else if (ol_flags & PKT_TX_UDP_CKSUM)
+			sqe.hdr.csum_l4 = SEND_L4_CSUM_UDP;
+		else
+			sqe.hdr.csum_l4 = SEND_L4_CSUM_DISABLE;
+		sqe.hdr.l4_offset = pkt->l3_len + pkt->l2_len;
+
+		/* L3 cksum */
+		if (ol_flags & PKT_TX_IP_CKSUM) {
+			sqe.hdr.csum_l3 = 1;
+			sqe.hdr.l3_offset = pkt->l2_len;
+		}
+	}
+
+	entry->buff[0] = sqe.buff[0];
+}
+
+void __hot
+nicvf_single_pool_free_xmited_buffers(struct nicvf_txq *sq)
+{
+	int j = 0;
+	uint32_t curr_head;
+	uint32_t head = sq->head;
+	struct rte_mbuf **txbuffs = sq->txbuffs;
+	void *obj_p[NICVF_MAX_TX_FREE_THRESH] __rte_cache_aligned;
+
+	curr_head = nicvf_addr_read(sq->sq_head) >> 4;
+	while (head != curr_head) {
+		if (txbuffs[head])
+			obj_p[j++] = txbuffs[head];
+
+		head = (head + 1) & sq->qlen_mask;
+	}
+
+	rte_mempool_put_bulk(sq->pool, obj_p, j);
+	sq->head = curr_head;
+	sq->xmit_bufs -= j;
+	NICVF_TX_ASSERT(sq->xmit_bufs >= 0);
+}
+
+void __hot
+nicvf_multi_pool_free_xmited_buffers(struct nicvf_txq *sq)
+{
+	uint32_t n = 0;
+	uint32_t curr_head;
+	uint32_t head = sq->head;
+	struct rte_mbuf **txbuffs = sq->txbuffs;
+
+	curr_head = nicvf_addr_read(sq->sq_head) >> 4;
+	while (head != curr_head) {
+		if (txbuffs[head]) {
+			rte_pktmbuf_free_seg(txbuffs[head]);
+			n++;
+		}
+
+		head = (head + 1) & sq->qlen_mask;
+	}
+
+	sq->head = curr_head;
+	sq->xmit_bufs -= n;
+	NICVF_TX_ASSERT(sq->xmit_bufs >= 0);
+}
+
+static inline uint32_t __hot
+nicvf_free_tx_desc(struct nicvf_txq *sq)
+{
+	return ((sq->head - sq->tail - 1) & sq->qlen_mask);
+}
+
+/* Send Header + Packet */
+#define TX_DESC_PER_PKT 2
+
+static inline uint32_t __hot
+nicvf_free_xmitted_buffers(struct nicvf_txq *sq, struct rte_mbuf **tx_pkts,
+			    uint16_t nb_pkts)
+{
+	uint32_t free_desc = nicvf_free_tx_desc(sq);
+
+	if (free_desc < nb_pkts * TX_DESC_PER_PKT ||
+			sq->xmit_bufs > sq->tx_free_thresh) {
+		if (unlikely(sq->pool == NULL))
+			sq->pool = tx_pkts[0]->pool;
+
+		sq->pool_free(sq);
+		/* Freed now, let see the number of free descs again */
+		free_desc = nicvf_free_tx_desc(sq);
+	}
+	return free_desc;
+}
+
+uint16_t __hot
+nicvf_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
+{
+	int i;
+	uint32_t free_desc;
+	uint32_t tail;
+	struct nicvf_txq *sq = tx_queue;
+	union sq_entry_t *desc_ptr = sq->desc;
+	struct rte_mbuf **txbuffs = sq->txbuffs;
+	struct rte_mbuf *pkt;
+	uint32_t qlen_mask = sq->qlen_mask;
+
+	tail = sq->tail;
+	free_desc = nicvf_free_xmitted_buffers(sq, tx_pkts, nb_pkts);
+
+	for (i = 0; i < nb_pkts && (int)free_desc >= TX_DESC_PER_PKT; i++) {
+		pkt = tx_pkts[i];
+
+		txbuffs[tail] = NULL;
+		fill_sq_desc_header(desc_ptr + tail, pkt);
+		tail = (tail + 1) & qlen_mask;
+
+		txbuffs[tail] = pkt;
+		fill_sq_desc_gather(desc_ptr + tail, pkt);
+		tail = (tail + 1) & qlen_mask;
+		free_desc -= TX_DESC_PER_PKT;
+	}
+
+	sq->tail = tail;
+	sq->xmit_bufs += i;
+	rte_wmb();
+
+	/* Inform HW to xmit the packets */
+	nicvf_addr_write(sq->sq_door, i * TX_DESC_PER_PKT);
+	return i;
+}
+
+uint16_t __hot
+nicvf_xmit_pkts_multiseg(void *tx_queue, struct rte_mbuf **tx_pkts,
+			 uint16_t nb_pkts)
+{
+	int i, k;
+	uint32_t used_desc, next_used_desc, used_bufs, free_desc, tail;
+	struct nicvf_txq *sq = tx_queue;
+	union sq_entry_t *desc_ptr = sq->desc;
+	struct rte_mbuf **txbuffs = sq->txbuffs;
+	struct rte_mbuf *pkt, *seg;
