DPDK patches and discussions
 help / color / mirror / Atom feed
From: Bruce Richardson <bruce.richardson@intel.com>
To: dev@dpdk.org
Subject: [dpdk-dev] [PATCH v3 2/5] distributor: new packet distributor library
Date: Tue,  3 Jun 2014 19:04:12 +0100	[thread overview]
Message-ID: <1401818652-11466-1-git-send-email-bruce.richardson@intel.com> (raw)
In-Reply-To: <1401358338-23455-3-git-send-email-bruce.richardson@intel.com>

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.

Changes in V2 patch:
* added support for a partial distributor flush when process() API
  called without any new mbufs
* Removed unused "future use" parameters from functions
* Improved comments to be clearer about thread safety
* Add locks around the tailq add in create() API fn
* Stylistic improvements for issues flagged by checkpatch

Changes in V3 patch:
* Flush function rewritten as calls to process API
* Additional doxygen comments on thread-safety of APIs

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
 lib/librte_distributor/Makefile          |  50 ++++
 lib/librte_distributor/rte_distributor.c | 425 +++++++++++++++++++++++++++++++
 lib/librte_distributor/rte_distributor.h | 194 ++++++++++++++
 3 files changed, 669 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..5eee442
--- /dev/null
+++ b/lib/librte_distributor/rte_distributor.c
@@ -0,0 +1,425 @@
+/*-
+ *   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       /**< empty flags: no buffer requested */
+#define RTE_DISTRIB_GET_BUF (1)    /**< worker requests a buffer, returns old */
+#define RTE_DISTRIB_RETURN_BUF (2) /**< worker returns a buffer, no request */
+
+#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)
+
+/**
+ * Buffer structure used to pass the pointer data between cores. This is cache
+ * line aligned, but to improve performance and prevent adjacent cache-line
+ * prefetches of buffers for other workers, e.g. when worker 1's buffer is on
+ * the next cache line to worker 0, we pad this out to three cache lines.
+ * Only 64-bits of the memory is actually used though.
+ */
+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)
+{
+	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 signed, 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 = (((int64_t)(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);
+}
+
+static inline void
+handle_worker_shutdown(struct rte_distributor *d, unsigned wkr)
+{
+	d->in_flight_tags[wkr] = 0;
+	d->bufs[wkr].bufptr64 = 0;
+	if (unlikely(d->backlog[wkr].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[wkr];
+
+		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;
+	}
+}
+
+/* this function is called when process() fn is called without any new
+ * packets. It goes through all the workers and clears any returned packets
+ * to do a partial flush.
+ */
+static int
+process_returns(struct rte_distributor *d)
+{
+	unsigned wkr;
+	unsigned flushed = 0;
+	unsigned ret_start = d->returns.start,
+			ret_count = d->returns.count;
+
+	for (wkr = 0; wkr < d->num_workers; wkr++) {
+
+		const int64_t data = d->bufs[wkr].bufptr64;
+		uintptr_t oldbuf = 0;
+
+		if (data & RTE_DISTRIB_GET_BUF) {
+			flushed++;
+			if (d->backlog[wkr].count)
+				d->bufs[wkr].bufptr64 =
+						backlog_pop(&d->backlog[wkr]);
+			else {
+				d->bufs[wkr].bufptr64 = RTE_DISTRIB_GET_BUF;
+				d->in_flight_tags[wkr] = 0;
+			}
+			oldbuf = data >> RTE_DISTRIB_FLAG_BITS;
+		} else if (data & RTE_DISTRIB_RETURN_BUF) {
+			handle_worker_shutdown(d, wkr);
+			oldbuf = data >> RTE_DISTRIB_FLAG_BITS;
+		}
+
+		store_return(oldbuf, d, &ret_start, &ret_count);
+	}
+
+	d->returns.start = ret_start;
+	d->returns.count = ret_count;
+
+	return flushed;
+}
+
+/* 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 wkr = 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;
+
+	if (unlikely(num_mbufs == 0))
+		return process_returns(d);
+
+	while (next_idx < num_mbufs || next_mb != NULL) {
+
+		int64_t data = d->bufs[wkr].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[wkr].count || next_mb)) {
+
+			if (d->backlog[wkr].count)
+				d->bufs[wkr].bufptr64 =
+						backlog_pop(&d->backlog[wkr]);
+
+			else {
+				d->bufs[wkr].bufptr64 = next_value;
+				d->in_flight_tags[wkr] = new_tag;
+				next_mb = NULL;
+			}
+			oldbuf = data >> RTE_DISTRIB_FLAG_BITS;
+		} else if (data & RTE_DISTRIB_RETURN_BUF) {
+			handle_worker_shutdown(d, wkr);
+			oldbuf = data >> RTE_DISTRIB_FLAG_BITS;
+		}
+
+		/* store returns in a circular buffer */
