drivers/net/qede/qede_rxtx.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright (c) 2016 - 2018 Cavium Inc.
   3  * All rights reserved.
   4  * www.cavium.com
   5  */
   6
   7 #include <rte_net.h>
   8 #include "qede_rxtx.h"
   9
  10 static inline int qede_alloc_rx_buffer(struct qede_rx_queue *rxq)
  11 {
  12         struct rte_mbuf *new_mb = NULL;
  13         struct eth_rx_bd *rx_bd;
  14         dma_addr_t mapping;
  15         uint16_t idx = rxq->sw_rx_prod & NUM_RX_BDS(rxq);
  16
  17         new_mb = rte_mbuf_raw_alloc(rxq->mb_pool);
  18         if (unlikely(!new_mb)) {
  19                 PMD_RX_LOG(ERR, rxq,
  20                            "Failed to allocate rx buffer "
  21                            "sw_rx_prod %u sw_rx_cons %u mp entries %u free %u",
  22                            idx, rxq->sw_rx_cons & NUM_RX_BDS(rxq),
  23                            rte_mempool_avail_count(rxq->mb_pool),
  24                            rte_mempool_in_use_count(rxq->mb_pool));
  25                 return -ENOMEM;
  26         }
  27         rxq->sw_rx_ring[idx].mbuf = new_mb;
  28         rxq->sw_rx_ring[idx].page_offset = 0;
  29         mapping = rte_mbuf_data_iova_default(new_mb);
  30         /* Advance PROD and get BD pointer */
  31         rx_bd = (struct eth_rx_bd *)ecore_chain_produce(&rxq->rx_bd_ring);
  32         rx_bd->addr.hi = rte_cpu_to_le_32(U64_HI(mapping));
  33         rx_bd->addr.lo = rte_cpu_to_le_32(U64_LO(mapping));
  34         rxq->sw_rx_prod++;
  35         return 0;
  36 }
  37
  38 #define QEDE_MAX_BULK_ALLOC_COUNT 512
  39
  40 static inline int qede_alloc_rx_bulk_mbufs(struct qede_rx_queue *rxq, int count)
  41 {
  42         void *obj_p[QEDE_MAX_BULK_ALLOC_COUNT] __rte_cache_aligned;
  43         struct rte_mbuf *mbuf = NULL;
  44         struct eth_rx_bd *rx_bd;
  45         dma_addr_t mapping;
  46         int i, ret = 0;
  47         uint16_t idx;
  48
  49         idx = rxq->sw_rx_prod & NUM_RX_BDS(rxq);
  50
  51         if (count > QEDE_MAX_BULK_ALLOC_COUNT)
  52                 count = QEDE_MAX_BULK_ALLOC_COUNT;
  53
  54         ret = rte_mempool_get_bulk(rxq->mb_pool, obj_p, count);
  55         if (unlikely(ret)) {
  56                 PMD_RX_LOG(ERR, rxq,
  57                            "Failed to allocate %d rx buffers "
  58                             "sw_rx_prod %u sw_rx_cons %u mp entries %u free %u",
  59                             count, idx, rxq->sw_rx_cons & NUM_RX_BDS(rxq),
  60                             rte_mempool_avail_count(rxq->mb_pool),
  61                             rte_mempool_in_use_count(rxq->mb_pool));
  62                 return -ENOMEM;
  63         }
  64
  65         for (i = 0; i < count; i++) {
  66                 mbuf = obj_p[i];
  67                 if (likely(i < count - 1))
  68                         rte_prefetch0(obj_p[i + 1]);
  69
  70                 idx = rxq->sw_rx_prod & NUM_RX_BDS(rxq);
  71                 rxq->sw_rx_ring[idx].mbuf = mbuf;
  72                 rxq->sw_rx_ring[idx].page_offset = 0;
  73                 mapping = rte_mbuf_data_iova_default(mbuf);
  74                 rx_bd = (struct eth_rx_bd *)
  75                         ecore_chain_produce(&rxq->rx_bd_ring);
  76                 rx_bd->addr.hi = rte_cpu_to_le_32(U64_HI(mapping));
  77                 rx_bd->addr.lo = rte_cpu_to_le_32(U64_LO(mapping));
  78                 rxq->sw_rx_prod++;
  79         }
  80
  81         return 0;
  82 }
  83
  84 /* Criterias for calculating Rx buffer size -
  85  * 1) rx_buf_size should not exceed the size of mbuf
  86  * 2) In scattered_rx mode - minimum rx_buf_size should be
  87  *    (MTU + Maximum L2 Header Size + 2) / ETH_RX_MAX_BUFF_PER_PKT
  88  * 3) In regular mode - minimum rx_buf_size should be
  89  *    (MTU + Maximum L2 Header Size + 2)
  90  *    In above cases +2 corrosponds to 2 bytes padding in front of L2
  91  *    header.
  92  * 4) rx_buf_size should be cacheline-size aligned. So considering
  93  *    criteria 1, we need to adjust the size to floor instead of ceil,
  94  *    so that we don't exceed mbuf size while ceiling rx_buf_size.
  95  */
  96 int
  97 qede_calc_rx_buf_size(struct rte_eth_dev *dev, uint16_t mbufsz,
  98                       uint16_t max_frame_size)
  99 {
 100         struct qede_dev *qdev = QEDE_INIT_QDEV(dev);
 101         struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
 102         int rx_buf_size;
 103
 104         if (dev->data->scattered_rx) {
 105                 /* per HW limitation, only ETH_RX_MAX_BUFF_PER_PKT number of
 106                  * bufferes can be used for single packet. So need to make sure
 107                  * mbuf size is sufficient enough for this.
 108                  */
 109                 if ((mbufsz * ETH_RX_MAX_BUFF_PER_PKT) <
 110                      (max_frame_size + QEDE_ETH_OVERHEAD)) {
 111                         DP_ERR(edev, "mbuf %d size is not enough to hold max fragments (%d) for max rx packet length (%d)\n",
 112                                mbufsz, ETH_RX_MAX_BUFF_PER_PKT, max_frame_size);
 113                         return -EINVAL;
 114                 }
 115
 116                 rx_buf_size = RTE_MAX(mbufsz,
 117                                       (max_frame_size + QEDE_ETH_OVERHEAD) /
 118                                        ETH_RX_MAX_BUFF_PER_PKT);
 119         } else {
 120                 rx_buf_size = max_frame_size + QEDE_ETH_OVERHEAD;
 121         }
 122
 123         /* Align to cache-line size if needed */
 124         return QEDE_FLOOR_TO_CACHE_LINE_SIZE(rx_buf_size);
 125 }
 126
 127 int
 128 qede_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
 129                     uint16_t nb_desc, unsigned int socket_id,
 130                     __rte_unused const struct rte_eth_rxconf *rx_conf,
 131                     struct rte_mempool *mp)
 132 {
 133         struct qede_dev *qdev = QEDE_INIT_QDEV(dev);
 134         struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
 135         struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
 136         struct qede_rx_queue *rxq;
 137         uint16_t max_rx_pkt_len;
 138         uint16_t bufsz;
 139         size_t size;
 140         int rc;
 141
 142         PMD_INIT_FUNC_TRACE(edev);
 143
 144         /* Note: Ring size/align is controlled by struct rte_eth_desc_lim */
 145         if (!rte_is_power_of_2(nb_desc)) {
 146                 DP_ERR(edev, "Ring size %u is not power of 2\n",
 147                           nb_desc);
 148                 return -EINVAL;
 149         }
 150
 151         /* Free memory prior to re-allocation if needed... */
 152         if (dev->data->rx_queues[queue_idx] != NULL) {
 153                 qede_rx_queue_release(dev->data->rx_queues[queue_idx]);
 154                 dev->data->rx_queues[queue_idx] = NULL;
 155         }
 156
 157         /* First allocate the rx queue data structure */
 158         rxq = rte_zmalloc_socket("qede_rx_queue", sizeof(struct qede_rx_queue),
 159                                  RTE_CACHE_LINE_SIZE, socket_id);
 160
 161         if (!rxq) {
 162                 DP_ERR(edev, "Unable to allocate memory for rxq on socket %u",
 163                           socket_id);
 164                 return -ENOMEM;
 165         }
 166
 167         rxq->qdev = qdev;
 168         rxq->mb_pool = mp;
 169         rxq->nb_rx_desc = nb_desc;
 170         rxq->queue_id = queue_idx;
 171         rxq->port_id = dev->data->port_id;
 172
 173         max_rx_pkt_len = (uint16_t)rxmode->max_rx_pkt_len;
 174
 175         /* Fix up RX buffer size */
 176         bufsz = (uint16_t)rte_pktmbuf_data_room_size(mp) - RTE_PKTMBUF_HEADROOM;
 177         /* cache align the mbuf size to simplfy rx_buf_size calculation */
 178         bufsz = QEDE_FLOOR_TO_CACHE_LINE_SIZE(bufsz);
 179         if ((rxmode->offloads & DEV_RX_OFFLOAD_SCATTER) ||
 180             (max_rx_pkt_len + QEDE_ETH_OVERHEAD) > bufsz) {
 181                 if (!dev->data->scattered_rx) {
 182                         DP_INFO(edev, "Forcing scatter-gather mode\n");
 183                         dev->data->scattered_rx = 1;
 184                 }
 185         }
 186
 187         rc = qede_calc_rx_buf_size(dev, bufsz, max_rx_pkt_len);
 188         if (rc < 0) {
 189                 rte_free(rxq);
 190                 return rc;
 191         }
 192
 193         rxq->rx_buf_size = rc;
 194
 195         DP_INFO(edev, "mtu %u mbufsz %u bd_max_bytes %u scatter_mode %d\n",
 196                 qdev->mtu, bufsz, rxq->rx_buf_size, dev->data->scattered_rx);
 197
 198         /* Allocate the parallel driver ring for Rx buffers */
 199         size = sizeof(*rxq->sw_rx_ring) * rxq->nb_rx_desc;
 200         rxq->sw_rx_ring = rte_zmalloc_socket("sw_rx_ring", size,
 201                                              RTE_CACHE_LINE_SIZE, socket_id);
 202         if (!rxq->sw_rx_ring) {
 203                 DP_ERR(edev, "Memory allocation fails for sw_rx_ring on"
 204                        " socket %u\n", socket_id);
 205                 rte_free(rxq);
 206                 return -ENOMEM;
 207         }
 208
 209         /* Allocate FW Rx ring  */
 210         rc = qdev->ops->common->chain_alloc(edev,
 211                                             ECORE_CHAIN_USE_TO_CONSUME_PRODUCE,
 212                                             ECORE_CHAIN_MODE_NEXT_PTR,
 213                                             ECORE_CHAIN_CNT_TYPE_U16,
 214                                             rxq->nb_rx_desc,
 215                                             sizeof(struct eth_rx_bd),
 216                                             &rxq->rx_bd_ring,
 217                                             NULL);
 218
 219         if (rc != ECORE_SUCCESS) {
 220                 DP_ERR(edev, "Memory allocation fails for RX BD ring"
 221                        " on socket %u\n", socket_id);
 222                 rte_free(rxq->sw_rx_ring);
 223                 rte_free(rxq);
 224                 return -ENOMEM;
 225         }
 226
 227         /* Allocate FW completion ring */
 228         rc = qdev->ops->common->chain_alloc(edev,
 229                                             ECORE_CHAIN_USE_TO_CONSUME,
 230                                             ECORE_CHAIN_MODE_PBL,
 231                                             ECORE_CHAIN_CNT_TYPE_U16,
 232                                             rxq->nb_rx_desc,
 233                                             sizeof(union eth_rx_cqe),
 234                                             &rxq->rx_comp_ring,
 235                                             NULL);
 236
 237         if (rc != ECORE_SUCCESS) {
 238                 DP_ERR(edev, "Memory allocation fails for RX CQE ring"
 239                        " on socket %u\n", socket_id);
 240                 qdev->ops->common->chain_free(edev, &rxq->rx_bd_ring);
 241                 rte_free(rxq->sw_rx_ring);
 242                 rte_free(rxq);
 243                 return -ENOMEM;
 244         }
 245
 246         dev->data->rx_queues[queue_idx] = rxq;
 247         qdev->fp_array[queue_idx].rxq = rxq;
 248
 249         DP_INFO(edev, "rxq %d num_desc %u rx_buf_size=%u socket %u\n",
 250                   queue_idx, nb_desc, rxq->rx_buf_size, socket_id);
 251
 252         return 0;
 253 }
 254
 255 static void
 256 qede_rx_queue_reset(__rte_unused struct qede_dev *qdev,
 257                     struct qede_rx_queue *rxq)
 258 {
 259         DP_INFO(&qdev->edev, "Reset RX queue %u\n", rxq->queue_id);
 260         ecore_chain_reset(&rxq->rx_bd_ring);
 261         ecore_chain_reset(&rxq->rx_comp_ring);
 262         rxq->sw_rx_prod = 0;
 263         rxq->sw_rx_cons = 0;
 264         *rxq->hw_cons_ptr = 0;
 265 }
 266
 267 static void qede_rx_queue_release_mbufs(struct qede_rx_queue *rxq)
 268 {
 269         uint16_t i;
 270
 271         if (rxq->sw_rx_ring) {
 272                 for (i = 0; i < rxq->nb_rx_desc; i++) {
 273                         if (rxq->sw_rx_ring[i].mbuf) {
 274                                 rte_pktmbuf_free(rxq->sw_rx_ring[i].mbuf);
 275                                 rxq->sw_rx_ring[i].mbuf = NULL;
 276                         }
 277                 }
 278         }
 279 }
 280
 281 void qede_rx_queue_release(void *rx_queue)
 282 {
 283         struct qede_rx_queue *rxq = rx_queue;
 284         struct qede_dev *qdev;
 285         struct ecore_dev *edev;
 286
 287         if (rxq) {
 288                 qdev = rxq->qdev;
 289                 edev = QEDE_INIT_EDEV(qdev);
 290                 PMD_INIT_FUNC_TRACE(edev);
 291                 qede_rx_queue_release_mbufs(rxq);
 292                 qdev->ops->common->chain_free(edev, &rxq->rx_bd_ring);
 293                 qdev->ops->common->chain_free(edev, &rxq->rx_comp_ring);
 294                 rte_free(rxq->sw_rx_ring);
 295                 rte_free(rxq);
 296         }
 297 }
 298
