1 /*- 2 * BSD LICENSE 3 * 4 * Copyright(c) 2010-2015 Intel Corporation. All rights reserved. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * * Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * * Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * * Neither the name of Intel Corporation nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #include <stdint.h> 35 #include <stdio.h> 36 #include <stdlib.h> 37 #include <string.h> 38 #include <errno.h> 39 40 #include <tmmintrin.h> 41 42 #include <rte_byteorder.h> 43 #include <rte_branch_prediction.h> 44 #include <rte_cycles.h> 45 #include <rte_ether.h> 46 #include <rte_ethdev.h> 47 #include <rte_errno.h> 48 #include <rte_memory.h> 49 #include <rte_mempool.h> 50 #include <rte_malloc.h> 51 #include <rte_mbuf.h> 52 #include <rte_prefetch.h> 53 #include <rte_string_fns.h> 54 55 #include "virtio_rxtx_simple.h" 56 57 #define RTE_VIRTIO_VPMD_RX_BURST 32 58 #define RTE_VIRTIO_DESC_PER_LOOP 8 59 #define RTE_VIRTIO_VPMD_RX_REARM_THRESH RTE_VIRTIO_VPMD_RX_BURST 60 61 /* virtio vPMD receive routine, only accept(nb_pkts >= RTE_VIRTIO_DESC_PER_LOOP) 62 * 63 * This routine is for non-mergeable RX, one desc for each guest buffer. 64 * This routine is based on the RX ring layout optimization. Each entry in the 65 * avail ring points to the desc with the same index in the desc ring and this 66 * will never be changed in the driver. 67 * 68 * - nb_pkts < RTE_VIRTIO_DESC_PER_LOOP, just return no packet 69 */ 70 uint16_t 71 virtio_recv_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts, 72 uint16_t nb_pkts) 73 { 74 struct virtnet_rx *rxvq = rx_queue; 75 struct virtqueue *vq = rxvq->vq; 76 struct virtio_hw *hw = vq->hw; 77 uint16_t nb_used; 78 uint16_t desc_idx; 79 struct vring_used_elem *rused; 80 struct rte_mbuf **sw_ring; 81 struct rte_mbuf **sw_ring_end; 82 uint16_t nb_pkts_received = 0; 83 __m128i shuf_msk1, shuf_msk2, len_adjust; 84 85 shuf_msk1 = _mm_set_epi8( 86 0xFF, 0xFF, 0xFF, 0xFF, 87 0xFF, 0xFF, /* vlan tci */ 88 5, 4, /* dat len */ 89 0xFF, 0xFF, 5, 4, /* pkt len */ 90 0xFF, 0xFF, 0xFF, 0xFF /* packet type */ 91 92 ); 93 94 shuf_msk2 = _mm_set_epi8( 95 0xFF, 0xFF, 0xFF, 0xFF, 96 0xFF, 0xFF, /* vlan tci */ 97 13, 12, /* dat len */ 98 0xFF, 0xFF, 13, 12, /* pkt len */ 99 0xFF, 0xFF, 0xFF, 0xFF /* packet type */ 100 ); 101 102 /* Subtract the header length. 103 * In which case do we need the header length in used->len ? 104 */ 105 len_adjust = _mm_set_epi16( 106 0, 0, 107 0, 108 (uint16_t)-vq->hw->vtnet_hdr_size, 109 0, (uint16_t)-vq->hw->vtnet_hdr_size, 110 0, 0); 111 112 if (unlikely(hw->started == 0)) 113 return nb_pkts_received; 114 115 if (unlikely(nb_pkts < RTE_VIRTIO_DESC_PER_LOOP)) 116 return 0; 117 118 nb_used = VIRTQUEUE_NUSED(vq); 119 120 rte_compiler_barrier(); 121 122 if (unlikely(nb_used == 0)) 123 return 0; 124 125 nb_pkts = RTE_ALIGN_FLOOR(nb_pkts, RTE_VIRTIO_DESC_PER_LOOP); 126 nb_used = RTE_MIN(nb_used, nb_pkts); 127 128 desc_idx = (uint16_t)(vq->vq_used_cons_idx & (vq->vq_nentries - 1)); 129 rused = &vq->vq_ring.used->ring[desc_idx]; 130 sw_ring = &vq->sw_ring[desc_idx]; 131 sw_ring_end = &vq->sw_ring[vq->vq_nentries]; 132 133 rte_prefetch0(rused); 134 135 if (vq->vq_free_cnt >= RTE_VIRTIO_VPMD_RX_REARM_THRESH) { 136 virtio_rxq_rearm_vec(rxvq); 