+		store_return(oldbuf, d, &ret_start, &ret_count);
+
+		if (++wkr == d->num_workers)
+			wkr = 0;
+	}
+	/* to finish, check all workers for backlog and schedule work for them
+	 * if they are ready */
+	for (wkr = 0; wkr < d->num_workers; wkr++)
+		if (d->backlog[wkr].count &&
+				(d->bufs[wkr].bufptr64 & RTE_DISTRIB_GET_BUF)) {
+
+			int64_t oldbuf = d->bufs[wkr].bufptr64 >>
+					RTE_DISTRIB_FLAG_BITS;
+			store_return(oldbuf, d, &ret_start, &ret_count);
+
+			d->bufs[wkr].bufptr64 = backlog_pop(&d->backlog[wkr]);
+		}
+
+	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++) {
+		unsigned idx = (returns->start + i) & RTE_DISTRIB_RETURNS_MASK;
+		mbufs[i] = returns->mbufs[idx];
+	}
+	returns->start += i;
+	returns->count -= i;
+
+	return retval;
+}
+
+/* return the number of packets in-flight in a distributor, i.e. packets
+ * being workered on or queued up in a backlog. */
+static inline unsigned
+total_outstanding(const struct rte_distributor *d)
+{
+	unsigned wkr, total_outstanding = 0;
+
+	for (wkr = 0; wkr < d->num_workers; wkr++)
+		total_outstanding += d->backlog[wkr].count +
+				!!(d->in_flight_tags[wkr]);
+	return total_outstanding;
+}
+
+/* flush the distributor, so that there are no outstanding packets in flight or
+ * queued up. */
+int
+rte_distributor_flush(struct rte_distributor *d)
+{
+	const unsigned flushed = total_outstanding(d);
+
+	while (total_outstanding(d) > 0)
+		rte_distributor_process(d, NULL, 0);
+
+	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 *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;
+	}
+
+	/* check that we have an initialised tail queue */
+	distributor_list = RTE_TAILQ_LOOKUP_BY_IDX(RTE_TAILQ_DISTRIBUTOR,
+				rte_distributor_list);
+	if (distributor_list == NULL) {
+		rte_errno = E_RTE_NO_TAILQ;
+		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;
+	}
+
+	d = mz->addr;
+	rte_snprintf(d->name, sizeof(d->name), "%s", name);
+	d->num_workers = num_workers;
+
+	rte_rwlock_write_lock(RTE_EAL_TAILQ_RWLOCK);
+	TAILQ_INSERT_TAIL(distributor_list, d, next);
+	rte_rwlock_write_unlock(RTE_EAL_TAILQ_RWLOCK);
+
+	return d;
+}
diff --git a/lib/librte_distributor/rte_distributor.h b/lib/librte_distributor/rte_distributor.h
new file mode 100644
index 0000000..bc70f0f
--- /dev/null
+++ b/lib/librte_distributor/rte_distributor.h
@@ -0,0 +1,194 @@
+/*-
+ *   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;
+
+/**
+ * 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
+ * @return
+ *   The newly created distributor instance
+ */
+struct rte_distributor *
+rte_distributor_create(const char *name, unsigned socket_id,
+		unsigned num_workers);
+
+/*  *** APIS to be called on the distributor lcore ***  */
+/*
+ * The following APIs are the public APIs which are designed for use on a
+ * single lcore which acts as the distributor lcore for a given distributor
+ * instance. These functions cannot be called on multiple cores simultaneously
+ * without using locking to protect access to the internals of the distributor.
+ *
+ * NOTE: a given lcore cannot act as both a distributor lcore and a worker lcore
+ * for the same distributor instance, otherwise deadlock will result.
+ */
+
+/**
+ * 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.
+ *
+ * This is not multi-thread safe and should only be called on a single lcore.
+ *
+ * @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
+ *
+ * This should only be called on the same lcore as rte_distributor_process()
+ *
+ * @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
+ *
+ * This should only be called on the same lcore as rte_distributor_process()
+ *
+ * @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.
+ *
+ * This should only be called on the same lcore as rte_distributor_process()
+ *
+ * @param d
+ *   The distributor instance to be used
+ */
+void
+rte_distributor_clear_returns(struct rte_distributor *d);
+
+/*  *** APIS to be called on the worker lcores ***  */
+/*
+ * The following APIs are the public APIs which are designed for use on
+ * multiple lcores which act as workers for a distributor. Each lcore should use
+ * a unique worker id when requesting packets.
+ *
+ * NOTE: a given lcore cannot act as both a distributor lcore and a worker lcore
+ * for the same distributor instance, otherwise deadlock will result.
+ */
+
+/**
+ * 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
+ *
+ * @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);
+
+/**
+ * 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
-- 
1.9.3

