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_memzone.h> 50 #include <rte_mempool.h> 51 #include <rte_malloc.h> 52 #include <rte_mbuf.h> 53 #include <rte_prefetch.h> 54 #include <rte_string_fns.h> 55 56 #include "virtio_rxtx_simple.h" 57 58 #define RTE_VIRTIO_VPMD_RX_BURST 32 59 #define RTE_VIRTIO_DESC_PER_LOOP 8 60 #define RTE_VIRTIO_VPMD_RX_REARM_THRESH RTE_VIRTIO_VPMD_RX_BURST 61 62 /* virtio vPMD receive routine, only accept(nb_pkts >= RTE_VIRTIO_DESC_PER_LOOP) 63 * 64 * This routine is for non-mergeable RX, one desc for each guest buffer. 65 * This routine is based on the RX ring layout optimization. Each entry in the 66 * avail ring points to the desc with the same index in the desc ring and this 67 * will never be changed in the driver. 68 * 69 * - nb_pkts < RTE_VIRTIO_DESC_PER_LOOP, just return no packet 70 */ 71 uint16_t 72 virtio_recv_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts, 73 uint16_t nb_pkts) 74 { 75 struct virtnet_rx *rxvq = rx_queue; 76 struct virtqueue *vq = rxvq->vq; 77 struct virtio_hw *hw = vq->hw; 78 uint16_t nb_used; 79 uint16_t desc_idx; 80 struct vring_used_elem *rused; 81 struct rte_mbuf **sw_ring; 82 struct rte_mbuf **sw_ring_end; 83 uint16_t nb_pkts_received = 0; 84 __m128i shuf_msk1, shuf_msk2, len_adjust; 85 86 shuf_msk1 = _mm_set_epi8( 87 0xFF, 0xFF, 0xFF, 0xFF, 88 0xFF, 0xFF, /* vlan tci */ 89 5, 4, /* dat len */ 90 0xFF, 0xFF, 5, 4, /* pkt len */ 91 0xFF, 0xFF, 0xFF, 0xFF /* packet type */ 92 93 ); 94 95 shuf_msk2 = _mm_set_epi8( 96 0xFF, 0xFF, 0xFF, 0xFF, 97 0xFF, 0xFF, /* vlan tci */ 98 13, 12, /* dat len */ 99 0xFF, 0xFF, 13, 12, /* pkt len */ 100 0xFF, 0xFF, 0xFF, 0xFF /* packet type */ 101 ); 102 103 /* Subtract the header length. 104 * In which case do we need the header length in used->len ? 105 */ 106 len_adjust = _mm_set_epi16( 107 0, 0, 108 0, 109 (uint16_t)-vq->hw->vtnet_hdr_size, 110 0, (uint16_t)-vq->hw->vtnet_hdr_size, 111 0, 0); 112 113 if (unlikely(hw->started == 0)) 114 return nb_pkts_received; 115 116 if (unlikely(nb_pkts < RTE_VIRTIO_DESC_PER_LOOP)) 117 return 0; 118 119 nb_used = VIRTQUEUE_NUSED(vq); 120 121 rte_compiler_barrier(); 122 123 if (unlikely(nb_used == 0)) 124 return 0; 125 126 nb_pkts = RTE_ALIGN_FLOOR(nb_pkts, RTE_VIRTIO_DESC_PER_LOOP); 127 nb_used = RTE_MIN(nb_used, nb_pkts); 128 129 desc_idx = (uint16_t)(vq->vq_used_cons_idx & (vq->vq_nentries - 1)); 130 rused = &vq->vq_ring.used->ring[desc_idx]; 131 sw_ring = &vq->sw_ring[desc_idx]; 132 sw_ring_end = &vq->sw_ring[vq->vq_nentries]; 133 134 rte_prefetch0(rused); 135 136 if (vq->vq_free_cnt >= RTE_VIRTIO_VPMD_RX_REARM_THRESH) { 137 virtio_rxq_rearm_vec(rxvq); 138 if (unlikely(virtqueue_kick_prepare(vq))) 139 virtqueue_notify(vq); 140 } 141 142 for (nb_pkts_received = 0; 143 nb_pkts_received < nb_used;) { 144 __m128i desc[RTE_VIRTIO_DESC_PER_LOOP / 2]; 145 __m128i mbp[RTE_VIRTIO_DESC_PER_LOOP / 2]; 146 __m128i pkt_mb[RTE_VIRTIO_DESC_PER_LOOP]; 147 148 