137 if (unlikely(virtqueue_kick_prepare(vq))) 138 virtqueue_notify(vq); 139 } 140 141 for (nb_pkts_received = 0; 142 nb_pkts_received < nb_used;) { 143 __m128i desc[RTE_VIRTIO_DESC_PER_LOOP / 2]; 144 __m128i mbp[RTE_VIRTIO_DESC_PER_LOOP / 2]; 145 __m128i pkt_mb[RTE_VIRTIO_DESC_PER_LOOP]; 146 147 mbp[0] = _mm_loadu_si128((__m128i *)(sw_ring + 0)); 148 desc[0] = _mm_loadu_si128((__m128i *)(rused + 0)); 149 _mm_storeu_si128((__m128i *)&rx_pkts[0], mbp[0]); 150 151 mbp[1] = _mm_loadu_si128((__m128i *)(sw_ring + 2)); 152 desc[1] = _mm_loadu_si128((__m128i *)(rused + 2)); 153 _mm_storeu_si128((__m128i *)&rx_pkts[2], mbp[1]); 154 155 mbp[2] = _mm_loadu_si128((__m128i *)(sw_ring + 4)); 156 desc[2] = _mm_loadu_si128((__m128i *)(rused + 4)); 157 _mm_storeu_si128((__m128i *)&rx_pkts[4], mbp[2]); 158 159 mbp[3] = _mm_loadu_si128((__m128i *)(sw_ring + 6)); 160 desc[3] = _mm_loadu_si128((__m128i *)(rused + 6)); 161 _mm_storeu_si128((__m128i *)&rx_pkts[6], mbp[3]); 162 163 pkt_mb[1] = _mm_shuffle_epi8(desc[0], shuf_msk2); 164 pkt_mb[0] = _mm_shuffle_epi8(desc[0], shuf_msk1); 165 pkt_mb[1] = _mm_add_epi16(pkt_mb[1], len_adjust); 166 pkt_mb[0] = _mm_add_epi16(pkt_mb[0], len_adjust); 167 _mm_storeu_si128((void *)&rx_pkts[1]->rx_descriptor_fields1, 168 pkt_mb[1]); 169 _mm_storeu_si128((void *)&rx_pkts[0]->rx_descriptor_fields1, 170 pkt_mb[0]); 171 172 pkt_mb[3] = _mm_shuffle_epi8(desc[1], shuf_msk2); 173 pkt_mb[2] = _mm_shuffle_epi8(desc[1], shuf_msk1); 174 pkt_mb[3] = _mm_add_epi16(pkt_mb[3], len_adjust); 175 pkt_mb[2] = _mm_add_epi16(pkt_mb[2], len_adjust); 176 _mm_storeu_si128((void *)&rx_pkts[3]->rx_descriptor_fields1, 177 pkt_mb[3]); 178 _mm_storeu_si128((void *)&rx_pkts[2]->rx_descriptor_fields1, 179 pkt_mb[2]); 180 181 pkt_mb[5] = _mm_shuffle_epi8(desc[2], shuf_msk2); 182 pkt_mb[4] = _mm_shuffle_epi8(desc[2], shuf_msk1); 183 pkt_mb[5] = _mm_add_epi16(pkt_mb[5], len_adjust); 184 pkt_mb[4] = _mm_add_epi16(pkt_mb[4], len_adjust); 185 _mm_storeu_si128((void *)&rx_pkts[5]->rx_descriptor_fields1, 186 pkt_mb[5]); 187 _mm_storeu_si128((void *)&rx_pkts[4]->rx_descriptor_fields1, 188 pkt_mb[4]); 189 190 pkt_mb[7] = _mm_shuffle_epi8(desc[3], shuf_msk2); 191 pkt_mb[6] = _mm_shuffle_epi8(desc[3], shuf_msk1); 192 pkt_mb[7] = _mm_add_epi16(pkt_mb[7], len_adjust); 193 pkt_mb[6] = _mm_add_epi16(pkt_mb[6], len_adjust); 194 _mm_storeu_si128((void *)&rx_pkts[7]->rx_descriptor_fields1, 195 pkt_mb[7]); 196 _mm_storeu_si128((void *)&rx_pkts[6]->rx_descriptor_fields1, 197 pkt_mb[6]); 198 199 if (unlikely(nb_used <= RTE_VIRTIO_DESC_PER_LOOP)) { 200 if (sw_ring + nb_used <= sw_ring_end) 201 nb_pkts_received += nb_used; 202 else 203 nb_pkts_received += sw_ring_end - sw_ring; 204 break; 205 } else { 206 if (unlikely(sw_ring + RTE_VIRTIO_DESC_PER_LOOP >= 207 sw_ring_end)) { 208 nb_pkts_received += sw_ring_end - sw_ring; 209 break; 210 } else { 211 nb_pkts_received += RTE_VIRTIO_DESC_PER_LOOP; 212 213 rx_pkts += RTE_VIRTIO_DESC_PER_LOOP; 214 sw_ring += RTE_VIRTIO_DESC_PER_LOOP; 215 rused += RTE_VIRTIO_DESC_PER_LOOP; 216 nb_used -= RTE_VIRTIO_DESC_PER_LOOP; 217 } 218 } 219 } 220 221 vq->vq_used_cons_idx += nb_pkts_received; 222 vq->vq_free_cnt += nb_pkts_received; 223 rxvq->stats.packets += nb_pkts_received; 224 return nb_pkts_received; 225 } 226