From mboxrd@z Thu Jan 1 00:00:00 1970 From: David Hunt Subject: [PATCH v7 01/17] lib: rename legacy distributor lib files Date: Tue, 21 Feb 2017 03:17:37 +0000 Message-ID: <1487647073-129064-2-git-send-email-david.hunt@intel.com> References: <1485163480-156507-2-git-send-email-david.hunt@intel.com> <1487647073-129064-1-git-send-email-david.hunt@intel.com> Cc: bruce.richardson@intel.com, David Hunt To: dev@dpdk.org Return-path: Received: from mga04.intel.com (mga04.intel.com [192.55.52.120]) by dpdk.org (Postfix) with ESMTP id EA98CFE5 for ; Tue, 21 Feb 2017 11:17:15 +0100 (CET) In-Reply-To: <1487647073-129064-1-git-send-email-david.hunt@intel.com> List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Sender: "dev" Move files out of the way so that we can replace with new versions of the distributor libtrary. Files are named in such a way as to match the symbol versioning that we will apply for backward ABI compatibility. Signed-off-by: David Hunt --- app/test/test_distributor.c | 2 +- app/test/test_distributor_perf.c | 2 +- examples/distributor/main.c | 2 +- lib/librte_distributor/Makefile | 4 +- lib/librte_distributor/rte_distributor.c | 487 --------------------------- lib/librte_distributor/rte_distributor.h | 247 -------------- lib/librte_distributor/rte_distributor_v20.c | 487 +++++++++++++++++++++++++++ lib/librte_distributor/rte_distributor_v20.h | 247 ++++++++++++++ 8 files changed, 739 insertions(+), 739 deletions(-) delete mode 100644 lib/librte_distributor/rte_distributor.c delete mode 100644 lib/librte_distributor/rte_distributor.h create mode 100644 lib/librte_distributor/rte_distributor_v20.c create mode 100644 lib/librte_distributor/rte_distributor_v20.h diff --git a/app/test/test_distributor.c b/app/test/test_distributor.c index 85cb8f3..ba402e2 100644 --- a/app/test/test_distributor.c +++ b/app/test/test_distributor.c @@ -39,7 +39,7 @@ #include #include #include -#include +#include #define ITER_POWER 20 /* log 2 of how many iterations we do when timing. */ #define BURST 32 diff --git a/app/test/test_distributor_perf.c b/app/test/test_distributor_perf.c index 7947fe9..fe0c97d 100644 --- a/app/test/test_distributor_perf.c +++ b/app/test/test_distributor_perf.c @@ -39,7 +39,7 @@ #include #include #include -#include +#include #define ITER_POWER 20 /* log 2 of how many iterations we do when timing. */ #define BURST 32 diff --git a/examples/distributor/main.c b/examples/distributor/main.c index e7641d2..fba5446 100644 --- a/examples/distributor/main.c +++ b/examples/distributor/main.c @@ -43,7 +43,7 @@ #include #include #include -#include +#include #define RX_RING_SIZE 256 #define TX_RING_SIZE 512 diff --git a/lib/librte_distributor/Makefile b/lib/librte_distributor/Makefile index 4c9af17..60837ed 100644 --- a/lib/librte_distributor/Makefile +++ b/lib/librte_distributor/Makefile @@ -42,10 +42,10 @@ EXPORT_MAP := rte_distributor_version.map LIBABIVER := 1 # all source are stored in SRCS-y -SRCS-$(CONFIG_RTE_LIBRTE_DISTRIBUTOR) := rte_distributor.c +SRCS-$(CONFIG_RTE_LIBRTE_DISTRIBUTOR) := rte_distributor_v20.c # install this header file -SYMLINK-$(CONFIG_RTE_LIBRTE_DISTRIBUTOR)-include := rte_distributor.h +SYMLINK-$(CONFIG_RTE_LIBRTE_DISTRIBUTOR)-include := rte_distributor_v20.h # this lib needs eal DEPDIRS-$(CONFIG_RTE_LIBRTE_DISTRIBUTOR) += lib/librte_eal diff --git a/lib/librte_distributor/rte_distributor.c b/lib/librte_distributor/rte_distributor.c deleted file mode 100644 index f3f778c..0000000 --- a/lib/librte_distributor/rte_distributor.c +++ /dev/null @@ -1,487 +0,0 @@ -/*- - * 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 -#include -#include -#include -#include -#include -#include -#include -#include -#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) - -/** - * Maximum number of workers allowed. - * Be aware of increasing the limit, becaus it is limited by how we track - * in-flight tags. See @in_flight_bitmask and @rte_distributor_process - */ -#define RTE_DISTRIB_MAX_WORKERS 64 - -/** - * 