[dpdk-dev] [PATCH 4/4] virtio: add neon support

Jerin Jacob jerin.jacob at caviumnetworks.com
Mon Jun 27 13:54:08 CEST 2016


Added neon based Rx vector implementation for virtio.
Selected neon based virtio implementation for ARM64 as
default and updated the MAINTAINERS file.

Signed-off-by: Jerin Jacob <jerin.jacob at caviumnetworks.com>
---
 MAINTAINERS                                  |   1 +
 config/defconfig_arm64-armv8a-linuxapp-gcc   |   1 -
 doc/guides/rel_notes/release_16_07.rst       |   2 +
 drivers/net/virtio/virtio_rxtx.c             |   2 +-
 drivers/net/virtio/virtio_rxtx_simple.c      |   2 +
 drivers/net/virtio/virtio_rxtx_simple_neon.h | 238 +++++++++++++++++++++++++++
 6 files changed, 244 insertions(+), 2 deletions(-)
 create mode 100644 drivers/net/virtio/virtio_rxtx_simple_neon.h

diff --git a/MAINTAINERS b/MAINTAINERS
index f6c0d3d..2bb12aa 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -142,6 +142,7 @@ F: lib/librte_eal/common/include/arch/arm/*_64.h
 F: lib/librte_acl/acl_run_neon.*
 F: lib/librte_lpm/rte_lpm_neon.h
 F: lib/librte_hash/rte*_arm64.h
+F: drivers/net/virtio/virtio_rxtx_simple_neon.h
 
 EZchip TILE-Gx
 M: Zhigang Lu <zlu at ezchip.com>
diff --git a/config/defconfig_arm64-armv8a-linuxapp-gcc b/config/defconfig_arm64-armv8a-linuxapp-gcc
index 95ed30e..a786562 100644
--- a/config/defconfig_arm64-armv8a-linuxapp-gcc
+++ b/config/defconfig_arm64-armv8a-linuxapp-gcc
@@ -48,6 +48,5 @@ CONFIG_RTE_IXGBE_INC_VECTOR=n
 CONFIG_RTE_LIBRTE_IVSHMEM=n
 CONFIG_RTE_LIBRTE_FM10K_PMD=n
 CONFIG_RTE_LIBRTE_I40E_PMD=n
-CONFIG_RTE_LIBRTE_VIRTIO_INC_VECTOR=n
 
 CONFIG_RTE_SCHED_VECTOR=n
diff --git a/doc/guides/rel_notes/release_16_07.rst b/doc/guides/rel_notes/release_16_07.rst
index 2694f50..3187a33 100644
--- a/doc/guides/rel_notes/release_16_07.rst
+++ b/doc/guides/rel_notes/release_16_07.rst
@@ -118,6 +118,8 @@ New Features
   * Root privilege is a must for sorting hugepages by physical address.
   * Can only be used with vhost user backend.
 
+* **Virtio NEON support for ARM.**
+
 Resolved Issues
 ---------------
 
diff --git a/drivers/net/virtio/virtio_rxtx.c b/drivers/net/virtio/virtio_rxtx.c
index e9b42f3..ca25db3 100644
--- a/drivers/net/virtio/virtio_rxtx.c
+++ b/drivers/net/virtio/virtio_rxtx.c
@@ -500,7 +500,7 @@ virtio_dev_tx_queue_setup(struct rte_eth_dev *dev,
 	}
 
 #ifdef RTE_LIBRTE_VIRTIO_INC_VECTOR
-#ifdef RTE_MACHINE_CPUFLAG_SSSE3
+#if defined RTE_MACHINE_CPUFLAG_SSSE3 || defined RTE_MACHINE_CPUFLAG_NEON
 	struct virtio_hw *hw = dev->data->dev_private;
 	/* Use simple rx/tx func if single segment and no offloads */
 	if ((tx_conf->txq_flags & VIRTIO_SIMPLE_FLAGS) == VIRTIO_SIMPLE_FLAGS &&
diff --git a/drivers/net/virtio/virtio_rxtx_simple.c b/drivers/net/virtio/virtio_rxtx_simple.c
index ca87605..e5dc010 100644
--- a/drivers/net/virtio/virtio_rxtx_simple.c
+++ b/drivers/net/virtio/virtio_rxtx_simple.c
@@ -131,6 +131,8 @@ virtio_rxq_rearm_vec(struct virtnet_rx *rxvq)
 
