From patchwork Fri Jun 18 10:37:09 2021 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Nithin Dabilpuram X-Patchwork-Id: 94440 X-Patchwork-Delegate: jerinj@marvell.com Return-Path: X-Original-To: patchwork@inbox.dpdk.org Delivered-To: patchwork@inbox.dpdk.org Received: from mails.dpdk.org (mails.dpdk.org [217.70.189.124]) by inbox.dpdk.org (Postfix) with ESMTP id EF4AFA0C46; Fri, 18 Jun 2021 12:43:10 +0200 (CEST) Received: from [217.70.189.124] (localhost [127.0.0.1]) by mails.dpdk.org (Postfix) with ESMTP id 3BCDF4114C; Fri, 18 Jun 2021 12:40:41 +0200 (CEST) Received: from mx0b-0016f401.pphosted.com (mx0b-0016f401.pphosted.com [67.231.156.173]) by mails.dpdk.org (Postfix) with ESMTP id CFF6D41191 for ; Fri, 18 Jun 2021 12:40:38 +0200 (CEST) Received: from pps.filterd (m0045851.ppops.net [127.0.0.1]) by mx0b-0016f401.pphosted.com (8.16.0.43/8.16.0.43) with SMTP id 15IAZPAi003484 for ; Fri, 18 Jun 2021 03:40:38 -0700 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=marvell.com; h=from : to : cc : subject : date : message-id : in-reply-to : references : mime-version : content-type; s=pfpt0220; bh=M+p6xdBvr6HWYgHSjPg2NHuPOvHEKHV6qhK7cK62Lew=; b=CxMwyazO+Ux70KkTrjBva6VeDvEbFHeETH7UC3JqYNSfHo+lDwBummJDDocyLAmXK8NE Yy86TTXipfktLr12/GlPKxS+0Ivof2Z3i1RD8bpQJ6gWxmtDWzT7AyQ/EhkoWw0vLNkS 2WBFB3u9Vtro1DaCPHWSr6aaqj0QT5A619MO4Z4nVCfQtCEDsrP+e3X/dUzwoWOM1ga3 u9zPAaly6sv8QWeH6s4P1ojIHO00Mtn3GJUoUc3fn5sN4WoQOxI1Ayf3GIxEE5DTVHTH qboWJ/01jUzdnYJZxp+EmG9poS6XNcLelQK6e2y6zgxeLzuJJezL6sQxinmeP1Q3Z8CO Xw== Received: from dc5-exch01.marvell.com ([199.233.59.181]) by mx0b-0016f401.pphosted.com with ESMTP id 397udry7hj-10 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-SHA384 bits=256 verify=NOT) for ; Fri, 18 Jun 2021 03:40:38 -0700 Received: from DC5-EXCH01.marvell.com (10.69.176.38) by DC5-EXCH01.marvell.com (10.69.176.38) with Microsoft SMTP Server (TLS) id 15.0.1497.18; Fri, 18 Jun 2021 03:40:34 -0700 Received: from maili.marvell.com (10.69.176.80) by DC5-EXCH01.marvell.com (10.69.176.38) with Microsoft SMTP Server id 15.0.1497.18 via Frontend Transport; Fri, 18 Jun 2021 03:40:34 -0700 Received: from hyd1588t430.marvell.com (unknown [10.29.52.204]) by maili.marvell.com (Postfix) with ESMTP id 497E55B6946; Fri, 18 Jun 2021 03:40:06 -0700 (PDT) From: Nithin Dabilpuram To: CC: , , , , , , , , "Nithin Dabilpuram" Date: Fri, 18 Jun 2021 16:07:09 +0530 Message-ID: <20210618103741.26526-31-ndabilpuram@marvell.com> X-Mailer: git-send-email 2.8.4 In-Reply-To: <20210618103741.26526-1-ndabilpuram@marvell.com> References: <20210306153404.10781-1-ndabilpuram@marvell.com> <20210618103741.26526-1-ndabilpuram@marvell.com> MIME-Version: 1.0 X-Proofpoint-ORIG-GUID: AZ-jqNLrC3hl7uNg-AiVI1us1vU1eXR4 X-Proofpoint-GUID: AZ-jqNLrC3hl7uNg-AiVI1us1vU1eXR4 X-Proofpoint-Virus-Version: vendor=fsecure engine=2.50.10434:6.0.391, 18.0.790 definitions=2021-06-18_04:2021-06-18, 2021-06-18 signatures=0 Subject: [dpdk-dev] [PATCH v3 30/62] net/cnxk: add Tx vector version for cn10k 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 Sender: "dev" Add Tx burst vector version for CN10K. Signed-off-by: Nithin Dabilpuram Signed-off-by: Pavan Nikhilesh --- drivers/net/cnxk/cn10k_tx.c | 10 + drivers/net/cnxk/cn10k_tx.h | 815 ++++++++++++++++++++++++++++++++++++++++ drivers/net/cnxk/cn10k_tx_vec.c | 25 ++ drivers/net/cnxk/meson.build | 3 +- 4 files changed, 852 insertions(+), 1 deletion(-) create mode 100644 drivers/net/cnxk/cn10k_tx_vec.c diff --git a/drivers/net/cnxk/cn10k_tx.c b/drivers/net/cnxk/cn10k_tx.c index 9803002..e6eb101 100644 --- a/drivers/net/cnxk/cn10k_tx.c +++ b/drivers/net/cnxk/cn10k_tx.c @@ -58,10 +58,20 @@ cn10k_eth_set_tx_function(struct rte_eth_dev *eth_dev) #undef T }; + const eth_tx_burst_t