[dpdk-dev] [PATCH v3] port: Eventdev port support added in the rte_port library

[Rahul Shah <rahul.r.shah@intel.com>]

Adding a new port type called eventdev to the
rte_port library.

Signed-off-by: Rahul Shah <rahul.r.shah@intel.com>
---
 lib/librte_port/Makefile             |   4 +-
 lib/librte_port/meson.build          |   8 +-
 lib/librte_port/rte_port_eventdev.c  | 585 +++++++++++++++++++++++++++
 lib/librte_port/rte_port_eventdev.h  | 100 +++++
 lib/librte_port/rte_port_version.map |   9 +
 5 files changed, 702 insertions(+), 4 deletions(-)
 create mode 100644 lib/librte_port/rte_port_eventdev.c
 create mode 100644 lib/librte_port/rte_port_eventdev.h

diff --git a/lib/librte_port/Makefile b/lib/librte_port/Makefile
index 1b83f6f2b..de6f0428a 100644
--- a/lib/librte_port/Makefile
+++ b/lib/librte_port/Makefile
@@ -11,7 +11,7 @@ ifeq ($(CONFIG_RTE_PORT_PCAP),y)
 LDLIBS += -lpcap
 endif
 LDLIBS += -lrte_eal -lrte_mempool -lrte_mbuf -lrte_ethdev
-LDLIBS += -lrte_ip_frag -lrte_sched -lrte_cryptodev
+LDLIBS += -lrte_ip_frag -lrte_sched -lrte_cryptodev -lrte_eventdev
 ifeq ($(CONFIG_RTE_LIBRTE_KNI),y)
 LDLIBS += -lrte_kni
 endif
@@ -39,6 +39,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_PORT) += rte_port_kni.c
 endif
 SRCS-$(CONFIG_RTE_LIBRTE_PORT) += rte_port_source_sink.c
 SRCS-$(CONFIG_RTE_LIBRTE_PORT) += rte_port_sym_crypto.c
+SRCS-$(CONFIG_RTE_LIBRTE_PORT) += rte_port_eventdev.c
 
 # install includes
 SYMLINK-$(CONFIG_RTE_LIBRTE_PORT)-include += rte_port.h
@@ -55,5 +56,6 @@ SYMLINK-$(CONFIG_RTE_LIBRTE_PORT)-include += rte_port_kni.h
 endif
 SYMLINK-$(CONFIG_RTE_LIBRTE_PORT)-include += rte_port_source_sink.h
 SYMLINK-$(CONFIG_RTE_LIBRTE_PORT)-include += rte_port_sym_crypto.h
+SYMLINK-$(CONFIG_RTE_LIBRTE_PORT)-include += rte_port_eventdev.h
 
 include $(RTE_SDK)/mk/rte.lib.mk
diff --git a/lib/librte_port/meson.build b/lib/librte_port/meson.build
index 0d11456f0..ae5f3b7d0 100644
--- a/lib/librte_port/meson.build
+++ b/lib/librte_port/meson.build
@@ -10,7 +10,8 @@ sources = files(
 	'rte_port_ring.c',
 	'rte_port_sched.c',
 	'rte_port_source_sink.c',
-	'rte_port_sym_crypto.c')
+	'rte_port_sym_crypto.c',
+	'rte_port_eventdev.c')
 headers = files(
 	'rte_port_ethdev.h',
 	'rte_port_fd.h',
@@ -20,8 +21,9 @@ headers = files(
 	'rte_port_ring.h',
 	'rte_port_sched.h',
 	'rte_port_source_sink.h',
-	'rte_port_sym_crypto.h')
-deps += ['ethdev', 'sched', 'ip_frag', 'cryptodev']
+	'rte_port_sym_crypto.h',
+	'rte_port_eventdev.h')
+deps += ['ethdev', 'sched', 'ip_frag', 'cryptodev', 'eventdev']
 
 if dpdk_conf.has('RTE_LIBRTE_KNI')
 	sources += files('rte_port_kni.c')
diff --git a/lib/librte_port/rte_port_eventdev.c b/lib/librte_port/rte_port_eventdev.c
new file mode 100644
index 000000000..4feb24863
--- /dev/null
+++ b/lib/librte_port/rte_port_eventdev.c
@@ -0,0 +1,585 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2010-2014 Intel Corporation
+ */
+
+ #include <string.h>
+#include <stdint.h>
+