 299 /* Stops a given RX queue in the HW */
 300 static int qede_rx_queue_stop(struct rte_eth_dev *eth_dev, uint16_t rx_queue_id)
 301 {
 302         struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
 303         struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
 304         struct ecore_hwfn *p_hwfn;
 305         struct qede_rx_queue *rxq;
 306         int hwfn_index;
 307         int rc;
 308
 309         if (rx_queue_id < eth_dev->data->nb_rx_queues) {
 310                 rxq = eth_dev->data->rx_queues[rx_queue_id];
 311                 hwfn_index = rx_queue_id % edev->num_hwfns;
 312                 p_hwfn = &edev->hwfns[hwfn_index];
 313                 rc = ecore_eth_rx_queue_stop(p_hwfn, rxq->handle,
 314                                 true, false);
 315                 if (rc != ECORE_SUCCESS) {
 316                         DP_ERR(edev, "RX queue %u stop fails\n", rx_queue_id);
 317                         return -1;
 318                 }
 319                 qede_rx_queue_release_mbufs(rxq);
 320                 qede_rx_queue_reset(qdev, rxq);
 321                 eth_dev->data->rx_queue_state[rx_queue_id] =
 322                         RTE_ETH_QUEUE_STATE_STOPPED;
 323                 DP_INFO(edev, "RX queue %u stopped\n", rx_queue_id);
 324         } else {
 325                 DP_ERR(edev, "RX queue %u is not in range\n", rx_queue_id);
 326                 rc = -EINVAL;
 327         }
 328
 329         return rc;
 330 }
 331
 332 int
 333 qede_tx_queue_setup(struct rte_eth_dev *dev,
 334                     uint16_t queue_idx,
 335                     uint16_t nb_desc,
 336                     unsigned int socket_id,
 337                     const struct rte_eth_txconf *tx_conf)
 338 {
 339         struct qede_dev *qdev = dev->data->dev_private;
 340         struct ecore_dev *edev = &qdev->edev;
 341         struct qede_tx_queue *txq;
 342         int rc;
 343
 344         PMD_INIT_FUNC_TRACE(edev);
 345
 346         if (!rte_is_power_of_2(nb_desc)) {
 347                 DP_ERR(edev, "Ring size %u is not power of 2\n",
 348                        nb_desc);
 349                 return -EINVAL;
 350         }
 351
 352         /* Free memory prior to re-allocation if needed... */
 353         if (dev->data->tx_queues[queue_idx] != NULL) {
 354                 qede_tx_queue_release(dev->data->tx_queues[queue_idx]);
 355                 dev->data->tx_queues[queue_idx] = NULL;
 356         }
 357
 358         txq = rte_zmalloc_socket("qede_tx_queue", sizeof(struct qede_tx_queue),
 359                                  RTE_CACHE_LINE_SIZE, socket_id);
 360
 361         if (txq == NULL) {
 362                 DP_ERR(edev,
 363                        "Unable to allocate memory for txq on socket %u",
 364                        socket_id);
 365                 return -ENOMEM;
 366         }
 367
 368         txq->nb_tx_desc = nb_desc;
 369         txq->qdev = qdev;
 370         txq->port_id = dev->data->port_id;
 371
 372         rc = qdev->ops->common->chain_alloc(edev,
 373                                             ECORE_CHAIN_USE_TO_CONSUME_PRODUCE,
 374                                             ECORE_CHAIN_MODE_PBL,
 375                                             ECORE_CHAIN_CNT_TYPE_U16,
 376                                             txq->nb_tx_desc,
 377                                             sizeof(union eth_tx_bd_types),
 378                                             &txq->tx_pbl,
 379                                             NULL);
 380         if (rc != ECORE_SUCCESS) {
 381                 DP_ERR(edev,
 382                        "Unable to allocate memory for txbd ring on socket %u",
 383                        socket_id);
 384                 qede_tx_queue_release(txq);
 385                 return -ENOMEM;
 386         }
 387
 388         /* Allocate software ring */
 389         txq->sw_tx_ring = rte_zmalloc_socket("txq->sw_tx_ring",
 390                                              (sizeof(struct qede_tx_entry) *
 391                                               txq->nb_tx_desc),
 392                                              RTE_CACHE_LINE_SIZE, socket_id);
 393
 394         if (!txq->sw_tx_ring) {
 395                 DP_ERR(edev,
 396                        "Unable to allocate memory for txbd ring on socket %u",
 397                        socket_id);
 398                 qdev->ops->common->chain_free(edev, &txq->tx_pbl);
 399                 qede_tx_queue_release(txq);
 400                 return -ENOMEM;
 401         }
 402
 403         txq->queue_id = queue_idx;
 404
 405         txq->nb_tx_avail = txq->nb_tx_desc;
 406
 407         txq->tx_free_thresh =
 408             tx_conf->tx_free_thresh ? tx_conf->tx_free_thresh :
 409             (txq->nb_tx_desc - QEDE_DEFAULT_TX_FREE_THRESH);
 410
 411         dev->data->tx_queues[queue_idx] = txq;
 412         qdev->fp_array[queue_idx].txq = txq;
 413
 414         DP_INFO(edev,
 415                   "txq %u num_desc %u tx_free_thresh %u socket %u\n",
 416                   queue_idx, nb_desc, txq->tx_free_thresh, socket_id);
 417
 418         return 0;
 419 }
 420
 421 static void
 422 qede_tx_queue_reset(__rte_unused struct qede_dev *qdev,
 423                     struct qede_tx_queue *txq)
 424 {
 425         DP_INFO(&qdev->edev, "Reset TX queue %u\n", txq->queue_id);
 426         ecore_chain_reset(&txq->tx_pbl);
 427         txq->sw_tx_cons = 0;
 428         txq->sw_tx_prod = 0;
 429         *txq->hw_cons_ptr = 0;
 430 }
 431
 432 static void qede_tx_queue_release_mbufs(struct qede_tx_queue *txq)
 433 {
 434         uint16_t i;
 435
 436         if (txq->sw_tx_ring) {
 437                 for (i = 0; i < txq->nb_tx_desc; i++) {
 438                         if (txq->sw_tx_ring[i].mbuf) {
 439                                 rte_pktmbuf_free(txq->sw_tx_ring[i].mbuf);
 440                                 txq->sw_tx_ring[i].mbuf = NULL;
 441                         }
 442                 }
 443         }
 444 }
 445
 446 void qede_tx_queue_release(void *tx_queue)
 447 {
 448         struct qede_tx_queue *txq = tx_queue;
 449         struct qede_dev *qdev;
 450         struct ecore_dev *edev;
 451
 452         if (txq) {
 453                 qdev = txq->qdev;
 454                 edev = QEDE_INIT_EDEV(qdev);
 455                 PMD_INIT_FUNC_TRACE(edev);
 456                 qede_tx_queue_release_mbufs(txq);
 457                 qdev->ops->common->chain_free(edev, &txq->tx_pbl);
 458                 rte_free(txq->sw_tx_ring);
 459                 rte_free(txq);
 460         }
 461 }
 462
 463 /* This function allocates fast-path status block memory */
 464 static int
 465 qede_alloc_mem_sb(struct qede_dev *qdev, struct ecore_sb_info *sb_info,
 466                   uint16_t sb_id)
 467 {
 468         struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
 469         struct status_block_e4 *sb_virt;
 470         dma_addr_t sb_phys;
 471         int rc;
 472
 473         sb_virt = OSAL_DMA_ALLOC_COHERENT(edev, &sb_phys,
 474                                           sizeof(struct status_block_e4));
 475         if (!sb_virt) {
 476                 DP_ERR(edev, "Status block allocation failed\n");
 477                 return -ENOMEM;
 478         }
 479         rc = qdev->ops->common->sb_init(edev, sb_info, sb_virt,
 480                                         sb_phys, sb_id);
 481         if (rc) {
 482                 DP_ERR(edev, "Status block initialization failed\n");
 483                 OSAL_DMA_FREE_COHERENT(edev, sb_virt, sb_phys,
 484                                        sizeof(struct status_block_e4));
 485                 return rc;
 486         }
 487
 488         return 0;
 489 }
 490
 491 int qede_alloc_fp_resc(struct qede_dev *qdev)
 492 {
 493         struct ecore_dev *edev = &qdev->edev;
 494         struct qede_fastpath *fp;
 495         uint32_t num_sbs;
 496         uint16_t sb_idx;
 497
 498         if (IS_VF(edev))
 499                 ecore_vf_get_num_sbs(ECORE_LEADING_HWFN(edev), &num_sbs);
 500         else
 501                 num_sbs = ecore_cxt_get_proto_cid_count
 502                           (ECORE_LEADING_HWFN(edev), PROTOCOLID_ETH, NULL);
 503
 504         if (num_sbs == 0) {
 505                 DP_ERR(edev, "No status blocks available\n");
 506                 return -EINVAL;
 507         }
 508
 509         qdev->fp_array = rte_calloc("fp", QEDE_RXTX_MAX(qdev),
 510                                 sizeof(*qdev->fp_array), RTE_CACHE_LINE_SIZE);
 511
 512         if (!qdev->fp_array) {
 513                 DP_ERR(edev, "fp array allocation failed\n");
 514                 return -ENOMEM;
 515         }
 516
 517         memset((void *)qdev->fp_array, 0, QEDE_RXTX_MAX(qdev) *
 518                         sizeof(*qdev->fp_array));
 519
 520         for (sb_idx = 0; sb_idx < QEDE_RXTX_MAX(qdev); sb_idx++) {
 521                 fp = &qdev->fp_array[sb_idx];
 522                 if (!fp)
 523                         continue;
 524                 fp->sb_info = rte_calloc("sb", 1, sizeof(struct ecore_sb_info),
 525                                 RTE_CACHE_LINE_SIZE);
 526                 if (!fp->sb_info) {
 527                         DP_ERR(edev, "FP sb_info allocation fails\n");
 528                         return -1;
 529                 }
 530                 if (qede_alloc_mem_sb(qdev, fp->sb_info, sb_idx)) {
 531                         DP_ERR(edev, "FP status block allocation fails\n");
 532                         return -1;
 533                 }
 534                 DP_INFO(edev, "sb_info idx 0x%x initialized\n",
 535                                 fp->sb_info->igu_sb_id);
 536         }
 537
 538         return 0;
 539 }
 540
 541 void qede_dealloc_fp_resc(struct rte_eth_dev *eth_dev)
 542 {
 543         struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
 544         struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
 545         struct qede_fastpath *fp;
 546         uint16_t sb_idx;
 547         uint8_t i;
 548
 549         PMD_INIT_FUNC_TRACE(edev);
 550
 551         for (sb_idx = 0; sb_idx < QEDE_RXTX_MAX(qdev); sb_idx++) {
 552                 fp = &qdev->fp_array[sb_idx];
 553                 if (!fp)
 554                         continue;
 555                 DP_INFO(edev, "Free sb_info index 0x%x\n",
 556                                 fp->sb_info->igu_sb_id);
 557                 if (fp->sb_info) {
 558                         OSAL_DMA_FREE_COHERENT(edev, fp->sb_info->sb_virt,
 559                                 fp->sb_info->sb_phys,
 560                                 sizeof(struct status_block_e4));
 561                         rte_free(fp->sb_info);
 562                         fp->sb_info = NULL;
 563                 }
 564         }
 565
 566         /* Free packet buffers and ring memories */
 567         for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
 568                 if (eth_dev->data->rx_queues[i]) {
 569                         qede_rx_queue_release(eth_dev->data->rx_queues[i]);
 570                         eth_dev->data->rx_queues[i] = NULL;
 571                 }
 572         }
 573
 574         for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
 575                 if (eth_dev->data->tx_queues[i]) {
 576                         qede_tx_queue_release(eth_dev->data->tx_queues[i]);
 577                         eth_dev->data->tx_queues[i] = NULL;
 578                 }
 579         }
 580
 581         if (qdev->fp_array)
 582                 rte_free(qdev->fp_array);
 583         qdev->fp_array = NULL;
 584 }
 585
 586 static inline void
 587 qede_update_rx_prod(__rte_unused struct qede_dev *edev,
 588                     struct qede_rx_queue *rxq)
 589 {
 590         uint16_t bd_prod = ecore_chain_get_prod_idx(&rxq->rx_bd_ring);
 591         uint16_t cqe_prod = ecore_chain_get_prod_idx(&rxq->rx_comp_ring);
 592         struct eth_rx_prod_data rx_prods = { 0 };
 593
 594         /* Update producers */
 595         rx_prods.bd_prod = rte_cpu_to_le_16(bd_prod);
 596         rx_prods.cqe_prod = rte_cpu_to_le_16(cqe_prod);
 597
 598         /* Make sure that the BD and SGE data is updated before updating the
 599          * producers since FW might read the BD/SGE right after the producer
 600          * is updated.
 601          */
 602         rte_wmb();
 603
 604         internal_ram_wr(rxq->hw_rxq_prod_addr, sizeof(rx_prods),
 605                         (uint32_t *)&rx_prods);
 606
 607         /* mmiowb is needed to synchronize doorbell writes from more than one
 608          * processor. It guarantees that the write arrives to the device before
 609          * the napi lock is released and another qede_poll is called (possibly
 610          * on another CPU). Without this barrier, the next doorbell can bypass
 611          * this doorbell. This is applicable to IA64/Altix systems.
