[dpdk-dev] [PATCH 2/4] distributor: new packet distributor library

[Bruce Richardson]

This adds the code for a new Intel DPDK library for packet distribution.
The distributor is a component which is designed to pass packets
one-at-a-time to workers, with dynamic load balancing. Using the RSS
field in the mbuf as a tag, the distributor tracks what packet tag is
being processed by what worker and then ensures that no two packets with
the same tag are in-flight simultaneously. Once a tag is not in-flight,
then the next packet with that tag will be sent to the next available
core.

Signed-off-by: Bruce Richardson <bruce.richardson at intel.com>
---
 lib/librte_distributor/Makefile          |  50 ++++
 lib/librte_distributor/rte_distributor.c | 417 +++++++++++++++++++++++++++++++
 lib/librte_distributor/rte_distributor.h | 173 +++++++++++++
 3 files changed, 640 insertions(+)
 create mode 100644 lib/librte_distributor/Makefile
 create mode 100644 lib/librte_distributor/rte_distributor.c
 create mode 100644 lib/librte_distributor/rte_distributor.h

diff --git a/lib/librte_distributor/Makefile b/lib/librte_distributor/Makefile
new file mode 100644
index 0000000..36699f8
--- /dev/null
+++ b/lib/librte_distributor/Makefile
@@ -0,0 +1,50 @@
+#   BSD LICENSE
+#
+#   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
+#   All rights reserved.
+#
+#   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 Intel Corporation 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 $(RTE_SDK)/mk/rte.vars.mk
+
+# library name
+LIB = librte_distributor.a
+
+CFLAGS += -O3
+CFLAGS += $(WERROR_FLAGS) -I$(SRCDIR)
+
+# all source are stored in SRCS-y
+SRCS-$(CONFIG_RTE_LIBRTE_DISTRIBUTOR) := rte_distributor.c
+
+# install this header file
+SYMLINK-$(CONFIG_RTE_LIBRTE_DISTRIBUTOR)-include := rte_distributor.h
+
+# this lib needs eal
+DEPDIRS-$(CONFIG_RTE_LIBRTE_DISTRIBUTOR) += lib/librte_eal
+DEPDIRS-$(CONFIG_RTE_LIBRTE_DISTRIBUTOR) += lib/librte_mbuf
+
+include $(RTE_SDK)/mk/rte.lib.mk
diff --git a/lib/librte_distributor/rte_distributor.c b/lib/librte_distributor/rte_distributor.c
new file mode 100644
index 0000000..cc8384e
--- /dev/null
+++ b/lib/librte_distributor/rte_distributor.c
@@ -0,0 +1,417 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
+ *   All rights reserved.
+ *
+ *   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 Intel Corporation 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 <stdio.h>
+#include <sys/queue.h>
+#include <string.h>
+#include <rte_mbuf.h>
+#include <rte_memzone.h>
+#include <rte_errno.h>
+#include <rte_string_fns.h>
+#include <rte_tailq.h>
+#include <rte_eal_memconfig.h>
+#include "rte_distributor.h"
+
+#define NO_FLAGS 0
+#define RTE_DISTRIB_PREFIX "DT_"
+
+/* we will use the bottom four bits of pointer for flags, shifting out
+ * the top four bits to make room (since a 64-bit pointer actually only uses
+ * 48 bits). An arithmetic-right-shift will then appropriately restore the
+ * original pointer value with proper sign extension into the top bits. */
+#define RTE_DISTRIB_FLAG_BITS 4
+#define RTE_DISTRIB_FLAGS_MASK (0x0F)
+#define RTE_DISTRIB_NO_BUF 0
+#define RTE_DISTRIB_GET_BUF (1)
+#define RTE_DISTRIB_RETURN_BUF (2)
+
+#define RTE_DISTRIB_BACKLOG_SIZE 8
+#define RTE_DISTRIB_BACKLOG_MASK (RTE_DISTRIB_BACKLOG_SIZE - 1)
+
+#define RTE_DISTRIB_MAX_RETURNS 128
+#define RTE_DISTRIB_RETURNS_MASK (RTE_DISTRIB_MAX_RETURNS - 1)
+
+union rte_distributor_buffer {
+	volatile int64_t bufptr64;
+	char pad[CACHE_LINE_SIZE*3];
+} __rte_cache_aligned;
+
+struct rte_distributor_backlog {
+	unsigned start;
+	unsigned count;
+	int64_t pkts[RTE_DISTRIB_BACKLOG_SIZE];
+};
+
+struct rte_distributor_returned_pkts {
+	unsigned start;
+	unsigned count;
+	struct rte_mbuf *mbufs[RTE_DISTRIB_MAX_RETURNS];
+};
+
+struct rte_distributor {
+	TAILQ_ENTRY(rte_distributor) next;    /**< Next in list. */
+
+	char name[RTE_DISTRIBUTOR_NAMESIZE];   /**< Name of the ring. */
+	unsigned num_workers;                 /**< Number of workers polling */