+#include <rte_mbuf.h>
+#include <rte_malloc.h>
+
+#include "rte_port_eventdev.h"
+
+/*
+ * Port EVENTDEV Reader
+ */
+#ifdef RTE_PORT_STATS_COLLECT
+
+#define RTE_PORT_EVENTDEV_READER_STATS_PKTS_IN_ADD(port, val) \
+	do {port->stats.n_pkts_in += val} while (0)
+#define RTE_PORT_EVENTDEV_READER_STATS_PKTS_DROP_ADD(port, val) \
+	do {port->stats.n_pkts_drop += val} while (0)
+
+#else
+
+#define RTE_PORT_EVENTDEV_READER_STATS_PKTS_IN_ADD(port, val)
+#define RTE_PORT_EVENTDEV_READER_STATS_PKTS_DROP_ADD(port, val)
+
+#endif
+
+struct rte_port_eventdev_reader {
+	struct rte_port_in_stats stats;
+
+	uint8_t  eventdev_id;
+	uint16_t port_id;
+
+	struct rte_event ev[RTE_PORT_IN_BURST_SIZE_MAX];
+};
+
+static void *
+rte_port_eventdev_reader_create(void *params, int socket_id)
+{
+	struct rte_port_eventdev_reader_params *conf =
+			params;
+	struct rte_port_eventdev_reader *port;
+
+	/* Check input parameters */
+	if (conf == NULL) {
+		RTE_LOG(ERR, PORT, "%s: params is NULL\n", __func__);
+		return NULL;
+	}
+
+	/* Memory allocation */
+	port = rte_zmalloc_socket("PORT", sizeof(*port),
+		RTE_CACHE_LINE_SIZE, socket_id);
+	if (port == NULL) {
+		RTE_LOG(ERR, PORT, "%s: Failed to allocate port\n", __func__);
+		return NULL;
+	}
+
+	/* Initialization */
+	port->eventdev_id = conf->eventdev_id;
+	port->port_id = conf->port_id;
+
+	return port;
+}
+
+static int
+rte_port_eventdev_reader_rx(void *port, struct rte_mbuf **pkts, uint32_t n_pkts)
+{
+	struct rte_port_eventdev_reader *p = port;
+	uint16_t rx_evts_cnt, i;
+
+	rx_evts_cnt = rte_event_dequeue_burst(p->eventdev_id, p->port_id,
+			p->ev, n_pkts, 0);
+
+	for (i = 0; i < rx_evts_cnt; i++)
+		pkts[i] = p->ev[i].mbuf;
+
+	RTE_PORT_EVENTDEV_READER_STATS_PKTS_IN_ADD(p, rx_evts_cnt);
+
+	return rx_evts_cnt;
+}
+
+static int
+rte_port_eventdev_reader_free(void *port)
+{
+	if (port == NULL) {
+		RTE_LOG(ERR, PORT, "%s: port is NULL\n", __func__);
+		return -EINVAL;
+	}
+
+	rte_free(port);
+
+	return 0;
+}
+
+static int rte_port_eventdev_reader_stats_read(void *port,
+	struct rte_port_in_stats *stats, int clear)
+{
+	struct rte_port_eventdev_reader *p =
+			port;
+
+	if (stats != NULL)
+		memcpy(stats, &p->stats, sizeof(p->stats));
+
+	if (clear)
+		memset(&p->stats, 0, sizeof(p->stats));
+
+	return 0;
+}
+
+/*
+ * Port EVENTDEV Writer
+ */
+#ifdef RTE_PORT_STATS_COLLECT
+
+#define RTE_PORT_EVENTDEV_WRITER_STATS_PKTS_IN_ADD(port, val) \
+	do {port->stats.n_pkts_in += val} while (0)
+#define RTE_PORT_EVENTDEV_WRITER_STATS_PKTS_DROP_ADD(port, val) \
+	do {port->stats.n_pkts_drop += val} while (0)
+
+#else
+
+#define RTE_PORT_EVENTDEV_WRITER_STATS_PKTS_IN_ADD(port, val)
+#define RTE_PORT_EVENTDEV_WRITER_STATS_PKTS_DROP_ADD(port, val)
+
+#endif
+
+struct rte_port_eventdev_writer {
+	struct rte_port_out_stats stats;
+
+	struct rte_event ev[2 * RTE_PORT_IN_BURST_SIZE_MAX];
+
+	uint32_t enq_burst_sz;
+	uint32_t enq_buf_count;
+	uint64_t bsz_mask;
+
+	uint8_t eventdev_id;
+	uint8_t port_id;