+	uint32_t qlen_mask = sq->qlen_mask;
+	uint16_t nb_segs;
+
+	tail = sq->tail;
+	used_desc = 0;
+	used_bufs = 0;
+
+	free_desc = nicvf_free_xmitted_buffers(sq, tx_pkts, nb_pkts);
+
+	for (i = 0; i < nb_pkts; i++) {
+		pkt = tx_pkts[i];
+
+		nb_segs = pkt->nb_segs;
+
+		next_used_desc = used_desc + nb_segs + 1;
+		if (next_used_desc > free_desc)
+			break;
+		used_desc = next_used_desc;
+		used_bufs += nb_segs;
+
+		txbuffs[tail] = NULL;
+		fill_sq_desc_header(desc_ptr + tail, pkt);
+		tail = (tail + 1) & qlen_mask;
+
+		txbuffs[tail] = pkt;
+		fill_sq_desc_gather(desc_ptr + tail, pkt);
+		tail = (tail + 1) & qlen_mask;
+
+		seg = pkt->next;
+		for (k = 1; k < nb_segs; k++) {
+			txbuffs[tail] = seg;
+			fill_sq_desc_gather(desc_ptr + tail, seg);
+			tail = (tail + 1) & qlen_mask;
+			seg = seg->next;
+		}
+	}
+
+	sq->tail = tail;
+	sq->xmit_bufs += used_bufs;
+	rte_wmb();
+
+	/* Inform HW to xmit the packets */
+	nicvf_addr_write(sq->sq_door, used_desc);
+	return nb_pkts;
+}
diff --git a/drivers/net/thunderx/nicvf_rxtx.h b/drivers/net/thunderx/nicvf_rxtx.h
new file mode 100644
index 0000000..b1fdc69
--- /dev/null
+++ b/drivers/net/thunderx/nicvf_rxtx.h
@@ -0,0 +1,93 @@
+/*
+ *   BSD LICENSE
+ *
+ *   Copyright (C) Cavium networks Ltd. 2016.
+ *
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
+ *   are met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in
+ *       the documentation and/or other materials provided with the
+ *       distribution.
+ *     * Neither the name of Cavium networks nor the names of its
+ *       contributors may be used to endorse or promote products derived
+ *       from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef __THUNDERX_NICVF_RXTX_H__
+#define __THUNDERX_NICVF_RXTX_H__
+
+#include <rte_ethdev.h>
+
+#define NICVF_TX_OFFLOAD_MASK (PKT_TX_IP_CKSUM | PKT_TX_L4_MASK)
+
+#ifndef __hot
+#define __hot	__attribute__((hot))
+#endif
+
+#if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
+static inline uint16_t __attribute__((const))
+nicvf_frag_num(uint16_t i)
+{
+	return (i & ~3) + 3 - (i & 3);
+}
+
+static inline void __hot
+fill_sq_desc_gather(union sq_entry_t *entry, struct rte_mbuf *pkt)
+{
+	/* Local variable sqe to avoid read from sq desc memory*/
+	union sq_entry_t sqe;
+
+	/* Fill the SQ gather entry */
+	sqe.buff[0] = 0; sqe.buff[1] = 0;
+	sqe.gather.subdesc_type = SQ_DESC_TYPE_GATHER;
+	sqe.gather.ld_type = NIC_SEND_LD_TYPE_E_LDT;
+	sqe.gather.size = pkt->data_len;
+	sqe.gather.addr = rte_mbuf_data_dma_addr(pkt);
+
+	entry->buff[0] = sqe.buff[0];
+	entry->buff[1] = sqe.buff[1];
+}
+
+#else
+
+static inline uint16_t __attribute__((const))
+nicvf_frag_num(uint16_t i)
+{
+	return i;
+}
+
+static inline void __hot
+fill_sq_desc_gather(union sq_entry_t *entry, struct rte_mbuf *pkt)
+{
+	entry->buff[0] = (uint64_t)SQ_DESC_TYPE_GATHER << 60 |
+			 (uint64_t)NIC_SEND_LD_TYPE_E_LDT << 58 |
+			 pkt->data_len;
+	entry->buff[1] = rte_mbuf_data_dma_addr(pkt);
+}
+#endif
+
+uint16_t nicvf_xmit_pkts(void *txq, struct rte_mbuf **tx_pkts, uint16_t pkts);
+uint16_t nicvf_xmit_pkts_multiseg(void *txq, struct rte_mbuf **tx_pkts,
+				  uint16_t pkts);
+
+void nicvf_single_pool_free_xmited_buffers(struct nicvf_txq *sq);
+void nicvf_multi_pool_free_xmited_buffers(struct nicvf_txq *sq);
+
+#endif /* __THUNDERX_NICVF_RXTX_H__  */
-- 
2.5.5