 612          */
 613         rte_wmb();
 614
 615         PMD_RX_LOG(DEBUG, rxq, "bd_prod %u  cqe_prod %u", bd_prod, cqe_prod);
 616 }
 617
 618 /* Starts a given RX queue in HW */
 619 static int
 620 qede_rx_queue_start(struct rte_eth_dev *eth_dev, uint16_t rx_queue_id)
 621 {
 622         struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
 623         struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
 624         struct ecore_queue_start_common_params params;
 625         struct ecore_rxq_start_ret_params ret_params;
 626         struct qede_rx_queue *rxq;
 627         struct qede_fastpath *fp;
 628         struct ecore_hwfn *p_hwfn;
 629         dma_addr_t p_phys_table;
 630         uint16_t page_cnt;
 631         uint16_t j;
 632         int hwfn_index;
 633         int rc;
 634
 635         if (rx_queue_id < eth_dev->data->nb_rx_queues) {
 636                 fp = &qdev->fp_array[rx_queue_id];
 637                 rxq = eth_dev->data->rx_queues[rx_queue_id];
 638                 /* Allocate buffers for the Rx ring */
 639                 for (j = 0; j < rxq->nb_rx_desc; j++) {
 640                         rc = qede_alloc_rx_buffer(rxq);
 641                         if (rc) {
 642                                 DP_ERR(edev, "RX buffer allocation failed"
 643                                                 " for rxq = %u\n", rx_queue_id);
 644                                 return -ENOMEM;
 645                         }
 646                 }
 647                 /* disable interrupts */
 648                 ecore_sb_ack(fp->sb_info, IGU_INT_DISABLE, 0);
 649                 /* Prepare ramrod */
 650                 memset(&params, 0, sizeof(params));
 651                 params.queue_id = rx_queue_id / edev->num_hwfns;
 652                 params.vport_id = 0;
 653                 params.stats_id = params.vport_id;
 654                 params.p_sb = fp->sb_info;
 655                 DP_INFO(edev, "rxq %u igu_sb_id 0x%x\n",
 656                                 fp->rxq->queue_id, fp->sb_info->igu_sb_id);
 657                 params.sb_idx = RX_PI;
 658                 hwfn_index = rx_queue_id % edev->num_hwfns;
 659                 p_hwfn = &edev->hwfns[hwfn_index];
 660                 p_phys_table = ecore_chain_get_pbl_phys(&fp->rxq->rx_comp_ring);
 661                 page_cnt = ecore_chain_get_page_cnt(&fp->rxq->rx_comp_ring);
 662                 memset(&ret_params, 0, sizeof(ret_params));
 663                 rc = ecore_eth_rx_queue_start(p_hwfn,
 664                                 p_hwfn->hw_info.opaque_fid,
 665                                 &params, fp->rxq->rx_buf_size,
 666                                 fp->rxq->rx_bd_ring.p_phys_addr,
 667                                 p_phys_table, page_cnt,
 668                                 &ret_params);
 669                 if (rc) {
 670                         DP_ERR(edev, "RX queue %u could not be started, rc = %d\n",
 671                                         rx_queue_id, rc);
 672                         return -1;
 673                 }
 674                 /* Update with the returned parameters */
 675                 fp->rxq->hw_rxq_prod_addr = ret_params.p_prod;
 676                 fp->rxq->handle = ret_params.p_handle;
 677
 678                 fp->rxq->hw_cons_ptr = &fp->sb_info->sb_virt->pi_array[RX_PI];
 679                 qede_update_rx_prod(qdev, fp->rxq);
 680                 eth_dev->data->rx_queue_state[rx_queue_id] =
 681                         RTE_ETH_QUEUE_STATE_STARTED;
 682                 DP_INFO(edev, "RX queue %u started\n", rx_queue_id);
 683         } else {
 684                 DP_ERR(edev, "RX queue %u is not in range\n", rx_queue_id);
 685                 rc = -EINVAL;
 686         }
 687
 688         return rc;
 689 }
 690
 691 static int
 692 qede_tx_queue_start(struct rte_eth_dev *eth_dev, uint16_t tx_queue_id)
 693 {
 694         struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
 695         struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
 696         struct ecore_queue_start_common_params params;
 697         struct ecore_txq_start_ret_params ret_params;
 698         struct ecore_hwfn *p_hwfn;
 699         dma_addr_t p_phys_table;
 700         struct qede_tx_queue *txq;
 701         struct qede_fastpath *fp;
 702         uint16_t page_cnt;
 703         int hwfn_index;
 704         int rc;
 705
 706         if (tx_queue_id < eth_dev->data->nb_tx_queues) {
 707                 txq = eth_dev->data->tx_queues[tx_queue_id];
 708                 fp = &qdev->fp_array[tx_queue_id];
 709                 memset(&params, 0, sizeof(params));
 710                 params.queue_id = tx_queue_id / edev->num_hwfns;
 711                 params.vport_id = 0;
 712                 params.stats_id = params.vport_id;
 713                 params.p_sb = fp->sb_info;
 714                 DP_INFO(edev, "txq %u igu_sb_id 0x%x\n",
 715                                 fp->txq->queue_id, fp->sb_info->igu_sb_id);
 716                 params.sb_idx = TX_PI(0); /* tc = 0 */
 717                 p_phys_table = ecore_chain_get_pbl_phys(&txq->tx_pbl);
 718                 page_cnt = ecore_chain_get_page_cnt(&txq->tx_pbl);
 719                 hwfn_index = tx_queue_id % edev->num_hwfns;
 720                 p_hwfn = &edev->hwfns[hwfn_index];
 721                 if (qdev->dev_info.is_legacy)
 722                         fp->txq->is_legacy = true;
 723                 rc = ecore_eth_tx_queue_start(p_hwfn,
 724                                 p_hwfn->hw_info.opaque_fid,
 725                                 &params, 0 /* tc */,
 726                                 p_phys_table, page_cnt,
 727                                 &ret_params);
 728                 if (rc != ECORE_SUCCESS) {
 729                         DP_ERR(edev, "TX queue %u couldn't be started, rc=%d\n",
 730                                         tx_queue_id, rc);
 731                         return -1;
 732                 }
 733                 txq->doorbell_addr = ret_params.p_doorbell;
 734                 txq->handle = ret_params.p_handle;
 735
 736                 txq->hw_cons_ptr = &fp->sb_info->sb_virt->pi_array[TX_PI(0)];
 737                 SET_FIELD(txq->tx_db.data.params, ETH_DB_DATA_DEST,
 738                                 DB_DEST_XCM);
 739                 SET_FIELD(txq->tx_db.data.params, ETH_DB_DATA_AGG_CMD,
 740                                 DB_AGG_CMD_SET);
 741                 SET_FIELD(txq->tx_db.data.params,
 742                                 ETH_DB_DATA_AGG_VAL_SEL,
 743                                 DQ_XCM_ETH_TX_BD_PROD_CMD);
 744                 txq->tx_db.data.agg_flags = DQ_XCM_ETH_DQ_CF_CMD;
 745                 eth_dev->data->tx_queue_state[tx_queue_id] =
 746                         RTE_ETH_QUEUE_STATE_STARTED;
 747                 DP_INFO(edev, "TX queue %u started\n", tx_queue_id);
 748         } else {
 749                 DP_ERR(edev, "TX queue %u is not in range\n", tx_queue_id);
 750                 rc = -EINVAL;
 751         }
 752
 753         return rc;
 754 }
 755
 756 static inline void
 757 qede_free_tx_pkt(struct qede_tx_queue *txq)
 758 {
 759         struct rte_mbuf *mbuf;
 760         uint16_t nb_segs;
 761         uint16_t idx;
 762
 763         idx = TX_CONS(txq);
 764         mbuf = txq->sw_tx_ring[idx].mbuf;
 765         if (mbuf) {
 766                 nb_segs = mbuf->nb_segs;
 767                 PMD_TX_LOG(DEBUG, txq, "nb_segs to free %u\n", nb_segs);
 768                 while (nb_segs) {
 769                         /* It's like consuming rxbuf in recv() */
 770                         ecore_chain_consume(&txq->tx_pbl);
 771                         txq->nb_tx_avail++;
 772                         nb_segs--;
 773                 }
 774                 rte_pktmbuf_free(mbuf);
 775                 txq->sw_tx_ring[idx].mbuf = NULL;
 776                 txq->sw_tx_cons++;
 777                 PMD_TX_LOG(DEBUG, txq, "Freed tx packet\n");
 778         } else {
 779                 ecore_chain_consume(&txq->tx_pbl);
 780                 txq->nb_tx_avail++;
 781         }
 782 }
 783
 784 static inline void
 785 qede_process_tx_compl(__rte_unused struct ecore_dev *edev,
 786                       struct qede_tx_queue *txq)
 787 {
 788         uint16_t hw_bd_cons;
 789 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
 790         uint16_t sw_tx_cons;
 791 #endif
 792
 793         rte_compiler_barrier();
 794         hw_bd_cons = rte_le_to_cpu_16(*txq->hw_cons_ptr);
 795 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
 796         sw_tx_cons = ecore_chain_get_cons_idx(&txq->tx_pbl);
 797         PMD_TX_LOG(DEBUG, txq, "Tx Completions = %u\n",
 798                    abs(hw_bd_cons - sw_tx_cons));
 799 #endif
 800         while (hw_bd_cons !=  ecore_chain_get_cons_idx(&txq->tx_pbl))
 801                 qede_free_tx_pkt(txq);
 802 }
 803
 804 static int qede_drain_txq(struct qede_dev *qdev,
 805                           struct qede_tx_queue *txq, bool allow_drain)
 806 {
 807         struct ecore_dev *edev = &qdev->edev;
 808         int rc, cnt = 1000;
 809
 810         while (txq->sw_tx_cons != txq->sw_tx_prod) {
 811                 qede_process_tx_compl(edev, txq);
 812                 if (!cnt) {
 813                         if (allow_drain) {
 814                                 DP_ERR(edev, "Tx queue[%u] is stuck,"
 815                                           "requesting MCP to drain\n",
 816                                           txq->queue_id);
 817                                 rc = qdev->ops->common->drain(edev);
 818                                 if (rc)
 819                                         return rc;
 820                                 return qede_drain_txq(qdev, txq, false);
 821                         }
 822                         DP_ERR(edev, "Timeout waiting for tx queue[%d]:"
 823                                   "PROD=%d, CONS=%d\n",
 824                                   txq->queue_id, txq->sw_tx_prod,
 825                                   txq->sw_tx_cons);
 826                         return -1;
 827                 }
 828                 cnt--;
 829                 DELAY(1000);
 830                 rte_compiler_barrier();
 831         }
 832
 833         /* FW finished processing, wait for HW to transmit all tx packets */
 834         DELAY(2000);
 835
 836         return 0;
 837 }
 838
 839 /* Stops a given TX queue in the HW */
 840 static int qede_tx_queue_stop(struct rte_eth_dev *eth_dev, uint16_t tx_queue_id)
 841 {
 842         struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
 843         struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
 844         struct ecore_hwfn *p_hwfn;
 845         struct qede_tx_queue *txq;
 846         int hwfn_index;
 847         int rc;
 848
 849         if (tx_queue_id < eth_dev->data->nb_tx_queues) {
 850                 txq = eth_dev->data->tx_queues[tx_queue_id];
 851                 /* Drain txq */
 852                 if (qede_drain_txq(qdev, txq, true))
 853                         return -1; /* For the lack of retcodes */
 854                 /* Stop txq */
 855                 hwfn_index = tx_queue_id % edev->num_hwfns;
 856                 p_hwfn = &edev->hwfns[hwfn_index];
 857                 rc = ecore_eth_tx_queue_stop(p_hwfn, txq->handle);
 858                 if (rc != ECORE_SUCCESS) {
 859                         DP_ERR(edev, "TX queue %u stop fails\n", tx_queue_id);
 860                         return -1;
 861                 }
 862                 qede_tx_queue_release_mbufs(txq);
 863                 qede_tx_queue_reset(qdev, txq);
 864                 eth_dev->data->tx_queue_state[tx_queue_id] =
 865                         RTE_ETH_QUEUE_STATE_STOPPED;
 866                 DP_INFO(edev, "TX queue %u stopped\n", tx_queue_id);
 867         } else {
 868                 DP_ERR(edev, "TX queue %u is not in range\n", tx_queue_id);
 869                 rc = -EINVAL;
 870         }
 871
 872         return rc;
 873 }
 874
 875 int qede_start_queues(struct rte_eth_dev *eth_dev)
 876 {
 877         struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
 878         uint8_t id;
 879         int rc = -1;
 880
 881         for_each_rss(id) {
 882                 rc = qede_rx_queue_start(eth_dev, id);
 883                 if (rc != ECORE_SUCCESS)
 884                         return -1;
 885         }
 886
 887         for_each_tss(id) {
 888                 rc = qede_tx_queue_start(eth_dev, id);
 889                 if (rc != ECORE_SUCCESS)
 890                         return -1;
 891         }
 892
 893         return rc;
 894 }
 895
 896 void qede_stop_queues(struct rte_eth_dev *eth_dev)
 897 {
 898         struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
 899         uint8_t id;
 900
 901         /* Stopping RX/TX queues */
 902         for_each_tss(id) {
 903                 qede_tx_queue_stop(eth_dev, id);
 904         }
 905
 906         for_each_rss(id) {
 907                 qede_rx_queue_stop(eth_dev, id);
 908         }
 909 }
 910
 911 static inline bool qede_tunn_exist(uint16_t flag)
 912 {
 913         return !!((PARSING_AND_ERR_FLAGS_TUNNELEXIST_MASK <<
 914                     PARSING_AND_ERR_FLAGS_TUNNELEXIST_SHIFT) & flag);
 915 }
 916
 917 static inline uint8_t qede_check_tunn_csum_l3(uint16_t flag)
 918 {
 919         return !!((PARSING_AND_ERR_FLAGS_TUNNELIPHDRERROR_MASK <<
 920                 PARSING_AND_ERR_FLAGS_TUNNELIPHDRERROR_SHIFT) & flag);
 921 }
 922
 923 /*
 924  * qede_check_tunn_csum_l4:
 925  * Returns:
 926  * 1 : If L4 csum is enabled AND if the validation has failed.