+
+	uint32_t in_flight_tags[RTE_MAX_LCORE];
+	struct rte_distributor_backlog backlog[RTE_MAX_LCORE];
+
+	union rte_distributor_buffer bufs[RTE_MAX_LCORE];
+
+	struct rte_distributor_returned_pkts returns;
+};
+
+TAILQ_HEAD(rte_distributor_list, rte_distributor);
+
+/**** APIs called by workers ****/
+
+struct rte_mbuf *
+rte_distributor_get_pkt(struct rte_distributor *d,
+		unsigned worker_id, struct rte_mbuf *oldpkt,
+		unsigned reserved __rte_unused)
+{
+	union rte_distributor_buffer *buf = &d->bufs[worker_id];
+	int64_t req = (((int64_t)(uintptr_t)oldpkt) << RTE_DISTRIB_FLAG_BITS) | \
+			RTE_DISTRIB_GET_BUF;
+	while (unlikely(buf->bufptr64 & RTE_DISTRIB_FLAGS_MASK))
+		rte_pause();
+	buf->bufptr64 = req;
+	while (buf->bufptr64 & RTE_DISTRIB_GET_BUF)
+		rte_pause();
+	/* since bufptr64 is a signed value, this should be an arithmetic shift */
+	int64_t ret = buf->bufptr64 >> RTE_DISTRIB_FLAG_BITS;
+	return (struct rte_mbuf *)((uintptr_t)ret);
+}
+
+int
+rte_distributor_return_pkt(struct rte_distributor *d,
+		unsigned worker_id, struct rte_mbuf *oldpkt)
+{
+	union rte_distributor_buffer *buf = &d->bufs[worker_id];
+	uint64_t req = ((uintptr_t)oldpkt << RTE_DISTRIB_FLAG_BITS) | \
+			RTE_DISTRIB_RETURN_BUF;
+	buf->bufptr64 = req;
+	return 0;
+}
+
+/**** APIs called on distributor core ***/
+
+/* as name suggests, adds a packet to the backlog for a particular worker */
+static int
+add_to_backlog(struct rte_distributor_backlog *bl, int64_t item)
+{
+	if (bl->count == RTE_DISTRIB_BACKLOG_SIZE)
+		return -1;
+
+	bl->pkts[(bl->start + bl->count++) &  (RTE_DISTRIB_BACKLOG_MASK)] = item;
+	return 0;
+}
+
+/* takes the next packet for a worker off the backlog */
+static int64_t
+backlog_pop(struct rte_distributor_backlog *bl)
+{
+	bl->count--;
+	return bl->pkts[bl->start++ & RTE_DISTRIB_BACKLOG_MASK];
+}
+
+/* stores a packet returned from a worker inside the returns array */
+static inline void
+store_return(uintptr_t oldbuf, struct rte_distributor *d,
+		unsigned *ret_start, unsigned *ret_count)
+{
+	/* store returns in a circular buffer - code is branch-free */
+	d->returns.mbufs[(*ret_start + *ret_count)
+	                 & RTE_DISTRIB_RETURNS_MASK] = (void *)oldbuf;
+	*ret_start += (*ret_count == RTE_DISTRIB_RETURNS_MASK) & !!(oldbuf);
+	*ret_count += (*ret_count != RTE_DISTRIB_RETURNS_MASK) & !!(oldbuf);
+}
+
+/* process a set of packets to distribute them to workers */
+int
+rte_distributor_process(struct rte_distributor *d,
+		struct rte_mbuf **mbufs, unsigned num_mbufs)
+{
+	unsigned next_idx = 0;
+	unsigned worker = 0;
+	struct rte_mbuf *next_mb = NULL;
+	int64_t next_value = 0;
+	uint32_t new_tag = 0;
+	unsigned ret_start = d->returns.start,
+			ret_count = d->returns.count;
+
+	while (next_idx < num_mbufs || next_mb != NULL) {
+
+		int64_t data = d->bufs[worker].bufptr64;
+		uintptr_t oldbuf = 0;
+
+		if (!next_mb) {
+			next_mb = mbufs[next_idx++];
+			next_value = (((int64_t)(uintptr_t)next_mb) << RTE_DISTRIB_FLAG_BITS);
+			new_tag = (next_mb->pkt.hash.rss | 1);
+
+			uint32_t match = 0;
+			unsigned i;
+			for (i = 0; i < d->num_workers; i++)
+				match |= (!(d->in_flight_tags[i] ^ new_tag) << i);
+
+			if (match) {
+				next_mb = NULL;
+				unsigned worker = __builtin_ctz(match);
+				if (add_to_backlog(&d->backlog[worker], next_value) < 0)
+					next_idx--;
+			}
+		}
+
+		if ((data & RTE_DISTRIB_GET_BUF) &&
+				(d->backlog[worker].count || next_mb)) {
+
+			if (d->backlog[worker].count)
+				d->bufs[worker].bufptr64 =
+						backlog_pop(&d->backlog[worker]);
+
+			else {
+				d->bufs[worker].bufptr64 = next_value;
+				d->in_flight_tags[worker] = new_tag;
+				next_mb = NULL;
+			}
+			oldbuf = data >> RTE_DISTRIB_FLAG_BITS;
+		}
+		else if (data & RTE_DISTRIB_RETURN_BUF) {
+			d->in_flight_tags[worker] = 0;
+			d->bufs[worker].bufptr64 = 0;
+			if (unlikely(d->backlog[worker].count != 0)) {
+				/* On return of a packet, we need to move the queued packets
+				 * for this core elsewhere.