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[RTE_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_DISTRIB_MAX_WORKERS]; - /**< Tracks the tag being processed per core */ - uint64_t in_flight_bitmask; - /**< on/off bits for in-flight tags. - * Note that if RTE_DISTRIB_MAX_WORKERS is larger than 64 then - * the bitmask has to expand. - */ - - struct rte_distributor_backlog backlog[RTE_DISTRIB_MAX_WORKERS]; - - union rte_distributor_buffer bufs[RTE_DISTRIB_MAX_WORKERS]; - - struct rte_distributor_returned_pkts returns; -}; - -TAILQ_HEAD(rte_distributor_list, rte_distributor); - -static struct rte_tailq_elem rte_distributor_tailq = { - .name = "RTE_DISTRIBUTOR", -}; -EAL_REGISTER_TAILQ(rte_distributor_tailq) - -/**** APIs called by workers ****/ - -void -rte_distributor_request_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; -} - -struct rte_mbuf * -rte_distributor_poll_pkt(struct rte_distributor *d, - unsigned worker_id) -{ - union rte_distributor_buffer *buf = &d->bufs[worker_id]; - if (buf->bufptr64 & RTE_DISTRIB_GET_BUF) - return NULL; - - /* 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); -} - -struct rte_mbuf * -rte_distributor_get_pkt(struct rte_distributor *d, - unsigned worker_id, struct rte_mbuf *oldpkt) -{ - struct rte_mbuf *ret; - rte_distributor_request_pkt(d, worker_id, oldpkt); - while ((ret = rte_distributor_poll_pkt(d, worker_id)) == NULL) - rte_pause(); - return 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->in_flight_bitmask &= ~(1UL << wkr); - 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. - * Note that the tags were set before first level call - * to rte_distributor_process. - */ - 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; - d->in_flight_bitmask &= ~(1UL << wkr); - } - 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); - /* - * User is advocated to set tag vaue for each - * mbuf before calling rte_distributor_process. - * User defined tags are used to identify flows, - * or sessions. - */ - new_tag = next_mb->hash.usr; - - /* - * Note that if RTE_DISTRIB_MAX_WORKERS is larger than 64 - * then the size of match has to be expanded. - */ - uint64_t match = 0; - unsigned i; - /* - * to scan for a match use "xor" and "not" to get a 0/1 - * value, then use shifting to merge to single "match" - * variable, where a one-bit indicates a match for the - * worker given by the bit-position - */ - for (i = 0; i < d->num_workers; i++) - match |= (!(d->in_flight_tags[i] ^ new_tag) - << i); - - /* Only turned-on bits are considered as match */ - match &= d->in_flight_bitmask; - - if (match) { - next_mb = NULL; - unsigned worker = __builtin_ctzl(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; - d->in_flight_bitmask |= (1UL << wkr); - 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; - - total_outstanding = __builtin_popcountl(d->in_flight_bitmask); - - for (wkr = 0; wkr < d->num_workers; wkr++) - total_outstanding += d->backlog[wkr].count; - - 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) & RTE_CACHE_LINE_MASK) != 0); - RTE_BUILD_BUG_ON((RTE_DISTRIB_MAX_WORKERS & 7) != 0); - RTE_BUILD_BUG_ON(RTE_DISTRIB_MAX_WORKERS > - sizeof(d->in_flight_bitmask) * CHAR_BIT); - - if (name == NULL || num_workers >= RTE_DISTRIB_MAX_WORKERS) { - rte_errno = EINVAL; - return NULL; - } - - 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; - snprintf(d->name, sizeof(d->name), "%s", name); - d->num_workers = num_workers; - - distributor_list = RTE_TAILQ_CAST(rte_distributor_tailq.head, - rte_distributor_list); - - 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 deleted file mode 100644 index 7d36bc8..0000000 --- a/lib/librte_distributor/rte_distributor.h +++ /dev/null @@ -1,247 +0,0 @@ -/*- - * 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_ - -/** - * @file - * RTE distributor - * - * The distributor is a component which is designed to pass packets - * one-at-a-time to workers, with dynamic load balancing. - */ - -#ifdef __cplusplus -extern "C" { -#endif - -#define RTE_DISTRIBUTOR_NAMESIZE 32 /**< Length of name for