 927  * 0 : Otherwise
 928  */
 929 static inline uint8_t qede_check_tunn_csum_l4(uint16_t flag)
 930 {
 931         if ((PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMWASCALCULATED_MASK <<
 932              PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMWASCALCULATED_SHIFT) & flag)
 933                 return !!((PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMERROR_MASK <<
 934                         PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMERROR_SHIFT) & flag);
 935
 936         return 0;
 937 }
 938
 939 static inline uint8_t qede_check_notunn_csum_l4(uint16_t flag)
 940 {
 941         if ((PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_MASK <<
 942              PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_SHIFT) & flag)
 943                 return !!((PARSING_AND_ERR_FLAGS_L4CHKSMERROR_MASK <<
 944                            PARSING_AND_ERR_FLAGS_L4CHKSMERROR_SHIFT) & flag);
 945
 946         return 0;
 947 }
 948
 949 /* Returns outer L2, L3 and L4 packet_type for tunneled packets */
 950 static inline uint32_t qede_rx_cqe_to_pkt_type_outer(struct rte_mbuf *m)
 951 {
 952         uint32_t packet_type = RTE_PTYPE_UNKNOWN;
 953         struct ether_hdr *eth_hdr;
 954         struct ipv4_hdr *ipv4_hdr;
 955         struct ipv6_hdr *ipv6_hdr;
 956         struct vlan_hdr *vlan_hdr;
 957         uint16_t ethertype;
 958         bool vlan_tagged = 0;
 959         uint16_t len;
 960
 961         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 962         len = sizeof(struct ether_hdr);
 963         ethertype = rte_cpu_to_be_16(eth_hdr->ether_type);
 964
 965          /* Note: Valid only if VLAN stripping is disabled */
 966         if (ethertype == ETHER_TYPE_VLAN) {
 967                 vlan_tagged = 1;
 968                 vlan_hdr = (struct vlan_hdr *)(eth_hdr + 1);
 969                 len += sizeof(struct vlan_hdr);
 970                 ethertype = rte_cpu_to_be_16(vlan_hdr->eth_proto);
 971         }
 972
 973         if (ethertype == ETHER_TYPE_IPv4) {
 974                 packet_type |= RTE_PTYPE_L3_IPV4;
 975                 ipv4_hdr = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *, len);
 976                 if (ipv4_hdr->next_proto_id == IPPROTO_TCP)
 977                         packet_type |= RTE_PTYPE_L4_TCP;
 978                 else if (ipv4_hdr->next_proto_id == IPPROTO_UDP)
 979                         packet_type |= RTE_PTYPE_L4_UDP;
 980         } else if (ethertype == ETHER_TYPE_IPv6) {
 981                 packet_type |= RTE_PTYPE_L3_IPV6;
 982                 ipv6_hdr = rte_pktmbuf_mtod_offset(m, struct ipv6_hdr *, len);
 983                 if (ipv6_hdr->proto == IPPROTO_TCP)
 984                         packet_type |= RTE_PTYPE_L4_TCP;
 985                 else if (ipv6_hdr->proto == IPPROTO_UDP)
 986                         packet_type |= RTE_PTYPE_L4_UDP;
 987         }
 988
 989         if (vlan_tagged)
 990                 packet_type |= RTE_PTYPE_L2_ETHER_VLAN;
 991         else
 992                 packet_type |= RTE_PTYPE_L2_ETHER;
 993
 994         return packet_type;
 995 }
 996
 997 static inline uint32_t qede_rx_cqe_to_pkt_type_inner(uint16_t flags)
 998 {
 999         uint16_t val;
1000
1001         /* Lookup table */
1002         static const uint32_t
1003         ptype_lkup_tbl[QEDE_PKT_TYPE_MAX] __rte_cache_aligned = {
1004                 [QEDE_PKT_TYPE_IPV4] = RTE_PTYPE_INNER_L3_IPV4          |
1005                                        RTE_PTYPE_INNER_L2_ETHER,
1006                 [QEDE_PKT_TYPE_IPV6] = RTE_PTYPE_INNER_L3_IPV6          |
1007                                        RTE_PTYPE_INNER_L2_ETHER,
1008                 [QEDE_PKT_TYPE_IPV4_TCP] = RTE_PTYPE_INNER_L3_IPV4      |
1009                                            RTE_PTYPE_INNER_L4_TCP       |
1010                                            RTE_PTYPE_INNER_L2_ETHER,
1011                 [QEDE_PKT_TYPE_IPV6_TCP] = RTE_PTYPE_INNER_L3_IPV6      |
1012                                            RTE_PTYPE_INNER_L4_TCP       |
1013                                            RTE_PTYPE_INNER_L2_ETHER,
1014                 [QEDE_PKT_TYPE_IPV4_UDP] = RTE_PTYPE_INNER_L3_IPV4      |
1015                                            RTE_PTYPE_INNER_L4_UDP       |
1016                                            RTE_PTYPE_INNER_L2_ETHER,
1017                 [QEDE_PKT_TYPE_IPV6_UDP] = RTE_PTYPE_INNER_L3_IPV6      |
1018                                            RTE_PTYPE_INNER_L4_UDP       |
1019                                            RTE_PTYPE_INNER_L2_ETHER,
1020                 /* Frags with no VLAN */
1021                 [QEDE_PKT_TYPE_IPV4_FRAG] = RTE_PTYPE_INNER_L3_IPV4     |
1022                                             RTE_PTYPE_INNER_L4_FRAG     |
1023                                             RTE_PTYPE_INNER_L2_ETHER,
1024                 [QEDE_PKT_TYPE_IPV6_FRAG] = RTE_PTYPE_INNER_L3_IPV6     |
1025                                             RTE_PTYPE_INNER_L4_FRAG     |
1026                                             RTE_PTYPE_INNER_L2_ETHER,
1027                 /* VLANs */
1028                 [QEDE_PKT_TYPE_IPV4_VLAN] = RTE_PTYPE_INNER_L3_IPV4     |
1029                                             RTE_PTYPE_INNER_L2_ETHER_VLAN,
1030                 [QEDE_PKT_TYPE_IPV6_VLAN] = RTE_PTYPE_INNER_L3_IPV6     |
1031                                             RTE_PTYPE_INNER_L2_ETHER_VLAN,
1032                 [QEDE_PKT_TYPE_IPV4_TCP_VLAN] = RTE_PTYPE_INNER_L3_IPV4 |
1033                                                 RTE_PTYPE_INNER_L4_TCP  |
1034                                                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1035                 [QEDE_PKT_TYPE_IPV6_TCP_VLAN] = RTE_PTYPE_INNER_L3_IPV6 |
1036                                                 RTE_PTYPE_INNER_L4_TCP  |
1037                                                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1038                 [QEDE_PKT_TYPE_IPV4_UDP_VLAN] = RTE_PTYPE_INNER_L3_IPV4 |
1039                                                 RTE_PTYPE_INNER_L4_UDP  |
1040                                                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1041                 [QEDE_PKT_TYPE_IPV6_UDP_VLAN] = RTE_PTYPE_INNER_L3_IPV6 |
1042                                                 RTE_PTYPE_INNER_L4_UDP  |
1043                                                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1044                 /* Frags with VLAN */
1045                 [QEDE_PKT_TYPE_IPV4_VLAN_FRAG] = RTE_PTYPE_INNER_L3_IPV4 |
1046                                                  RTE_PTYPE_INNER_L4_FRAG |
1047                                                  RTE_PTYPE_INNER_L2_ETHER_VLAN,
1048                 [QEDE_PKT_TYPE_IPV6_VLAN_FRAG] = RTE_PTYPE_INNER_L3_IPV6 |
1049                                                  RTE_PTYPE_INNER_L4_FRAG |
1050                                                  RTE_PTYPE_INNER_L2_ETHER_VLAN,
1051         };
1052
1053         /* Bits (0..3) provides L3/L4 protocol type */
1054         /* Bits (4,5) provides frag and VLAN info */
1055         val = ((PARSING_AND_ERR_FLAGS_L3TYPE_MASK <<
1056                PARSING_AND_ERR_FLAGS_L3TYPE_SHIFT) |
1057                (PARSING_AND_ERR_FLAGS_L4PROTOCOL_MASK <<
1058                 PARSING_AND_ERR_FLAGS_L4PROTOCOL_SHIFT) |
1059                (PARSING_AND_ERR_FLAGS_IPV4FRAG_MASK <<
1060                 PARSING_AND_ERR_FLAGS_IPV4FRAG_SHIFT) |
1061                 (PARSING_AND_ERR_FLAGS_TAG8021QEXIST_MASK <<
1062                  PARSING_AND_ERR_FLAGS_TAG8021QEXIST_SHIFT)) & flags;
1063
1064         if (val < QEDE_PKT_TYPE_MAX)
1065                 return ptype_lkup_tbl[val];
1066
1067         return RTE_PTYPE_UNKNOWN;
1068 }
1069
1070 static inline uint32_t qede_rx_cqe_to_pkt_type(uint16_t flags)
1071 {
1072         uint16_t val;
1073
1074         /* Lookup table */
1075         static const uint32_t
1076         ptype_lkup_tbl[QEDE_PKT_TYPE_MAX] __rte_cache_aligned = {
1077                 [QEDE_PKT_TYPE_IPV4] = RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L2_ETHER,
1078                 [QEDE_PKT_TYPE_IPV6] = RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L2_ETHER,
1079                 [QEDE_PKT_TYPE_IPV4_TCP] = RTE_PTYPE_L3_IPV4    |
1080                                            RTE_PTYPE_L4_TCP     |
1081                                            RTE_PTYPE_L2_ETHER,
1082                 [QEDE_PKT_TYPE_IPV6_TCP] = RTE_PTYPE_L3_IPV6    |
1083                                            RTE_PTYPE_L4_TCP     |
1084                                            RTE_PTYPE_L2_ETHER,
1085                 [QEDE_PKT_TYPE_IPV4_UDP] = RTE_PTYPE_L3_IPV4    |
1086                                            RTE_PTYPE_L4_UDP     |
1087                                            RTE_PTYPE_L2_ETHER,
1088                 [QEDE_PKT_TYPE_IPV6_UDP] = RTE_PTYPE_L3_IPV6    |
1089                                            RTE_PTYPE_L4_UDP     |
1090                                            RTE_PTYPE_L2_ETHER,
1091                 /* Frags with no VLAN */
1092                 [QEDE_PKT_TYPE_IPV4_FRAG] = RTE_PTYPE_L3_IPV4   |
1093                                             RTE_PTYPE_L4_FRAG   |
1094                                             RTE_PTYPE_L2_ETHER,
1095                 [QEDE_PKT_TYPE_IPV6_FRAG] = RTE_PTYPE_L3_IPV6   |
1096                                             RTE_PTYPE_L4_FRAG   |
1097                                             RTE_PTYPE_L2_ETHER,
1098                 /* VLANs */
1099                 [QEDE_PKT_TYPE_IPV4_VLAN] = RTE_PTYPE_L3_IPV4           |
1100                                             RTE_PTYPE_L2_ETHER_VLAN,
1101                 [QEDE_PKT_TYPE_IPV6_VLAN] = RTE_PTYPE_L3_IPV6           |
1102                                             RTE_PTYPE_L2_ETHER_VLAN,
1103                 [QEDE_PKT_TYPE_IPV4_TCP_VLAN] = RTE_PTYPE_L3_IPV4       |
1104                                                 RTE_PTYPE_L4_TCP        |
1105                                                 RTE_PTYPE_L2_ETHER_VLAN,
1106                 [QEDE_PKT_TYPE_IPV6_TCP_VLAN] = RTE_PTYPE_L3_IPV6       |
1107                                                 RTE_PTYPE_L4_TCP        |
1108                                                 RTE_PTYPE_L2_ETHER_VLAN,
1109                 [QEDE_PKT_TYPE_IPV4_UDP_VLAN] = RTE_PTYPE_L3_IPV4       |
1110                                                 RTE_PTYPE_L4_UDP        |
1111                                                 RTE_PTYPE_L2_ETHER_VLAN,
1112                 [QEDE_PKT_TYPE_IPV6_UDP_VLAN] = RTE_PTYPE_L3_IPV6       |
1113                                                 RTE_PTYPE_L4_UDP        |
1114                                                 RTE_PTYPE_L2_ETHER_VLAN,
1115                 /* Frags with VLAN */
1116                 [QEDE_PKT_TYPE_IPV4_VLAN_FRAG] = RTE_PTYPE_L3_IPV4      |
1117                                                  RTE_PTYPE_L4_FRAG      |
1118                                                  RTE_PTYPE_L2_ETHER_VLAN,
1119                 [QEDE_PKT_TYPE_IPV6_VLAN_FRAG] = RTE_PTYPE_L3_IPV6      |
1120                                                  RTE_PTYPE_L4_FRAG      |
1121                                                  RTE_PTYPE_L2_ETHER_VLAN,
1122         };
1123
1124         /* Bits (0..3) provides L3/L4 protocol type */
1125         /* Bits (4,5) provides frag and VLAN info */
1126         val = ((PARSING_AND_ERR_FLAGS_L3TYPE_MASK <<
1127                PARSING_AND_ERR_FLAGS_L3TYPE_SHIFT) |
1128                (PARSING_AND_ERR_FLAGS_L4PROTOCOL_MASK <<
1129                 PARSING_AND_ERR_FLAGS_L4PROTOCOL_SHIFT) |
1130                (PARSING_AND_ERR_FLAGS_IPV4FRAG_MASK <<
1131                 PARSING_AND_ERR_FLAGS_IPV4FRAG_SHIFT) |
1132                 (PARSING_AND_ERR_FLAGS_TAG8021QEXIST_MASK <<
1133                  PARSING_AND_ERR_FLAGS_TAG8021QEXIST_SHIFT)) & flags;
1134
1135         if (val < QEDE_PKT_TYPE_MAX)
1136                 return ptype_lkup_tbl[val];
1137
1138         return RTE_PTYPE_UNKNOWN;
1139 }
1140
1141 static inline uint8_t
1142 qede_check_notunn_csum_l3(struct rte_mbuf *m, uint16_t flag)
1143 {
1144         struct ipv4_hdr *ip;
1145         uint16_t pkt_csum;
1146         uint16_t calc_csum;
1147         uint16_t val;
1148
1149         val = ((PARSING_AND_ERR_FLAGS_IPHDRERROR_MASK <<
1150                 PARSING_AND_ERR_FLAGS_IPHDRERROR_SHIFT) & flag);
1151
1152         if (unlikely(val)) {
1153                 m->packet_type = qede_rx_cqe_to_pkt_type(flag);
1154                 if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
1155                         ip = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
1156                                            sizeof(struct ether_hdr));
1157                         pkt_csum = ip->hdr_checksum;
1158                         ip->hdr_checksum = 0;
1159                         calc_csum = rte_ipv4_cksum(ip);
1160                         ip->hdr_checksum = pkt_csum;
1161                         return (calc_csum != pkt_csum);
1162                 } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
1163                         return 1;
1164                 }
1165         }
1166         return 0;
1167 }
1168
1169 static inline void qede_rx_bd_ring_consume(struct qede_rx_queue *rxq)
1170 {
1171         ecore_chain_consume(&rxq->rx_bd_ring);
1172         rxq->sw_rx_cons++;
1173 }
1174
1175 static inline void
1176 qede_reuse_page(__rte_unused struct qede_dev *qdev,
1177                 struct qede_rx_queue *rxq, struct qede_rx_entry *curr_cons)
1178 {
1179         struct eth_rx_bd *rx_bd_prod = ecore_chain_produce(&rxq->rx_bd_ring);
1180         uint16_t idx = rxq->sw_rx_prod & NUM_RX_BDS(rxq);
1181         struct qede_rx_entry *curr_prod;
1182         dma_addr_t new_mapping;
1183
1184         curr_prod = &rxq->sw_rx_ring[idx];
1185         *curr_prod = *curr_cons;
1186
1187         new_mapping = rte_mbuf_data_iova_default(curr_prod->mbuf) +
1188                       curr_prod->page_offset;
1189
1190         rx_bd_prod->addr.hi = rte_cpu_to_le_32(U64_HI(new_mapping));
1191         rx_bd_prod->addr.lo = rte_cpu_to_le_32(U64_LO(new_mapping));
1192
1193         rxq->sw_rx_prod++;
1194 }
1195
1196 static inline void
1197 qede_recycle_rx_bd_ring(struct qede_rx_queue *rxq,
1198                         struct qede_dev *qdev, uint8_t count)
1199 {
1200         struct qede_rx_entry *curr_cons;
1201
1202         for (; count > 0; count--) {
1203                 curr_cons = &rxq->sw_rx_ring[rxq->sw_rx_cons & NUM_RX_BDS(rxq)];
1204                 qede_reuse_page(qdev, rxq, curr_cons);
1205                 qede_rx_bd_ring_consume(rxq);
1206         }
1207 }
1208
1209 static inline void
1210 qede_rx_process_tpa_cmn_cont_end_cqe(__rte_unused struct qede_dev *qdev,
1211                                      struct qede_rx_queue *rxq,
1212                                      uint8_t agg_index, uint16_t len)
1213 {
1214         struct qede_agg_info *tpa_info;
1215         struct rte_mbuf *curr_frag; /* Pointer to currently filled TPA seg */
1216         uint16_t cons_idx;
1217
1218         /* Under certain conditions it is possible that FW may not consume
1219          * additional or new BD. So decision to consume the BD must be made
1220          * based on len_list[0].