+				 * Easiest solution is to set things up for
+				 * a recursive call. That will cause those packets to be queued
+				 * up for the next free core, i.e. it will return as soon as a
+				 * core becomes free to accept the first packet, as subsequent
+				 * ones will be added to the backlog for that core.
+				 */
+				struct rte_mbuf *pkts[RTE_DISTRIB_BACKLOG_SIZE];
+				unsigned i;
+				struct rte_distributor_backlog *bl = &d->backlog[worker];
+
+				for (i = 0; i < bl->count; i++) {
+					unsigned idx = (bl->start + i) & RTE_DISTRIB_BACKLOG_MASK;
+					pkts[i] = (void *)((uintptr_t)
+							(bl->pkts[idx] >> RTE_DISTRIB_FLAG_BITS));
+				}
+				/* recursive call */
+				rte_distributor_process(d, pkts, i);
+				bl->count = bl->start = 0;
+			}
+			oldbuf = data >> RTE_DISTRIB_FLAG_BITS;
+		}
+
+		/* store returns in a circular buffer */
+		store_return(oldbuf, d, &ret_start, &ret_count);
+
+		if (++worker == d->num_workers)
+			worker = 0;
+	}
+	/* to finish, check all workers for backlog and schedule work for them
+	 * if they are ready */
+	for (worker = 0; worker < d->num_workers; worker++)
+		if (d->backlog[worker].count &&
+				(d->bufs[worker].bufptr64 & RTE_DISTRIB_GET_BUF)) {
+
+			int64_t oldbuf = d->bufs[worker].bufptr64 >> RTE_DISTRIB_FLAG_BITS;
+			store_return(oldbuf, d, &ret_start, &ret_count);
+
+			d->bufs[worker].bufptr64 =
+					backlog_pop(&d->backlog[worker]);
+		}
+
+	d->returns.start = ret_start;
+	d->returns.count = ret_count;
+	return num_mbufs;
+}
+
+/* return to the caller, packets returned from workers */
+int
+rte_distributor_returned_pkts(struct rte_distributor *d,
+		struct rte_mbuf **mbufs, unsigned max_mbufs)
+{
+	struct rte_distributor_returned_pkts *returns = &d->returns;
+	unsigned retval = max_mbufs < returns->count ? max_mbufs : returns->count;
+	unsigned i;
+
+	for (i = 0; i < retval; i++)
+		mbufs[i] = returns->mbufs[(returns->start + i) &
+		                          RTE_DISTRIB_RETURNS_MASK];
+	returns->start += i;
+	returns->count -= i;
+
+	return retval;
+}
+
+/* local function used by the flush function only, to reassign a backlog for
+ * a shutdown core. The process function uses a recursive call for this, but
+ * that is not done in flush, as we need to track the outstanding packets count.
