Adding PROX(Packet pROcessing eXecution engine) VNF to sampleVNF

JIRA: SAMPLEVNF-55

PROX is a DPDK-based application implementing Telco use-cases such as
a simplified BRAS/BNG, light-weight AFTR... It also allows configuring
finer grained network functions like QoS, Routing, load-balancing...

(We are moving PROX version v039 to sampleVNF
https://01.org/intel-data-plane-performance-demonstrators/prox-overview)

Change-Id: Ia3cb02cf0e49ac5596e922c197ff7e010293d033
Signed-off-by: Deepak S <deepak.s@linux.intel.com>

1 files changed, 665 insertions, 0 deletions

diff --git a/VNFs/DPPD-PROX/tx_pkt.c b/VNFs/DPPD-PROX/tx_pkt.c
new file mode 100644
index 00000000..c6f6010c
--- /dev/null
+++ b/VNFs/DPPD-PROX/tx_pkt.c
@@ -0,0 +1,665 @@
+/*
+// Copyright (c) 2010-2017 Intel Corporation
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+*/
+
+#include <rte_ethdev.h>
+#include <rte_version.h>
+
+#include "rx_pkt.h"
+#include "tx_pkt.h"
+#include "task_base.h"
+#include "stats.h"
+#include "prefetch.h"
+#include "prox_assert.h"
+#include "log.h"
+#include "mbuf_utils.h"
+
+static void buf_pkt_single(struct task_base *tbase, struct rte_mbuf *mbuf, const uint8_t out)
+{
+	const uint16_t prod = tbase->ws_mbuf->idx[out].prod++;
+	tbase->ws_mbuf->mbuf[out][prod & WS_MBUF_MASK] = mbuf;
+}
+
+static inline void buf_pkt_all(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts, uint8_t *out)
+{
+	for (uint16_t j = 0; j < n_pkts; ++j) {
+		if (unlikely(out[j] >= OUT_HANDLED)) {
+			rte_pktmbuf_free(mbufs[j]);
+			if (out[j] == OUT_HANDLED)
+				TASK_STATS_ADD_DROP_HANDLED(&tbase->aux->stats, 1);
+			else
+				TASK_STATS_ADD_DROP_DISCARD(&tbase->aux->stats, 1);
+		}
+		else {
+			buf_pkt_single(tbase, mbufs[j], out[j]);
+		}
+	}
+}
+#define MAX_PMD_TX 32
+
+/* The following help functions also report stats. Therefore we need
+   to pass the task_base struct. */
+static inline int txhw_drop(const struct port_queue *port_queue, struct rte_mbuf **mbufs, uint16_t n_pkts, __attribute__((unused)) struct task_base *tbase)
+{
+	uint16_t ntx;
+	int ret;
+
+	/* TX vector mode can't transmit more than 32 packets */
+	if (n_pkts > MAX_PMD_TX) {
+		ntx = rte_eth_tx_burst(port_queue->port, port_queue->queue, mbufs, MAX_PMD_TX);
+		ntx += rte_eth_tx_burst(port_queue->port, port_queue->queue, mbufs + ntx, n_pkts - ntx);
+	} else {
+		ntx = rte_eth_tx_burst(port_queue->port, port_queue->queue, mbufs, n_pkts);
+	}
+
+	TASK_STATS_ADD_TX(&tbase->aux->stats, ntx);
+	ret =  n_pkts - ntx;
+	if (ntx < n_pkts) {
+		TASK_STATS_ADD_DROP_TX_FAIL(&tbase->aux->stats, n_pkts - ntx);
+		if (tbase->tx_pkt == tx_pkt_bw) {
+			uint32_t drop_bytes = 0;
+			do {
+				drop_bytes += mbuf_wire_size(mbufs[ntx]);
+				rte_pktmbuf_free(mbufs[ntx++]);
+			} while (ntx < n_pkts);