1221          */
1222         if (rte_le_to_cpu_16(len)) {
1223                 tpa_info = &rxq->tpa_info[agg_index];
1224                 cons_idx = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
1225                 curr_frag = rxq->sw_rx_ring[cons_idx].mbuf;
1226                 assert(curr_frag);
1227                 curr_frag->nb_segs = 1;
1228                 curr_frag->pkt_len = rte_le_to_cpu_16(len);
1229                 curr_frag->data_len = curr_frag->pkt_len;
1230                 tpa_info->tpa_tail->next = curr_frag;
1231                 tpa_info->tpa_tail = curr_frag;
1232                 qede_rx_bd_ring_consume(rxq);
1233                 if (unlikely(qede_alloc_rx_buffer(rxq) != 0)) {
1234                         PMD_RX_LOG(ERR, rxq, "mbuf allocation fails\n");
1235                         rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed++;
1236                         rxq->rx_alloc_errors++;
1237                 }
1238         }
1239 }
1240
1241 static inline void
1242 qede_rx_process_tpa_cont_cqe(struct qede_dev *qdev,
1243                              struct qede_rx_queue *rxq,
1244                              struct eth_fast_path_rx_tpa_cont_cqe *cqe)
1245 {
1246         PMD_RX_LOG(INFO, rxq, "TPA cont[%d] - len [%d]\n",
1247                    cqe->tpa_agg_index, rte_le_to_cpu_16(cqe->len_list[0]));
1248         /* only len_list[0] will have value */
1249         qede_rx_process_tpa_cmn_cont_end_cqe(qdev, rxq, cqe->tpa_agg_index,
1250                                              cqe->len_list[0]);
1251 }
1252
1253 static inline void
1254 qede_rx_process_tpa_end_cqe(struct qede_dev *qdev,
1255                             struct qede_rx_queue *rxq,
1256                             struct eth_fast_path_rx_tpa_end_cqe *cqe)
1257 {
1258         struct rte_mbuf *rx_mb; /* Pointer to head of the chained agg */
1259
1260         qede_rx_process_tpa_cmn_cont_end_cqe(qdev, rxq, cqe->tpa_agg_index,
1261                                              cqe->len_list[0]);
1262         /* Update total length and frags based on end TPA */
1263         rx_mb = rxq->tpa_info[cqe->tpa_agg_index].tpa_head;
1264         /* TODO:  Add Sanity Checks */
1265         rx_mb->nb_segs = cqe->num_of_bds;
1266         rx_mb->pkt_len = cqe->total_packet_len;
1267
1268         PMD_RX_LOG(INFO, rxq, "TPA End[%d] reason %d cqe_len %d nb_segs %d"
1269                    " pkt_len %d\n", cqe->tpa_agg_index, cqe->end_reason,
1270                    rte_le_to_cpu_16(cqe->len_list[0]), rx_mb->nb_segs,
1271                    rx_mb->pkt_len);
1272 }
1273
1274 static inline uint32_t qede_rx_cqe_to_tunn_pkt_type(uint16_t flags)
1275 {
1276         uint32_t val;
1277
1278         /* Lookup table */
1279         static const uint32_t
1280         ptype_tunn_lkup_tbl[QEDE_PKT_TYPE_TUNN_MAX_TYPE] __rte_cache_aligned = {
1281                 [QEDE_PKT_TYPE_UNKNOWN] = RTE_PTYPE_UNKNOWN,
1282                 [QEDE_PKT_TYPE_TUNN_GENEVE] = RTE_PTYPE_TUNNEL_GENEVE,
1283                 [QEDE_PKT_TYPE_TUNN_GRE] = RTE_PTYPE_TUNNEL_GRE,
1284                 [QEDE_PKT_TYPE_TUNN_VXLAN] = RTE_PTYPE_TUNNEL_VXLAN,
1285                 [QEDE_PKT_TYPE_TUNN_L2_TENID_NOEXIST_GENEVE] =
1286                                 RTE_PTYPE_TUNNEL_GENEVE,
1287                 [QEDE_PKT_TYPE_TUNN_L2_TENID_NOEXIST_GRE] =
1288                                 RTE_PTYPE_TUNNEL_GRE,
1289                 [QEDE_PKT_TYPE_TUNN_L2_TENID_NOEXIST_VXLAN] =
1290                                 RTE_PTYPE_TUNNEL_VXLAN,
1291                 [QEDE_PKT_TYPE_TUNN_L2_TENID_EXIST_GENEVE] =
1292                                 RTE_PTYPE_TUNNEL_GENEVE,
1293                 [QEDE_PKT_TYPE_TUNN_L2_TENID_EXIST_GRE] =
1294                                 RTE_PTYPE_TUNNEL_GRE,
1295                 [QEDE_PKT_TYPE_TUNN_L2_TENID_EXIST_VXLAN] =
1296                                 RTE_PTYPE_TUNNEL_VXLAN,
1297                 [QEDE_PKT_TYPE_TUNN_IPV4_TENID_NOEXIST_GENEVE] =
1298                                 RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV4,
1299                 [QEDE_PKT_TYPE_TUNN_IPV4_TENID_NOEXIST_GRE] =
1300                                 RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV4,
1301                 [QEDE_PKT_TYPE_TUNN_IPV4_TENID_NOEXIST_VXLAN] =
1302                                 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV4,
1303                 [QEDE_PKT_TYPE_TUNN_IPV4_TENID_EXIST_GENEVE] =
1304                                 RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV4,
1305                 [QEDE_PKT_TYPE_TUNN_IPV4_TENID_EXIST_GRE] =
1306                                 RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV4,
1307                 [QEDE_PKT_TYPE_TUNN_IPV4_TENID_EXIST_VXLAN] =
1308                                 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV4,
1309                 [QEDE_PKT_TYPE_TUNN_IPV6_TENID_NOEXIST_GENEVE] =
1310                                 RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV6,
1311                 [QEDE_PKT_TYPE_TUNN_IPV6_TENID_NOEXIST_GRE] =
1312                                 RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV6,
1313                 [QEDE_PKT_TYPE_TUNN_IPV6_TENID_NOEXIST_VXLAN] =
1314                                 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV6,
1315                 [QEDE_PKT_TYPE_TUNN_IPV6_TENID_EXIST_GENEVE] =
1316                                 RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV6,
1317                 [QEDE_PKT_TYPE_TUNN_IPV6_TENID_EXIST_GRE] =
1318                                 RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV6,
1319                 [QEDE_PKT_TYPE_TUNN_IPV6_TENID_EXIST_VXLAN] =
1320                                 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV6,
1321         };
1322
1323         /* Cover bits[4-0] to include tunn_type and next protocol */
1324         val = ((ETH_TUNNEL_PARSING_FLAGS_TYPE_MASK <<
1325                 ETH_TUNNEL_PARSING_FLAGS_TYPE_SHIFT) |
1326                 (ETH_TUNNEL_PARSING_FLAGS_NEXT_PROTOCOL_MASK <<
1327                 ETH_TUNNEL_PARSING_FLAGS_NEXT_PROTOCOL_SHIFT)) & flags;
1328
1329         if (val < QEDE_PKT_TYPE_TUNN_MAX_TYPE)
1330                 return ptype_tunn_lkup_tbl[val];
1331         else
1332                 return RTE_PTYPE_UNKNOWN;
1333 }
1334
1335 static inline int
1336 qede_process_sg_pkts(void *p_rxq,  struct rte_mbuf *rx_mb,
1337                      uint8_t num_segs, uint16_t pkt_len)
1338 {
1339         struct qede_rx_queue *rxq = p_rxq;
1340         struct qede_dev *qdev = rxq->qdev;
1341         register struct rte_mbuf *seg1 = NULL;
1342         register struct rte_mbuf *seg2 = NULL;
1343         uint16_t sw_rx_index;
1344         uint16_t cur_size;
1345
1346         seg1 = rx_mb;
1347         while (num_segs) {
1348                 cur_size = pkt_len > rxq->rx_buf_size ? rxq->rx_buf_size :
1349                                                         pkt_len;
1350                 if (unlikely(!cur_size)) {
1351                         PMD_RX_LOG(ERR, rxq, "Length is 0 while %u BDs"
1352                                    " left for mapping jumbo\n", num_segs);
1353                         qede_recycle_rx_bd_ring(rxq, qdev, num_segs);
1354                         return -EINVAL;
1355                 }
1356                 sw_rx_index = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
1357                 seg2 = rxq->sw_rx_ring[sw_rx_index].mbuf;
1358                 qede_rx_bd_ring_consume(rxq);
1359                 pkt_len -= cur_size;
1360                 seg2->data_len = cur_size;
1361                 seg1->next = seg2;
1362                 seg1 = seg1->next;
1363                 num_segs--;
1364                 rxq->rx_segs++;
1365         }
1366
1367         return 0;
1368 }
1369
1370 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1371 static inline void
1372 print_rx_bd_info(struct rte_mbuf *m, struct qede_rx_queue *rxq,
1373                  uint8_t bitfield)
1374 {
1375         PMD_RX_LOG(INFO, rxq,
1376                 "len 0x%04x bf 0x%04x hash_val 0x%x"
1377                 " ol_flags 0x%04lx l2=%s l3=%s l4=%s tunn=%s"
1378                 " inner_l2=%s inner_l3=%s inner_l4=%s\n",
1379                 m->data_len, bitfield, m->hash.rss,
1380                 (unsigned long)m->ol_flags,
1381                 rte_get_ptype_l2_name(m->packet_type),
1382                 rte_get_ptype_l3_name(m->packet_type),
1383                 rte_get_ptype_l4_name(m->packet_type),
1384                 rte_get_ptype_tunnel_name(m->packet_type),
1385                 rte_get_ptype_inner_l2_name(m->packet_type),
1386                 rte_get_ptype_inner_l3_name(m->packet_type),
1387                 rte_get_ptype_inner_l4_name(m->packet_type));
1388 }
1389 #endif
1390
1391 uint16_t
1392 qede_recv_pkts(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
1393 {
1394         struct qede_rx_queue *rxq = p_rxq;
1395         struct qede_dev *qdev = rxq->qdev;
1396         struct ecore_dev *edev = &qdev->edev;
1397         uint16_t hw_comp_cons, sw_comp_cons, sw_rx_index;
1398         uint16_t rx_pkt = 0;
1399         union eth_rx_cqe *cqe;
1400         struct eth_fast_path_rx_reg_cqe *fp_cqe = NULL;
1401         register struct rte_mbuf *rx_mb = NULL;
1402         register struct rte_mbuf *seg1 = NULL;
1403         enum eth_rx_cqe_type cqe_type;