+	uint8_t queue_id;
+	uint8_t sched_type;
+	uint8_t evt_op;
+};
+
+static void *
+rte_port_eventdev_writer_create(void *params, int socket_id)
+{
+	struct rte_port_eventdev_writer_params *conf =
+			params;
+	struct rte_port_eventdev_writer *port;
+	unsigned int i;
+
+	/* Check input parameters */
+	if ((conf == NULL) ||
+		(conf->enq_burst_sz == 0) ||
+		(conf->enq_burst_sz > RTE_PORT_IN_BURST_SIZE_MAX) ||
+		(!rte_is_power_of_2(conf->enq_burst_sz))) {
+		RTE_LOG(ERR, PORT, "%s: Invalid input parameters\n", __func__);
+		return NULL;
+	}
+
+	/* Memory allocation */
+	port = rte_zmalloc_socket("PORT", sizeof(*port),
+		RTE_CACHE_LINE_SIZE, socket_id);
+	if (port == NULL) {
+		RTE_LOG(ERR, PORT, "%s: Failed to allocate port\n", __func__);
+		return NULL;
+	}
+
+	/* Initialization */
+	port->enq_burst_sz = conf->enq_burst_sz;
+	port->enq_buf_count = 0;
+	port->bsz_mask = 1LLU << (conf->enq_burst_sz - 1);
+
+	port->eventdev_id = conf->eventdev_id;
+	port->port_id = conf->port_id;
+	port->queue_id = conf->queue_id;
+	port->sched_type = conf->sched_type;
+	port->evt_op = conf->evt_op;
+	memset(&port->ev, 0, sizeof(port->ev));
+
+	for (i = 0; i < ARRAY_SIZE(port->ev); i++) {
+		port->ev[i].queue_id = port->queue_id;
+		port->ev[i].sched_type = port->sched_type;
+		port->ev[i].op = port->evt_op;
+	}
+
+	return port;
+}
+
+static inline void
+send_burst(struct rte_port_eventdev_writer *p)
+{
+	uint32_t nb_enq;
+
+	nb_enq = rte_event_enqueue_burst(p->eventdev_id, p->port_id,
+			p->ev, p->enq_buf_count);
+
+	RTE_PORT_EVENTDEV_WRITER_STATS_PKTS_DROP_ADD(p, p->enq_buf_count -
+			nb_enq);
+
+	for (; nb_enq < p->enq_buf_count; nb_enq++)
+		rte_pktmbuf_free(p->ev[nb_enq].mbuf);
+
+	p->enq_buf_count = 0;
+}
+
+static int
+rte_port_eventdev_writer_tx(void *port, struct rte_mbuf *pkt)
+{
+	struct rte_port_eventdev_writer *p = port;
+
+	p->ev[p->enq_buf_count++].mbuf  = pkt;
+	RTE_PORT_EVENTDEV_WRITER_STATS_PKTS_IN_ADD(p, 1);
+	if (p->enq_buf_count >= p->enq_burst_sz)
+		send_burst(p);
+
+	return 0;
+}
+
+static int
+rte_port_eventdev_writer_tx_bulk(void *port,
+	struct rte_mbuf **pkts,
+	uint64_t pkts_mask)
+{
+	struct rte_port_eventdev_writer *p =
+			port;
+	uint64_t bsz_mask = p->bsz_mask;
+	uint32_t enq_buf_count = p->enq_buf_count;
+	uint64_t expr = (pkts_mask & (pkts_mask + 1)) |
+					((pkts_mask & bsz_mask) ^ bsz_mask);
+
+	if (expr == 0) {
+		uint64_t n_pkts = __builtin_popcountll(pkts_mask);
+		uint32_t i, n_enq_ok;
+
+		if (enq_buf_count)
+			send_burst(p);
+
+		RTE_PORT_EVENTDEV_WRITER_STATS_PKTS_IN_ADD(p, n_pkts);
+
+		struct rte_event events[2 * RTE_PORT_IN_BURST_SIZE_MAX] = {};
+		for (i = 0; i < n_pkts; i++) {
+			events[i].mbuf = pkts[i];
+			events[i].queue_id = p->queue_id;
+			events[i].sched_type = p->sched_type;
+			events[i].op = p->evt_op;
+		}
+
+		n_enq_ok = rte_event_enqueue_burst(p->eventdev_id, p->port_id,
+				events, n_pkts);
+
+		RTE_PORT_EVENTDEV_WRITER_STATS_PKTS_DROP_ADD(p,