+			TASK_STATS_ADD_DROP_BYTES(&tbase->aux->stats, drop_bytes);
+		}
+		else {
+			do {
+				rte_pktmbuf_free(mbufs[ntx++]);
+			} while (ntx < n_pkts);
+		}
+	}
+	return ret;
+}
+
+static inline int txhw_no_drop(const struct port_queue *port_queue, struct rte_mbuf **mbufs, uint16_t n_pkts, __attribute__((unused)) struct task_base *tbase)
+{
+	uint16_t ret;
+	uint16_t n = n_pkts;
+
+	TASK_STATS_ADD_TX(&tbase->aux->stats, n_pkts);
+
+	do {
+		ret = rte_eth_tx_burst(port_queue->port, port_queue->queue, mbufs, n_pkts);
+		mbufs += ret;
+		n_pkts -= ret;
+	}
+	while (n_pkts);
+	return (n != ret);
+}
+
+static inline int ring_enq_drop(struct rte_ring *ring, struct rte_mbuf *const *mbufs, uint16_t n_pkts, __attribute__((unused)) struct task_base *tbase)
+{
+	int ret = 0;
+	/* return 0 on succes, -ENOBUFS on failure */
+	// Rings can be single or multiproducer (ctrl rings are multi producer)
+#if RTE_VERSION < RTE_VERSION_NUM(17,5,0,1)
+	if (unlikely(rte_ring_enqueue_bulk(ring, (void *const *)mbufs, n_pkts))) {
+#else
+	if (unlikely(rte_ring_enqueue_bulk(ring, (void *const *)mbufs, n_pkts, NULL) == 0)) {
+#endif
+		ret = n_pkts;
+		if (tbase->tx_pkt == tx_pkt_bw) {
+			uint32_t drop_bytes = 0;
+			for (uint16_t i = 0; i < n_pkts; ++i) {
+				drop_bytes += mbuf_wire_size(mbufs[i]);
+				rte_pktmbuf_free(mbufs[i]);
+			}
+			TASK_STATS_ADD_DROP_BYTES(&tbase->aux->stats, drop_bytes);
+			TASK_STATS_ADD_DROP_TX_FAIL(&tbase->aux->stats, n_pkts);
+		}
+		else {
+			for (uint16_t i = 0; i < n_pkts; ++i)
+				rte_pktmbuf_free(mbufs[i]);
+			TASK_STATS_ADD_DROP_TX_FAIL(&tbase->aux->stats, n_pkts);
+		}
+	}
+	else {
+		TASK_STATS_ADD_TX(&tbase->aux->stats, n_pkts);
+	}
+	return ret;
+}
+
+static inline int ring_enq_no_drop(struct rte_ring *ring, struct rte_mbuf *const *mbufs, uint16_t n_pkts, __attribute__((unused)) struct task_base *tbase)
+{
+	int i = 0;
+#if RTE_VERSION < RTE_VERSION_NUM(17,5,0,1)
+	while (rte_ring_enqueue_bulk(ring, (void *const *)mbufs, n_pkts)) {
+#else
+	while (rte_ring_enqueue_bulk(ring, (void *const *)mbufs, n_pkts, NULL) == 0) {
+#endif
+		i++;
+	};
+	TASK_STATS_ADD_TX(&tbase->aux->stats, n_pkts);
+	return (i != 0);
+}
+
+void flush_queues_hw(struct task_base *tbase)
+{
+	uint16_t prod, cons;
+
+	for (uint8_t i = 0; i < tbase->tx_params_hw.nb_txports; ++i) {
+		prod = tbase->ws_mbuf->idx[i].prod;
+		cons = tbase->ws_mbuf->idx[i].cons;
+
+		if (prod != cons) {
+			tbase->ws_mbuf->idx[i].prod = 0;
+			tbase->ws_mbuf->idx[i].cons = 0;
+			txhw_drop(&tbase->tx_params_hw.tx_port_queue[i], tbase->ws_mbuf->mbuf[i] + (cons & WS_MBUF_MASK), prod - cons, tbase);
+		}
+	}
+
+	tbase->flags &= ~FLAG_TX_FLUSH;
+}
+
+void flush_queues_sw(struct task_base *tbase)
+{
+	uint16_t prod, cons;
+
+	for (uint8_t i = 0; i < tbase->tx_params_sw.nb_txrings; ++i) {