1404         uint16_t pkt_len = 0; /* Sum of all BD segments */
1405         uint16_t len; /* Length of first BD */
1406         uint8_t num_segs = 1;
1407         uint16_t preload_idx;
1408         uint16_t parse_flag;
1409 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1410         uint8_t bitfield_val;
1411 #endif
1412         uint8_t tunn_parse_flag;
1413         struct eth_fast_path_rx_tpa_start_cqe *cqe_start_tpa;
1414         uint64_t ol_flags;
1415         uint32_t packet_type;
1416         uint16_t vlan_tci;
1417         bool tpa_start_flg;
1418         uint8_t offset, tpa_agg_idx, flags;
1419         struct qede_agg_info *tpa_info = NULL;
1420         uint32_t rss_hash;
1421         int rx_alloc_count = 0;
1422
1423         hw_comp_cons = rte_le_to_cpu_16(*rxq->hw_cons_ptr);
1424         sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
1425
1426         rte_rmb();
1427
1428         if (hw_comp_cons == sw_comp_cons)
1429                 return 0;
1430
1431         /* Allocate buffers that we used in previous loop */
1432         if (rxq->rx_alloc_count) {
1433                 if (unlikely(qede_alloc_rx_bulk_mbufs(rxq,
1434                              rxq->rx_alloc_count))) {
1435                         struct rte_eth_dev *dev;
1436
1437                         PMD_RX_LOG(ERR, rxq,
1438                                    "New buffer allocation failed,"
1439                                    "dropping incoming packetn");
1440                         dev = &rte_eth_devices[rxq->port_id];
1441                         dev->data->rx_mbuf_alloc_failed +=
1442                                                         rxq->rx_alloc_count;
1443                         rxq->rx_alloc_errors += rxq->rx_alloc_count;
1444                         return 0;
1445                 }
1446                 qede_update_rx_prod(qdev, rxq);
1447                 rxq->rx_alloc_count = 0;
1448         }
1449
1450         while (sw_comp_cons != hw_comp_cons) {
1451                 ol_flags = 0;
1452                 packet_type = RTE_PTYPE_UNKNOWN;
1453                 vlan_tci = 0;
1454                 tpa_start_flg = false;
1455                 rss_hash = 0;
1456
1457                 /* Get the CQE from the completion ring */
1458                 cqe =
1459                     (union eth_rx_cqe *)ecore_chain_consume(&rxq->rx_comp_ring);
1460                 cqe_type = cqe->fast_path_regular.type;
1461                 PMD_RX_LOG(INFO, rxq, "Rx CQE type %d\n", cqe_type);
1462
1463                 switch (cqe_type) {
1464                 case ETH_RX_CQE_TYPE_REGULAR:
1465                         fp_cqe = &cqe->fast_path_regular;
1466                 break;
1467                 case ETH_RX_CQE_TYPE_TPA_START:
1468                         cqe_start_tpa = &cqe->fast_path_tpa_start;
1469                         tpa_info = &rxq->tpa_info[cqe_start_tpa->tpa_agg_index];
1470                         tpa_start_flg = true;
1471                         /* Mark it as LRO packet */
1472                         ol_flags |= PKT_RX_LRO;
1473                         /* In split mode,  seg_len is same as len_on_first_bd
1474                          * and ext_bd_len_list will be empty since there are
1475                          * no additional buffers
1476                          */
1477                         PMD_RX_LOG(INFO, rxq,
1478                             "TPA start[%d] - len_on_first_bd %d header %d"
1479                             " [bd_list[0] %d], [seg_len %d]\n",
1480                             cqe_start_tpa->tpa_agg_index,
1481                             rte_le_to_cpu_16(cqe_start_tpa->len_on_first_bd),
1482                             cqe_start_tpa->header_len,
1483                             rte_le_to_cpu_16(cqe_start_tpa->ext_bd_len_list[0]),
1484                             rte_le_to_cpu_16(cqe_start_tpa->seg_len));
1485
1486                 break;
1487                 case ETH_RX_CQE_TYPE_TPA_CONT:
1488                         qede_rx_process_tpa_cont_cqe(qdev, rxq,
1489                                                      &cqe->fast_path_tpa_cont);
1490                         goto next_cqe;
1491                 case ETH_RX_CQE_TYPE_TPA_END:
1492                         qede_rx_process_tpa_end_cqe(qdev, rxq,
1493                                                     &cqe->fast_path_tpa_end);
1494                         tpa_agg_idx = cqe->fast_path_tpa_end.tpa_agg_index;
1495                         tpa_info = &rxq->tpa_info[tpa_agg_idx];
1496                         rx_mb = rxq->tpa_info[tpa_agg_idx].tpa_head;
1497                         goto tpa_end;
1498                 case ETH_RX_CQE_TYPE_SLOW_PATH:
1499                         PMD_RX_LOG(INFO, rxq, "Got unexpected slowpath CQE\n");
1500                         ecore_eth_cqe_completion(
1501                                 &edev->hwfns[rxq->queue_id % edev->num_hwfns],
1502                                 (struct eth_slow_path_rx_cqe *)cqe);
1503                         /* fall-thru */
1504                 default:
1505                         goto next_cqe;
1506                 }
1507
1508                 /* Get the data from the SW ring */
1509                 sw_rx_index = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
1510                 rx_mb = rxq->sw_rx_ring[sw_rx_index].mbuf;
1511                 assert(rx_mb != NULL);
1512
1513                 /* Handle regular CQE or TPA start CQE */
1514                 if (!tpa_start_flg) {
1515                         parse_flag = rte_le_to_cpu_16(fp_cqe->pars_flags.flags);
1516                         offset = fp_cqe->placement_offset;
1517                         len = rte_le_to_cpu_16(fp_cqe->len_on_first_bd);
1518                         pkt_len = rte_le_to_cpu_16(fp_cqe->pkt_len);
1519                         vlan_tci = rte_le_to_cpu_16(fp_cqe->vlan_tag);
1520                         rss_hash = rte_le_to_cpu_32(fp_cqe->rss_hash);
1521 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1522                         bitfield_val = fp_cqe->bitfields;
1523 #endif
1524                 } else {
1525                         parse_flag =
1526                             rte_le_to_cpu_16(cqe_start_tpa->pars_flags.flags);
1527                         offset = cqe_start_tpa->placement_offset;
1528                         /* seg_len = len_on_first_bd */
1529                         len = rte_le_to_cpu_16(cqe_start_tpa->len_on_first_bd);
1530                         vlan_tci = rte_le_to_cpu_16(cqe_start_tpa->vlan_tag);
1531 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1532                         bitfield_val = cqe_start_tpa->bitfields;
1533 #endif
1534                         rss_hash = rte_le_to_cpu_32(cqe_start_tpa->rss_hash);
1535                 }
1536                 if (qede_tunn_exist(parse_flag)) {
1537                         PMD_RX_LOG(INFO, rxq, "Rx tunneled packet\n");
1538                         if (unlikely(qede_check_tunn_csum_l4(parse_flag))) {
1539                                 PMD_RX_LOG(ERR, rxq,
1540                                             "L4 csum failed, flags = 0x%x\n",
1541                                             parse_flag);
1542                                 rxq->rx_hw_errors++;
1543                                 ol_flags |= PKT_RX_L4_CKSUM_BAD;
1544                         } else {
1545                                 ol_flags |= PKT_RX_L4_CKSUM_GOOD;
1546                         }
1547
1548                         if (unlikely(qede_check_tunn_csum_l3(parse_flag))) {
1549                                 PMD_RX_LOG(ERR, rxq,
1550                                         "Outer L3 csum failed, flags = 0x%x\n",
1551                                         parse_flag);
1552                                   rxq->rx_hw_errors++;
1553                                   ol_flags |= PKT_RX_EIP_CKSUM_BAD;
1554                         } else {
1555                                   ol_flags |= PKT_RX_IP_CKSUM_GOOD;
1556                         }
1557
1558                         if (tpa_start_flg)
1559                                 flags = cqe_start_tpa->tunnel_pars_flags.flags;
1560                         else
1561                                 flags = fp_cqe->tunnel_pars_flags.flags;
1562                         tunn_parse_flag = flags;
1563
1564                         /* Tunnel_type */
1565                         packet_type =
1566                                 qede_rx_cqe_to_tunn_pkt_type(tunn_parse_flag);
1567
1568                         /* Inner header */
1569                         packet_type |=
1570                               qede_rx_cqe_to_pkt_type_inner(parse_flag);
1571
1572                         /* Outer L3/L4 types is not available in CQE */
1573                         packet_type |= qede_rx_cqe_to_pkt_type_outer(rx_mb);
1574
1575                         /* Outer L3/L4 types is not available in CQE.
1576                          * Need to add offset to parse correctly,
1577                          */
1578                         rx_mb->data_off = offset + RTE_PKTMBUF_HEADROOM;
1579                         packet_type |= qede_rx_cqe_to_pkt_type_outer(rx_mb);
1580                 } else {
1581                         packet_type |= qede_rx_cqe_to_pkt_type(parse_flag);
1582                 }
1583
1584                 /* Common handling for non-tunnel packets and for inner
1585                  * headers in the case of tunnel.
