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 A9569A0350; Tue, 22 Feb 2022 07:46:53 +0100 (CET) Received: from [217.70.189.124] (localhost [127.0.0.1]) by mails.dpdk.org (Postfix) with ESMTP id 3D52740DF4; Tue, 22 Feb 2022 07:46:53 +0100 (CET) Received: from mga12.intel.com (mga12.intel.com [192.55.52.136]) by mails.dpdk.org (Postfix) with ESMTP id 33B6C4068C for ; Tue, 22 Feb 2022 07:46:49 +0100 (CET) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1645512411; x=1677048411; h=from:to:cc:subject:date:message-id:in-reply-to: references:mime-version:content-transfer-encoding; bh=uIQZME2ScpVCLuQ3eyHLABm/W5iunpIoOUR3zNbvoaY=; b=a6fbL5cSLvbhHlCJwWN/YPtupxa80VWU6vl4DERgr+/hRlZd0Np+mf6/ tW4G+I/6oBvFoQh9fGM480Bu0s1wjcI5/o8S8PN39BmKDdFphRU3HjLvJ 6q3bBMDFNCTcaCMXPlxmm4C1+Ms9YjyIcnlZMHlOxPkl6hJ4vUQEAWs8k HA78zkrgVg84zyfCaxlg4pEzCTjCa6bIsbYiEA3QNPwYIEMbSFpVaZh/n +GhBAFGocvT75ILT8SbePQpMcIq7jjFH6OPrXCh81QCX8jVIBJkoFcgHF Oj1IqtPsogKhs5me/tP52yLoBWma37s95DfEgDoTt50dp3l7tFoJfrJsv g==; X-IronPort-AV: E=McAfee;i="6200,9189,10265"; a="231592361" X-IronPort-AV: E=Sophos;i="5.88,387,1635231600"; d="scan'208";a="231592361" Received: from orsmga007.jf.intel.com ([10.7.209.58]) by fmsmga106.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 21 Feb 2022 22:46:44 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.88,387,1635231600"; d="scan'208";a="532109194" Received: from npg-wuwenjun-dpdk-01.sh.intel.com ([10.67.110.181]) by orsmga007.jf.intel.com with ESMTP; 21 Feb 2022 22:46:42 -0800 From: Wenjun Wu To: dev@dpdk.org, qiming.yang@intel.com, qi.z.zhang@intel.com Cc: harry.van.haaren@intel.com, simei.su@intel.com, Wenjun Wu Subject: [PATCH v3] net/ice: improve performance of RX timestamp offload Date: Tue, 22 Feb 2022 14:26:12 +0800 Message-Id: <20220222062612.335622-1-wenjun1.wu@intel.com> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20220222055016.298763-1-wenjun1.wu@intel.com> References: <20220222055016.298763-1-wenjun1.wu@intel.com> MIME-Version: 1.0 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 Previously, each time a burst of packets is received, SW reads HW register and assembles it and the timestamp from descriptor together to get the complete 64 bits timestamp. This patch optimizes the algorithm. The SW only needs to check the monotonicity of the low 32bits timestamp to avoid crossing borders. Each time before SW receives a burst of packets, it should check the time difference between current time and last update time to avoid the low 32 bits timestamp cycling twice. Signed-off-by: Wenjun Wu --- v3: add missing conditional compilation v2: add conditional compilation --- drivers/net/ice/ice_ethdev.h | 3 + drivers/net/ice/ice_rxtx.c | 140 ++++++++++++++++++++++++++--------- 2 files changed, 110 insertions(+), 33 deletions(-) diff --git a/drivers/net/ice/ice_ethdev.h b/drivers/net/ice/ice_ethdev.h index 3ed580d438..6778941d7d 100644 --- a/drivers/net/ice/ice_ethdev.h +++ b/drivers/net/ice/ice_ethdev.h @@ -554,6 +554,9 @@ struct ice_adapter { struct rte_timecounter tx_tstamp_tc; bool ptp_ena; uint64_t time_hw; + uint32_t hw_time_high; /* high 32 bits of timestamp */ + uint32_t hw_time_low; /* low 32 bits of timestamp */ + uint64_t hw_time_update; /* SW time of HW record updating */ struct ice_fdir_prof_info fdir_prof_info[ICE_MAX_PTGS]; struct ice_rss_prof_info rss_prof_info[ICE_MAX_PTGS]; /* True if DCF state of the associated PF is on */ diff --git a/drivers/net/ice/ice_rxtx.c b/drivers/net/ice/ice_rxtx.c index 4f218bcd0d..981347b618 100644 --- a/drivers/net/ice/ice_rxtx.c +++ b/drivers/net/ice/ice_rxtx.c @@ -1576,7 +1576,6 @@ ice_rx_scan_hw_ring(struct ice_rx_queue *rxq) #ifndef RTE_LIBRTE_ICE_16BYTE_RX_DESC struct ice_vsi *vsi = rxq->vsi; struct ice_hw *hw = ICE_VSI_TO_HW(vsi); - uint64_t ts_ns; struct ice_adapter *ad = rxq->vsi->adapter; #endif rxdp = &rxq->rx_ring[rxq->rx_tail]; @@ -1588,8 +1587,17 @@ ice_rx_scan_hw_ring(struct ice_rx_queue *rxq) if (!