+		prod = tbase->ws_mbuf->idx[i].prod;
+		cons = tbase->ws_mbuf->idx[i].cons;
+
+		if (prod != cons) {
+			tbase->ws_mbuf->idx[i].prod = 0;
+			tbase->ws_mbuf->idx[i].cons = 0;
+			ring_enq_drop(tbase->tx_params_sw.tx_rings[i], tbase->ws_mbuf->mbuf[i] + (cons & WS_MBUF_MASK), prod - cons, tbase);
+		}
+	}
+	tbase->flags &= ~FLAG_TX_FLUSH;
+}
+
+void flush_queues_no_drop_hw(struct task_base *tbase)
+{
+	uint16_t prod, cons;
+
+	for (uint8_t i = 0; i < tbase->tx_params_hw.nb_txports; ++i) {
+		prod = tbase->ws_mbuf->idx[i].prod;
+		cons = tbase->ws_mbuf->idx[i].cons;
+
+		if (prod != cons) {
+			tbase->ws_mbuf->idx[i].prod = 0;
+			tbase->ws_mbuf->idx[i].cons = 0;
+			txhw_no_drop(&tbase->tx_params_hw.tx_port_queue[i], tbase->ws_mbuf->mbuf[i] + (cons & WS_MBUF_MASK), prod - cons, tbase);
+		}
+	}
+
+	tbase->flags &= ~FLAG_TX_FLUSH;
+}
+
+void flush_queues_no_drop_sw(struct task_base *tbase)
+{
+	uint16_t prod, cons;
+
+	for (uint8_t i = 0; i < tbase->tx_params_sw.nb_txrings; ++i) {
+		prod = tbase->ws_mbuf->idx[i].prod;
+		cons = tbase->ws_mbuf->idx[i].cons;
+
+		if (prod != cons) {
+			tbase->ws_mbuf->idx[i].prod = 0;
+			tbase->ws_mbuf->idx[i].cons = 0;
+			ring_enq_no_drop(tbase->tx_params_sw.tx_rings[i], tbase->ws_mbuf->mbuf[i] + (cons & WS_MBUF_MASK), prod - cons, tbase);
+		}
+	}
+	tbase->flags &= ~FLAG_TX_FLUSH;
+}
+
+/* "try" functions try to send packets to sw/hw w/o failing or blocking;
+   They return if ring/queue is full and are used by aggregators.
+   "try" functions do not have drop/no drop flavors
+   They are only implemented in never_discard mode (as by default they
+   use only one outgoing ring. */
+uint16_t tx_try_self(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts)
+{
+	if (n_pkts < 64) {
+		tx_pkt_never_discard_self(tbase, mbufs, n_pkts, NULL);
+		return n_pkts;
+	} else {
+		tx_pkt_never_discard_self(tbase, mbufs, 64, NULL);
+		return 64;
+	}
+}
+
+uint16_t tx_try_sw1(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts)
+{
+	const int bulk_size = 64;
+	uint16_t ret = bulk_size, sent = 0, n_bulks;
+	n_bulks = n_pkts >> __builtin_ctz(bulk_size);
+
+	for (int i = 0; i < n_bulks; i++) {
+#if RTE_VERSION < RTE_VERSION_NUM(17,5,0,1)
+		ret = rte_ring_enqueue_burst(tbase->tx_params_sw.tx_rings[0], (void *const *)mbufs, bulk_size);
+#else
+		ret = rte_ring_enqueue_burst(tbase->tx_params_sw.tx_rings[0], (void *const *)mbufs, bulk_size, NULL);
+#endif
+		mbufs += ret;
+		sent += ret;
+		if (ret != bulk_size)
+			break;
+	}
+	if ((ret == bulk_size) && (n_pkts & (bulk_size - 1))) {
+#if RTE_VERSION < RTE_VERSION_NUM(17,5,0,1)
+		ret = rte_ring_enqueue_burst(tbase->tx_params_sw.tx_rings[0], (void *const *)mbufs, (n_pkts & (bulk_size - 1)));
+#else
+		ret = rte_ring_enqueue_burst(tbase->tx_params_sw.tx_rings[0], (void *const *)mbufs, (n_pkts & (bulk_size - 1)), NULL);
