From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from dpdk.org (dpdk.org [92.243.14.124]) by inbox.dpdk.org (Postfix) with ESMTP id 95326A0531; Mon, 10 Feb 2020 00:07:37 +0100 (CET) Received: from [92.243.14.124] (localhost [127.0.0.1]) by dpdk.org (Postfix) with ESMTP id 10BD91BEE2; Mon, 10 Feb 2020 00:07:36 +0100 (CET) Received: from mellanox.co.il (mail-il-dmz.mellanox.com [193.47.165.129]) by dpdk.org (Postfix) with ESMTP id 15C2C1BEC4 for ; Mon, 10 Feb 2020 00:07:33 +0100 (CET) Received: from Internal Mail-Server by MTLPINE1 (envelope-from viacheslavo@mellanox.com) with ESMTPS (AES256-SHA encrypted); 10 Feb 2020 01:07:30 +0200 Received: from pegasus11.mtr.labs.mlnx (pegasus11.mtr.labs.mlnx [10.210.16.104]) by labmailer.mlnx (8.13.8/8.13.8) with ESMTP id 019N7TWt027806; Mon, 10 Feb 2020 01:07:29 +0200 Received: from pegasus11.mtr.labs.mlnx (localhost [127.0.0.1]) by pegasus11.mtr.labs.mlnx (8.14.7/8.14.7) with ESMTP id 019MssrO001921; Sun, 9 Feb 2020 22:54:54 GMT Received: (from viacheslavo@localhost) by pegasus11.mtr.labs.mlnx (8.14.7/8.14.7/Submit) id 019MssfU001920; Sun, 9 Feb 2020 22:54:54 GMT X-Authentication-Warning: pegasus11.mtr.labs.mlnx: viacheslavo set sender to viacheslavo@mellanox.com using -f From: Viacheslav Ovsiienko To: dev@dpdk.org Cc: matan@mellanox.com, rasland@mellanox.com, orika@mellanox.com, stable@dpdk.org Date: Sun, 9 Feb 2020 22:54:53 +0000 Message-Id: <1581288893-1878-1-git-send-email-viacheslavo@mellanox.com> X-Mailer: git-send-email 1.8.3.1 Subject: [dpdk-dev] [PATCH] net/mlx5: fix the legacy multi packet write session 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" To provide the better PCIe bandwidth utilization the ConnectX-4LX NIC supports the multi-packet write (MPW) sessions allowing to pack multiple packets into one descriptor (WQE). This is legacy feature and it has some limitations on the packets and data description segments. To provide the best performance all inline packets must be put into shared data segment and the total length of MPW session must be limited. The limit is controlled with txq_inline_mpw devarg. Fixes: 82e75f8323bf ("net/mlx5: fix legacy multi-packet Tx descriptors") Cc: stable@dpdk.org Signed-off-by: Viacheslav Ovsiienko --- drivers/common/mlx5/mlx5_prm.h | 2 +- drivers/net/mlx5/mlx5_rxtx.c | 90 ++++++++++++++++++++++++++++++++++++------ 2 files changed, 80 insertions(+), 12 deletions(-) diff --git a/drivers/common/mlx5/mlx5_prm.h b/drivers/common/mlx5/mlx5_prm.h index 855b37a..4ac3d4b 100644 --- a/drivers/common/mlx5/mlx5_prm.h +++ b/drivers/common/mlx5/mlx5_prm.h @@ -100,7 +100,7 @@ */ #define MLX5_EMPW_MAX_PACKETS MLX5_TX_COMP_THRESH #define MLX5_MPW_MAX_PACKETS 6 -#define MLX5_MPW_INLINE_MAX_PACKETS 2 +#define MLX5_MPW_INLINE_MAX_PACKETS 6 /* * Default packet length threshold to be inlined with diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c index 5eea932..86e0fab 100644 --- a/drivers/net/mlx5/mlx5_rxtx.c +++ b/drivers/net/mlx5/mlx5_rxtx.c @@ -2949,8 +2949,14 @@ enum mlx5_txcmp_code { unsigned int part; uint8_t *pdst; - dseg->bcount = rte_cpu_to_be_32(len | MLX5_ETH_WQE_DATA_INLINE); - pdst = &dseg->inline_data[0]; + if (!MLX5_TXOFF_CONFIG(MPW)) { + /* Store the descriptor byte counter for eMPW sessions. */ + dseg->bcount = rte_cpu_to_be_32(len | MLX5_ETH_WQE_DATA_INLINE); + pdst = &dseg->inline_data[0]; + } else { + /* The entire legacy MPW session counter is stored on close. */ + pdst = (uint8_t *)dseg; + } /* * The WQEBB space availability is checked by caller. * Here we should be aware of WQE ring buffer wraparound only. @@ -2962,7 +2968,8 @@ enum mlx5_txcmp_code { len -= part; if (likely(!len)) { pdst += part; - pdst = RTE_PTR_ALIGN(pdst, MLX5_WSEG_SIZE); + if (!MLX5_TXOFF_CONFIG(MPW)) + pdst = RTE_PTR_ALIGN(pdst, MLX5_WSEG_SIZE); /* Note: no final wraparound check here. */ return (struct mlx5_wqe_dseg *)pdst; } @@ -3010,9 +3017,16 @@ enum mlx5_txcmp_code { static_assert(MLX5_DSEG_MIN_INLINE_SIZE == (2 * RTE_ETHER_ADDR_LEN), "invalid Data Segment data size"); - dseg->bcount = rte_cpu_to_be_32((len + sizeof(struct rte_vlan_hdr)) | - MLX5_ETH_WQE_DATA_INLINE); - pdst = &dseg->inline_data[0]; + if (!MLX5_TXOFF_CONFIG(MPW)) { + /* Store the descriptor byte counter for eMPW sessions. */ + dseg->bcount = rte_cpu_to_be_32 + ((len + sizeof(struct rte_vlan_hdr)) | + MLX5_ETH_WQE_DATA_INLINE); + pdst = &dseg->inline_data[0]; + } else { + /* The entire legacy MPW session counter is stored on close. */ + pdst = (uint8_t *)dseg; + } memcpy(pdst, buf, MLX5_DSEG_MIN_INLINE_SIZE); buf += MLX5_DSEG_MIN_INLINE_SIZE; pdst += MLX5_DSEG_MIN_INLINE_SIZE; @@ -3035,7 +3049,8 @@ enum mlx5_txcmp_code { len -= part; if (likely(!len)) { pdst += part; - pdst = RTE_PTR_ALIGN(pdst, MLX5_WSEG_SIZE); + if (!MLX5_TXOFF_CONFIG(MPW)) + pdst = RTE_PTR_ALIGN(pdst, MLX5_WSEG_SIZE); /* Note: no final wraparound check here. */ return (struct mlx5_wqe_dseg *)pdst; } @@ -3921,15 +3936,33 @@ enum mlx5_txcmp_code { unsigned int slen, unsigned int olx __rte_unused) { + struct mlx5_wqe_dseg *dseg = &loc->wqe_last->dseg[0]; + MLX5_ASSERT(MLX5_TXOFF_CONFIG(INLINE)); - MLX5_ASSERT((len % MLX5_WSEG_SIZE) == 0); #ifdef MLX5_PMD_SOFT_COUNTERS /* Update sent data bytes counter. */ txq->stats.obytes += slen; #else (void)slen; #endif - len = len / MLX5_WSEG_SIZE + 2; + if (MLX5_TXOFF_CONFIG(MPW) && dseg->bcount == RTE_BE32(0)) { + /* + * If the legacy MPW session contains the inline packets + * we should set the only inline data segment length + * and align the total length to the segment size. + */ + MLX5_ASSERT(len > sizeof(dseg->bcount)); + dseg->bcount = rte_cpu_to_be_32((len - sizeof(dseg->bcount)) | + MLX5_ETH_WQE_DATA_INLINE); + len = (len + MLX5_WSEG_SIZE - 1) / MLX5_WSEG_SIZE + 2; + } else { + /* + * The session is not legacy MPW or contains the + * data buffer pointer segments. + */ + MLX5_ASSERT((len % MLX5_WSEG_SIZE) == 0); + len = len / MLX5_WSEG_SIZE + 2; + } loc->wqe_last->cseg.sq_ds = rte_cpu_to_be_32(txq->qp_num_8s | len); txq->wqe_ci += (len + 3) / 4; loc->wqe_free -= (len + 3) / 4; @@ -4208,6 +4241,15 @@ enum mlx5_txcmp_code { loc->wqe_free) * MLX5_WQE_SIZE - MLX5_WQE_CSEG_SIZE - MLX5_WQE_ESEG_SIZE; + /* Limit the room for legacy MPW sessions for performance. */ + if (MLX5_TXOFF_CONFIG(MPW)) + room = RTE_MIN(room, + RTE_MAX(txq->inlen_empw + + sizeof(dseg->bcount) + + (MLX5_TXOFF_CONFIG(VLAN) ? + sizeof(struct rte_vlan_hdr) : 0), + MLX5_MPW_INLINE_MAX_PACKETS * + MLX5_WQE_DSEG_SIZE)); /* Build WQE till we have space, packets and resources. */ part = room; for (;;) { @@ -4238,8 +4280,26 @@ enum mlx5_txcmp_code { if (dlen > txq->inlen_empw || loc->mbuf->ol_flags & PKT_TX_DYNF_NOINLINE) goto pointer_empw; + if (MLX5_TXOFF_CONFIG(MPW)) { + tlen = dlen; + if (part == room) { + /* Open new inline MPW session. */ + tlen += sizeof(dseg->bcount); + dseg->bcount = RTE_BE32(0); + dseg = RTE_PTR_ADD + (dseg, sizeof(dseg->bcount)); + } else { + /* + * No pointer and inline descriptor + * intermix for legacy MPW sessions. + */ + if (loc->wqe_last->dseg[0].bcount) + break; + } + } else { + tlen = sizeof(dseg->bcount) + dlen; + } /* Inline entire packet, optional VLAN insertion. */ - tlen = sizeof(dseg->bcount) + dlen; if (MLX5_TXOFF_CONFIG(VLAN) && loc->mbuf->ol_flags & PKT_TX_VLAN_PKT) { /* @@ -4265,7 +4325,8 @@ enum mlx5_txcmp_code { dseg = mlx5_tx_dseg_empw(txq, loc, dseg, dptr, dlen, olx); } - tlen = RTE_ALIGN(tlen, MLX5_WSEG_SIZE); + if (!MLX5_TXOFF_CONFIG(MPW)) + tlen = RTE_ALIGN(tlen, MLX5_WSEG_SIZE); MLX5_ASSERT(room >= tlen); room -= tlen; /* @@ -4276,6 +4337,13 @@ enum mlx5_txcmp_code { goto next_mbuf; pointer_empw: /* + * No pointer and inline descriptor + * intermix for legacy MPW sessions. + */ + if (MLX5_TXOFF_CONFIG(MPW) && + loc->wqe_last->dseg[0].bcount == RTE_BE32(0)) + break; + /* * Not inlinable VLAN packets are * proceeded outside of this routine. */ -- 1.8.3.1