From patchwork Fri Dec 18 17:14:56 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Slava Ovsiienko X-Patchwork-Id: 85483 X-Patchwork-Delegate: rasland@nvidia.com Return-Path: X-Original-To: patchwork@inbox.dpdk.org Delivered-To: patchwork@inbox.dpdk.org Received: from dpdk.org (dpdk.org [92.243.14.124]) by inbox.dpdk.org (Postfix) with ESMTP id 5743FA09FD; Fri, 18 Dec 2020 18:15:43 +0100 (CET) Received: from [92.243.14.124] (localhost [127.0.0.1]) by dpdk.org (Postfix) with ESMTP id DAE6CCB12; Fri, 18 Dec 2020 18:15:08 +0100 (CET) Received: from mellanox.co.il (mail-il-dmz.mellanox.com [193.47.165.129]) by dpdk.org (Postfix) with ESMTP id A9D5FCAFF for ; Fri, 18 Dec 2020 18:15:05 +0100 (CET) Received: from Internal Mail-Server by MTLPINE1 (envelope-from viacheslavo@nvidia.com) with SMTP; 18 Dec 2020 19:15:00 +0200 Received: from nvidia.com (pegasus12.mtr.labs.mlnx [10.210.17.40]) by labmailer.mlnx (8.13.8/8.13.8) with ESMTP id 0BIHEwIR006669; Fri, 18 Dec 2020 19:14:59 +0200 From: Viacheslav Ovsiienko To: dev@dpdk.org Cc: rasland@nvidia.com, matan@nvidia.com, orika@nvidia.com, thomas@monjalon.net, stable@dpdk.org Date: Fri, 18 Dec 2020 17:14:56 +0000 Message-Id: <1608311697-31529-2-git-send-email-viacheslavo@nvidia.com> X-Mailer: git-send-email 1.8.3.1 In-Reply-To: <1608311697-31529-1-git-send-email-viacheslavo@nvidia.com> References: <1608311697-31529-1-git-send-email-viacheslavo@nvidia.com> Subject: [dpdk-dev] [PATCH 1/2] net/mlx5: optimize inline mbuf freeing X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Sender: "dev" The mlx5 PMD supports packet data inlining by pushing data to the transmit descriptor. If packet is short enough and all data are inline, the mbuf is not needed for data send anymore and can be freed. The mbuf free was performed in the most inner loop building the transmit descriptors. This patch postpones the mbuf free transaction to the tx_burst routine exit, optimizing the loop and allowing the bulk freeing for the multiple mbufs in single pool API call. Cc: stable@dpdk.org Signed-off-by: Viacheslav Ovsiienko --- drivers/net/mlx5/mlx5_rxtx.c | 38 ++++++++++++++++++++++++++++++++++---- drivers/net/mlx5/mlx5_rxtx.h | 1 + 2 files changed, 35 insertions(+), 4 deletions(-) diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c index d12d746..e8c8783 100644 --- a/drivers/net/mlx5/mlx5_rxtx.c +++ b/drivers/net/mlx5/mlx5_rxtx.c @@ -1940,6 +1940,17 @@ enum mlx5_txcmp_code { } } } +/* + * No inline version to free buffers for optimal call + * on the tx_burst completion. + */ +static __rte_noinline void +__mlx5_tx_free_mbuf(struct rte_mbuf **__rte_restrict pkts, + unsigned int pkts_n, + unsigned int olx __rte_unused) +{ + mlx5_tx_free_mbuf(pkts, pkts_n, olx); +} /** * Free the mbuf from the elts ring buffer till new tail. @@ -4392,10 +4403,25 @@ enum mlx5_txcmp_code { MLX5_ASSERT(room >= tlen); room -= tlen; /* - * Packet data are completely inlined, - * free the packet immediately. + * Packet data are completely inline, + * we can try to free the packet. + */ + if (likely(loc->pkts_sent == loc->mbuf_free)) { + /* + * All the packets from the burst beginning + * are inline, we can free mbufs directly + * from the origin array on tx_burst exit(). + */ + loc->mbuf_free++; + goto next_mbuf; + } + /* + * In order no to call rte_pktmbuf_free_seg() here, + * in the most inner loop (that might be very + * expensive) we just save the mbuf in elts. */ - rte_pktmbuf_free_seg(loc->mbuf); + txq->elts[txq->elts_head++ & txq->elts_m] = loc->mbuf; + loc->elts_free--; goto next_mbuf; pointer_empw: /* @@ -4417,6 +4443,7 @@ enum mlx5_txcmp_code { mlx5_tx_dseg_ptr(txq, loc, dseg, dptr, dlen, olx); /* We have to store mbuf in elts.*/ txq->elts[txq->elts_head++ & txq->elts_m] = loc->mbuf; + loc->elts_free--; room -= MLX5_WQE_DSEG_SIZE; /* Ring buffer wraparound is checked at the loop end.*/ ++dseg; @@ -4426,7 +4453,6 @@ enum mlx5_txcmp_code { slen += dlen; #endif loc->pkts_sent++; - loc->elts_free--; pkts_n--; if (unlikely(!pkts_n || !loc->elts_free)) { /* @@ -4880,6 +4906,8 @@ enum mlx5_txcmp_code { MLX5_ASSERT(txq->wqe_s >= (uint16_t)(txq->wqe_ci - txq->wqe_pi)); if (unlikely(!pkts_n)) return 0; + if (MLX5_TXOFF_CONFIG(INLINE)) + loc.mbuf_free = 0; loc.pkts_sent = 0; loc.pkts_copy = 0; loc.wqe_last = NULL; @@ -5143,6 +5171,8 @@ enum mlx5_txcmp_code { /* Increment sent packets counter. */ txq->stats.opackets += loc.pkts_sent; #endif + if (MLX5_TXOFF_CONFIG(INLINE) && loc.mbuf_free) + __mlx5_tx_free_mbuf(pkts, loc.mbuf_free, olx); return loc.pkts_sent; } diff --git a/drivers/net/mlx5/mlx5_rxtx.h b/drivers/net/mlx5/mlx5_rxtx.h index 7989a50..fc5cc2e 100644 --- a/drivers/net/mlx5/mlx5_rxtx.h +++ b/drivers/net/mlx5/mlx5_rxtx.h @@ -217,6 +217,7 @@ struct mlx5_txq_local { uint16_t wqe_free; /* available wqe remain. */ uint16_t mbuf_off; /* data offset in current mbuf. */ uint16_t mbuf_nseg; /* number of remaining mbuf. */ + uint16_t mbuf_free; /* number of inline mbufs to free. */ }; /* TX queue descriptor. */