 #ifdef RTE_MACHINE_CPUFLAG_SSSE3
 #include "virtio_rxtx_simple_sse.h"
+#elif RTE_MACHINE_CPUFLAG_NEON
+#include "virtio_rxtx_simple_neon.h"
 #endif
 
 #define VIRTIO_TX_FREE_THRESH 32
diff --git a/drivers/net/virtio/virtio_rxtx_simple_neon.h b/drivers/net/virtio/virtio_rxtx_simple_neon.h
new file mode 100644
index 0000000..41f347d
--- /dev/null
+++ b/drivers/net/virtio/virtio_rxtx_simple_neon.h
@@ -0,0 +1,238 @@
+/*
+ *   BSD LICENSE
+ *
+ *   Copyright (C) Cavium networks Ltd. 2016
+ *
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
+ *   are met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in
+ *       the documentation and/or other materials provided with the
+ *       distribution.
+ *     * Neither the name of Cavium networks nor the names of its
+ *       contributors may be used to endorse or promote products derived
+ *       from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <errno.h>
+
+#include <rte_byteorder.h>
+#include <rte_branch_prediction.h>
+#include <rte_cycles.h>
+#include <rte_ether.h>
+#include <rte_ethdev.h>
+#include <rte_errno.h>
+#include <rte_memory.h>
+#include <rte_memzone.h>
+#include <rte_mempool.h>
+#include <rte_malloc.h>
+#include <rte_mbuf.h>
+#include <rte_prefetch.h>
+#include <rte_string_fns.h>
+#include <rte_vect.h>
+
+#include "virtio_logs.h"
+#include "virtio_ethdev.h"
+#include "virtqueue.h"
+#include "virtio_rxtx.h"
+
+#define RTE_VIRTIO_VPMD_RX_BURST 32
+#define RTE_VIRTIO_DESC_PER_LOOP 8
+#define RTE_VIRTIO_VPMD_RX_REARM_THRESH RTE_VIRTIO_VPMD_RX_BURST
+
+/* virtio vPMD receive routine, only accept(nb_pkts >= RTE_VIRTIO_DESC_PER_LOOP)
+ *
+ * This routine is for non-mergeable RX, one desc for each guest buffer.
+ * This routine is based on the RX ring layout optimization. Each entry in the
+ * avail ring points to the desc with the same index in the desc ring and this
+ * will never be changed in the driver.
+ *
+ * - nb_pkts < RTE_VIRTIO_DESC_PER_LOOP, just return no packet
+ */
+uint16_t
+virtio_recv_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts,
+	uint16_t nb_pkts)
+{
+	struct virtnet_rx *rxvq = rx_queue;
+	struct virtqueue *vq = rxvq->vq;
+	uint16_t nb_used;
+	uint16_t desc_idx;
+	struct vring_used_elem *rused;
+	struct rte_mbuf **sw_ring;
+	struct rte_mbuf **sw_ring_end;
+	uint16_t nb_pkts_received;
+
+	uint8x16_t shuf_msk1 = {
+		0xFF, 0xFF, 0xFF, 0xFF, /* packet type */
+		4, 5, 0xFF, 0xFF,       /* pkt len */
+		4, 5,                   /* dat len */