+ */
+static inline int
+move_worker_backlog(struct rte_distributor *d, unsigned worker)
+{
+	struct rte_distributor_backlog *bl = &d->backlog[worker];
+	unsigned i;
+
+	for (i = 0; i < d->num_workers; i++) {
+		if (i == worker)
+			continue;
+		/* check worker is active and then if backlog will fit */
+		if ((d->in_flight_tags[i] != 0 ||
+				(d->bufs[i].bufptr64 & RTE_DISTRIB_GET_BUF)) &&
+				(bl->count + d->backlog[i].count) <= RTE_DISTRIB_BACKLOG_SIZE) {
+			while (bl->count)
+				add_to_backlog(&d->backlog[i], backlog_pop(bl));
+			return 0;
+		}
+	}
+	return -1;
+}
+
+/* flush the distributor, so that there are no outstanding packets in flight or
+ * queued up. */
+int
+rte_distributor_flush(struct rte_distributor *d)
+{
+	unsigned worker, total_outstanding = 0;
+	unsigned flushed = 0;
+	unsigned ret_start = d->returns.start,
+			ret_count = d->returns.count;
+
+	for (worker = 0; worker < d->num_workers; worker++)
+		total_outstanding += d->backlog[worker].count +
+				!!(d->in_flight_tags[worker]);
+
+	worker = 0;
+	while (flushed < total_outstanding) {
+
+		if (d->in_flight_tags[worker] != 0 || d->backlog[worker].count) {
+			const int64_t data = d->bufs[worker].bufptr64;
+			uintptr_t oldbuf = 0;
+
+			if (data & RTE_DISTRIB_GET_BUF) {
+				flushed += (d->in_flight_tags[worker] != 0);
+				if (d->backlog[worker].count) {
+					d->bufs[worker].bufptr64 =
+							backlog_pop(&d->backlog[worker]);
+					/* we need to mark something as being in-flight, but it
+					 * doesn't matter what as we never check it except
+					 * to check for non-zero.
+					 */
+					d->in_flight_tags[worker] = 1;
+				} else {
+					d->bufs[worker].bufptr64 = RTE_DISTRIB_GET_BUF;
+					d->in_flight_tags[worker] = 0;
+				}
+				oldbuf = data >> RTE_DISTRIB_FLAG_BITS;
+			}
+			else if (data & RTE_DISTRIB_RETURN_BUF) {
+				if (d->backlog[worker].count == 0 ||
+						move_worker_backlog(d, worker) == 0) {
+					/* only if we move backlog, process this packet */
+					d->bufs[worker].bufptr64 = 0;
+					oldbuf = data >> RTE_DISTRIB_FLAG_BITS;
+					flushed ++;
+					d->in_flight_tags[worker] = 0;
+				}
+			}
+
+			store_return(oldbuf, d, &ret_start, &ret_count);
+		}
+
+		if (++worker == d->num_workers)
+			worker = 0;
+	}
+	d->returns.start = ret_start;
+	d->returns.count = ret_count;
+
+	return flushed;
+}
+
+/* clears the internal returns array in the distributor */
+void
+rte_distributor_clear_returns(struct rte_distributor *d)
+{
+	d->returns.start = d->returns.count = 0;
+#ifndef __OPTIMIZE__
+	memset(d->returns.mbufs, 0, sizeof(d->returns.mbufs));
+#endif
+}
+
+/* creates a distributor instance */
+struct rte_distributor *
+rte_distributor_create(const char *name,
+		unsigned socket_id,
+		unsigned num_workers,
+		struct rte_distributor_extra_args *args __rte_unused)
+{
+	struct rte_distributor *d;
+	struct rte_distributor_list *distributor_list;
+	char mz_name[RTE_MEMZONE_NAMESIZE];
+	const struct rte_memzone *mz;
+
+	/* compilation-time checks */
+	RTE_BUILD_BUG_ON((sizeof(*d) & CACHE_LINE_MASK) != 0);
+	RTE_BUILD_BUG_ON((RTE_MAX_LCORE & 7) != 0);
+
+	if (name == NULL || num_workers >= RTE_MAX_LCORE) {
+		rte_errno = EINVAL;
+		return NULL;
+	}
+	rte_snprintf(mz_name, sizeof(mz_name), RTE_DISTRIB_PREFIX"%s", name);
+	mz = rte_memzone_reserve(mz_name, sizeof(*d), socket_id, NO_FLAGS);
+	if (mz == NULL) {
+		rte_errno = ENOMEM;
+		return NULL;
+	}
+
+	/* check that we have an initialised tail queue */
+	if ((distributor_list = RTE_TAILQ_LOOKUP_BY_IDX(RTE_TAILQ_DISTRIBUTOR,
+			rte_distributor_list)) == NULL) {
+		rte_errno = E_RTE_NO_TAILQ;
+		return NULL;
+	}
+
+	d = mz->addr;
+	rte_snprintf(d->name, sizeof(d->name), "%s", name);
+	d->num_workers = num_workers;
+	TAILQ_INSERT_TAIL(distributor_list, d, next);
+
+	return d;
+}
+
diff --git a/lib/librte_distributor/rte_distributor.h b/lib/librte_distributor/rte_distributor.h
new file mode 100644
index 0000000..d684ff9
--- /dev/null
+++ b/lib/librte_distributor/rte_distributor.h
@@ -0,0 +1,173 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
+ *   All rights reserved.