+				n_pkts - n_enq_ok);
+		for (; n_enq_ok < n_pkts; n_enq_ok++)
+			rte_pktmbuf_free(pkts[n_enq_ok]);
+
+	} else {
+		for (; pkts_mask;) {
+			uint32_t pkt_index = __builtin_ctzll(pkts_mask);
+			uint64_t pkt_mask = 1LLU << pkt_index;
+
+			p->ev[enq_buf_count++].mbuf = pkts[pkt_index];
+
+			RTE_PORT_EVENTDEV_WRITER_STATS_PKTS_IN_ADD(p, 1);
+			pkts_mask &= ~pkt_mask;
+		}
+
+		p->enq_buf_count = enq_buf_count;
+		if (enq_buf_count >= p->enq_burst_sz)
+			send_burst(p);
+	}
+
+	return 0;
+}
+
+static int
+rte_port_eventdev_writer_flush(void *port)
+{
+	struct rte_port_eventdev_writer *p =
+			port;
+
+	if (p->enq_buf_count > 0)
+		send_burst(p);
+
+	return 0;
+}
+
+static int
+rte_port_eventdev_writer_free(void *port)
+{
+	if (port == NULL) {
+		RTE_LOG(ERR, PORT, "%s: Port is NULL\n", __func__);
+		return -EINVAL;
+	}
+
+	rte_port_eventdev_writer_flush(port);
+	rte_free(port);
+
+	return 0;
+}
+
+static int rte_port_eventdev_writer_stats_read(void *port,
+	struct rte_port_out_stats *stats, int clear)
+{
+	struct rte_port_eventdev_writer *p =
+			port;
+
+	if (stats != NULL)
+		memcpy(stats, &p->stats, sizeof(p->stats));
+
+	if (clear)
+		memset(&p->stats, 0, sizeof(p->stats));
+
+	return 0;
+}
+
+/*
+ * Port EVENTDEV Writer Nodrop
+ */
+#ifdef RTE_PORT_STATS_COLLECT
+
+#define RTE_PORT_EVENTDEV_WRITER_NODROP_STATS_PKTS_IN_ADD(port, val) \
+	do {port->stats.n_pkts_in += val} while (0)
+#define RTE_PORT_EVENTDEV_WRITER_NODROP_STATS_PKTS_DROP_ADD(port, val) \
+	do {port->stats.n_pkts_drop += val} while (0)
+
+#else
+
+#define RTE_PORT_EVENTDEV_WRITER_NODROP_STATS_PKTS_IN_ADD(port, val)
+#define RTE_PORT_EVENTDEV_WRITER_NODROP_STATS_PKTS_DROP_ADD(port, val)
+
+#endif
+
+struct rte_port_eventdev_writer_nodrop {
+	struct rte_port_out_stats stats;
+
+	struct rte_event ev[2 * RTE_PORT_IN_BURST_SIZE_MAX];
+
+	uint32_t enq_burst_sz;
+	uint32_t enq_buf_count;
+	uint64_t bsz_mask;
+	uint64_t n_retries;
+	uint8_t eventdev_id;
+	uint8_t port_id;
+	uint8_t queue_id;
+	uint8_t sched_type;
+	uint8_t evt_op;
+};
+
+
+static void *
+rte_port_eventdev_writer_nodrop_create(void *params, int socket_id)
+{
+	struct rte_port_eventdev_writer_nodrop_params *conf =
+			params;
+	struct rte_port_eventdev_writer_nodrop *port;
+	unsigned int i;
+
+	/* Check input parameters */
+	if ((conf == NULL) ||
+		(conf->enq_burst_sz == 0) ||
+		(conf->enq_burst_sz > RTE_PORT_IN_BURST_SIZE_MAX) ||
+		(!rte_is_power_of_2(conf->enq_burst_sz))) {
+		RTE_LOG(ERR, PORT, "%s: Invalid input parameters\n", __func__);
+		return NULL;
+	}
+
+	/* Memory allocation */
+	port = rte_zmalloc_socket("PORT", sizeof(*port),
+		RTE_CACHE_LINE_SIZE, socket_id);
+	if (port == NULL) {
+		RTE_LOG(ERR, PORT, "%s: Failed to allocate port\n", __func__);
+		return NULL;
+	}
+
+	/* Initialization */
+	port->enq_burst_sz = conf->enq_burst_sz;
+	port->enq_buf_count = 0;
+	port->bsz_mask = 1LLU << (conf->enq_burst_sz - 1);
+
+	port->eventdev_id = conf->eventdev_id;
+	port->port_id = conf->port_id;