+#endif
+		mbufs += ret;
+		sent += ret;
+	}
+	TASK_STATS_ADD_TX(&tbase->aux->stats, sent);
+	return sent;
+}
+
+uint16_t tx_try_hw1(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts)
+{
+	const struct port_queue *port_queue = &tbase->tx_params_hw.tx_port_queue[0];
+	const int bulk_size = 64;
+	uint16_t ret = bulk_size, n_bulks, sent = 0;
+	n_bulks = n_pkts >>  __builtin_ctz(bulk_size);
+
+	for (int i = 0; i < n_bulks; i++) {
+		ret = rte_eth_tx_burst(port_queue->port, port_queue->queue, mbufs, bulk_size);
+		mbufs += ret;
+		sent += ret;
+		if (ret != bulk_size)
+			break;
+	}
+	if ((ret == bulk_size) && (n_pkts & (bulk_size - 1))) {
+		ret = rte_eth_tx_burst(port_queue->port, port_queue->queue, mbufs, (n_pkts & (bulk_size - 1)));
+		mbufs += ret;
+		sent += ret;
+	}
+	TASK_STATS_ADD_TX(&tbase->aux->stats, sent);
+	return sent;
+}
+
+int tx_pkt_no_drop_never_discard_hw1_lat_opt(struct task_base *tbase, struct rte_mbuf **mbufs, const uint16_t n_pkts, __attribute__((unused)) uint8_t *out)
+{
+	return txhw_no_drop(&tbase->tx_params_hw.tx_port_queue[0], mbufs, n_pkts, tbase);
+}
+
+int tx_pkt_no_drop_never_discard_hw1_thrpt_opt(struct task_base *tbase, struct rte_mbuf **mbufs, const uint16_t n_pkts, __attribute__((unused)) uint8_t *out)
+{
+	static uint8_t fake_out[MAX_PKT_BURST] = {0};
+	int ret = 0;
+	if (n_pkts == MAX_PKT_BURST) {
+		// First xmit what was queued
+        	uint16_t prod, cons;
+
+               	prod = tbase->ws_mbuf->idx[0].prod;
+               	cons = tbase->ws_mbuf->idx[0].cons;
+
+		if ((uint16_t)(prod - cons)){
+                	tbase->flags &= ~FLAG_TX_FLUSH;
+                	tbase->ws_mbuf->idx[0].prod = 0;
+                	tbase->ws_mbuf->idx[0].cons = 0;
+                	ret+= txhw_no_drop(&tbase->tx_params_hw.tx_port_queue[0], tbase->ws_mbuf->mbuf[0] + (cons & WS_MBUF_MASK), (uint16_t)(prod - cons), tbase);
+		}
+		ret+= txhw_no_drop(&tbase->tx_params_hw.tx_port_queue[0], mbufs, n_pkts, tbase);
+	} else {
+		ret+= tx_pkt_no_drop_hw(tbase, mbufs, n_pkts, fake_out);
+	}
+	return ret;
+}
+
+int tx_pkt_never_discard_hw1_lat_opt(struct task_base *tbase, struct rte_mbuf **mbufs, const uint16_t n_pkts, __attribute__((unused)) uint8_t *out)
+{
+	return txhw_drop(&tbase->tx_params_hw.tx_port_queue[0], mbufs, n_pkts, tbase);
+}
+
+int tx_pkt_never_discard_hw1_thrpt_opt(struct task_base *tbase, struct rte_mbuf **mbufs, const uint16_t n_pkts, __attribute__((unused)) uint8_t *out)
+{
+	static uint8_t fake_out[MAX_PKT_BURST] = {0};
+	int ret = 0;
+	if (n_pkts == MAX_PKT_BURST) {
+		// First xmit what was queued
+        	uint16_t prod, cons;
+
+               	prod = tbase->ws_mbuf->idx[0].prod;
+               	cons = tbase->ws_mbuf->idx[0].cons;
+
+		if ((uint16_t)(prod - cons)){
