1 /*- 2 * BSD LICENSE 3 * 4 * Copyright 2017 6WIND S.A. 5 * Copyright 2017 Mellanox. 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 6WIND S.A. 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 <assert.h> 35 #include <stdint.h> 36 #include <string.h> 37 #include <stdlib.h> 38 39 /* Verbs header. */ 40 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */ 41 #ifdef PEDANTIC 42 #pragma GCC diagnostic ignored "-Wpedantic" 43 #endif 44 #include <infiniband/verbs.h> 45 #include <infiniband/mlx5dv.h> 46 #ifdef PEDANTIC 47 #pragma GCC diagnostic error "-Wpedantic" 48 #endif 49 50 #include <rte_mbuf.h> 51 #include <rte_mempool.h> 52 #include <rte_prefetch.h> 53 54 #include "mlx5.h" 55 #include "mlx5_utils.h" 56 #include "mlx5_rxtx.h" 57 #include "mlx5_rxtx_vec.h" 58 #include "mlx5_autoconf.h" 59 #include "mlx5_defs.h" 60 #include "mlx5_prm.h" 61 62 #if defined RTE_ARCH_X86_64 63 #include "mlx5_rxtx_vec_sse.h" 64 #elif defined RTE_ARCH_ARM64 65 #include "mlx5_rxtx_vec_neon.h" 66 #else 67 #error "This should not be compiled if SIMD instructions are not supported." 68 #endif 69 70 /** 71 * Count the number of continuous single segment packets. 72 * 73 * @param pkts 74 * Pointer to array of packets. 75 * @param pkts_n 76 * Number of packets. 77 * 78 * @return 79 * Number of continuous single segment packets. 80 */ 81 static inline unsigned int 82 txq_check_multiseg(struct rte_mbuf **pkts, uint16_t pkts_n) 83 { 84 unsigned int pos; 85 86 if (!pkts_n) 87 return 0; 88 /* Count the number of continuous single segment packets. */ 89 for (pos = 0; pos < pkts_n; ++pos) 90 if (NB_SEGS(pkts[pos]) > 1) 91 break; 92 return pos; 93 } 94 95 /** 96 * Count the number of packets having same ol_flags and calculate cs_flags. 97 * 98 * @param txq 99 * Pointer to TX queue structure. 100 * @param pkts 101 * Pointer to array of packets. 102 * @param pkts_n 103 * Number of packets. 104 * @param cs_flags 105 * Pointer of flags to be returned. 106 * 107 * @return 108 * Number of packets having same ol_flags. 109 */ 110 static inline unsigned int 111 txq_calc_offload(struct mlx5_txq_data *txq, struct rte_mbuf **pkts, 112 uint16_t pkts_n, uint8_t *cs_flags) 113 { 114 unsigned int pos; 115 const uint64_t ol_mask = 116 PKT_TX_IP_CKSUM | PKT_TX_TCP_CKSUM | 117 PKT_TX_UDP_CKSUM | PKT_TX_TUNNEL_GRE | 118 PKT_TX_TUNNEL_VXLAN | PKT_TX_OUTER_IP_CKSUM; 119 120 if (!pkts_n) 121 return 0; 122 /* Count the number of packets having same ol_flags. */ 123 for (pos = 1; pos < pkts_n; ++pos) 124 if ((pkts[pos]->ol_flags ^ pkts[0]->ol_flags) & ol_mask) 125 break; 126 *cs_flags = txq_ol_cksum_to_cs(txq, pkts[0]); 127 return pos; 128 } 129 130 /** 131 * DPDK callback for vectorized TX. 132 * 133 * @param dpdk_txq 134 * Generic pointer to TX queue structure. 135 * @param[in] pkts 136 * Packets to transmit. 137 * @param pkts_n 138 * Number of packets in array. 139 * 140 * @return 141 * Number of packets successfully transmitted (<= pkts_n). 