(stat_err0 & (1 << ICE_RX_FLEX_DESC_STATUS0_DD_S))) return 0; - if (rxq->offloads & RTE_ETH_RX_OFFLOAD_TIMESTAMP) - rxq->hw_register_set = 1; +#ifndef RTE_LIBRTE_ICE_16BYTE_RX_DESC + if (rxq->offloads & RTE_ETH_RX_OFFLOAD_TIMESTAMP) { + uint64_t sw_cur_time = rte_get_timer_cycles() / (rte_get_timer_hz() / 1000); + + if (sw_cur_time - ad->hw_time_update > 4) { + ad->hw_time_high = ICE_READ_REG(hw, GLTSYN_TIME_H(0)); + ad->hw_time_low = 0; + ad->hw_time_update = sw_cur_time; + } + } +#endif /** * Scan LOOK_AHEAD descriptors at a time to determine which @@ -1625,14 +1633,25 @@ ice_rx_scan_hw_ring(struct ice_rx_queue *rxq) rxd_to_pkt_fields_ops[rxq->rxdid](rxq, mb, &rxdp[j]); #ifndef RTE_LIBRTE_ICE_16BYTE_RX_DESC if (ice_timestamp_dynflag > 0) { - ts_ns = ice_tstamp_convert_32b_64b(hw, ad, - rxq->hw_register_set, - rte_le_to_cpu_32(rxdp[j].wb.flex_ts.ts_high)); - rxq->hw_register_set = 0; + rxq->time_high = + rte_le_to_cpu_32(rxdp[j].wb.flex_ts.ts_high); *RTE_MBUF_DYNFIELD(mb, - ice_timestamp_dynfield_offset, - rte_mbuf_timestamp_t *) = ts_ns; - mb->ol_flags |= ice_timestamp_dynflag; + ice_timestamp_dynfield_offset, + uint32_t *) = rxq->time_high; + if (rxq->time_high > ad->hw_time_low) { + *RTE_MBUF_DYNFIELD(mb, + (ice_timestamp_dynfield_offset + 4), + uint32_t *) = ad->hw_time_high; + } else { + ad->hw_time_high += 1; + *RTE_MBUF_DYNFIELD(mb, + (ice_timestamp_dynfield_offset + 4), + uint32_t *) = ad->hw_time_high; + ad->hw_time_update = + rte_get_timer_cycles() / + (rte_get_timer_hz() / 1000); + } + pkt_flags |= ice_timestamp_dynflag; } if (ad->ptp_ena && ((mb->packet_type & @@ -1657,6 +1676,13 @@ ice_rx_scan_hw_ring(struct ice_rx_queue *rxq) break; } +#ifndef RTE_LIBRTE_ICE_16BYTE_RX_DESC + if (nb_rx > 0 && rxq->offloads & RTE_ETH_RX_OFFLOAD_TIMESTAMP) + ad->hw_time_low = *RTE_MBUF_DYNFIELD(rxq->rx_stage[nb_rx - 1], + ice_timestamp_dynfield_offset, + uint32_t *); +#endif + /* Clear software ring entries */ for (i = 0; i < nb_rx; i++) rxq->sw_ring[rxq->rx_tail + i].mbuf = NULL; @@ -1833,12 +1859,18 @@ ice_recv_scattered_pkts(void *rx_queue, #ifndef RTE_LIBRTE_ICE_16BYTE_RX_DESC struct ice_vsi *vsi = rxq->vsi; struct ice_hw *hw = ICE_VSI_TO_HW(vsi); - uint64_t ts_ns; struct ice_adapter *ad = rxq->vsi->adapter; -#endif - if (rxq->offloads & RTE_ETH_RX_OFFLOAD_TIMESTAMP) - rxq->hw_register_set = 1; + if (rxq->offloads & RTE_ETH_RX_OFFLOAD_TIMESTAMP) { + uint64_t sw_cur_time = rte_get_timer_cycles() / (rte_get_timer_hz() / 1000); + + if (sw_cur_time - ad->hw_time_update > 4) { + ad->hw_time_high = ICE_READ_REG(hw, GLTSYN_TIME_H(0)); + ad->hw_time_low = 0; + ad->hw_time_update = sw_cur_time; + } + } +#endif while (nb_rx < nb_pkts) { rxdp = &rx_ring[rx_id]; @@ -1951,14 +1983,25 @@ ice_recv_scattered_pkts(void *rx_queue, pkt_flags = ice_rxd_error_to_pkt_flags(rx_stat_err0); #ifndef RTE_LIBRTE_ICE_16BYTE_RX_DESC if (ice_timestamp_dynflag > 0) { - ts_ns = ice_tstamp_convert_32b_64b(hw, ad, - rxq->hw_register_set, - rte_le_to_cpu_32(rxd.wb.flex_ts.ts_high)); - rxq->hw_register_set = 0; - *RTE_MBUF_DYNFIELD(first_seg, - ice_timestamp_dynfield_offset, - rte_mbuf_timestamp_t *) = ts_ns; - first_seg->ol_flags |= ice_timestamp_dynflag; + rxq->time_high = + rte_le_to_cpu_32(rxd.wb.flex_ts.ts_high); + *RTE_MBUF_DYNFIELD(rxm, + ice_timestamp_dynfield_offset, + uint32_t *) = rxq->time_high; + if (rxq->time_high > ad->hw_time_low) { + *RTE_MBUF_DYNFIELD(rxm, + (ice_timestamp_dynfield_offset + 4), + uint32_t *) = ad->hw_time_high; + } else { + ad->hw_time_high += 1; + *RTE_MBUF_DYNFIELD(rxm, + (ice_timestamp_dynfield_offset + 4), + uint32_t *) = ad->hw_time_high; + ad->hw_time_update = + rte_get_timer_cycles() / + (rte_get_timer_hz() / 1000); + } + pkt_flags |= ice_timestamp_dynflag; } if (ad->ptp_ena && ((first_seg->packet_type & RTE_PTYPE_L2_MASK) @@ -1977,6 +2020,13 @@ ice_recv_scattered_pkts(void *rx_queue, first_seg = NULL; } +#ifndef RTE_LIBRTE_ICE_16BYTE_RX_DESC + if (nb_rx > 0 && rxq->offloads & RTE_ETH_RX_OFFLOAD_TIMESTAMP) + ad->hw_time_low = *RTE_MBUF_DYNFIELD(rx_pkts[nb_rx - 1], + ice_timestamp_dynfield_offset, + uint32_t *); +#endif + /* Record index of the next RX descriptor to probe. */ rxq->rx_tail = rx_id; rxq->pkt_first_seg = first_seg; @@ -2327,12 +2377,18 @@ ice_recv_pkts(void *rx_queue, #ifndef RTE_LIBRTE_ICE_16BYTE_RX_DESC struct ice_vsi *vsi = rxq->vsi; struct ice_hw *hw = ICE_VSI_TO_HW(vsi); - uint64_t ts_ns; struct ice_adapter *ad = rxq->vsi->adapter; -#endif - if (rxq->offloads & RTE_ETH_RX_OFFLOAD_TIMESTAMP) - rxq->hw_register_set = 1; + if (rxq->offloads & RTE_ETH_RX_OFFLOAD_TIMESTAMP) { + uint64_t sw_cur_time = rte_get_timer_cycles() / (rte_get_timer_hz() / 1000); + + if (sw_cur_time - ad->hw_time_update > 4) { + ad->hw_time_high = ICE_READ_REG(hw, GLTSYN_TIME_H(0)); + ad->hw_time_low = 0; + ad->hw_time_update = sw_cur_time; + } + } +#endif while (nb_rx < nb_pkts) { rxdp = &rx_ring[rx_id]; @@ -2386,14 +2442,24 @@ ice_recv_pkts(void *rx_queue, pkt_flags = ice_rxd_error_to_pkt_flags(rx_stat_err0); #ifndef RTE_LIBRTE_ICE_16BYTE_RX_DESC if (ice_timestamp_dynflag > 0) { - ts_ns = ice_tstamp_convert_32b_64b(hw, ad, - rxq->hw_register_set, - rte_le_to_cpu_32(rxd.wb.flex_ts.ts_high)); - rxq->hw_register_set = 0; + rxq->time_high = rte_le_to_cpu_32(rxd.wb.flex_ts.ts_high); *RTE_MBUF_DYNFIELD(rxm, - ice_timestamp_dynfield_offset, - rte_mbuf_timestamp_t *) = ts_ns; - rxm->ol_flags |= ice_timestamp_dynflag; + ice_timestamp_dynfield_offset, + uint32_t *) = rxq->time_high; + if (rxq->time_high > ad->hw_time_low) { + *RTE_MBUF_DYNFIELD(rxm, + (ice_timestamp_dynfield_offset + 4), + uint32_t *) = ad->hw_time_high; + } else { + ad->hw_time_high += 1; + *RTE_MBUF_DYNFIELD(rxm, + (ice_timestamp_dynfield_offset + 4), + uint32_t *) = ad->hw_time_high; + ad->hw_time_update = + rte_get_timer_cycles() / + (rte_get_timer_hz() / 1000); + } + pkt_flags |= ice_timestamp_dynflag; } if (ad->ptp_ena && ((rxm->packet_type & RTE_PTYPE_L2_MASK) == @@ -2408,6 +2474,14 @@ ice_recv_pkts(void *rx_queue, /* copy old mbuf to rx_pkts */ rx_pkts[nb_rx++] = rxm; } + +#ifndef RTE_LIBRTE_ICE_16BYTE_RX_DESC + if (nb_rx > 0 && rxq->offloads & RTE_ETH_RX_OFFLOAD_TIMESTAMP) + ad->hw_time_low = *RTE_MBUF_DYNFIELD(rx_pkts[nb_rx - 1], + ice_timestamp_dynfield_offset, + uint32_t *); +#endif + rxq->rx_tail = rx_id; /** * If the number of free RX descriptors is greater than the RX free -- 2.25.1