nix_eth_tx_vec_burst[2][2][2][2][2] = { +#define T(name, f4, f3, f2, f1, f0, sz, flags) \ + [f4][f3][f2][f1][f0] = cn10k_nix_xmit_pkts_vec_##name, + + NIX_TX_FASTPATH_MODES +#undef T + }; + if (dev->scalar_ena || (dev->tx_offload_flags & (NIX_TX_OFFLOAD_VLAN_QINQ_F | NIX_TX_OFFLOAD_TSO_F))) pick_tx_func(eth_dev, nix_eth_tx_burst); + else + pick_tx_func(eth_dev, nix_eth_tx_vec_burst); if (dev->tx_offloads & DEV_TX_OFFLOAD_MULTI_SEGS) pick_tx_func(eth_dev, nix_eth_tx_burst_mseg); diff --git a/drivers/net/cnxk/cn10k_tx.h b/drivers/net/cnxk/cn10k_tx.h index 63e9848..b74df10 100644 --- a/drivers/net/cnxk/cn10k_tx.h +++ b/drivers/net/cnxk/cn10k_tx.h @@ -4,6 +4,8 @@ #ifndef __CN10K_TX_H__ #define __CN10K_TX_H__ +#include + #define NIX_TX_OFFLOAD_NONE (0) #define NIX_TX_OFFLOAD_L3_L4_CSUM_F BIT(0) #define NIX_TX_OFFLOAD_OL3_OL4_CSUM_F BIT(1) @@ -38,6 +40,9 @@ } \ } while (0) +#define LMT_OFF(lmt_addr, lmt_num, offset) \ + (void *)((lmt_addr) + ((lmt_num) << ROC_LMT_LINE_SIZE_LOG2) + (offset)) + /* Function to determine no of tx subdesc required in case ext * sub desc is enabled. */ @@ -48,6 +53,14 @@ cn10k_nix_tx_ext_subs(const uint16_t flags) (NIX_TX_OFFLOAD_VLAN_QINQ_F | NIX_TX_OFFLOAD_TSO_F)) ? 1 : 0; } +static __rte_always_inline uint8_t +cn10k_nix_pkts_per_vec_brst(const uint16_t flags) +{ + RTE_SET_USED(flags); + /* We can pack up to 4 packets per LMTLINE if there are no offloads. */ + return 4 << ROC_LMT_LINES_PER_CORE_LOG2; +} + static __rte_always_inline uint64_t cn10k_nix_tx_steor_data(const uint16_t flags) { @@ -76,6 +89,35 @@ cn10k_nix_tx_steor_data(const uint16_t flags) return data; } +static __rte_always_inline uint64_t +cn10k_nix_tx_steor_vec_data(const uint16_t flags) +{ + const uint64_t dw_m1 = 0x7; + uint64_t data; + + RTE_SET_USED(flags); + /* This will be moved to addr area */ + data = dw_m1; + /* 15 vector sizes for single seg */ + data |= dw_m1 << 19; + data |= dw_m1 << 22; + data |= dw_m1 << 25; + data |= dw_m1 << 28; + data |= dw_m1 << 31; + data |= dw_m1 << 34; + data |= dw_m1 << 37; + data |= dw_m1 << 40; + data |= dw_m1 << 43; + data |= dw_m1 << 46; + data |= dw_m1 << 49; + data |= dw_m1 << 52; + data |= dw_m1 << 55; + data |= dw_m1 << 58; + data |= dw_m1 << 61; + + return data; +} + static __rte_always_inline void cn10k_nix_tx_skeleton(const struct cn10k_eth_txq *txq, uint64_t *cmd, const uint16_t flags) @@ -589,6 +631,776 @@ cn10k_nix_xmit_pkts_mseg(void *tx_queue, struct rte_mbuf **tx_pkts, return pkts; } +#if defined(RTE_ARCH_ARM64) + +#define NIX_DESCS_PER_LOOP 4 +static __rte_always_inline uint16_t +cn10k_nix_xmit_pkts_vector(void *tx_queue, struct rte_mbuf **tx_pkts, + uint16_t pkts, uint64_t *cmd, const uint16_t flags) +{ + uint64x2_t dataoff_iova0, dataoff_iova1, dataoff_iova2, dataoff_iova3; + uint64x2_t len_olflags0, len_olflags1, len_olflags2, len_olflags3; + uint64x2_t cmd0[NIX_DESCS_PER_LOOP], cmd1[NIX_DESCS_PER_LOOP]; + uint64_t *mbuf0, *mbuf1, *mbuf2, *mbuf3, data, pa; + uint64x2_t senddesc01_w0, senddesc23_w0; + uint64x2_t senddesc01_w1, senddesc23_w1; + uint16_t left, scalar, burst, i, lmt_id; + uint64x2_t sgdesc01_w0, sgdesc23_w0; + uint64x2_t sgdesc01_w1, sgdesc23_w1; + struct cn10k_eth_txq *txq = tx_queue; + uintptr_t laddr = txq->lmt_base; + rte_iova_t io_addr = txq->io_addr; + uint64x2_t ltypes01, ltypes23; + uint64x2_t xtmp128, ytmp128; + uint64x2_t xmask01, xmask23; + uint8_t lnum; + + NIX_XMIT_FC_OR_RETURN(txq, pkts); + + scalar = pkts & (NIX_DESCS_PER_LOOP - 1); + pkts = RTE_ALIGN_FLOOR(pkts, NIX_DESCS_PER_LOOP); + + /* Reduce the cached count */ + txq->fc_cache_pkts -= pkts; + + senddesc01_w0 = vld1q_dup_u64(&txq->send_hdr_w0); + senddesc23_w0 = senddesc01_w0; + senddesc01_w1 = vdupq_n_u64(0); + senddesc23_w1 = senddesc01_w1; + sgdesc01_w0 = vld1q_dup_u64(&txq->sg_w0); + sgdesc23_w0 = sgdesc01_w0; + + /* Get LMT base address and LMT ID as lcore id */ + ROC_LMT_BASE_ID_GET(laddr, lmt_id); + left = pkts; +again: + /* Number of packets to prepare depends on offloads enabled. */ + burst = left > cn10k_nix_pkts_per_vec_brst(flags) ? + cn10k_nix_pkts_per_vec_brst(flags) : + left; + lnum = 0; + for (i = 0; i < burst; i += NIX_DESCS_PER_LOOP) { + /* Clear lower 32bit of SEND_HDR_W0 and SEND_SG_W0 */ + senddesc01_w0 = + vbicq_u64(senddesc01_w0, vdupq_n_u64(0xFFFFFFFF)); + sgdesc01_w0 = vbicq_u64(sgdesc01_w0, vdupq_n_u64(0xFFFFFFFF)); + + senddesc23_w0 = senddesc01_w0; + sgdesc23_w0 = sgdesc01_w0; + + /* Move mbufs to iova */ + mbuf0 = (uint64_t *)tx_pkts[0]; + mbuf1 = (uint64_t *)tx_pkts[1]; + mbuf2 = (uint64_t *)tx_pkts[2]; + mbuf3 = (uint64_t *)tx_pkts[3]; + + mbuf0 = (uint64_t *)((uintptr_t)mbuf0 + + offsetof(struct rte_mbuf, buf_iova)); + mbuf1 = (uint64_t *)((uintptr_t)mbuf1 + + offsetof(struct rte_mbuf, buf_iova)); + mbuf2 = (uint64_t *)((uintptr_t)mbuf2 + + offsetof(struct rte_mbuf, buf_iova)); + mbuf3 = (uint64_t *)((uintptr_t)mbuf3 + + offsetof(struct rte_mbuf, buf_iova)); + /* + * Get mbuf's, olflags, iova, pktlen, dataoff + * dataoff_iovaX.D[0] = iova, + * dataoff_iovaX.D[1](15:0) = mbuf->dataoff + * len_olflagsX.D[0] = ol_flags, + * len_olflagsX.D[1](63:32) = mbuf->pkt_len + */ + dataoff_iova0 = vld1q_u64(mbuf0); + len_olflags0 = vld1q_u64(mbuf0 + 2); + dataoff_iova1 = vld1q_u64(mbuf1); + len_olflags1 = vld1q_u64(mbuf1 + 2); + dataoff_iova2 = vld1q_u64(mbuf2); + len_olflags2 = vld1q_u64(mbuf2 + 2); + dataoff_iova3 = vld1q_u64(mbuf3); + len_olflags3 = vld1q_u64(mbuf3 + 2); + + /* Move mbufs to point pool */ + mbuf0 = (uint64_t *)((uintptr_t)mbuf0 + + offsetof(struct rte_mbuf, pool) - + offsetof(struct rte_mbuf, buf_iova)); + mbuf1 = (uint64_t *)((uintptr_t)mbuf1 + + offsetof(struct rte_mbuf, pool) - + offsetof(struct rte_mbuf, buf_iova)); + mbuf2 = (uint64_t *)((uintptr_t)mbuf2 + + offsetof(struct rte_mbuf, pool) - + offsetof(struct rte_mbuf, buf_iova)); + mbuf3 = (uint64_t *)((uintptr_t)mbuf3 + + offsetof(struct rte_mbuf, pool) - + offsetof(struct rte_mbuf, buf_iova)); + + if (flags & (NIX_TX_OFFLOAD_OL3_OL4_CSUM_F | + NIX_TX_OFFLOAD_L3_L4_CSUM_F)) { + /* Get tx_offload for ol2, ol3, l2, l3 lengths */ + /* + * E(8):OL2_LEN(7):OL3_LEN(9):E(24):L3_LEN(9):L2_LEN(7) + * E(8):OL2_LEN(7):OL3_LEN(9):E(24):L3_LEN(9):L2_LEN(7) + */ + + asm volatile("LD1 {%[a].D}[0],[%[in]]\n\t" + : [a] "+w"(senddesc01_w1) + : [in] "r"(mbuf0 + 2) + : "memory"); + + asm volatile("LD1 {%[a].D}[1],[%[in]]\n\t" + : [a] "+w"(senddesc01_w1) + : [in] "r"(mbuf1 + 2) + : "memory"); + + asm volatile("LD1 {%[b].D}[0],[%[in]]\n\t" + : [b] "+w"(senddesc23_w1) + : [in] "r"(mbuf2 + 2) + : "memory"); + + asm volatile("LD1 {%[b].D}[1],[%[in]]\n\t" + : [b] "+w"(senddesc23_w1) + : [in] "r"(mbuf3 + 2) + : "memory"); + + /* Get pool pointer alone */ + mbuf0 = (uint64_t *)*mbuf0; + mbuf1 = (uint64_t *)*mbuf1; + mbuf2 = (uint64_t *)*mbuf2; + mbuf3 = (uint64_t *)*mbuf3; + } else { + /* Get pool pointer alone */ + mbuf0 = (uint64_t *)*mbuf0; + mbuf1 = (uint64_t *)*mbuf1; + mbuf2 = (uint64_t *)*mbuf2; + mbuf3 = (uint64_t *)*mbuf3; + } + + const uint8x16_t shuf_mask2 = { + 0x4, 0x5, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, + 0xc, 0xd, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, + }; + xtmp128 = vzip2q_u64(len_olflags0, len_olflags1); + ytmp128 = vzip2q_u64(len_olflags2, len_olflags3); + + /* Clear dataoff_iovaX.D[1] bits other than dataoff(15:0) */ + const uint64x2_t and_mask0 = { + 0xFFFFFFFFFFFFFFFF, + 0x000000000000FFFF, + }; + + dataoff_iova0 = vandq_u64(dataoff_iova0, and_mask0); + dataoff_iova1 = vandq_u64(dataoff_iova1, and_mask0); + dataoff_iova2 = vandq_u64(dataoff_iova2, and_mask0); + dataoff_iova3 = vandq_u64(dataoff_iova3, and_mask0); + + /* + * Pick only 16 bits of pktlen preset at bits 63:32 + * and place them at bits 15:0. + */ + xtmp128 = vqtbl1q_u8(xtmp128, shuf_mask2); + ytmp128 = vqtbl1q_u8(ytmp128, shuf_mask2); + + /* Add pairwise to get dataoff + iova in sgdesc_w1 */ + sgdesc01_w1 = vpaddq_u64(dataoff_iova0, dataoff_iova1); + sgdesc23_w1 = vpaddq_u64(dataoff_iova2, dataoff_iova3); + + /* Orr both sgdesc_w0 and senddesc_w0 with 16 bits of + * pktlen at 15:0 position. + */ + sgdesc01_w0 = vorrq_u64(sgdesc01_w0, xtmp128); + sgdesc23_w0 = vorrq_u64(sgdesc23_w0, ytmp128); + senddesc01_w0 = vorrq_u64(senddesc01_w0, xtmp128); + senddesc23_w0 = vorrq_u64(senddesc23_w0, ytmp128); + + /* Move mbuf to point to pool_id. */ + mbuf0 = (uint64_t *)((uintptr_t)mbuf0 + + offsetof(struct rte_mempool, pool_id)); + mbuf1 = (uint64_t *)((uintptr_t)mbuf1 + + offsetof(struct rte_mempool, pool_id)); + mbuf2 = (uint64_t *)((uintptr_t)mbuf2 + + offsetof(struct rte_mempool, pool_id)); + mbuf3 = (uint64_t *)((uintptr_t)mbuf3 + + offsetof(struct rte_mempool, pool_id)); + + if ((flags & NIX_TX_OFFLOAD_L3_L4_CSUM_F) && + !(flags & NIX_TX_OFFLOAD_OL3_OL4_CSUM_F)) { + /* + * Lookup table to translate ol_flags to + * il3/il4 types. But we still use ol3/ol4 types in + * senddesc_w1 as only one header processing is enabled. + */ + const uint8x16_t tbl = { + /* [0-15] = il4type:il3type */ + 0x04, /* none (IPv6 assumed) */ + 0x14, /* PKT_TX_TCP_CKSUM (IPv6 assumed) */ + 0x24, /* PKT_TX_SCTP_CKSUM (IPv6 assumed) */ + 0x34, /* PKT_TX_UDP_CKSUM (IPv6 assumed) */ + 0x03, /* PKT_TX_IP_CKSUM */ + 0x13, /* PKT_TX_IP_CKSUM | PKT_TX_TCP_CKSUM */ + 0x23, /* PKT_TX_IP_CKSUM | PKT_TX_SCTP_CKSUM */ + 0x33, /* PKT_TX_IP_CKSUM | PKT_TX_UDP_CKSUM */ + 0x02, /* PKT_TX_IPV4 */ + 0x12, /* PKT_TX_IPV4 | PKT_TX_TCP_CKSUM */ + 0x22, /* PKT_TX_IPV4 | PKT_TX_SCTP_CKSUM */ + 0x32, /* PKT_TX_IPV4 | PKT_TX_UDP_CKSUM */ + 0x03, /* PKT_TX_IPV4 | PKT_TX_IP_CKSUM */ + 0x13, /* PKT_TX_IPV4 | PKT_TX_IP_CKSUM | + * PKT_TX_TCP_CKSUM + */ + 0x23, /* PKT_TX_IPV4 | PKT_TX_IP_CKSUM | + * PKT_TX_SCTP_CKSUM + */ + 0x33, /* PKT_TX_IPV4 | PKT_TX_IP_CKSUM | + * PKT_TX_UDP_CKSUM + */ + }; + + /* Extract olflags to translate to iltypes */ + xtmp128 = vzip1q_u64(len_olflags0, len_olflags1); + ytmp128 = vzip1q_u64(len_olflags2, len_olflags3); + + /* + * E(47):L3_LEN(9):L2_LEN(7+z) + * E(47):L3_LEN(9):L2_LEN(7+z) + */ + senddesc01_w1 = vshlq_n_u64(senddesc01_w1, 1); + senddesc23_w1 = vshlq_n_u64(senddesc23_w1, 1); + + /* Move OLFLAGS bits 55:52 to 51:48 + * with zeros preprended on the byte and rest + * don't care + */ + xtmp128 = vshrq_n_u8(xtmp128, 4); + ytmp128 = vshrq_n_u8(ytmp128, 4); + /* + * E(48):L3_LEN(8):L2_LEN(z+7) + * E(48):L3_LEN(8):L2_LEN(z+7) + */ + const int8x16_t tshft3 = { + -1, 0, 8, 8, 8, 8, 8, 8, + -1, 0, 8, 8, 8, 8, 8, 8, + }; + + senddesc01_w1 = vshlq_u8(senddesc01_w1, tshft3); + senddesc23_w1 = vshlq_u8(senddesc23_w1, tshft3); + + /* Do the lookup */ + ltypes01 = vqtbl1q_u8(tbl, xtmp128); + ltypes23 = vqtbl1q_u8(tbl, ytmp128); + + /* Pick only relevant fields i.e Bit 48:55 of iltype + * and place it in ol3/ol4type of senddesc_w1 + */ + const uint8x16_t shuf_mask0 = { + 0xFF, 0xFF, 0xFF, 0xFF, 0x6, 0xFF, 0xFF, 0xFF, + 0xFF, 0xFF, 0xFF, 0xFF, 0xE, 0xFF, 0xFF, 0xFF, + }; + + ltypes01 = vqtbl1q_u8(ltypes01, shuf_mask0); + ltypes23 = vqtbl1q_u8(ltypes23, shuf_mask0); + + /* Prepare ol4ptr, ol3ptr from ol3len, ol2len. + * a [E(32):E(16):OL3(8):OL2(8)] + * a = a + (a << 8) + * a [E(32):E(16):(OL3+OL2):OL2] + * => E(32):E(16)::OL4PTR(8):OL3PTR(8) + */ + senddesc01_w1 = vaddq_u8(senddesc01_w1, + vshlq_n_u16(senddesc01_w1, 8)); + senddesc23_w1 = vaddq_u8(senddesc23_w1, + vshlq_n_u16(senddesc23_w1, 8)); + + /* Move ltypes to senddesc*_w1 */ + senddesc01_w1 = vorrq_u64(senddesc01_w1, ltypes01); + senddesc23_w1 = vorrq_u64(senddesc23_w1, ltypes23); + } else if (!