DPDK patches and discussions
 help / color / mirror / Atom feed
From: Neil Horman <nhorman@tuxdriver.com>
To: Bruce Richardson <bruce.richardson@intel.com>
Cc: dev@dpdk.org
Subject: Re: [dpdk-dev] [PATCH v3 2/5] distributor: new packet distributor library
Date: Tue, 3 Jun 2014 14:38:23 -0400	[thread overview]
Message-ID: <20140603183823.GD20038@hmsreliant.think-freely.org> (raw)
In-Reply-To: <1401818652-11466-1-git-send-email-bruce.richardson@intel.com>

On Tue, Jun 03, 2014 at 07:04:12PM +0100, Bruce Richardson wrote:
> 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
> 
> 
Acked-by: Neil Horman <nhorman@tuxdriver.com>

  reply	other threads:[~2014-06-03 18:38 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   ` [dpdk-dev] [PATCH v3 " Bruce Richardson
2014-06-03 18:38     ` Neil Horman [this message]
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=20140603183823.GD20038@hmsreliant.think-freely.org \
    --to=nhorman@tuxdriver.com \
    --cc=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).