1586                  */
1587                 if (unlikely(qede_check_notunn_csum_l4(parse_flag))) {
1588                         PMD_RX_LOG(ERR, rxq,
1589                                     "L4 csum failed, flags = 0x%x\n",
1590                                     parse_flag);
1591                         rxq->rx_hw_errors++;
1592                         ol_flags |= PKT_RX_L4_CKSUM_BAD;
1593                 } else {
1594                         ol_flags |= PKT_RX_L4_CKSUM_GOOD;
1595                 }
1596                 if (unlikely(qede_check_notunn_csum_l3(rx_mb, parse_flag))) {
1597                         PMD_RX_LOG(ERR, rxq, "IP csum failed, flags = 0x%x\n",
1598                                    parse_flag);
1599                         rxq->rx_hw_errors++;
1600                         ol_flags |= PKT_RX_IP_CKSUM_BAD;
1601                 } else {
1602                         ol_flags |= PKT_RX_IP_CKSUM_GOOD;
1603                 }
1604
1605                 if (CQE_HAS_VLAN(parse_flag) ||
1606                     CQE_HAS_OUTER_VLAN(parse_flag)) {
1607                         /* Note: FW doesn't indicate Q-in-Q packet */
1608                         ol_flags |= PKT_RX_VLAN;
1609                         if (qdev->vlan_strip_flg) {
1610                                 ol_flags |= PKT_RX_VLAN_STRIPPED;
1611                                 rx_mb->vlan_tci = vlan_tci;
1612                         }
1613                 }
1614
1615                 /* RSS Hash */
1616                 if (qdev->rss_enable) {
1617                         ol_flags |= PKT_RX_RSS_HASH;
1618                         rx_mb->hash.rss = rss_hash;
1619                 }
1620
1621                 rx_alloc_count++;
1622                 qede_rx_bd_ring_consume(rxq);
1623
1624                 if (!tpa_start_flg && fp_cqe->bd_num > 1) {
1625                         PMD_RX_LOG(DEBUG, rxq, "Jumbo-over-BD packet: %02x BDs"
1626                                    " len on first: %04x Total Len: %04x",
1627                                    fp_cqe->bd_num, len, pkt_len);
1628                         num_segs = fp_cqe->bd_num - 1;
1629                         seg1 = rx_mb;
1630                         if (qede_process_sg_pkts(p_rxq, seg1, num_segs,
1631                                                  pkt_len - len))
1632                                 goto next_cqe;
1633
1634                         rx_alloc_count += num_segs;
1635                         rxq->rx_segs += num_segs;
1636                 }
1637                 rxq->rx_segs++; /* for the first segment */
1638
1639                 /* Prefetch next mbuf while processing current one. */
1640                 preload_idx = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
1641                 rte_prefetch0(rxq->sw_rx_ring[preload_idx].mbuf);
1642
1643                 /* Update rest of the MBUF fields */
1644                 rx_mb->data_off = offset + RTE_PKTMBUF_HEADROOM;
1645                 rx_mb->port = rxq->port_id;
1646                 rx_mb->ol_flags = ol_flags;
1647                 rx_mb->data_len = len;
1648                 rx_mb->packet_type = packet_type;
1649 #ifdef RTE_LIBRTE_QEDE_DEBUG_RX
1650                 print_rx_bd_info(rx_mb, rxq, bitfield_val);
1651 #endif
1652                 if (!tpa_start_flg) {
1653                         rx_mb->nb_segs = fp_cqe->bd_num;
1654                         rx_mb->pkt_len = pkt_len;
1655                 } else {
1656                         /* store ref to the updated mbuf */
1657                         tpa_info->tpa_head = rx_mb;
1658                         tpa_info->tpa_tail = tpa_info->tpa_head;
1659                 }
1660                 rte_prefetch1(rte_pktmbuf_mtod(rx_mb, void *));
1661 tpa_end:
1662                 if (!tpa_start_flg) {
1663                         rx_pkts[rx_pkt] = rx_mb;
1664                         rx_pkt++;
1665                 }
1666 next_cqe:
1667                 ecore_chain_recycle_consumed(&rxq->rx_comp_ring);
1668                 sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
1669                 if (rx_pkt == nb_pkts) {
1670                         PMD_RX_LOG(DEBUG, rxq,
1671                                    "Budget reached nb_pkts=%u received=%u",
1672                                    rx_pkt, nb_pkts);
1673                         break;
1674                 }
1675         }
1676
1677         /* Request number of bufferes to be allocated in next loop */
1678         rxq->rx_alloc_count = rx_alloc_count;
1679
1680         rxq->rcv_pkts += rx_pkt;
1681
1682         PMD_RX_LOG(DEBUG, rxq, "rx_pkts=%u core=%d", rx_pkt, rte_lcore_id());
1683
1684         return rx_pkt;
1685 }
1686
1687
1688 /* Populate scatter gather buffer descriptor fields */
1689 static inline uint16_t
1690 qede_encode_sg_bd(struct qede_tx_queue *p_txq, struct rte_mbuf *m_seg,
1691                   struct eth_tx_2nd_bd **bd2, struct eth_tx_3rd_bd **bd3,
1692                   uint16_t start_seg)
1693 {
1694         struct qede_tx_queue *txq = p_txq;
1695         struct eth_tx_bd *tx_bd = NULL;
1696         dma_addr_t mapping;
1697         uint16_t nb_segs = 0;
1698
1699         /* Check for scattered buffers */
1700         while (m_seg) {
1701                 if (start_seg == 0) {
1702                         if (!*bd2) {
1703                                 *bd2 = (struct eth_tx_2nd_bd *)
1704                                         ecore_chain_produce(&txq->tx_pbl);
1705                                 memset(*bd2, 0, sizeof(struct eth_tx_2nd_bd));
1706                                 nb_segs++;
1707                         }
1708                         mapping = rte_mbuf_data_iova(m_seg);
1709                         QEDE_BD_SET_ADDR_LEN(*bd2, mapping, m_seg->data_len);
1710                         PMD_TX_LOG(DEBUG, txq, "BD2 len %04x", m_seg->data_len);
1711                 } else if (start_seg == 1) {
1712                         if (!*bd3) {
1713                                 *bd3 = (struct eth_tx_3rd_bd *)
1714                                         ecore_chain_produce(&txq->tx_pbl);
1715                                 memset(*bd3, 0, sizeof(struct eth_tx_3rd_bd));
1716                                 nb_segs++;
1717                         }
1718                         mapping = rte_mbuf_data_iova(m_seg);
1719                         QEDE_BD_SET_ADDR_LEN(*bd3, mapping, m_seg->data_len);
1720                         PMD_TX_LOG(DEBUG, txq, "BD3 len %04x", m_seg->data_len);
1721                 } else {
1722                         tx_bd = (struct eth_tx_bd *)
1723                                 ecore_chain_produce(&txq->tx_pbl);
1724                         memset(tx_bd, 0, sizeof(*tx_bd));
1725                         nb_segs++;
1726                         mapping = rte_mbuf_data_iova(m_seg);
1727                         QEDE_BD_SET_ADDR_LEN(tx_bd, mapping, m_seg->data_len);
1728                         PMD_TX_LOG(DEBUG, txq, "BD len %04x", m_seg->data_len);
1729                 }
1730                 start_seg++;
1731                 m_seg = m_seg->next;
1732         }
1733
1734         /* Return total scattered buffers */
1735         return nb_segs;
1736 }
1737
1738 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
1739 static inline void
1740 print_tx_bd_info(struct qede_tx_queue *txq,
1741                  struct eth_tx_1st_bd *bd1,
1742                  struct eth_tx_2nd_bd *bd2,
1743                  struct eth_tx_3rd_bd *bd3,
1744                  uint64_t tx_ol_flags)
1745 {
1746         char ol_buf[256] = { 0 }; /* for verbose prints */
1747
1748         if (bd1)
1749                 PMD_TX_LOG(INFO, txq,
1750                    "BD1: nbytes=0x%04x nbds=0x%04x bd_flags=0x%04x bf=0x%04x",
1751                    rte_cpu_to_le_16(bd1->nbytes), bd1->data.nbds,
1752                    bd1->data.bd_flags.bitfields,
1753                    rte_cpu_to_le_16(bd1->data.bitfields));
1754         if (bd2)
1755                 PMD_TX_LOG(INFO, txq,
1756                    "BD2: nbytes=0x%04x bf1=0x%04x bf2=0x%04x tunn_ip=0x%04x\n",
1757                    rte_cpu_to_le_16(bd2->nbytes), bd2->data.bitfields1,
1758                    bd2->data.bitfields2, bd2->data.tunn_ip_size);
1759         if (bd3)
1760                 PMD_TX_LOG(INFO, txq,
1761                    "BD3: nbytes=0x%04x bf=0x%04x MSS=0x%04x "
1762                    "tunn_l4_hdr_start_offset_w=0x%04x tunn_hdr_size=0x%04x\n",
1763                    rte_cpu_to_le_16(bd3->nbytes),
1764                    rte_cpu_to_le_16(bd3->data.bitfields),
1765                    rte_cpu_to_le_16(bd3->data.lso_mss),
1766                    bd3->data.tunn_l4_hdr_start_offset_w,
1767                    bd3->data.tunn_hdr_size_w);
1768
1769         rte_get_tx_ol_flag_list(tx_ol_flags, ol_buf, sizeof(ol_buf));
1770         PMD_TX_LOG(INFO, txq, "TX offloads = %s\n", ol_buf);
1771 }
1772 #endif
1773
1774 /* TX prepare to check packets meets TX conditions */
1775 uint16_t
1776 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
1777 qede_xmit_prep_pkts(void *p_txq, struct rte_mbuf **tx_pkts,
1778                     uint16_t nb_pkts)
1779 {
1780         struct qede_tx_queue *txq = p_txq;
1781 #else
1782 qede_xmit_prep_pkts(__rte_unused void *p_txq, struct rte_mbuf **tx_pkts,
1783                     uint16_t nb_pkts)
1784 {
1785 #endif
1786         uint64_t ol_flags;
1787         struct rte_mbuf *m;
1788         uint16_t i;
1789 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
1790         int ret;
1791 #endif
1792
1793         for (i = 0; i < nb_pkts; i++) {
1794                 m = tx_pkts[i];
1795                 ol_flags = m->ol_flags;
1796                 if (ol_flags & PKT_TX_TCP_SEG) {
1797                         if (m->nb_segs >= ETH_TX_MAX_BDS_PER_LSO_PACKET) {
1798                                 rte_errno = -EINVAL;
1799                                 break;
1800                         }
1801                         /* TBD: confirm its ~9700B for both ? */
1802                         if (m->tso_segsz > ETH_TX_MAX_NON_LSO_PKT_LEN) {
1803                                 rte_errno = -EINVAL;
1804                                 break;
1805                         }
1806                 } else {
1807                         if (m->nb_segs >= ETH_TX_MAX_BDS_PER_NON_LSO_PACKET) {
1808                                 rte_errno = -EINVAL;
1809                                 break;
1810                         }
1811                 }
1812                 if (ol_flags & QEDE_TX_OFFLOAD_NOTSUP_MASK) {
1813                         /* We support only limited tunnel protocols */
1814                         if (ol_flags & PKT_TX_TUNNEL_MASK) {
1815                                 uint64_t temp;
1816
1817                                 temp = ol_flags & PKT_TX_TUNNEL_MASK;
1818                                 if (temp == PKT_TX_TUNNEL_VXLAN ||
1819                                     temp == PKT_TX_TUNNEL_GENEVE ||
1820                                     temp == PKT_TX_TUNNEL_MPLSINUDP ||
1821                                     temp == PKT_TX_TUNNEL_GRE)
1822                                         break;
1823                         }
1824
1825                         rte_errno = -ENOTSUP;
1826                         break;
1827                 }
1828
1829 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
1830                 ret = rte_validate_tx_offload(m);
1831                 if (ret != 0) {
1832                         rte_errno = ret;
1833                         break;
1834                 }
1835 #endif
1836         }
1837
1838 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
1839         if (unlikely(i != nb_pkts))
1840                 PMD_TX_LOG(ERR, txq, "TX prepare failed for %u\n",
1841                            nb_pkts - i);
1842 #endif
1843         return i;
1844 }
1845
1846 #define MPLSINUDP_HDR_SIZE                      (12)
1847
1848 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
1849 static inline void
1850 qede_mpls_tunn_tx_sanity_check(struct rte_mbuf *mbuf,
1851                                struct qede_tx_queue *txq)
1852 {
1853         if (((mbuf->outer_l2_len + mbuf->outer_l3_len) / 2) > 0xff)
1854                 PMD_TX_LOG(ERR, txq, "tunn_l4_hdr_start_offset overflow\n");
1855         if (((mbuf->outer_l2_len + mbuf->outer_l3_len +
1856                 MPLSINUDP_HDR_SIZE) / 2) > 0xff)
1857                 PMD_TX_LOG(ERR, txq, "tunn_hdr_size overflow\n");
1858         if (((mbuf->l2_len - MPLSINUDP_HDR_SIZE) / 2) >
1859                 ETH_TX_DATA_2ND_BD_TUNN_INNER_L2_HDR_SIZE_W_MASK)
1860                 PMD_TX_LOG(ERR, txq, "inner_l2_hdr_size overflow\n");
1861         if (((mbuf->l2_len - MPLSINUDP_HDR_SIZE + mbuf->l3_len) / 2) >
1862                 ETH_TX_DATA_2ND_BD_L4_HDR_START_OFFSET_W_MASK)
1863                 PMD_TX_LOG(ERR, txq, "inner_l2_hdr_size overflow\n");
1864 }
1865 #endif
1866
1867 uint16_t
1868 qede_xmit_pkts(void *p_txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
1869 {
1870         struct qede_tx_queue *txq = p_txq;
1871         struct qede_dev *qdev = txq->qdev;
1872         struct ecore_dev *edev = &qdev->edev;
1873         struct rte_mbuf *mbuf;
1874         struct rte_mbuf *m_seg = NULL;
1875         uint16_t nb_tx_pkts;
1876         uint16_t bd_prod;
1877         uint16_t idx;
1878         uint16_t nb_frags;
1879         uint16_t nb_pkt_sent = 0;
1880         uint8_t nbds;
1881         bool lso_flg;
1882         bool mplsoudp_flg;
1883         __rte_unused bool tunn_flg;
1884         bool tunn_ipv6_ext_flg;
1885         struct eth_tx_1st_bd *bd1;
1886         struct eth_tx_2nd_bd *bd2;
1887         struct eth_tx_3rd_bd *bd3;
1888         uint64_t tx_ol_flags;
1889         uint16_t hdr_size;
1890         /* BD1 */
1891         uint16_t bd1_bf;
1892         uint8_t bd1_bd_flags_bf;
1893         uint16_t vlan;
1894         /* BD2 */
1895         uint16_t bd2_bf1;
1896         uint16_t bd2_bf2;
1897         /* BD3 */
1898         uint16_t mss;
1899         uint16_t bd3_bf;
1900
1901         uint8_t tunn_l4_hdr_start_offset;
1902         uint8_t tunn_hdr_size;
1903         uint8_t inner_l2_hdr_size;
1904         uint16_t inner_l4_hdr_offset;
1905
1906         if (unlikely(txq->nb_tx_avail < txq->tx_free_thresh)) {
1907                 PMD_TX_LOG(DEBUG, txq, "send=%u avail=%u free_thresh=%u",
1908                            nb_pkts, txq->nb_tx_avail, txq->tx_free_thresh);
1909                 qede_process_tx_compl(edev, txq);
1910         }
1911
1912         nb_tx_pkts  = nb_pkts;
1913         bd_prod = rte_cpu_to_le_16(ecore_chain_get_prod_idx(&txq->tx_pbl));
1914         while (nb_tx_pkts--) {
1915                 /* Init flags/values */
1916                 tunn_flg = false;
1917                 lso_flg = false;
1918                 nbds = 0;
1919                 vlan = 0;
1920                 bd1 = NULL;
1921                 bd2 = NULL;
1922                 bd3 = NULL;
1923                 hdr_size = 0;
1924                 bd1_bf = 0;
1925                 bd1_bd_flags_bf = 0;
1926                 bd2_bf1 = 0;
1927                 bd2_bf2 = 0;
1928                 mss = 0;
1929                 bd3_bf = 0;
1930                 mplsoudp_flg = false;
1931                 tunn_ipv6_ext_flg = false;
1932                 tunn_hdr_size = 0;
1933                 tunn_l4_hdr_start_offset = 0;
1934
1935                 mbuf = *tx_pkts++;
1936                 assert(mbuf);
1937
1938                 /* Check minimum TX BDS availability against available BDs */
1939                 if (unlikely(txq->nb_tx_avail < mbuf->nb_segs))
1940                         break;
1941
1942                 tx_ol_flags = mbuf->ol_flags;
1943                 bd1_bd_flags_bf |= 1 << ETH_TX_1ST_BD_FLAGS_START_BD_SHIFT;
1944
1945                 /* TX prepare would have already checked supported tunnel Tx
1946                  * offloads. Don't rely on pkt_type marked by Rx, instead use
1947                  * tx_ol_flags to decide.
1948                  */
1949                 tunn_flg = !!(tx_ol_flags & PKT_TX_TUNNEL_MASK);
1950
1951                 if (tunn_flg) {
1952                         /* Check against max which is Tunnel IPv6 + ext */
1953                         if (unlikely(txq->nb_tx_avail <
1954                                 ETH_TX_MIN_BDS_PER_TUNN_IPV6_WITH_EXT_PKT))
1955                                         break;
1956
1957                         /* First indicate its a tunnel pkt */
1958                         bd1_bf |= ETH_TX_DATA_1ST_BD_TUNN_FLAG_MASK <<
1959                                   ETH_TX_DATA_1ST_BD_TUNN_FLAG_SHIFT;
1960                         /* Legacy FW had flipped behavior in regard to this bit
1961                          * i.e. it needed to set to prevent FW from touching
1962                          * encapsulated packets when it didn't need to.