+                	tbase->flags &= ~FLAG_TX_FLUSH;
+                	tbase->ws_mbuf->idx[0].prod = 0;
+                	tbase->ws_mbuf->idx[0].cons = 0;
+                	ret+= txhw_drop(&tbase->tx_params_hw.tx_port_queue[0], tbase->ws_mbuf->mbuf[0] + (cons & WS_MBUF_MASK), (uint16_t)(prod - cons), tbase);
+		}
+		ret+= txhw_drop(&tbase->tx_params_hw.tx_port_queue[0], mbufs, n_pkts, tbase);
+	} else {
+		ret+= tx_pkt_hw(tbase, mbufs, n_pkts, fake_out);
+	}
+	return ret;
+}
+
+/* Transmit to hw using tx_params_hw_sw structure
+   This function is used  to transmit to hw when tx_params_hw_sw should be used
+   i.e. when the task needs to transmit both to hw and sw */
+int tx_pkt_no_drop_never_discard_hw1_no_pointer(struct task_base *tbase, struct rte_mbuf **mbufs, const uint16_t n_pkts, __attribute__((unused)) uint8_t *out)
+{
+	txhw_no_drop(&tbase->tx_params_hw_sw.tx_port_queue, mbufs, n_pkts, tbase);
+	return 0;
+}
+
+int tx_pkt_no_drop_never_discard_sw1(struct task_base *tbase, struct rte_mbuf **mbufs, const uint16_t n_pkts, __attribute__((unused)) uint8_t *out)
+{
+	return ring_enq_no_drop(tbase->tx_params_sw.tx_rings[0], mbufs, n_pkts, tbase);
+}
+
+int tx_pkt_never_discard_sw1(struct task_base *tbase, struct rte_mbuf **mbufs, const uint16_t n_pkts, __attribute__((unused)) uint8_t *out)
+{
+	return ring_enq_drop(tbase->tx_params_sw.tx_rings[0], mbufs, n_pkts, tbase);
+}
+
+static uint16_t tx_pkt_free_dropped(__attribute__((unused)) struct task_base *tbase, struct rte_mbuf **mbufs, const uint16_t n_pkts, uint8_t *out)
+{
+	uint64_t v = 0;
+	uint16_t i;
+	/* The most probable and most important optimize case is if
+	   the no packets should be dropped. */
+	for (i = 0; i + 8 < n_pkts; i += 8) {
+		v |= *((uint64_t*)(&out[i]));
+	}
+	for (; i < n_pkts; ++i) {
+		v |= out[i];
+	}
+
+	if (unlikely(v)) {
+		/* At least some packets need to be dropped, so the
+		   mbufs array needs to be updated. */
+		uint16_t n_kept = 0;
+		uint16_t n_discard = 0;
+		for (uint16_t i = 0; i < n_pkts; ++i) {
+			if (unlikely(out[i] >= OUT_HANDLED)) {
+				rte_pktmbuf_free(mbufs[i]);
+				n_discard += out[i] == OUT_DISCARD;
+				continue;
+			}
+			mbufs[n_kept++] = mbufs[i];
+		}
+		TASK_STATS_ADD_DROP_DISCARD(&tbase->aux->stats, n_discard);
+		TASK_STATS_ADD_DROP_HANDLED(&tbase->aux->stats, n_pkts - n_kept - n_discard);
+		return n_kept;
+	}
+	return n_pkts;
+}
+
+int tx_pkt_no_drop_hw1(struct task_base *tbase, struct rte_mbuf **mbufs, const uint16_t n_pkts, uint8_t *out)
+{
+	const uint16_t n_kept = tx_pkt_free_dropped(tbase, mbufs, n_pkts, out);
+	int ret = 0;
+
+	if (likely(n_kept))
+		ret = txhw_no_drop(&tbase->tx_params_hw.tx_port_queue[0], mbufs, n_kept, tbase);
+	return ret;
+}
+