142 */ 143 uint16_t 144 mlx5_tx_burst_raw_vec(void *dpdk_txq, struct rte_mbuf **pkts, 145 uint16_t pkts_n) 146 { 147 struct mlx5_txq_data *txq = (struct mlx5_txq_data *)dpdk_txq; 148 uint16_t nb_tx = 0; 149 150 while (pkts_n > nb_tx) { 151 uint16_t n; 152 uint16_t ret; 153 154 n = RTE_MIN((uint16_t)(pkts_n - nb_tx), MLX5_VPMD_TX_MAX_BURST); 155 ret = txq_burst_v(txq, &pkts[nb_tx], n, 0); 156 nb_tx += ret; 157 if (!ret) 158 break; 159 } 160 return nb_tx; 161 } 162 163 /** 164 * DPDK callback for vectorized TX with multi-seg packets and offload. 165 * 166 * @param dpdk_txq 167 * Generic pointer to TX queue structure. 168 * @param[in] pkts 169 * Packets to transmit. 170 * @param pkts_n 171 * Number of packets in array. 172 * 173 * @return 174 * Number of packets successfully transmitted (<= pkts_n). 175 */ 176 uint16_t 177 mlx5_tx_burst_vec(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n) 178 { 179 struct mlx5_txq_data *txq = (struct mlx5_txq_data *)dpdk_txq; 180 uint16_t nb_tx = 0; 181 182 while (pkts_n > nb_tx) { 183 uint8_t cs_flags = 0; 184 uint16_t n; 185 uint16_t ret; 186 187 /* Transmit multi-seg packets in the head of pkts list. */ 188 if (!(txq->flags & ETH_TXQ_FLAGS_NOMULTSEGS) && 189 NB_SEGS(pkts[nb_tx]) > 1) 190 nb_tx += txq_scatter_v(txq, 191 &pkts[nb_tx], 192 pkts_n - nb_tx); 193 n = RTE_MIN((uint16_t)(pkts_n - nb_tx), MLX5_VPMD_TX_MAX_BURST); 194 if (!(txq->flags & ETH_TXQ_FLAGS_NOMULTSEGS)) 195 n = txq_check_multiseg(&pkts[nb_tx], n); 196 if (!(txq->flags & ETH_TXQ_FLAGS_NOOFFLOADS)) 197 n = txq_calc_offload(txq, &pkts[nb_tx], n, &cs_flags); 198 ret = txq_burst_v(txq, &pkts[nb_tx], n, cs_flags); 199 nb_tx += ret; 200 if (!ret) 201 break; 202 } 203 return nb_tx; 204 } 205 206 /** 207 * Skip error packets. 208 * 209 * @param rxq 210 * Pointer to RX queue structure. 211 * @param[out] pkts 212 * Array to store received packets. 213 * @param pkts_n 214 * Maximum number of packets in array. 215 * 216 * @return 217 * Number of packets successfully received (<= pkts_n). 218 */ 219 static uint16_t 220 rxq_handle_pending_error(struct mlx5_rxq_data *rxq, struct rte_mbuf **pkts, 221 uint16_t pkts_n) 222 { 223 uint16_t n = 0; 224 unsigned int i; 225 #ifdef MLX5_PMD_SOFT_COUNTERS 226 uint32_t err_bytes = 0; 227 #endif 228 229 for (i = 0; i < pkts_n; ++i) { 230 struct rte_mbuf *pkt = pkts[i]; 231 232 if (pkt->packet_type == RTE_PTYPE_ALL_MASK) { 233 #ifdef MLX5_PMD_SOFT_COUNTERS 234 err_bytes += PKT_LEN(pkt); 235 #endif 236 rte_pktmbuf_free_seg(pkt); 237 } else { 238 pkts[n++] = pkt; 239 } 240 } 241 rxq->stats.idropped += (pkts_n - n); 242 #ifdef MLX5_PMD_SOFT_COUNTERS 243 /* Correct counters of errored completions. */ 244 rxq->stats.ipackets -= (pkts_n - n); 245 rxq->stats.ibytes -= err_bytes; 246 #endif 247 return n; 248 } 249 250 /** 251 * DPDK callback for vectorized RX. 252 * 253 * @param dpdk_rxq 254 * Generic pointer to RX queue structure. 