+		0xFF, 0xFF,             /* vlan tci */
+		0xFF, 0xFF, 0xFF, 0xFF
+	};
+
+	uint8x16_t shuf_msk2 = {
+		0xFF, 0xFF, 0xFF, 0xFF, /* packet type */
+		12, 13, 0xFF, 0xFF,     /* pkt len */
+		12, 13,                 /* dat len */
+		0xFF, 0xFF,             /* vlan tci */
+		0xFF, 0xFF, 0xFF, 0xFF
+	};
+
+	/* Subtract the header length.
+	 *  In which case do we need the header length in used->len ?
+	 */
+	uint16x8_t len_adjust = {
+		0, 0,
+		(uint16_t)vq->hw->vtnet_hdr_size, 0,
+		(uint16_t)vq->hw->vtnet_hdr_size,
+		0,
+		0, 0
+	};
+
+	if (unlikely(nb_pkts < RTE_VIRTIO_DESC_PER_LOOP))
+		return 0;
+
+	nb_used = VIRTQUEUE_NUSED(vq);
+
+	rte_rmb();
+
+	if (unlikely(nb_used == 0))
+		return 0;
+
+	nb_pkts = RTE_ALIGN_FLOOR(nb_pkts, RTE_VIRTIO_DESC_PER_LOOP);
+	nb_used = RTE_MIN(nb_used, nb_pkts);
+
+	desc_idx = (uint16_t)(vq->vq_used_cons_idx & (vq->vq_nentries - 1));
+	rused = &vq->vq_ring.used->ring[desc_idx];
+	sw_ring  = &vq->sw_ring[desc_idx];
+	sw_ring_end = &vq->sw_ring[vq->vq_nentries];
+
+	rte_prefetch_non_temporal(rused);
+
+	if (vq->vq_free_cnt >= RTE_VIRTIO_VPMD_RX_REARM_THRESH) {
+		virtio_rxq_rearm_vec(rxvq);
+		if (unlikely(virtqueue_kick_prepare(vq)))
+			virtqueue_notify(vq);
+	}
+
+	for (nb_pkts_received = 0;
+		nb_pkts_received < nb_used;) {
+		uint64x2_t desc[RTE_VIRTIO_DESC_PER_LOOP / 2];
+		uint64x2_t mbp[RTE_VIRTIO_DESC_PER_LOOP / 2];
+		uint64x2_t pkt_mb[RTE_VIRTIO_DESC_PER_LOOP];
+
+		mbp[0] = vld1q_u64((uint64_t *)(sw_ring + 0));
+		desc[0] = vld1q_u64((uint64_t *)(rused + 0));
+		vst1q_u64((uint64_t *)&rx_pkts[0], mbp[0]);
+
+		mbp[1] = vld1q_u64((uint64_t *)(sw_ring + 2));
+		desc[1] = vld1q_u64((uint64_t *)(rused + 2));
+		vst1q_u64((uint64_t *)&rx_pkts[2], mbp[1]);
+
+		mbp[2] = vld1q_u64((uint64_t *)(sw_ring + 4));
+		desc[2] = vld1q_u64((uint64_t *)(rused + 4));
+		vst1q_u64((uint64_t *)&rx_pkts[4], mbp[2]);
+
+		mbp[3] = vld1q_u64((uint64_t *)(sw_ring + 6));
+		desc[3] = vld1q_u64((uint64_t *)(rused + 6));
+		vst1q_u64((uint64_t *)&rx_pkts[6], mbp[3]);
+
+		pkt_mb[1] = vreinterpretq_u64_u8(vqtbl1q_u8(
+				vreinterpretq_u8_u64(desc[0]), shuf_msk2));
+		pkt_mb[0] = vreinterpretq_u64_u8(vqtbl1q_u8(
+				vreinterpretq_u8_u64(desc[0]), shuf_msk1));
+		pkt_mb[1] = vreinterpretq_u64_u16(vsubq_u16(
+				vreinterpretq_u16_u64(pkt_mb[1]), len_adjust));
+		pkt_mb[0] = vreinterpretq_u64_u16(vsubq_u16(
+				vreinterpretq_u16_u64(pkt_mb[0]), len_adjust));
+		vst1q_u64((void *)&rx_pkts[1]->rx_descriptor_fields1,
+			pkt_mb[1]);