instance */ - -struct rte_distributor; -struct rte_mbuf; - -/** - * 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. - * - * The user is advocated to set tag for each mbuf before calling this function. - * If user doesn't set the tag, the tag value can be various values depending on - * driver implementation and configuration. - * - * 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); - -/** - * API called by a worker to request 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. - * Unlike rte_distributor_get_pkt(), this function does not wait for a new - * packet to be provided by the distributor. - * - * NOTE: after calling this function, rte_distributor_poll_pkt() should - * be used to poll for the packet requested. The rte_distributor_get_pkt() - * API should *not* be used to try and retrieve the new packet. - * - * @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 - */ -void -rte_distributor_request_pkt(struct rte_distributor *d, - unsigned worker_id, struct rte_mbuf *oldpkt); - -/** - * API called by a worker to check for a new packet that was previously - * requested by a call to rte_distributor_request_pkt(). It does not wait - * for the new packet to be available, but returns NULL if the request has - * not yet been fulfilled by the distributor. - * - * @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. - * - * @return - * A new packet to be processed by the worker thread, or NULL if no - * packet is yet available. - */ -struct rte_mbuf * -rte_distributor_poll_pkt(struct rte_distributor *d, - unsigned worker_id); - -#ifdef __cplusplus -} -#endif - -#endif diff --git a/lib/librte_distributor/rte_distributor_v20.c b/lib/librte_distributor/rte_distributor_v20.c new file mode 100644 index 0000000..b890947 --- /dev/null +++ b/lib/librte_distributor/rte_distributor_v20.c @@ -0,0 +1,487 @@ +/*- + * 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 +#include +#include +#include +#include +#include +#include +#include +#include +#include "rte_distributor_v20.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) + +/** + * Maximum number of workers allowed. + * Be aware of increasing the limit, becaus it is limited by how we track + * in-flight tags. See @in_flight_bitmask and @rte_distributor_process + */ +#define RTE_DISTRIB_MAX_WORKERS 64 + +/** + * 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[RTE_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_DISTRIB_MAX_WORKERS]; + /**< Tracks the tag being processed per core */ + uint64_t in_flight_bitmask; + /**< on/off bits for in-flight tags. + * Note that if RTE_DISTRIB_MAX_WORKERS is larger than 64 then + * the bitmask has to expand. + */ + + struct rte_distributor_backlog backlog[RTE_DISTRIB_MAX_WORKERS]; + + union rte_distributor_buffer bufs[RTE_DISTRIB_MAX_WORKERS]; + + struct rte_distributor_returned_pkts returns; +}; + +TAILQ_HEAD(rte_distributor_list, rte_distributor); + +static struct rte_tailq_elem rte_distributor_tailq = { + .name = "RTE_DISTRIBUTOR", +}; +EAL_REGISTER_TAILQ(rte_distributor_tailq) + +/**** APIs called by workers ****/ + +void +rte_distributor_request_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; +} + +struct rte_mbuf * +rte_distributor_poll_pkt(struct rte_distributor *d, + unsigned worker_id) +{ + union rte_distributor_buffer *buf = &d->bufs[worker_id]; + if (buf->bufptr64 & RTE_DISTRIB_GET_BUF) + return NULL; + + /* 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); +} + +struct rte_mbuf * +rte_distributor_get_pkt(struct rte_distributor *d, + unsigned worker_id, struct rte_mbuf *oldpkt) +{ + struct rte_mbuf *ret; + rte_distributor_request_pkt(d, worker_id, oldpkt); + while ((ret = rte_distributor_poll_pkt(d, worker_id)) == NULL) + rte_pause(); + return 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->in_flight_bitmask &= ~(1UL << wkr); + 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. + * Note that the tags were set before first level call + * to rte_distributor_process. + */ + 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; + d->in_flight_bitmask &= ~(1UL << wkr); + } + 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); + /* + * User is advocated to set tag vaue for each + * mbuf before calling rte_distributor_process. + * User defined tags are used to identify flows, + * or sessions. + */ + new_tag = next_mb->hash.usr; + + /* + * Note that if RTE_DISTRIB_MAX_WORKERS is larger than 64 + * then the size of match has to be expanded. + */ + uint64_t match = 0; + unsigned i; + /* + * to scan for a match use "xor" and "not" to get a 0/1 + * value, then use shifting to merge to single "match" + * variable, where a one-bit indicates a match for the + * worker given by the bit-position + */ + for (i = 0; i < d->num_workers; i++) + match |= (!(d->in_flight_tags[i] ^ new_tag) + << i); + + /* Only turned-on bits are considered as match */ + match &= d->in_flight_bitmask; + + if (match) { + next_mb = NULL; + unsigned worker = __builtin_ctzl(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; + d->in_flight_bitmask |= (1UL << wkr); + 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; + + total_outstanding = __builtin_popcountl(d->in_flight_bitmask); + + for (wkr = 0; wkr < d->num_workers; wkr++) + total_outstanding += d->backlog[wkr].count; + + 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) & RTE_CACHE_LINE_MASK) != 0); + RTE_BUILD_BUG_ON((RTE_DISTRIB_MAX_WORKERS & 7) != 0); + RTE_BUILD_BUG_ON(RTE_DISTRIB_MAX_WORKERS > + sizeof(d->in_flight_bitmask) * CHAR_BIT); + + if (name == NULL || num_workers >= RTE_DISTRIB_MAX_WORKERS) { + rte_errno = EINVAL; + return NULL; + } + + 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; + snprintf(d->name, sizeof(d->name), "%s", name); + d->num_workers = num_workers; + + distributor_list = RTE_TAILQ_CAST(rte_distributor_tailq.head, + rte_distributor_list); + + 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_v20.h b/lib/librte_distributor/rte_distributor_v20.h new file mode 100644 index 0000000..7d36bc8 --- /dev/null +++ b/lib/librte_distributor/rte_distributor_v20.h @@ -0,0 +1,247 @@ +/*- + * 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_ + +/** + * @file + * RTE distributor + * + * The distributor is a component which is designed to pass packets + * one-at-a-time to workers, with dynamic load balancing. + */ + +#ifdef __cplusplus +extern "C" { +#endif + +#define RTE_DISTRIBUTOR_NAMESIZE 32 /**< Length of name for instance */ + +struct rte_distributor; +struct rte_mbuf; + +/** + * 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. + * + * The user is advocated to set tag for each mbuf before calling this function. + * If user doesn't set the tag, the tag value can be various values depending on + * driver implementation and configuration. + * + * 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); + +/** + * API called by a worker to request 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. + * Unlike rte_distributor_get_pkt(), this function does not wait for a new + * packet to be provided by the distributor. + * + * NOTE: after calling this function, rte_distributor_poll_pkt() should + * be used to poll for the packet requested. The rte_distributor_get_pkt() + * API should *not* be used to try and retrieve the new packet. + * + * @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 + */ +void +rte_distributor_request_pkt(struct rte_distributor *d, + unsigned worker_id, struct rte_mbuf *oldpkt); + +/** + * API called by a worker to check for a new packet that was previously + * requested by a call to rte_distributor_request_pkt(). It does not wait + * for the new packet to be available, but returns NULL if the request has + * not yet been fulfilled by the distributor. + * + * @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. + * + * @return + * A new packet to be processed by the worker thread, or NULL if no + * packet is yet available. + */ +struct rte_mbuf * +rte_distributor_poll_pkt(struct rte_distributor *d, + unsigned worker_id); + +#ifdef __cplusplus +} +#endif + +#endif -- 2.7.4