From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mga01.intel.com (mga01.intel.com [192.55.52.88]) by dpdk.org (Postfix) with ESMTP id 8DADA7E23 for ; Thu, 12 Feb 2015 09:17:45 +0100 (CET) Received: from fmsmga003.fm.intel.com ([10.253.24.29]) by fmsmga101.fm.intel.com with ESMTP; 12 Feb 2015 00:17:44 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.09,564,1418112000"; d="scan'208";a="453566864" Received: from shvmail01.sh.intel.com ([10.239.29.42]) by FMSMGA003.fm.intel.com with ESMTP; 12 Feb 2015 00:03:01 -0800 Received: from shecgisg004.sh.intel.com (shecgisg004.sh.intel.com [10.239.29.89]) by shvmail01.sh.intel.com with ESMTP id t1C8HdIf013186; Thu, 12 Feb 2015 16:17:39 +0800 Received: from shecgisg004.sh.intel.com (localhost [127.0.0.1]) by shecgisg004.sh.intel.com (8.13.6/8.13.6/SuSE Linux 0.8) with ESMTP id t1C8HaqP003189; Thu, 12 Feb 2015 16:17:38 +0800 Received: (from cliang18@localhost) by shecgisg004.sh.intel.com (8.13.6/8.13.6/Submit) id t1C8Haw6003185; Thu, 12 Feb 2015 16:17:36 +0800 From: Cunming Liang To: dev@dpdk.org Date: Thu, 12 Feb 2015 16:16:35 +0800 Message-Id: <1423728996-3004-19-git-send-email-cunming.liang@intel.com> X-Mailer: git-send-email 1.7.4.1 In-Reply-To: <1423728996-3004-1-git-send-email-cunming.liang@intel.com> References: <1422842559-13617-1-git-send-email-cunming.liang@intel.com> <1423728996-3004-1-git-send-email-cunming.liang@intel.com> Subject: [dpdk-dev] [PATCH v5 18/19] ring: add sched_yield to avoid spin forever X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: patches and discussions about DPDK List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Thu, 12 Feb 2015 08:17:46 -0000 Add a sched_yield() syscall if the thread spins for too long, waiting other thread to finish its operations on the ring. That gives pre-empted thread a chance to proceed and finish with ring enqnue/dequeue operation. The purpose is to reduce contention on the ring. By ring_perf_test, it doesn't shows additional perf penalty. Signed-off-by: Cunming Liang --- v5 changes: add RTE_RING_PAUSE_REP to config file v4 changes: update and add more comments on sched_yield() v3 changes: new patch adding sched_yield() in rte_ring to avoid long spin config/common_bsdapp | 1 + config/common_linuxapp | 1 + lib/librte_ring/rte_ring.h | 31 +++++++++++++++++++++++++++---- 3 files changed, 29 insertions(+), 4 deletions(-) diff --git a/config/common_bsdapp b/config/common_bsdapp index 57bacb8..52c5143 100644 --- a/config/common_bsdapp +++ b/config/common_bsdapp @@ -234,6 +234,7 @@ CONFIG_RTE_PMD_PACKET_PREFETCH=y CONFIG_RTE_LIBRTE_RING=y CONFIG_RTE_LIBRTE_RING_DEBUG=n CONFIG_RTE_RING_SPLIT_PROD_CONS=n +CONFIG_RTE_RING_PAUSE_REP=n # # Compile librte_mempool diff --git a/config/common_linuxapp b/config/common_linuxapp index d428f84..0b4eb3c 100644 --- a/config/common_linuxapp +++ b/config/common_linuxapp @@ -242,6 +242,7 @@ CONFIG_RTE_PMD_PACKET_PREFETCH=y CONFIG_RTE_LIBRTE_RING=y CONFIG_RTE_LIBRTE_RING_DEBUG=n CONFIG_RTE_RING_SPLIT_PROD_CONS=n +CONFIG_RTE_RING_PAUSE_REP=n # # Compile librte_mempool diff --git a/lib/librte_ring/rte_ring.h b/lib/librte_ring/rte_ring.h index 553a880..eecba0b 100644 --- a/lib/librte_ring/rte_ring.h +++ b/lib/librte_ring/rte_ring.h @@ -127,6 +127,11 @@ struct rte_ring_debug_stats { #define RTE_RING_NAMESIZE 32 /**< The maximum length of a ring name. */ #define RTE_RING_MZ_PREFIX "RG_" +#ifndef RTE_RING_PAUSE_REP +#define RTE_RING_PAUSE_REP 0 /**< yield after pause num of times, + * no yield if RTE_RING_PAUSE_REP not defined. */ +#endif + /** * An RTE ring structure. * @@ -410,7 +415,7 @@ __rte_ring_mp_do_enqueue(struct rte_ring *r, void * const *obj_table, uint32_t cons_tail, free_entries; const unsigned max = n; int success; - unsigned i; + unsigned i, rep = 0; uint32_t mask = r->prod.mask; int ret; @@ -468,9 +473,18 @@ __rte_ring_mp_do_enqueue(struct rte_ring *r, void * const *obj_table, * If there are other enqueues in progress that preceded us, * we need to wait for them to complete */ - while (unlikely(r->prod.tail != prod_head)) + while (unlikely(r->prod.tail != prod_head)) { rte_pause(); + /* Set RTE_RING_PAUSE_REP to avoid spin too long waiting for + * other thread finish. It gives pre-empted thread a chance + * to proceed and finish with ring denqnue operation. */ + if (RTE_RING_PAUSE_REP && + ++rep == RTE_RING_PAUSE_REP) { + rep = 0; + sched_yield(); + } + } r->prod.tail = prod_next; return ret; } @@ -589,7 +603,7 @@ __rte_ring_mc_do_dequeue(struct rte_ring *r, void **obj_table, uint32_t cons_next, entries; const unsigned max = n; int success; - unsigned i; + unsigned i, rep = 0; uint32_t mask = r->prod.mask; /* move cons.head atomically */ @@ -634,9 +648,18 @@ __rte_ring_mc_do_dequeue(struct rte_ring *r, void **obj_table, * If there are other dequeues in progress that preceded us, * we need to wait for them to complete */ - while (unlikely(r->cons.tail != cons_head)) + while (unlikely(r->cons.tail != cons_head)) { rte_pause(); + /* Set RTE_RING_PAUSE_REP to avoid spin too long waiting for + * other thread finish. It gives pre-empted thread a chance + * to proceed and finish with ring denqnue operation. */ + if (RTE_RING_PAUSE_REP && + ++rep == RTE_RING_PAUSE_REP) { + rep = 0; + sched_yield(); + } + } __RING_STAT_ADD(r, deq_success, n); r->cons.tail = cons_next; -- 1.8.1.4