+	port->queue_id = conf->queue_id;
+	port->sched_type = conf->sched_type;
+	port->evt_op = conf->evt_op;
+	memset(&port->ev, 0, sizeof(port->ev));
+
+	for (i = 0; i < ARRAY_SIZE(port->ev); i++) {
+		port->ev[i].queue_id = port->queue_id;
+		port->ev[i].sched_type = port->sched_type;
+		port->ev[i].op = port->evt_op;
+	}
+	/*
+	 * When n_retries is 0 it means that we should wait for every event to
+	 * send no matter how many retries should it take. To limit number of
+	 * branches in fast path, we use UINT64_MAX instead of branching.
+	 */
+	port->n_retries = (conf->n_retries == 0) ? UINT64_MAX : conf->n_retries;
+
+	return port;
+}
+
+static inline void
+send_burst_nodrop(struct rte_port_eventdev_writer_nodrop *p)
+{
+	uint32_t nb_enq, i;
+
+	nb_enq = rte_event_enqueue_burst(p->eventdev_id, p->port_id,
+			p->ev, p->enq_buf_count);
+
+	/* We sent all the packets in a first try */
+	if (nb_enq >= p->enq_buf_count) {
+		p->enq_buf_count = 0;
+		return;
+	}
+
+	for (i = 0; i < p->n_retries; i++) {
+		nb_enq += rte_event_enqueue_burst(p->eventdev_id, p->port_id,
+							p->ev + nb_enq,
+							p->enq_buf_count - nb_enq);
+
+		/* We sent all the events in more than one try */
+		if (nb_enq >= p->enq_buf_count) {
+			p->enq_buf_count = 0;
+			return;
+		}
+	}
+	/* We didn't send the events in maximum allowed attempts */
+	RTE_PORT_EVENTDEV_WRITER_NODROP_STATS_PKTS_DROP_ADD(p,
+			p->enq_buf_count - nb_enq);
+	for (; nb_enq < p->enq_buf_count; nb_enq++)
+		rte_pktmbuf_free(p->ev[nb_enq].mbuf);
+
+	p->enq_buf_count = 0;
+}
+
+static int
+rte_port_eventdev_writer_nodrop_tx(void *port, struct rte_mbuf *pkt)
+{
+	struct rte_port_eventdev_writer_nodrop *p = port;
+
+	p->ev[p->enq_buf_count++].mbuf = pkt;
+
+	RTE_PORT_EVENTDEV_WRITER_NODROP_STATS_PKTS_IN_ADD(p, 1);
+	if (p->enq_buf_count >= p->enq_burst_sz)
+		send_burst_nodrop(p);
+
+	return 0;
+}
+
+static int
+rte_port_eventdev_writer_nodrop_tx_bulk(void *port,
+	struct rte_mbuf **pkts,
+	uint64_t pkts_mask)
+{
+	struct rte_port_eventdev_writer_nodrop *p =
+			port;
+
+	uint64_t bsz_mask = p->bsz_mask;
+	uint32_t enq_buf_count = p->enq_buf_count;
+	uint64_t expr = (pkts_mask & (pkts_mask + 1)) |
+					((pkts_mask & bsz_mask) ^ bsz_mask);
+
+	if (expr == 0) {
+		uint64_t n_pkts = __builtin_popcountll(pkts_mask);
+		uint32_t i, n_enq_ok;
+
+		if (enq_buf_count)
+			send_burst_nodrop(p);
+
+		RTE_PORT_EVENTDEV_WRITER_NODROP_STATS_PKTS_IN_ADD(p, n_pkts);
+
+		struct rte_event events[RTE_PORT_IN_BURST_SIZE_MAX] = {};
+
+		for (i = 0; i < n_pkts; i++) {
+			events[i].mbuf = pkts[i];
+			events[i].queue_id = p->queue_id;
+			events[i].sched_type = p->sched_type;
+			events[i].op = p->evt_op;
+		}
+
+		n_enq_ok = rte_event_enqueue_burst(p->eventdev_id, p->port_id,
+				events, n_pkts);
+
+		if (n_enq_ok >= n_pkts)
+			return 0;
+
+		/*
+		 * If we did not manage to enqueue all events in single burst,
+		 * move remaining events to the buffer and call send burst.