+		vst1q_u64((void *)&rx_pkts[0]->rx_descriptor_fields1,
+			pkt_mb[0]);
+
+		pkt_mb[3] = vreinterpretq_u64_u8(vqtbl1q_u8(
+				vreinterpretq_u8_u64(desc[1]), shuf_msk2));
+		pkt_mb[2] = vreinterpretq_u64_u8(vqtbl1q_u8(
+				vreinterpretq_u8_u64(desc[1]), shuf_msk1));
+		pkt_mb[3] = vreinterpretq_u64_u16(vsubq_u16(
+				vreinterpretq_u16_u64(pkt_mb[3]), len_adjust));
+		pkt_mb[2] = vreinterpretq_u64_u16(vsubq_u16(
+				vreinterpretq_u16_u64(pkt_mb[2]), len_adjust));
+		vst1q_u64((void *)&rx_pkts[3]->rx_descriptor_fields1,
+			pkt_mb[3]);
+		vst1q_u64((void *)&rx_pkts[2]->rx_descriptor_fields1,
+			pkt_mb[2]);
+
+		pkt_mb[5] = vreinterpretq_u64_u8(vqtbl1q_u8(
+				vreinterpretq_u8_u64(desc[2]), shuf_msk2));
+		pkt_mb[4] = vreinterpretq_u64_u8(vqtbl1q_u8(
+				vreinterpretq_u8_u64(desc[2]), shuf_msk1));
+		pkt_mb[5] = vreinterpretq_u64_u16(vsubq_u16(
+				vreinterpretq_u16_u64(pkt_mb[5]), len_adjust));
+		pkt_mb[4] = vreinterpretq_u64_u16(vsubq_u16(
+				vreinterpretq_u16_u64(pkt_mb[4]), len_adjust));
+		vst1q_u64((void *)&rx_pkts[5]->rx_descriptor_fields1,
+			pkt_mb[5]);
+		vst1q_u64((void *)&rx_pkts[4]->rx_descriptor_fields1,
+			pkt_mb[4]);
+
+		pkt_mb[7] = vreinterpretq_u64_u8(vqtbl1q_u8(
+				vreinterpretq_u8_u64(desc[3]), shuf_msk2));
+		pkt_mb[6] = vreinterpretq_u64_u8(vqtbl1q_u8(
+				vreinterpretq_u8_u64(desc[3]), shuf_msk1));
+		pkt_mb[7] = vreinterpretq_u64_u16(vsubq_u16(
+				vreinterpretq_u16_u64(pkt_mb[7]), len_adjust));
+		pkt_mb[6] = vreinterpretq_u64_u16(vsubq_u16(
+				vreinterpretq_u16_u64(pkt_mb[6]), len_adjust));
+		vst1q_u64((void *)&rx_pkts[7]->rx_descriptor_fields1,
+			pkt_mb[7]);
+		vst1q_u64((void *)&rx_pkts[6]->rx_descriptor_fields1,
+			pkt_mb[6]);
+
+		if (unlikely(nb_used <= RTE_VIRTIO_DESC_PER_LOOP)) {
+			if (sw_ring + nb_used <= sw_ring_end)
+				nb_pkts_received += nb_used;
+			else
+				nb_pkts_received += sw_ring_end - sw_ring;
+			break;
+		} else {
+			if (unlikely(sw_ring + RTE_VIRTIO_DESC_PER_LOOP >=
+				sw_ring_end)) {
+				nb_pkts_received += sw_ring_end - sw_ring;
+				break;
+			} else {
+				nb_pkts_received += RTE_VIRTIO_DESC_PER_LOOP;
+
+				rx_pkts += RTE_VIRTIO_DESC_PER_LOOP;
+				sw_ring += RTE_VIRTIO_DESC_PER_LOOP;
+				rused   += RTE_VIRTIO_DESC_PER_LOOP;
+				nb_used -= RTE_VIRTIO_DESC_PER_LOOP;
+			}
+		}
+	}
+
+	vq->vq_used_cons_idx += nb_pkts_received;
+	vq->vq_free_cnt += nb_pkts_received;
+	rxvq->stats.packets += nb_pkts_received;
+	return nb_pkts_received;
+}
-- 
2.5.5



More information about the dev mailing list