mbp[0] = _mm_loadu_si128((__m128i *)(sw_ring + 0)); 149 desc[0] = _mm_loadu_si128((__m128i *)(rused + 0)); 150 _mm_storeu_si128((__m128i *)&rx_pkts[0], mbp[0]); 151 152 mbp[1] = _mm_loadu_si128((__m128i *)(sw_ring + 2)); 153 desc[1] = _mm_loadu_si128((__m128i *)(rused + 2)); 154 _mm_storeu_si128((__m128i *)&rx_pkts[2], mbp[1]); 155 156 mbp[2] = _mm_loadu_si128((__m128i *)(sw_ring + 4)); 157 desc[2] = _mm_loadu_si128((__m128i *)(rused + 4)); 158 _mm_storeu_si128((__m128i *)&rx_pkts[4], mbp[2]); 159 160 mbp[3] = _mm_loadu_si128((__m128i *)(sw_ring + 6)); 161 desc[3] = _mm_loadu_si128((__m128i *)(rused + 6)); 162 _mm_storeu_si128((__m128i *)&rx_pkts[6], mbp[3]); 163 164 pkt_mb[1] = _mm_shuffle_epi8(desc[0], shuf_msk2); 165 pkt_mb[0] = _mm_shuffle_epi8(desc[0], shuf_msk1); 166 pkt_mb[1] = _mm_add_epi16(pkt_mb[1], len_adjust); 167 pkt_mb[0] = _mm_add_epi16(pkt_mb[0], len_adjust); 168 _mm_storeu_si128((void *)&rx_pkts[1]->rx_descriptor_fields1, 169 pkt_mb[1]); 170 _mm_storeu_si128((void *)&rx_pkts[0]->rx_descriptor_fields1, 171 pkt_mb[0]); 172 173 pkt_mb[3] = _mm_shuffle_epi8(desc[1], shuf_msk2); 174 pkt_mb[2] = _mm_shuffle_epi8(desc[1], shuf_msk1); 175 pkt_mb[3] = _mm_add_epi16(pkt_mb[3], len_adjust); 176 pkt_mb[2] = _mm_add_epi16(pkt_mb[2], len_adjust); 177 _mm_storeu_si128((void *)&rx_pkts[3]->rx_descriptor_fields1, 178 pkt_mb[3]); 179 _mm_storeu_si128((void *)&rx_pkts[2]->rx_descriptor_fields1, 180 pkt_mb[2]); 181 182 pkt_mb[5] = _mm_shuffle_epi8(desc[2], shuf_msk2); 183 pkt_mb[4] = _mm_shuffle_epi8(desc[2], shuf_msk1); 184 pkt_mb[5] = _mm_add_epi16(pkt_mb[5], len_adjust); 185 pkt_mb[4] = _mm_add_epi16(pkt_mb[4], len_adjust); 186 _mm_storeu_si128((void *)&rx_pkts[5]->rx_descriptor_fields1, 187 pkt_mb[5]); 188 _mm_storeu_si128((void *)&rx_pkts[4]->rx_descriptor_fields1, 189 pkt_mb[4]); 190 191 pkt_mb[7] = _mm_shuffle_epi8(desc[3], shuf_msk2); 192 pkt_mb[6] = _mm_shuffle_epi8(desc[3], shuf_msk1); 193 pkt_mb[7] = _mm_add_epi16(pkt_mb[7], len_adjust); 194 pkt_mb[6] = _mm_add_epi16(pkt_mb[6], len_adjust); 195 _mm_storeu_si128((void *)&rx_pkts[7]->rx_descriptor_fields1, 196 pkt_mb[7]); 197 _mm_storeu_si128((void *)&rx_pkts[6]->rx_descriptor_fields1, 198 pkt_mb[6]); 199 200 if (unlikely(nb_used <= RTE_VIRTIO_DESC_PER_LOOP)) { 201 if (sw_ring + nb_used <= sw_ring_end) 202 nb_pkts_received += nb_used; 203 else 204 nb_pkts_received += sw_ring_end - sw_ring; 205 break; 206 } else { 207 if (unlikely(sw_ring + RTE_VIRTIO_DESC_PER_LOOP >= 208 sw_ring_end)) { 209 nb_pkts_received += sw_ring_end - sw_ring; 210 break; 211 } else { 212 nb_pkts_received += RTE_VIRTIO_DESC_PER_LOOP; 213 214 rx_pkts += RTE_VIRTIO_DESC_PER_LOOP; 215 sw_ring += RTE_VIRTIO_DESC_PER_LOOP; 216 rused += RTE_VIRTIO_DESC_PER_LOOP; 217 nb_used -= RTE_VIRTIO_DESC_PER_LOOP; 218 } 219 } 220 } 221 222 vq->vq_used_cons_idx += nb_pkts_received; 223 vq->vq_free_cnt += nb_pkts_received; 224 rxvq->stats.packets += nb_pkts_received; 225 return nb_pkts_received; 226 } 227