+int tx_pkt_no_drop_sw1(struct task_base *tbase, struct rte_mbuf **mbufs, const uint16_t n_pkts, uint8_t *out)
+{
+	const uint16_t n_kept = tx_pkt_free_dropped(tbase, mbufs, n_pkts, out);
+	int ret = 0;
+
+	if (likely(n_kept))
+		ret = ring_enq_no_drop(tbase->tx_params_sw.tx_rings[0], mbufs, n_kept, tbase);
+	return ret;
+}
+
+int tx_pkt_hw1(struct task_base *tbase, struct rte_mbuf **mbufs, const uint16_t n_pkts, uint8_t *out)
+{
+	const uint16_t n_kept = tx_pkt_free_dropped(tbase, mbufs, n_pkts, out);
+
+	if (likely(n_kept))
+		return txhw_drop(&tbase->tx_params_hw.tx_port_queue[0], mbufs, n_kept, tbase);
+	return n_pkts;
+}
+
+int tx_pkt_sw1(struct task_base *tbase, struct rte_mbuf **mbufs, const uint16_t n_pkts, uint8_t *out)
+{
+	const uint16_t n_kept = tx_pkt_free_dropped(tbase, mbufs, n_pkts, out);
+
+	if (likely(n_kept))
+		return ring_enq_drop(tbase->tx_params_sw.tx_rings[0], mbufs, n_kept, tbase);
+	return 0;
+}
+
+int tx_pkt_self(struct task_base *tbase, struct rte_mbuf **mbufs, const uint16_t n_pkts, uint8_t *out)
+{
+	const uint16_t n_kept = tx_pkt_free_dropped(tbase, mbufs, n_pkts, out);
+
+	TASK_STATS_ADD_TX(&tbase->aux->stats, n_kept);
+	tbase->ws_mbuf->idx[0].nb_rx = n_kept;
+	struct rte_mbuf **tx_mbuf = tbase->ws_mbuf->mbuf[0] + (tbase->ws_mbuf->idx[0].prod & WS_MBUF_MASK);
+	for (uint16_t i = 0; i < n_kept; ++i) {
+		tx_mbuf[i] = mbufs[i];
+	}
+	return 0;
+}
+
+int tx_pkt_never_discard_self(struct task_base *tbase, struct rte_mbuf **mbufs, const uint16_t n_pkts, __attribute__((unused)) uint8_t *out)
+{
+	TASK_STATS_ADD_TX(&tbase->aux->stats, n_pkts);
+	tbase->ws_mbuf->idx[0].nb_rx = n_pkts;
+	struct rte_mbuf **tx_mbuf = tbase->ws_mbuf->mbuf[0] + (tbase->ws_mbuf->idx[0].prod & WS_MBUF_MASK);
+	for (uint16_t i = 0; i < n_pkts; ++i) {
+		tx_mbuf[i] = mbufs[i];
+	}
+	return 0;
+}
+
+int tx_pkt_no_drop_hw(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts, uint8_t *out)
+{
+	int ret = 0;
+	buf_pkt_all(tbase, mbufs, n_pkts, out);
+
+	const uint8_t nb_bufs = tbase->tx_params_hw.nb_txports;
+	uint16_t prod, cons;
+
+	for (uint8_t i = 0; i < nb_bufs; ++i) {
+		prod = tbase->ws_mbuf->idx[i].prod;
+		cons = tbase->ws_mbuf->idx[i].cons;
+
+		if (((uint16_t)(prod - cons)) >= MAX_PKT_BURST) {
+			tbase->flags &= ~FLAG_TX_FLUSH;
+			tbase->ws_mbuf->idx[i].cons = cons + MAX_PKT_BURST;
+			ret+= txhw_no_drop(&tbase->tx_params_hw.tx_port_queue[i], tbase->ws_mbuf->mbuf[i] + (cons & WS_MBUF_MASK), MAX_PKT_BURST, tbase);
+		}
+	}
+	return ret;
+}
+
+int tx_pkt_no_drop_sw(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts, uint8_t *out)
+{
+	int ret = 0;
+	buf_pkt_all(tbase, mbufs, n_pkts, out);
+
+	const uint8_t nb_bufs = tbase->tx_params_sw.nb_txrings;
+	uint16_t prod, cons;
+
+	for (uint8_t i = 0; i < nb_bufs; ++i) {
+		prod = tbase->ws_mbuf->idx[i].prod;
+		cons = tbase->ws_mbuf->idx[i].cons;