(flags & NIX_TX_OFFLOAD_L3_L4_CSUM_F) && + (flags & NIX_TX_OFFLOAD_OL3_OL4_CSUM_F)) { + /* + * Lookup table to translate ol_flags to + * ol3/ol4 types. + */ + + const uint8x16_t tbl = { + /* [0-15] = ol4type:ol3type */ + 0x00, /* none */ + 0x03, /* OUTER_IP_CKSUM */ + 0x02, /* OUTER_IPV4 */ + 0x03, /* OUTER_IPV4 | OUTER_IP_CKSUM */ + 0x04, /* OUTER_IPV6 */ + 0x00, /* OUTER_IPV6 | OUTER_IP_CKSUM */ + 0x00, /* OUTER_IPV6 | OUTER_IPV4 */ + 0x00, /* OUTER_IPV6 | OUTER_IPV4 | + * OUTER_IP_CKSUM + */ + 0x00, /* OUTER_UDP_CKSUM */ + 0x33, /* OUTER_UDP_CKSUM | OUTER_IP_CKSUM */ + 0x32, /* OUTER_UDP_CKSUM | OUTER_IPV4 */ + 0x33, /* OUTER_UDP_CKSUM | OUTER_IPV4 | + * OUTER_IP_CKSUM + */ + 0x34, /* OUTER_UDP_CKSUM | OUTER_IPV6 */ + 0x00, /* OUTER_UDP_CKSUM | OUTER_IPV6 | + * OUTER_IP_CKSUM + */ + 0x00, /* OUTER_UDP_CKSUM | OUTER_IPV6 | + * OUTER_IPV4 + */ + 0x00, /* OUTER_UDP_CKSUM | OUTER_IPV6 | + * OUTER_IPV4 | OUTER_IP_CKSUM + */ + }; + + /* Extract olflags to translate to iltypes */ + xtmp128 = vzip1q_u64(len_olflags0, len_olflags1); + ytmp128 = vzip1q_u64(len_olflags2, len_olflags3); + + /* + * E(47):OL3_LEN(9):OL2_LEN(7+z) + * E(47):OL3_LEN(9):OL2_LEN(7+z) + */ + const uint8x16_t shuf_mask5 = { + 0x6, 0x5, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, + 0xE, 0xD, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, + }; + senddesc01_w1 = vqtbl1q_u8(senddesc01_w1, shuf_mask5); + senddesc23_w1 = vqtbl1q_u8(senddesc23_w1, shuf_mask5); + + /* Extract outer ol flags only */ + const uint64x2_t o_cksum_mask = { + 0x1C00020000000000, + 0x1C00020000000000, + }; + + xtmp128 = vandq_u64(xtmp128, o_cksum_mask); + ytmp128 = vandq_u64(ytmp128, o_cksum_mask); + + /* Extract OUTER_UDP_CKSUM bit 41 and + * move it to bit 61 + */ + + xtmp128 = xtmp128 | vshlq_n_u64(xtmp128, 20); + ytmp128 = ytmp128 | vshlq_n_u64(ytmp128, 20); + + /* Shift oltype by 2 to start nibble from BIT(56) + * instead of BIT(58) + */ + xtmp128 = vshrq_n_u8(xtmp128, 2); + ytmp128 = vshrq_n_u8(ytmp128, 2); + /* + * E(48):L3_LEN(8):L2_LEN(z+7) + * E(48):L3_LEN(8):L2_LEN(z+7) + */ + const int8x16_t tshft3 = { + -1, 0, 8, 8, 8, 8, 8, 8, + -1, 0, 8, 8, 8, 8, 8, 8, + }; + + senddesc01_w1 = vshlq_u8(senddesc01_w1, tshft3); + senddesc23_w1 = vshlq_u8(senddesc23_w1, tshft3); + + /* Do the lookup */ + ltypes01 = vqtbl1q_u8(tbl, xtmp128); + ltypes23 = vqtbl1q_u8(tbl, ytmp128); + + /* Pick only relevant fields i.e Bit 56:63 of oltype + * and place it in ol3/ol4type of senddesc_w1 + */ + const uint8x16_t shuf_mask0 = { + 0xFF, 0xFF, 0xFF, 0xFF, 0x7, 0xFF, 0xFF, 0xFF, + 0xFF, 0xFF, 0xFF, 0xFF, 0xF, 0xFF, 0xFF, 0xFF, + }; + + ltypes01 = vqtbl1q_u8(ltypes01, shuf_mask0); + ltypes23 = vqtbl1q_u8(ltypes23, shuf_mask0); + + /* Prepare ol4ptr, ol3ptr from ol3len, ol2len. + * a [E(32):E(16):OL3(8):OL2(8)] + * a = a + (a << 8) + * a [E(32):E(16):(OL3+OL2):OL2] + * => E(32):E(16)::OL4PTR(8):OL3PTR(8) + */ + senddesc01_w1 = vaddq_u8(senddesc01_w1, + vshlq_n_u16(senddesc01_w1, 8)); + senddesc23_w1 = vaddq_u8(senddesc23_w1, + vshlq_n_u16(senddesc23_w1, 8)); + + /* Move ltypes to senddesc*_w1 */ + senddesc01_w1 = vorrq_u64(senddesc01_w1, ltypes01); + senddesc23_w1 = vorrq_u64(senddesc23_w1, ltypes23); + } else if ((flags & NIX_TX_OFFLOAD_L3_L4_CSUM_F) && + (flags & NIX_TX_OFFLOAD_OL3_OL4_CSUM_F)) { + /* Lookup table to translate ol_flags to + * ol4type, ol3type, il4type, il3type of senddesc_w1 + */ + const uint8x16x2_t tbl = {{ + { + /* [0-15] = il4type:il3type */ + 0x04, /* none (IPv6) */ + 0x14, /* PKT_TX_TCP_CKSUM (IPv6) */ + 0x24, /* PKT_TX_SCTP_CKSUM (IPv6) */ + 0x34, /* PKT_TX_UDP_CKSUM (IPv6) */ + 0x03, /* PKT_TX_IP_CKSUM */ + 0x13, /* PKT_TX_IP_CKSUM | + * PKT_TX_TCP_CKSUM + */ + 0x23, /* PKT_TX_IP_CKSUM | + * PKT_TX_SCTP_CKSUM + */ + 0x33, /* PKT_TX_IP_CKSUM | + * PKT_TX_UDP_CKSUM + */ + 0x02, /* PKT_TX_IPV4 */ + 0x12, /* PKT_TX_IPV4 | + * PKT_TX_TCP_CKSUM + */ + 0x22, /* PKT_TX_IPV4 | + * PKT_TX_SCTP_CKSUM + */ + 0x32, /* PKT_TX_IPV4 | + * PKT_TX_UDP_CKSUM + */ + 0x03, /* PKT_TX_IPV4 | + * PKT_TX_IP_CKSUM + */ + 0x13, /* PKT_TX_IPV4 | PKT_TX_IP_CKSUM | + * PKT_TX_TCP_CKSUM + */ + 0x23, /* PKT_TX_IPV4 | PKT_TX_IP_CKSUM | + * PKT_TX_SCTP_CKSUM + */ + 0x33, /* PKT_TX_IPV4 | PKT_TX_IP_CKSUM | + * PKT_TX_UDP_CKSUM + */ + }, + + { + /* [16-31] = ol4type:ol3type */ + 0x00, /* none */ + 0x03, /* OUTER_IP_CKSUM */ + 0x02, /* OUTER_IPV4 */ + 0x03, /* OUTER_IPV4 | OUTER_IP_CKSUM */ + 0x04, /* OUTER_IPV6 */ + 0x00, /* OUTER_IPV6 | OUTER_IP_CKSUM */ + 0x00, /* OUTER_IPV6 | OUTER_IPV4 */ + 0x00, /* OUTER_IPV6 | OUTER_IPV4 | + * OUTER_IP_CKSUM + */ + 0x00, /* OUTER_UDP_CKSUM */ + 0x33, /* OUTER_UDP_CKSUM | + * OUTER_IP_CKSUM + */ + 0x32, /* OUTER_UDP_CKSUM | + * OUTER_IPV4 + */ + 0x33, /* OUTER_UDP_CKSUM | + * OUTER_IPV4 | OUTER_IP_CKSUM + */ + 0x34, /* OUTER_UDP_CKSUM | + * OUTER_IPV6 + */ + 0x00, /* OUTER_UDP_CKSUM | OUTER_IPV6 | + * OUTER_IP_CKSUM + */ + 0x00, /* OUTER_UDP_CKSUM | OUTER_IPV6 | + * OUTER_IPV4 + */ + 0x00, /* OUTER_UDP_CKSUM | OUTER_IPV6 | + * OUTER_IPV4 | OUTER_IP_CKSUM + */ + }, + }}; + + /* Extract olflags to translate to oltype & iltype */ + xtmp128 = vzip1q_u64(len_olflags0, len_olflags1); + ytmp128 = vzip1q_u64(len_olflags2, len_olflags3); + + /* + * E(8):OL2_LN(7):OL3_LN(9):E(23):L3_LN(9):L2_LN(7+z) + * E(8):OL2_LN(7):OL3_LN(9):E(23):L3_LN(9):L2_LN(7+z) + */ + const uint32x4_t tshft_4 = { + 1, + 0, + 1, + 0, + }; + senddesc01_w1 = vshlq_u32(senddesc01_w1, tshft_4); + senddesc23_w1 = vshlq_u32(senddesc23_w1, tshft_4); + + /* + * E(32):L3_LEN(8):L2_LEN(7+Z):OL3_LEN(8):OL2_LEN(7+Z) + * E(32):L3_LEN(8):L2_LEN(7+Z):OL3_LEN(8):OL2_LEN(7+Z) + */ + const uint8x16_t shuf_mask5 = { + 0x6, 0x5, 0x0, 0x1, 0xFF, 0xFF, 0xFF, 0xFF, + 0xE, 0xD, 0x8, 0x9, 0xFF, 0xFF, 0xFF, 0xFF, + }; + senddesc01_w1 = vqtbl1q_u8(senddesc01_w1, shuf_mask5); + senddesc23_w1 = vqtbl1q_u8(senddesc23_w1, shuf_mask5); + + /* Extract outer and inner header ol_flags */ + const uint64x2_t oi_cksum_mask = { + 0x1CF0020000000000, + 0x1CF0020000000000, + }; + + xtmp128 = vandq_u64(xtmp128, oi_cksum_mask); + ytmp128 = vandq_u64(ytmp128, oi_cksum_mask); + + /* Extract OUTER_UDP_CKSUM bit 41 and + * move it to bit 61 + */ + + xtmp128 = xtmp128 | vshlq_n_u64(xtmp128, 20); + ytmp128 = ytmp128 | vshlq_n_u64(ytmp128, 20); + + /* Shift right oltype by 2 and iltype by 4 + * to start oltype nibble from BIT(58) + * instead of BIT(56) and iltype nibble from BIT(48) + * instead of BIT(52). + */ + const int8x16_t tshft5 = { + 8, 8, 8, 8, 8, 8, -4, -2, + 8, 8, 8, 8, 8, 8, -4, -2, + }; + + xtmp128 = vshlq_u8(xtmp128, tshft5); + ytmp128 = vshlq_u8(ytmp128, tshft5); + /* + * E(32):L3_LEN(8):L2_LEN(8):OL3_LEN(8):OL2_LEN(8) + * E(32):L3_LEN(8):L2_LEN(8):OL3_LEN(8):OL2_LEN(8) + */ + const int8x16_t tshft3 = { + -1, 0, -1, 0, 0, 0, 0, 0, + -1, 0, -1, 0, 0, 0, 0, 0, + }; + + senddesc01_w1 = vshlq_u8(senddesc01_w1, tshft3); + senddesc23_w1 = vshlq_u8(senddesc23_w1, tshft3); + + /* Mark Bit(4) of oltype */ + const uint64x2_t oi_cksum_mask2 = { + 0x1000000000000000, + 0x1000000000000000, + }; + + xtmp128 = vorrq_u64(xtmp128, oi_cksum_mask2); + ytmp128 = vorrq_u64(ytmp128, oi_cksum_mask2); + + /* Do the lookup */ + ltypes01 = vqtbl2q_u8(tbl, xtmp128); + ltypes23 = vqtbl2q_u8(tbl, ytmp128); + + /* Pick only relevant fields i.e Bit 48:55 of iltype and + * Bit 56:63 of oltype and place it in corresponding + * place in senddesc_w1. + */ + const uint8x16_t shuf_mask0 = { + 0xFF, 0xFF, 0xFF, 0xFF, 0x7, 0x6, 0xFF, 0xFF, + 0xFF, 0xFF, 0xFF, 0xFF, 0xF, 0xE, 0xFF, 0xFF, + }; + + ltypes01 = vqtbl1q_u8(ltypes01, shuf_mask0); + ltypes23 = vqtbl1q_u8(ltypes23, shuf_mask0); + + /* Prepare l4ptr, l3ptr, ol4ptr, ol3ptr from + * l3len, l2len, ol3len, ol2len. + * a [E(32):L3(8):L2(8):OL3(8):OL2(8)] + * a = a + (a << 8) + * a [E:(L3+L2):(L2+OL3):(OL3+OL2):OL2] + * a = a + (a << 16) + * a [E:(L3+L2+OL3+OL2):(L2+OL3+OL2):(OL3+OL2):OL2] + * => E(32):IL4PTR(8):IL3PTR(8):OL4PTR(8):OL3PTR(8) + */ + senddesc01_w1 = vaddq_u8(senddesc01_w1, + vshlq_n_u32(senddesc01_w1, 8)); + senddesc23_w1 = vaddq_u8(senddesc23_w1, + vshlq_n_u32(senddesc23_w1, 8)); + + /* Continue preparing l4ptr, l3ptr, ol4ptr, ol3ptr */ + senddesc01_w1 = vaddq_u8( + senddesc01_w1, vshlq_n_u32(senddesc01_w1, 16)); + senddesc23_w1 = vaddq_u8( + senddesc23_w1, vshlq_n_u32(senddesc23_w1, 16)); + + /* Move ltypes to senddesc*_w1 */ + senddesc01_w1 = vorrq_u64(senddesc01_w1, ltypes01); + senddesc23_w1 = vorrq_u64(senddesc23_w1, ltypes23); + } + + xmask01 = vdupq_n_u64(0); + xmask23 = xmask01; + asm volatile("LD1 {%[a].H}[0],[%[in]]\n\t" + : [a] "+w"(xmask01) + : [in] "r"(mbuf0) + : "memory"); + + asm volatile("LD1 {%[a].H}[4],[%[in]]\n\t" + : [a] "+w"(xmask01) + : [in] "r"(mbuf1) + : "memory"); + + asm volatile("LD1 {%[b].H}[0],[%[in]]\n\t" + : [b] "+w"(xmask23) + : [in] "r"(mbuf2) + : "memory"); + + asm volatile("LD1 {%[b].H}[4],[%[in]]\n\t" + : [b] "+w"(xmask23) + : [in] "r"(mbuf3) + : "memory"); + xmask01 = vshlq_n_u64(xmask01, 20); + xmask23 = vshlq_n_u64(xmask23, 20); + + senddesc01_w0 = vorrq_u64(senddesc01_w0, xmask01); + senddesc23_w0 = vorrq_u64(senddesc23_w0, xmask23); + + if (flags & NIX_TX_OFFLOAD_MBUF_NOFF_F) { + /* Set don't free bit if reference count > 1 */ + xmask01 = vdupq_n_u64(0); + xmask23 = xmask01; + + /* Move mbufs to iova */ + mbuf0 = (uint64_t *)tx_pkts[0]; + mbuf1 = (uint64_t *)tx_pkts[1]; + mbuf2 = (uint64_t *)tx_pkts[2]; + mbuf3 = (uint64_t *)tx_pkts[3]; + + if (cnxk_nix_prefree_seg((struct rte_mbuf *)mbuf0)) + vsetq_lane_u64(0x80000, xmask01, 0); + else + __mempool_check_cookies( + ((struct rte_mbuf *)mbuf0)->pool, + (void **)&mbuf0, 1, 0); + + if (cnxk_nix_prefree_seg((struct rte_mbuf *)mbuf1)) + vsetq_lane_u64(0x80000, xmask01, 1); + else + __mempool_check_cookies( + ((struct rte_mbuf *)mbuf1)->pool, + (void **)&mbuf1, 1, 0); + + if (cnxk_nix_prefree_seg((struct rte_mbuf *)mbuf2)) + vsetq_lane_u64(0x80000, xmask23, 0); + else + __mempool_check_cookies( + ((struct rte_mbuf *)mbuf2)->pool, + (void **)&mbuf2, 1, 0); + + if (cnxk_nix_prefree_seg((struct rte_mbuf *)mbuf3)) + vsetq_lane_u64(0x80000, xmask23, 1); + else + __mempool_check_cookies( + ((struct rte_mbuf *)mbuf3)->pool, + (void **)&mbuf3, 1, 0); + senddesc01_w0 = vorrq_u64(senddesc01_w0, xmask01); + senddesc23_w0 = vorrq_u64(senddesc23_w0, xmask23); + } else { + /* Move mbufs to iova */ + mbuf0 = (uint64_t *)tx_pkts[0]; + mbuf1 = (uint64_t *)tx_pkts[1]; + mbuf2 = (uint64_t *)tx_pkts[2]; + mbuf3 = (uint64_t *)tx_pkts[3]; + + /* Mark mempool object as "put" since + * it is freed by NIX + */ + __mempool_check_cookies( + ((struct rte_mbuf *)mbuf0)->pool, + (void **)&mbuf0, 1, 0); + + __mempool_check_cookies( + ((struct rte_mbuf *)mbuf1)->pool, + (void **)&mbuf1, 1, 0); + + __mempool_check_cookies( + ((struct rte_mbuf *)mbuf2)->pool, + (void **)&mbuf2, 1, 0); + + __mempool_check_cookies( + ((struct rte_mbuf *)mbuf3)->pool, + (void **)&mbuf3, 1, 0); + } + + /* Create 4W cmd for 4 mbufs (sendhdr, sgdesc) */ + cmd0[0] = vzip1q_u64(senddesc01_w0, senddesc01_w1); + cmd0[1] = vzip2q_u64(senddesc01_w0, senddesc01_w1); + cmd0[2] = vzip1q_u64(senddesc23_w0, senddesc23_w1); + cmd0[3] = vzip2q_u64(senddesc23_w0, senddesc23_w1); + + cmd1[0] = vzip1q_u64(sgdesc01_w0, sgdesc01_w1); + cmd1[1] = vzip2q_u64(sgdesc01_w0, sgdesc01_w1); + cmd1[2] = vzip1q_u64(sgdesc23_w0, sgdesc23_w1); + cmd1[3] = vzip2q_u64(sgdesc23_w0, sgdesc23_w1); + + /* Store the prepared send desc to LMT lines */ + vst1q_u64(LMT_OFF(laddr, lnum, 0), cmd0[0]); + vst1q_u64(LMT_OFF(laddr, lnum, 16), cmd1[0]); + vst1q_u64(LMT_OFF(laddr, lnum, 32), cmd0[1]); + vst1q_u64(LMT_OFF(laddr, lnum, 48), cmd1[1]); + vst1q_u64(LMT_OFF(laddr, lnum, 64), cmd0[2]); + vst1q_u64(LMT_OFF(laddr, lnum, 80), cmd1[2]); + vst1q_u64(LMT_OFF(laddr, lnum, 96), cmd0[3]); + vst1q_u64(LMT_OFF(laddr, lnum, 112), cmd1[3]); + lnum += 1; + + tx_pkts = tx_pkts + NIX_DESCS_PER_LOOP; + } + + /* Trigger LMTST */ + if (lnum > 16) { + data = cn10k_nix_tx_steor_vec_data(flags); + pa = io_addr | (data & 0x7) << 4; + data &= ~0x7ULL; + data |= (15ULL << 12); + data |= (uint64_t)lmt_id; + + /* STEOR0 */ + roc_lmt_submit_steorl(data, pa); + + data = cn10k_nix_tx_steor_vec_data(flags); + pa = io_addr | (data & 0x7) << 4; + data &= ~0x7ULL; + data |= ((uint64_t)(lnum - 17)) << 12; + data |= (uint64_t)(lmt_id + 16); + + /* STEOR1 */ + roc_lmt_submit_steorl(data, pa); + } else if (lnum) { + data = cn10k_nix_tx_steor_vec_data(flags); + pa = io_addr | (data & 0x7) << 4; + data &= ~0x7ULL; + data |= ((uint64_t)(lnum - 1)) << 12; + data |= lmt_id; + + /* STEOR0 */ + roc_lmt_submit_steorl(data, pa); + } + + left -= burst; + rte_io_wmb(); + if (left) + goto again; + + if (unlikely(scalar)) + pkts += cn10k_nix_xmit_pkts(tx_queue, tx_pkts, scalar, cmd, + flags); + + return pkts; +} + +#else +static __rte_always_inline uint16_t +cn10k_nix_xmit_pkts_vector(void *tx_queue, struct rte_mbuf **tx_pkts, + uint16_t pkts, uint64_t *cmd, const uint16_t flags) +{ + RTE_SET_USED(tx_queue); + RTE_SET_USED(tx_pkts); + RTE_SET_USED(pkts); + RTE_SET_USED(cmd); + RTE_SET_USED(flags); + return 0; +} +#endif + #define L3L4CSUM_F NIX_TX_OFFLOAD_L3_L4_CSUM_F #define OL3OL4CSUM_F NIX_TX_OFFLOAD_OL3_OL4_CSUM_F #define VLAN_F NIX_TX_OFFLOAD_VLAN_QINQ_F @@ -667,6 +1479,9 @@ T(tso_noff_vlan_ol3ol4csum_l3l4csum, 1, 1, 1, 1, 1, 6, \ void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t pkts); \ \ uint16_t __rte_noinline __rte_hot cn10k_nix_xmit_pkts_mseg_##name( \ + void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t pkts); \ + \ + uint16_t __rte_noinline __rte_hot cn10k_nix_xmit_pkts_vec_##name( \ void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t pkts); NIX_TX_FASTPATH_MODES diff --git a/drivers/net/cnxk/cn10k_tx_vec.c b/drivers/net/cnxk/cn10k_tx_vec.c new file mode 100644 index 0000000..42baeb5 --- /dev/null +++ b/drivers/net/cnxk/cn10k_tx_vec.c @@ -0,0 +1,25 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(C) 2021 Marvell. + */ + +#include "cn10k_ethdev.h" +#include "cn10k_tx.h" + +#define T(name, f4, f3, f2, f1, f0, sz, flags) \ + uint16_t __rte_noinline __rte_hot \ + cn10k_nix_xmit_pkts_vec_##name(void *tx_queue, \ + struct rte_mbuf **tx_pkts, \ + uint16_t pkts) \ + { \ + uint64_t cmd[sz]; \ + \ + /* VLAN, TSTMP, TSO is not supported by vec */ \ + if ((flags) & NIX_TX_OFFLOAD_VLAN_QINQ_F || \ + (flags) & NIX_TX_OFFLOAD_TSO_F) \ + return 0; \ + return cn10k_nix_xmit_pkts_vector(tx_queue, tx_pkts, pkts, cmd,\ + (flags)); \ + } + +NIX_TX_FASTPATH_MODES +#undef T diff --git a/drivers/net/cnxk/meson.build b/drivers/net/cnxk/meson.build index 21e5676..8f32dc7 100644 --- a/drivers/net/cnxk/meson.build +++ b/drivers/net/cnxk/meson.build @@ -28,7 +28,8 @@ sources += files('cn10k_ethdev.c', 'cn10k_rx_mseg.c', 'cn10k_rx_vec.c', 'cn10k_tx.c', - 'cn10k_tx_mseg.c') + 'cn10k_tx_mseg.c', + 'cn10k_tx_vec.c') deps += ['bus_pci', 'cryptodev', 'eventdev', 'security'] deps += ['common_cnxk', 'mempool_cnxk']