+		 */
+		for (; n_enq_ok < n_pkts; n_enq_ok++) {
+			struct rte_mbuf *pkt = pkts[n_enq_ok];
+			p->ev[p->enq_buf_count++].mbuf = pkt;
+		}
+		send_burst_nodrop(p);
+	} else {
+		for (; pkts_mask;) {
+			uint32_t pkt_index = __builtin_ctzll(pkts_mask);
+			uint64_t pkt_mask = 1LLU << pkt_index;
+
+			p->ev[enq_buf_count++].mbuf = pkts[pkt_index];
+
+			RTE_PORT_EVENTDEV_WRITER_STATS_PKTS_IN_ADD(p, 1);
+			pkts_mask &= ~pkt_mask;
+		}
+
+		p->enq_buf_count = enq_buf_count;
+		if (enq_buf_count >= p->enq_burst_sz)
+			send_burst_nodrop(p);
+	}
+
+	return 0;
+}
+
+static int
+rte_port_eventdev_writer_nodrop_flush(void *port)
+{
+	struct rte_port_eventdev_writer_nodrop *p =
+			port;
+
+	if (p->enq_buf_count > 0)
+		send_burst_nodrop(p);
+
+	return 0;
+}
+
+static int
+rte_port_eventdev_writer_nodrop_free(void *port)
+{
+	if (port == NULL) {
+		RTE_LOG(ERR, PORT, "%s: Port is NULL\n", __func__);
+		return -EINVAL;
+	}
+
+	rte_port_eventdev_writer_nodrop_flush(port);
+	rte_free(port);
+
+	return 0;
+}
+
+static int rte_port_eventdev_writer_nodrop_stats_read(void *port,
+	struct rte_port_out_stats *stats, int clear)
+{
+	struct rte_port_eventdev_writer_nodrop *p =
+			port;
+
+	if (stats != NULL)
+		memcpy(stats, &p->stats, sizeof(p->stats));
+
+	if (clear)
+		memset(&p->stats, 0, sizeof(p->stats));
+
+	return 0;
+}
+
+/*
+ * Summary of port operations
+ */
+struct rte_port_in_ops rte_port_eventdev_reader_ops = {
+	.f_create = rte_port_eventdev_reader_create,
+	.f_free = rte_port_eventdev_reader_free,
+	.f_rx = rte_port_eventdev_reader_rx,
+	.f_stats = rte_port_eventdev_reader_stats_read,
+};
+
+struct rte_port_out_ops rte_port_eventdev_writer_ops = {
+	.f_create = rte_port_eventdev_writer_create,
+	.f_free = rte_port_eventdev_writer_free,
+	.f_tx = rte_port_eventdev_writer_tx,
+	.f_tx_bulk = rte_port_eventdev_writer_tx_bulk,
+	.f_flush = rte_port_eventdev_writer_flush,
+	.f_stats = rte_port_eventdev_writer_stats_read,
+};
+
+struct rte_port_out_ops rte_port_eventdev_writer_nodrop_ops = {
+	.f_create = rte_port_eventdev_writer_nodrop_create,
+	.f_free = rte_port_eventdev_writer_nodrop_free,
+	.f_tx = rte_port_eventdev_writer_nodrop_tx,
+	.f_tx_bulk = rte_port_eventdev_writer_nodrop_tx_bulk,
+	.f_flush = rte_port_eventdev_writer_nodrop_flush,
+	.f_stats = rte_port_eventdev_writer_nodrop_stats_read,
+};
+
diff --git a/lib/librte_port/rte_port_eventdev.h b/lib/librte_port/rte_port_eventdev.h
new file mode 100644
index 000000000..946735646
--- /dev/null
+++ b/lib/librte_port/rte_port_eventdev.h
@@ -0,0 +1,100 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2010-2014 Intel Corporation