  parent reply	other threads:[~2014-06-03 18:12 UTC|newest]

Thread overview: 29+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2014-05-20 10:00 [dpdk-dev] [PATCH 0/4] New library: rte_distributor Bruce Richardson
2014-05-20 10:00 ` [dpdk-dev] [PATCH 1/4] eal: add tailq for new distributor component Bruce Richardson
2014-05-20 10:00 ` [dpdk-dev] [PATCH 2/4] distributor: new packet distributor library Bruce Richardson
2014-05-20 18:18   ` Neil Horman
2014-05-21 10:21     ` Richardson, Bruce
2014-05-21 15:23       ` Neil Horman
2014-05-20 10:00 ` [dpdk-dev] [PATCH 3/4] distributor: add distributor library to build Bruce Richardson
2014-05-20 10:00 ` [dpdk-dev] [PATCH 4/4] distributor: add unit tests for distributor lib Bruce Richardson
2014-05-20 10:38 ` [dpdk-dev] [PATCH 0/4] New library: rte_distributor Neil Horman
2014-05-20 11:02   ` Richardson, Bruce
2014-05-20 17:14     ` Neil Horman
2014-05-20 19:32       ` Richardson, Bruce
2014-05-27 22:32 ` Thomas Monjalon
2014-05-28  8:48   ` Richardson, Bruce
2014-05-29 10:12 ` [dpdk-dev] [PATCH v2 0/5] " Bruce Richardson
2014-06-05  1:58   ` Cao, Waterman
2014-06-12 13:57   ` Thomas Monjalon
2014-05-29 10:12 ` [dpdk-dev] [PATCH v2 1/5] eal: add tailq for new distributor component Bruce Richardson
2014-05-29 10:12 ` [dpdk-dev] [PATCH v2 2/5] distributor: new packet distributor library Bruce Richardson
2014-05-29 13:48   ` Neil Horman
2014-06-02 21:40     ` Richardson, Bruce
2014-06-03 11:01       ` Neil Horman
2014-06-03 14:33         ` Richardson, Bruce
2014-06-03 14:51           ` Neil Horman
2014-06-03 18:04   ` Bruce Richardson [this message]
2014-06-03 18:38     ` [dpdk-dev] [PATCH v3 " Neil Horman
2014-05-29 10:12 ` [dpdk-dev] [PATCH v2 3/5] distributor: add distributor library to build Bruce Richardson
2014-05-29 10:12 ` [dpdk-dev] [PATCH v2 4/5] distributor: add unit tests for distributor lib Bruce Richardson
2014-05-29 10:12 ` [dpdk-dev] [PATCH v2 5/5] docs: add distributor lib to API docs Bruce Richardson

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=1401818652-11466-1-git-send-email-bruce.richardson@intel.com \
    --to=bruce.richardson@intel.com \
    --cc=dev@dpdk.org \
    /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).