+
+		if (((uint16_t)(prod - cons)) >= MAX_PKT_BURST) {
+			tbase->flags &= ~FLAG_TX_FLUSH;
+			tbase->ws_mbuf->idx[i].cons = cons + MAX_PKT_BURST;
+			ret += ring_enq_no_drop(tbase->tx_params_sw.tx_rings[i], tbase->ws_mbuf->mbuf[i] + (cons & WS_MBUF_MASK), MAX_PKT_BURST, tbase);
+		}
+	}
+	return ret;
+}
+
+int tx_pkt_hw(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts, uint8_t *out)
+{
+	int ret = 0;
+	buf_pkt_all(tbase, mbufs, n_pkts, out);
+
+	const uint8_t nb_bufs = tbase->tx_params_hw.nb_txports;
+	uint16_t prod, cons;
+
+	for (uint8_t i = 0; i < nb_bufs; ++i) {
+		prod = tbase->ws_mbuf->idx[i].prod;
+		cons = tbase->ws_mbuf->idx[i].cons;
+
+		if (((uint16_t)(prod - cons)) >= MAX_PKT_BURST) {
+			tbase->flags &= ~FLAG_TX_FLUSH;
+			tbase->ws_mbuf->idx[i].cons = cons + MAX_PKT_BURST;
+			ret += txhw_drop(&tbase->tx_params_hw.tx_port_queue[i], tbase->ws_mbuf->mbuf[i] + (cons & WS_MBUF_MASK), MAX_PKT_BURST, tbase);
+		}
+	}
+	return ret;
+}
+
+int tx_pkt_sw(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts, uint8_t *out)
+{
+	int ret = 0;
+	buf_pkt_all(tbase, mbufs, n_pkts, out);
+
+	const uint8_t nb_bufs = tbase->tx_params_sw.nb_txrings;
+	uint16_t prod, cons;
+	for (uint8_t i = 0; i < nb_bufs; ++i) {
+		prod = tbase->ws_mbuf->idx[i].prod;
+		cons = tbase->ws_mbuf->idx[i].cons;
+
+		if (((uint16_t)(prod - cons)) >= MAX_PKT_BURST) {
+			tbase->flags &= ~FLAG_TX_FLUSH;
+			tbase->ws_mbuf->idx[i].cons = cons + MAX_PKT_BURST;
+			ret+= ring_enq_drop(tbase->tx_params_sw.tx_rings[i], tbase->ws_mbuf->mbuf[i] + (cons & WS_MBUF_MASK), MAX_PKT_BURST, tbase);
+		}
+	}
+	return ret;
+}
+
+int tx_pkt_trace(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts, uint8_t *out)
+{
+	int ret = 0;
+	if (tbase->aux->task_rt_dump.cur_trace == 0) {
+		// No packet received since dumping...
+		// So the transmitted packets should not be linked to received packets
+		tbase->aux->task_rt_dump.n_print_tx = tbase->aux->task_rt_dump.n_trace;
+		tbase->aux->task_rt_dump.n_trace = 0;
+		task_base_del_rx_pkt_function(tbase, rx_pkt_trace);
+		return tx_pkt_dump(tbase, mbufs, n_pkts, out);
+	}
+	plog_info("Tracing %d pkts\n", tbase->aux->task_rt_dump.cur_trace);
+
+	for (uint32_t i = 0; i < tbase->aux->task_rt_dump.cur_trace; ++i) {
+		struct rte_mbuf tmp;
+		/* For each packet being transmitted, find which
+		   buffer represent the packet as it was before
+		   processing. */
+		uint32_t j = 0;
+		uint32_t len = sizeof(tbase->aux->task_rt_dump.pkt_mbuf_addr)/sizeof(tbase->aux->task_rt_dump.pkt_mbuf_addr[0]);
+		for (;j < len; ++j) {
+			if (tbase->aux->task_rt_dump.pkt_mbuf_addr[j] == mbufs[i])
+				break;
+		}
+		if (j == len) {
+			plog_info("Trace RX: missing!\n");
+		}
+		else {
+#if RTE_VERSION >= RTE_VERSION_NUM(1,8,0,0)
+			tmp.data_off = 0;