+ *
+ *   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 Intel Corporation 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 _RTE_DISTRIBUTE_H_
+#define _RTE_DISTRIBUTE_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <rte_mbuf.h>
+
+#define RTE_DISTRIBUTOR_NAMESIZE 32 /**< Length of name for instance */
+
+struct rte_distributor;
+
+struct rte_distributor_extra_args { }; /**< reserved for future use*/
+
+/**
+ * Function to create a new distributor instance
+ *
+ * Reserves the memory needed for the distributor operation and
+ * initializes the distributor to work with the configured number of workers.
+ *
+ * @param name
+ *   The name to be given to the distributor instance.
+ * @param socket_id
+ *   The NUMA node on which the memory is to be allocated
+ * @param num_workers
+ *   The maximum number of workers that will request packets from this
+ *   distributor
+ * @param extra_args
+ *   Reserved for future use, should be passed in as NULL
+ * @return
+ *   The newly created distributor instance
+ */
+struct rte_distributor *
+rte_distributor_create(const char *name, unsigned socket_id,
+		unsigned num_workers, struct rte_distributor_extra_args *extra_args);
+
+/**
+ * Process a set of packets by distributing them among workers that request
+ * packets. The distributor will ensure that no two packets that have the
+ * same flow id, or tag, in the mbuf will be procesed at the same time.
+ *
+ * NOTE: This is not thread safe, should only be called in one thread at a time
+ *
+ * @param d
+ *   The distributor instance to be used
+ * @param mbufs
+ *   The mbufs to be distributed
+ * @param num_mbufs
+ *   The number of mbufs in the mbufs array
+ * @return
+ *   The number of mbufs processed.
+ */
+int
+rte_distributor_process(struct rte_distributor *d,
+		struct rte_mbuf **mbufs, unsigned num_mbufs);
+
+/**
+ * Get a set of mbufs that have been returned to the distributor by workers
+ *
+ * @param d
+ *   The distributor instance to be used
+ * @param mbufs
+ *   The mbufs pointer array to be filled in
+ * @param max_mbufs
+ *   The size of the mbufs array
+ * @return
+ *   The number of mbufs returned in the mbufs array.
+ */
+int
+rte_distributor_returned_pkts(struct rte_distributor *d,
+		struct rte_mbuf **mbufs, unsigned max_mbufs);
+
+/**
+ * Flush the distributor component, so that there are no in-flight or
+ * backlogged packets awaiting processing
+ *
+ * @param d
+ *   The distributor instance to be used
+ * @return
+ *   The number of queued/in-flight packets that were completed by this call.
+ */
+int
+rte_distributor_flush(struct rte_distributor *d);
+
+/**
+ * Clears the array of returned packets used as the source for the
+ * rte_distributor_returned_pkts() API call.
+ *
+ * @param d
+ *   The distributor instance to be used
+ */
+void
+rte_distributor_clear_returns(struct rte_distributor *d);
+
+/**
+ * API called by a worker to get a new packet to process. Any previous packet
+ * given to the worker is assumed to have completed processing, and may be
+ * optionally returned to the distributor via the oldpkt parameter.
+ *
+ * @param d
+ *   The distributor instance to be used
+ * @param worker_id
+ *   The worker instance number to use - must be less that num_workers passed
+ *   at distributor creation time.
+ * @param oldpkt
+ *   The previous packet, if any, being processed by the worker
+ * @param reserved
+ *   Reserved for future use, should be set to zero.
+ *
+ * @return
+ *   A new packet to be processed by the worker thread.
+ */
+struct rte_mbuf *
+rte_distributor_get_pkt(struct rte_distributor *d,
+		unsigned worker_id, struct rte_mbuf *oldpkt, unsigned reserved);
+
+/**
+ * API called by a worker to return a completed packet without requesting a
+ * new packet, for example, because a worker thread is shutting down
+ *
+ * @param d
+ *   The distributor instance to be used
+ * @param worker_id
+ *   The worker instance number to use - must be less that num_workers passed
+ *   at distributor creation time.
+ * @param mbuf
+ *   The previous packet being processed by the worker
+ */
+int
+rte_distributor_return_pkt(struct rte_distributor *d, unsigned worker_id,
+		struct rte_mbuf *mbuf);
+
+/******************************************/
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif
-- 
1.9.0