From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mga06.intel.com (mga06.intel.com [134.134.136.31]) by dpdk.org (Postfix) with ESMTP id CF3A78019 for ; Fri, 9 Sep 2016 12:47:28 +0200 (CEST) Received: from fmsmga002.fm.intel.com ([10.253.24.26]) by orsmga104.jf.intel.com with ESMTP; 09 Sep 2016 03:47:28 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.30,304,1470726000"; d="scan'208";a="1053571922" Received: from unknown (HELO dpdk5.sh.intel.com) ([10.239.129.118]) by fmsmga002.fm.intel.com with ESMTP; 09 Sep 2016 03:47:27 -0700 From: Zhihong Wang To: dev@dpdk.org Cc: maxime.coquelin@redhat.com, yuanhan.liu@linux.intel.com, thomas.monjalon@6wind.com, Zhihong Wang Date: Thu, 8 Sep 2016 23:39:28 -0400 Message-Id: <1473392368-84903-7-git-send-email-zhihong.wang@intel.com> X-Mailer: git-send-email 2.7.4 In-Reply-To: <1473392368-84903-1-git-send-email-zhihong.wang@intel.com> References: <1471319402-112998-1-git-send-email-zhihong.wang@intel.com> <1473392368-84903-1-git-send-email-zhihong.wang@intel.com> Subject: [dpdk-dev] [PATCH v5 6/6] vhost: optimize cache access X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: patches and discussions about DPDK List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Fri, 09 Sep 2016 10:47:29 -0000 This patch reorders the code to delay virtio header write to optimize cache access efficiency for cases where the mrg_rxbuf feature is turned on. It reduces CPU pipeline stall cycles significantly. Signed-off-by: Zhihong Wang --- Changes in v3: 1. Remove unnecessary memset which causes frontend stall on SNB & IVB. 2. Rename variables to follow naming convention. lib/librte_vhost/virtio_net.c | 20 ++++++++++++++------ 1 file changed, 14 insertions(+), 6 deletions(-) diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.c index e9f6353..0086bcb 100644 --- a/lib/librte_vhost/virtio_net.c +++ b/lib/librte_vhost/virtio_net.c @@ -197,6 +197,7 @@ enqueue_packet(struct virtio_net *dev, struct vhost_virtqueue *vq, uint32_t mbuf_len; uint32_t mbuf_avail; uint32_t cpy_len; + uint32_t copy_virtio_hdr; uint32_t num_buffers = 0; /* start with the first mbuf of the packet */ @@ -211,12 +212,12 @@ enqueue_packet(struct virtio_net *dev, struct vhost_virtqueue *vq, if (unlikely(!desc_addr)) goto error; - /* handle virtio header */ + /* + * handle virtio header, the actual write operation is delayed + * for cache optimization, to reduce CPU pipeline stall cycles. + */ virtio_hdr = (struct virtio_net_hdr_mrg_rxbuf *)(uintptr_t)desc_addr; - virtio_enqueue_offload(mbuf, &(virtio_hdr->hdr)); - if (is_mrg_rxbuf) - virtio_hdr->num_buffers = 1; - + copy_virtio_hdr = 1; vhost_log_write(dev, desc->addr, dev->vhost_hlen); PRINT_PACKET(dev, (uintptr_t)desc_addr, dev->vhost_hlen, 0); desc_offset = dev->vhost_hlen; @@ -266,8 +267,15 @@ enqueue_packet(struct virtio_net *dev, struct vhost_virtqueue *vq, goto error; } - /* copy mbuf data */ + /* copy virtio header and mbuf data */ cpy_len = RTE_MIN(desc->len - desc_offset, mbuf_avail); + if (copy_virtio_hdr) { + copy_virtio_hdr = 0; + virtio_enqueue_offload(mbuf, &(virtio_hdr->hdr)); + if (is_mrg_rxbuf) + virtio_hdr->num_buffers = num_buffers + 1; + } + rte_memcpy((void *)(uintptr_t)desc_addr, rte_pktmbuf_mtod_offset(mbuf, void *, mbuf_len - mbuf_avail), -- 2.7.4