+#endif
+			rte_pktmbuf_data_len(&tmp) = tbase->aux->task_rt_dump.pkt_cpy_len[j];
+			rte_pktmbuf_pkt_len(&tmp) = tbase->aux->task_rt_dump.pkt_cpy_len[j];
+			tmp.buf_addr = tbase->aux->task_rt_dump.pkt_cpy[j];
+			plogd_info(&tmp, "Trace RX: ");
+		}
+
+		if (out) {
+			if (out[i] != 0xFF)
+				plogd_info(mbufs[i], "Trace TX[%d]: ", out[i]);
+			else
+				plogd_info(mbufs[i], "Trace Dropped: ");
+		} else
+			plogd_info(mbufs[i], "Trace TX: ");
+	}
+	ret = tbase->aux->tx_pkt_orig(tbase, mbufs, n_pkts, out);
+
+	/* Unset by TX when n_trace = 0 */
+	if (0 == tbase->aux->task_rt_dump.n_trace) {
+		tbase->tx_pkt = tbase->aux->tx_pkt_orig;
+		tbase->aux->tx_pkt_orig = NULL;
+		task_base_del_rx_pkt_function(tbase, rx_pkt_trace);
+	}
+	return ret;
+}
+
+int tx_pkt_dump(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts, uint8_t *out)
+{
+	uint32_t n_dump = tbase->aux->task_rt_dump.n_print_tx;
+	int ret = 0;
+
+	n_dump = n_pkts < n_dump? n_pkts : n_dump;
+	for (uint32_t i = 0; i < n_dump; ++i) {
+		if (out)
+			plogd_info(mbufs[i], "TX[%d]: ", out[i]);
+		else
+			plogd_info(mbufs[i], "TX: ");
+	}
+	tbase->aux->task_rt_dump.n_print_tx -= n_dump;
+
+	ret = tbase->aux->tx_pkt_orig(tbase, mbufs, n_pkts, out);
+
+	if (0 == tbase->aux->task_rt_dump.n_print_tx) {
+		tbase->tx_pkt = tbase->aux->tx_pkt_orig;
+		tbase->aux->tx_pkt_orig = NULL;
+	}
+	return ret;
+}
+
+/* Gather the distribution of the number of packets that have been
+   xmitted from one TX call. Since the value is only modified by the
+   task that xmits the packet, no atomic operation is needed. */
+int tx_pkt_distr(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts, uint8_t *out)
+{
+	tbase->aux->tx_bucket[n_pkts]++;
+	return tbase->aux->tx_pkt_orig(tbase, mbufs, n_pkts, out);
+}
+
+int tx_pkt_bw(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts, uint8_t *out)
+{
+	uint32_t tx_bytes = 0;
+	uint32_t drop_bytes = 0;
+
+	for (uint16_t i = 0; i < n_pkts; ++i) {
+		if (!out || out[i] < OUT_HANDLED)
+			tx_bytes += mbuf_wire_size(mbufs[i]);
+		else
+			drop_bytes += mbuf_wire_size(mbufs[i]);
+	}
+
+	TASK_STATS_ADD_TX_BYTES(&tbase->aux->stats, tx_bytes);
+	TASK_STATS_ADD_DROP_BYTES(&tbase->aux->stats, drop_bytes);
+	return tbase->aux->tx_pkt_orig(tbase, mbufs, n_pkts, out);
+}
+
+int tx_pkt_drop_all(struct task_base *tbase, struct rte_mbuf **mbufs, uint16_t n_pkts, uint8_t *out)
+{
+	for (uint16_t j = 0; j < n_pkts; ++j) {
+		rte_pktmbuf_free(mbufs[j]);
+	}
+	if (out == NULL)
+		TASK_STATS_ADD_DROP_HANDLED(&tbase->aux->stats, n_pkts);
+	else {
+		for (uint16_t j = 0; j < n_pkts; ++j) {
+			if (out[j] == OUT_HANDLED)
+				TASK_STATS_ADD_DROP_HANDLED(&tbase->aux->stats, 1);
+			else
+				TASK_STATS_ADD_DROP_DISCARD(&tbase->aux->stats, 1);
+		}
+	}
+	return n_pkts;
+}