+ */
+
+#ifndef __INCLUDE_RTE_PORT_EVENTDEV_H__
+#define __INCLUDE_RTE_PORT_EVENTDEV_H__
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifndef ARRAY_SIZE
+#define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
+#endif
+
+/**
+ * @file
+ * RTE Port Eventdev Interface
+ *
+ * eventdev_reader: input port built on top of pre-initialized eventdev
+ * interface
+ * eventdev_writer: output port built on top of pre-initialized eventdev
+ * interface
+ *
+ **/
+
+#include <stdint.h>
+#include <rte_eventdev.h>
+
+#include "rte_port.h"
+
+/** Eventdev_reader port parameters */
+struct rte_port_eventdev_reader_params {
+	/** Eventdev Device ID */
+	uint8_t eventdev_id;
+
+	/** Eventdev Port ID */
+	uint8_t port_id;
+};
+
+/** Eventdev_reader port operations. */
+extern struct rte_port_in_ops rte_port_eventdev_reader_ops;
+
+/** Eventdev_writer port parameters. */
+struct rte_port_eventdev_writer_params {
+	/** Eventdev Device ID. */
+	uint8_t eventdev_id;
+
+	/** Eventdev Port ID. */
+	uint8_t port_id;
+
+	/** Eventdev Queue ID. */
+	uint8_t queue_id;
+
+	/** Burst size to eventdev interface. */
+	uint32_t enq_burst_sz;
+
+	/** Scheduler synchronization type (RTE_SCHED_TYPE_*)*/
+	uint8_t sched_type;
+
+	/** The type of eventdev enqueue operation - new/forward/release */
+	uint8_t evt_op;
+};
+
+/** Eventdev_writer port operations. */
+extern struct rte_port_out_ops rte_port_eventdev_writer_ops;
+
+/** Event_writer_nodrop port parameters. */
+struct rte_port_eventdev_writer_nodrop_params {
+	/** Eventdev Device ID. */
+	uint8_t eventdev_id;
+
+	/** Eventdev Port ID. */
+	uint16_t port_id;
+
+	/** Eventdev Queue ID. */
+	uint16_t queue_id;
+
+	/** Burst size to eventdev interface. */
+	uint32_t enq_burst_sz;
+
+	/** Scheduler synchronization type (RTE_SCHED_TYPE_*)*/
+	uint8_t sched_type;
+
+	/** The type of eventdev enqueue operation - new/forward/release */
+	uint8_t evt_op;
+
+	/** Maximum number of retries, 0 for no limit. */
+	uint32_t n_retries;
+};
+
+/** Eventdev_writer_nodrop port operations. */
+extern struct rte_port_out_ops rte_port_eventdev_writer_nodrop_ops;
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif
+
diff --git a/lib/librte_port/rte_port_version.map b/lib/librte_port/rte_port_version.map
index 609bcec3f..3906c57c8 100644
--- a/lib/librte_port/rte_port_version.map
+++ b/lib/librte_port/rte_port_version.map
@@ -60,3 +60,12 @@ DPDK_18.11 {
 	rte_port_sym_crypto_writer_nodrop_ops;
 
 } DPDK_16.11;
+
+EXPERIMENTAL {
+    global:
+
+   rte_port_eventdev_reader_ops;
+    rte_port_eventdev_writer_ops;
+    rte_port_eventdev_writer_nodrop_ops;
+
+};
-- 
2.20.1