255 * @param[out] pkts 256 * Array to store received packets. 257 * @param pkts_n 258 * Maximum number of packets in array. 259 * 260 * @return 261 * Number of packets successfully received (<= pkts_n). 262 */ 263 uint16_t 264 mlx5_rx_burst_vec(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n) 265 { 266 struct mlx5_rxq_data *rxq = dpdk_rxq; 267 uint16_t nb_rx; 268 uint64_t err = 0; 269 270 nb_rx = rxq_burst_v(rxq, pkts, pkts_n, &err); 271 if (unlikely(err)) 272 nb_rx = rxq_handle_pending_error(rxq, pkts, nb_rx); 273 return nb_rx; 274 } 275 276 /** 277 * Check Tx queue flags are set for raw vectorized Tx. 278 * 279 * @param dev 280 * Pointer to Ethernet device. 281 * 282 * @return 283 * 1 if supported, negative errno value if not. 284 */ 285 int __attribute__((cold)) 286 mlx5_check_raw_vec_tx_support(struct rte_eth_dev *dev) 287 { 288 struct priv *priv = dev->data->dev_private; 289 uint16_t i; 290 291 /* All the configured queues should support. */ 292 for (i = 0; i < priv->txqs_n; ++i) { 293 struct mlx5_txq_data *txq = (*priv->txqs)[i]; 294 295 if (!(txq->flags & ETH_TXQ_FLAGS_NOMULTSEGS) || 296 !(txq->flags & ETH_TXQ_FLAGS_NOOFFLOADS)) 297 break; 298 } 299 if (i != priv->txqs_n) 300 return -ENOTSUP; 301 return 1; 302 } 303 304 /** 305 * Check a device can support vectorized TX. 306 * 307 * @param dev 308 * Pointer to Ethernet device. 309 * 310 * @return 311 * 1 if supported, negative errno value if not. 312 */ 313 int __attribute__((cold)) 314 mlx5_check_vec_tx_support(struct rte_eth_dev *dev) 315 { 316 struct priv *priv = dev->data->dev_private; 317 318 if (!priv->tx_vec_en || 319 priv->txqs_n > MLX5_VPMD_MIN_TXQS || 320 priv->mps != MLX5_MPW_ENHANCED || 321 priv->tso) 322 return -ENOTSUP; 323 return 1; 324 } 325 326 /** 327 * Check a RX queue can support vectorized RX. 328 * 329 * @param rxq 330 * Pointer to RX queue. 331 * 332 * @return 333 * 1 if supported, negative errno value if not. 334 */ 335 int __attribute__((cold)) 336 mlx5_rxq_check_vec_support(struct mlx5_rxq_data *rxq) 337 { 338 struct mlx5_rxq_ctrl *ctrl = 339 container_of(rxq, struct mlx5_rxq_ctrl, rxq); 340 341 if (!ctrl->priv->rx_vec_en || rxq->sges_n != 0) 342 return -ENOTSUP; 343 return 1; 344 } 345 346 /** 347 * Check a device can support vectorized RX. 348 * 349 * @param dev 350 * Pointer to Ethernet device. 351 * 352 * @return 353 * 1 if supported, negative errno value if not. 354 */ 355 int __attribute__((cold)) 356 mlx5_check_vec_rx_support(struct rte_eth_dev *dev) 357 { 358 struct priv *priv = dev->data->dev_private; 359 uint16_t i; 360 361 if (!priv->rx_vec_en) 362 return -ENOTSUP; 363 /* All the configured queues should support. */ 364 for (i = 0; i < priv->rxqs_n; ++i) { 365 struct mlx5_rxq_data *rxq = (*priv->rxqs)[i]; 366 367 if (!rxq) 368 continue; 369 if (mlx5_rxq_check_vec_support(rxq) < 0) 370 break; 371 } 372 if (i != priv->rxqs_n) 373 return -ENOTSUP; 374 return 1; 375 } 376