From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mails.dpdk.org (mails.dpdk.org [217.70.189.124]) by inbox.dpdk.org (Postfix) with ESMTP id BBE96A00C5; Thu, 15 Sep 2022 08:58:45 +0200 (CEST) Received: from [217.70.189.124] (localhost [127.0.0.1]) by mails.dpdk.org (Postfix) with ESMTP id 671D34281A; Thu, 15 Sep 2022 08:58:41 +0200 (CEST) Received: from foss.arm.com (foss.arm.com [217.140.110.172]) by mails.dpdk.org (Postfix) with ESMTP id A53B440223 for ; Thu, 15 Sep 2022 08:58:39 +0200 (CEST) Received: from usa-sjc-imap-foss1.foss.arm.com (unknown [10.121.207.14]) by usa-sjc-mx-foss1.foss.arm.com (Postfix) with ESMTP id A027D1692; Wed, 14 Sep 2022 23:58:45 -0700 (PDT) Received: from net-arm-n1sdp-02.shanghai.arm.com (net-arm-n1sdp-02.shanghai.arm.com [10.169.210.140]) by usa-sjc-imap-foss1.foss.arm.com (Postfix) with ESMTPA id ACC5E3F73B; Wed, 14 Sep 2022 23:59:05 -0700 (PDT) From: Joyce Kong To: jgrajcia@cisco.com, stephen@networkplumber.org Cc: dev@dpdk.org, nd@arm.com, Joyce Kong Subject: [PATCH v4 1/2] net/memif: add a Rx fast path Date: Thu, 15 Sep 2022 14:58:24 +0800 Message-Id: <20220915065825.26169-2-joyce.kong@arm.com> X-Mailer: git-send-email 2.17.1 In-Reply-To: <20220915065825.26169-1-joyce.kong@arm.com> References: <20220412093243.3670187-1-joyce.kong@arm.com> <20220915065825.26169-1-joyce.kong@arm.com> MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org For memif non-zero-copy mode, there is a branch to compare the mbuf and memif buffer size during memory copying. Mbuf and memif buffer size is defined at compile time. If memif buf size <= mbuf size, add a fast Rx memory copy path by removing this branch and mbuf bulk alloc. The removal of the branch and bulk alloc lead to considerable performance uplift. Test with 1p1q on N1SDP AArch64 server, -------------------------------------------- buf size | memif <= mbuf | memif > mbuf | -------------------------------------------- non-zc gain | 26.85% | -0.37% | -------------------------------------------- zc gain | 8.57% | 3.04% | -------------------------------------------- Test with 1p1q on Cascade Lake Xeon X86 server, -------------------------------------------- buf size | memif <= mbuf | memif > mbuf | -------------------------------------------- non-zc gain | 17.54% | -0.42% | -------------------------------------------- zc gain | 10.67% | 0.26% | -------------------------------------------- Signed-off-by: Joyce Kong Reviewed-by: Ruifeng Wang Acked-by: Morten Brørup --- drivers/net/memif/rte_eth_memif.c | 146 +++++++++++++++++++++--------- drivers/net/memif/rte_eth_memif.h | 2 + 2 files changed, 107 insertions(+), 41 deletions(-) diff --git a/drivers/net/memif/rte_eth_memif.c b/drivers/net/memif/rte_eth_memif.c index dd951b8296..762293f636 100644 --- a/drivers/net/memif/rte_eth_memif.c +++ b/drivers/net/memif/rte_eth_memif.c @@ -297,7 +297,7 @@ eth_memif_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts) rte_eth_devices[mq->in_port].process_private; memif_ring_t *ring = memif_get_ring_from_queue(proc_private, mq); uint16_t cur_slot, last_slot, n_slots, ring_size, mask, s0; - uint16_t n_rx_pkts = 0; + uint16_t pkts, rx_pkts, n_rx_pkts = 0; uint16_t mbuf_size = rte_pktmbuf_data_room_size(mq->mempool) - RTE_PKTMBUF_HEADROOM; uint16_t src_len, src_off, dst_len, dst_off, cp_len; @@ -342,66 +342,131 @@ eth_memif_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts) goto refill; n_slots = last_slot - cur_slot; - while (n_slots && n_rx_pkts < nb_pkts) { - mbuf_head = rte_pktmbuf_alloc(mq->mempool); - if (unlikely(mbuf_head == NULL)) + if (likely(mbuf_size >= pmd->cfg.pkt_buffer_size)) { + struct rte_mbuf *mbufs[MAX_PKT_BURST]; +next_bulk: + ret = rte_pktmbuf_alloc_bulk(mq->mempool, mbufs, MAX_PKT_BURST); + if (unlikely(ret < 0)) goto no_free_bufs; - mbuf = mbuf_head; - mbuf->port = mq->in_port; - dst_off = 0; -next_slot: - s0 = cur_slot & mask; - d0 = &ring->desc[s0]; + rx_pkts = 0; + pkts = nb_pkts < MAX_PKT_BURST ? nb_pkts : MAX_PKT_BURST; + while (n_slots && rx_pkts < pkts) { + mbuf_head = mbufs[n_rx_pkts]; + mbuf = mbuf_head; - src_len = d0->length; - src_off = 0; +next_slot1: + mbuf->port = mq->in_port; + s0 = cur_slot & mask; + d0 = &ring->desc[s0]; - do { - dst_len = mbuf_size - dst_off; - if (dst_len == 0) { - dst_off = 0; - dst_len = mbuf_size; + cp_len = d0->length; + + rte_pktmbuf_data_len(mbuf) = cp_len; + rte_pktmbuf_pkt_len(mbuf) = cp_len; + if (mbuf != mbuf_head) + rte_pktmbuf_pkt_len(mbuf_head) += cp_len; + + rte_memcpy(rte_pktmbuf_mtod(mbuf, void *), + (uint8_t *)memif_get_buffer(proc_private, d0), cp_len); - /* store pointer to tail */ + cur_slot++; + n_slots--; + + if (d0->flags & MEMIF_DESC_FLAG_NEXT) { mbuf_tail = mbuf; mbuf = rte_pktmbuf_alloc(mq->mempool); - if (unlikely(mbuf == NULL)) + if (unlikely(mbuf == NULL)) { + rte_pktmbuf_free_bulk(mbufs + rx_pkts, + MAX_PKT_BURST - rx_pkts); goto no_free_bufs; - mbuf->port = mq->in_port; + } ret = memif_pktmbuf_chain(mbuf_head, mbuf_tail, mbuf); if (unlikely(ret < 0)) { MIF_LOG(ERR, "number-of-segments-overflow"); rte_pktmbuf_free(mbuf); + rte_pktmbuf_free_bulk(mbufs + rx_pkts, + MAX_PKT_BURST - rx_pkts); goto no_free_bufs; } + goto next_slot1; } - cp_len = RTE_MIN(dst_len, src_len); - rte_pktmbuf_data_len(mbuf) += cp_len; - rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf); - if (mbuf != mbuf_head) - rte_pktmbuf_pkt_len(mbuf_head) += cp_len; + mq->n_bytes += rte_pktmbuf_pkt_len(mbuf_head); + *bufs++ = mbuf_head; + rx_pkts++; + n_rx_pkts++; + } - rte_memcpy(rte_pktmbuf_mtod_offset(mbuf, void *, - dst_off), - (uint8_t *)memif_get_buffer(proc_private, d0) + - src_off, cp_len); + if (rx_pkts < MAX_PKT_BURST) { + rte_pktmbuf_free_bulk(mbufs + rx_pkts, MAX_PKT_BURST - rx_pkts); + } else { + nb_pkts -= rx_pkts; + if (nb_pkts) + goto next_bulk; + } + } else { + while (n_slots && n_rx_pkts < nb_pkts) { + mbuf_head = rte_pktmbuf_alloc(mq->mempool); + if (unlikely(mbuf_head == NULL)) + goto no_free_bufs; + mbuf = mbuf_head; + mbuf->port = mq->in_port; + +next_slot2: + s0 = cur_slot & mask; + d0 = &ring->desc[s0]; - src_off += cp_len; - dst_off += cp_len; - src_len -= cp_len; - } while (src_len); + src_len = d0->length; + dst_off = 0; + src_off = 0; - cur_slot++; - n_slots--; + do { + dst_len = mbuf_size - dst_off; + if (dst_len == 0) { + dst_off = 0; + dst_len = mbuf_size; + + /* store pointer to tail */ + mbuf_tail = mbuf; + mbuf = rte_pktmbuf_alloc(mq->mempool); + if (unlikely(mbuf == NULL)) + goto no_free_bufs; + mbuf->port = mq->in_port; + ret = memif_pktmbuf_chain(mbuf_head, mbuf_tail, mbuf); + if (unlikely(ret < 0)) { + MIF_LOG(ERR, "number-of-segments-overflow"); + rte_pktmbuf_free(mbuf); + goto no_free_bufs; + } + } + cp_len = RTE_MIN(dst_len, src_len); - if (d0->flags & MEMIF_DESC_FLAG_NEXT) - goto next_slot; + rte_pktmbuf_data_len(mbuf) += cp_len; + rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf); + if (mbuf != mbuf_head) + rte_pktmbuf_pkt_len(mbuf_head) += cp_len; - mq->n_bytes += rte_pktmbuf_pkt_len(mbuf_head); - *bufs++ = mbuf_head; - n_rx_pkts++; + rte_memcpy(rte_pktmbuf_mtod_offset(mbuf, void *, + dst_off), + (uint8_t *)memif_get_buffer(proc_private, d0) + + src_off, cp_len); + + src_off += cp_len; + dst_off += cp_len; + src_len -= cp_len; + } while (src_len); + + cur_slot++; + n_slots--; + + if (d0->flags & MEMIF_DESC_FLAG_NEXT) + goto next_slot2; + + mq->n_bytes += rte_pktmbuf_pkt_len(mbuf_head); + *bufs++ = mbuf_head; + n_rx_pkts++; + } } no_free_bufs: @@ -694,7 +759,6 @@ eth_memif_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts) return n_tx_pkts; } - static int memif_tx_one_zc(struct pmd_process_private *proc_private, struct memif_queue *mq, memif_ring_t *ring, struct rte_mbuf *mbuf, const uint16_t mask, diff --git a/drivers/net/memif/rte_eth_memif.h b/drivers/net/memif/rte_eth_memif.h index 81e7dceae0..09928ecc86 100644 --- a/drivers/net/memif/rte_eth_memif.h +++ b/drivers/net/memif/rte_eth_memif.h @@ -25,6 +25,8 @@ #define ETH_MEMIF_DISC_STRING_SIZE 96 #define ETH_MEMIF_SECRET_SIZE 24 +#define MAX_PKT_BURST 32 + extern int memif_logtype; #define MIF_LOG(level, fmt, args...) \ -- 2.17.1