1963                          */
1964                         if (unlikely(txq->is_legacy)) {
1965                                 bd1_bf ^= 1 <<
1966                                         ETH_TX_DATA_1ST_BD_TUNN_FLAG_SHIFT;
1967                         }
1968
1969                         /* Outer IP checksum offload */
1970                         if (tx_ol_flags & (PKT_TX_OUTER_IP_CKSUM |
1971                                            PKT_TX_OUTER_IPV4)) {
1972                                 bd1_bd_flags_bf |=
1973                                         ETH_TX_1ST_BD_FLAGS_TUNN_IP_CSUM_MASK <<
1974                                         ETH_TX_1ST_BD_FLAGS_TUNN_IP_CSUM_SHIFT;
1975                         }
1976
1977                         /**
1978                          * Currently, only inner checksum offload in MPLS-in-UDP
1979                          * tunnel with one MPLS label is supported. Both outer
1980                          * and inner layers  lengths need to be provided in
1981                          * mbuf.
1982                          */
1983                         if ((tx_ol_flags & PKT_TX_TUNNEL_MASK) ==
1984                                                 PKT_TX_TUNNEL_MPLSINUDP) {
1985                                 mplsoudp_flg = true;
1986 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
1987                                 qede_mpls_tunn_tx_sanity_check(mbuf, txq);
1988 #endif
1989                                 /* Outer L4 offset in two byte words */
1990                                 tunn_l4_hdr_start_offset =
1991                                   (mbuf->outer_l2_len + mbuf->outer_l3_len) / 2;
1992                                 /* Tunnel header size in two byte words */
1993                                 tunn_hdr_size = (mbuf->outer_l2_len +
1994                                                 mbuf->outer_l3_len +
1995                                                 MPLSINUDP_HDR_SIZE) / 2;
1996                                 /* Inner L2 header size in two byte words */
1997                                 inner_l2_hdr_size = (mbuf->l2_len -
1998                                                 MPLSINUDP_HDR_SIZE) / 2;
1999                                 /* Inner L4 header offset from the beggining
2000                                  * of inner packet in two byte words
2001                                  */
2002                                 inner_l4_hdr_offset = (mbuf->l2_len -
2003                                         MPLSINUDP_HDR_SIZE + mbuf->l3_len) / 2;
2004
2005                                 /* Inner L2 size and address type */
2006                                 bd2_bf1 |= (inner_l2_hdr_size &
2007                                         ETH_TX_DATA_2ND_BD_TUNN_INNER_L2_HDR_SIZE_W_MASK) <<
2008                                         ETH_TX_DATA_2ND_BD_TUNN_INNER_L2_HDR_SIZE_W_SHIFT;
2009                                 bd2_bf1 |= (UNICAST_ADDRESS &
2010                                         ETH_TX_DATA_2ND_BD_TUNN_INNER_ETH_TYPE_MASK) <<
2011                                         ETH_TX_DATA_2ND_BD_TUNN_INNER_ETH_TYPE_SHIFT;
2012                                 /* Treated as IPv6+Ext */
2013                                 bd2_bf1 |=
2014                                     1 << ETH_TX_DATA_2ND_BD_TUNN_IPV6_EXT_SHIFT;
2015
2016                                 /* Mark inner IPv6 if present */
2017                                 if (tx_ol_flags & PKT_TX_IPV6)
2018                                         bd2_bf1 |=
2019                                                 1 << ETH_TX_DATA_2ND_BD_TUNN_INNER_IPV6_SHIFT;
2020
2021                                 /* Inner L4 offsets */
2022                                 if ((tx_ol_flags & (PKT_TX_IPV4 | PKT_TX_IPV6)) &&
2023                                      (tx_ol_flags & (PKT_TX_UDP_CKSUM |
2024                                                         PKT_TX_TCP_CKSUM))) {
2025                                         /* Determines if BD3 is needed */
2026                                         tunn_ipv6_ext_flg = true;
2027                                         if ((tx_ol_flags & PKT_TX_L4_MASK) ==
2028                                                         PKT_TX_UDP_CKSUM) {
2029                                                 bd2_bf1 |=
2030                                                         1 << ETH_TX_DATA_2ND_BD_L4_UDP_SHIFT;
2031                                         }
2032
2033                                         /* TODO other pseudo checksum modes are
2034                                          * not supported
2035                                          */
2036                                         bd2_bf1 |=
2037                                         ETH_L4_PSEUDO_CSUM_CORRECT_LENGTH <<
2038                                         ETH_TX_DATA_2ND_BD_L4_PSEUDO_CSUM_MODE_SHIFT;
2039                                         bd2_bf2 |= (inner_l4_hdr_offset &
2040                                                 ETH_TX_DATA_2ND_BD_L4_HDR_START_OFFSET_W_MASK) <<
2041                                                 ETH_TX_DATA_2ND_BD_L4_HDR_START_OFFSET_W_SHIFT;
2042                                 }
2043                         } /* End MPLSoUDP */
2044                 } /* End Tunnel handling */
2045
2046                 if (tx_ol_flags & PKT_TX_TCP_SEG) {
2047                         lso_flg = true;
2048                         if (unlikely(txq->nb_tx_avail <
2049                                                 ETH_TX_MIN_BDS_PER_LSO_PKT))
2050                                 break;
2051                         /* For LSO, packet header and payload must reside on
2052                          * buffers pointed by different BDs. Using BD1 for HDR
2053                          * and BD2 onwards for data.
2054                          */
2055                         hdr_size = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
2056                         if (tunn_flg)
2057                                 hdr_size += mbuf->outer_l2_len +
2058                                             mbuf->outer_l3_len;
2059
2060                         bd1_bd_flags_bf |= 1 << ETH_TX_1ST_BD_FLAGS_LSO_SHIFT;
2061                         bd1_bd_flags_bf |=
2062                                         1 << ETH_TX_1ST_BD_FLAGS_IP_CSUM_SHIFT;
2063                         /* PKT_TX_TCP_SEG implies PKT_TX_TCP_CKSUM */
2064                         bd1_bd_flags_bf |=
2065                                         1 << ETH_TX_1ST_BD_FLAGS_L4_CSUM_SHIFT;
2066                         mss = rte_cpu_to_le_16(mbuf->tso_segsz);
2067                         /* Using one header BD */
2068                         bd3_bf |= rte_cpu_to_le_16(1 <<
2069                                         ETH_TX_DATA_3RD_BD_HDR_NBD_SHIFT);
2070                 } else {
2071                         if (unlikely(txq->nb_tx_avail <
2072                                         ETH_TX_MIN_BDS_PER_NON_LSO_PKT))
2073                                 break;
2074                         bd1_bf |=
2075                                (mbuf->pkt_len & ETH_TX_DATA_1ST_BD_PKT_LEN_MASK)
2076                                 << ETH_TX_DATA_1ST_BD_PKT_LEN_SHIFT;
2077                 }
2078
2079                 /* Descriptor based VLAN insertion */
2080                 if (tx_ol_flags & PKT_TX_VLAN_PKT) {
2081                         vlan = rte_cpu_to_le_16(mbuf->vlan_tci);
2082                         bd1_bd_flags_bf |=
2083                             1 << ETH_TX_1ST_BD_FLAGS_VLAN_INSERTION_SHIFT;
2084                 }
2085
2086                 /* Offload the IP checksum in the hardware */
2087                 if (tx_ol_flags & PKT_TX_IP_CKSUM) {
2088                         bd1_bd_flags_bf |=
2089                                 1 << ETH_TX_1ST_BD_FLAGS_IP_CSUM_SHIFT;
2090                         /* There's no DPDK flag to request outer-L4 csum
2091                          * offload. But in the case of tunnel if inner L3 or L4
2092                          * csum offload is requested then we need to force
2093                          * recalculation of L4 tunnel header csum also.
2094                          */
2095                         if (tunn_flg && ((tx_ol_flags & PKT_TX_TUNNEL_MASK) !=
2096                                                         PKT_TX_TUNNEL_GRE)) {
2097                                 bd1_bd_flags_bf |=
2098                                         ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_MASK <<
2099                                         ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_SHIFT;
2100                         }
2101                 }
2102
2103                 /* L4 checksum offload (tcp or udp) */
2104                 if ((tx_ol_flags & (PKT_TX_IPV4 | PKT_TX_IPV6)) &&
2105                     (tx_ol_flags & (PKT_TX_UDP_CKSUM | PKT_TX_TCP_CKSUM))) {
2106                         bd1_bd_flags_bf |=
2107                                 1 << ETH_TX_1ST_BD_FLAGS_L4_CSUM_SHIFT;
2108                         /* There's no DPDK flag to request outer-L4 csum
2109                          * offload. But in the case of tunnel if inner L3 or L4
2110                          * csum offload is requested then we need to force
2111                          * recalculation of L4 tunnel header csum also.
2112                          */
2113                         if (tunn_flg) {
2114                                 bd1_bd_flags_bf |=
2115                                         ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_MASK <<
2116                                         ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_SHIFT;
2117                         }
2118                 }
2119
2120                 /* Fill the entry in the SW ring and the BDs in the FW ring */
2121                 idx = TX_PROD(txq);
2122                 txq->sw_tx_ring[idx].mbuf = mbuf;
2123
2124                 /* BD1 */
2125                 bd1 = (struct eth_tx_1st_bd *)ecore_chain_produce(&txq->tx_pbl);
2126                 memset(bd1, 0, sizeof(struct eth_tx_1st_bd));
2127                 nbds++;
2128
2129                 /* Map MBUF linear data for DMA and set in the BD1 */
2130                 QEDE_BD_SET_ADDR_LEN(bd1, rte_mbuf_data_iova(mbuf),
2131                                      mbuf->data_len);
2132                 bd1->data.bitfields = rte_cpu_to_le_16(bd1_bf);
2133                 bd1->data.bd_flags.bitfields = bd1_bd_flags_bf;
2134                 bd1->data.vlan = vlan;
2135
2136                 if (lso_flg || mplsoudp_flg) {
2137                         bd2 = (struct eth_tx_2nd_bd *)ecore_chain_produce
2138                                                         (&txq->tx_pbl);
2139                         memset(bd2, 0, sizeof(struct eth_tx_2nd_bd));
2140                         nbds++;
2141
2142                         /* BD1 */
2143                         QEDE_BD_SET_ADDR_LEN(bd1, rte_mbuf_data_iova(mbuf),
2144                                              hdr_size);
2145                         /* BD2 */
2146                         QEDE_BD_SET_ADDR_LEN(bd2, (hdr_size +
2147                                              rte_mbuf_data_iova(mbuf)),
2148                                              mbuf->data_len - hdr_size);
2149                         bd2->data.bitfields1 = rte_cpu_to_le_16(bd2_bf1);
2150                         if (mplsoudp_flg) {
2151                                 bd2->data.bitfields2 =
2152                                         rte_cpu_to_le_16(bd2_bf2);
2153                                 /* Outer L3 size */
2154                                 bd2->data.tunn_ip_size =
2155                                         rte_cpu_to_le_16(mbuf->outer_l3_len);
2156                         }
2157                         /* BD3 */
2158                         if (lso_flg || (mplsoudp_flg && tunn_ipv6_ext_flg)) {
2159                                 bd3 = (struct eth_tx_3rd_bd *)
2160                                         ecore_chain_produce(&txq->tx_pbl);
2161                                 memset(bd3, 0, sizeof(struct eth_tx_3rd_bd));
2162                                 nbds++;
2163                                 bd3->data.bitfields = rte_cpu_to_le_16(bd3_bf);
2164                                 if (lso_flg)
2165                                         bd3->data.lso_mss = mss;
2166                                 if (mplsoudp_flg) {
2167                                         bd3->data.tunn_l4_hdr_start_offset_w =
2168                                                 tunn_l4_hdr_start_offset;
2169                                         bd3->data.tunn_hdr_size_w =
2170                                                 tunn_hdr_size;
2171                                 }
2172                         }
2173                 }
2174
2175                 /* Handle fragmented MBUF */
2176                 m_seg = mbuf->next;
2177
2178                 /* Encode scatter gather buffer descriptors if required */
2179                 nb_frags = qede_encode_sg_bd(txq, m_seg, &bd2, &bd3, nbds - 1);
2180                 bd1->data.nbds = nbds + nb_frags;
2181
2182                 txq->nb_tx_avail -= bd1->data.nbds;
2183                 txq->sw_tx_prod++;
2184                 bd_prod =
2185                     rte_cpu_to_le_16(ecore_chain_get_prod_idx(&txq->tx_pbl));
2186 #ifdef RTE_LIBRTE_QEDE_DEBUG_TX
2187                 print_tx_bd_info(txq, bd1, bd2, bd3, tx_ol_flags);
2188 #endif
2189                 nb_pkt_sent++;
2190                 txq->xmit_pkts++;
2191         }
2192
2193         /* Write value of prod idx into bd_prod */
2194         txq->tx_db.data.bd_prod = bd_prod;
2195         rte_wmb();
2196         rte_compiler_barrier();
2197         DIRECT_REG_WR_RELAXED(edev, txq->doorbell_addr, txq->tx_db.raw);
2198         rte_wmb();
2199
2200         /* Check again for Tx completions */
2201         qede_process_tx_compl(edev, txq);
2202
2203         PMD_TX_LOG(DEBUG, txq, "to_send=%u sent=%u bd_prod=%u core=%d",
2204                    nb_pkts, nb_pkt_sent, TX_PROD(txq), rte_lcore_id());
2205
2206         return nb_pkt_sent;
2207 }
2208
2209 uint16_t
2210 qede_rxtx_pkts_dummy(__rte_unused void *p_rxq,
2211                      __rte_unused struct rte_mbuf **pkts,
2212                      __rte_unused uint16_t nb_pkts)
2213 {
2214         return 0;
2215 }
2216
2217
2218 /* this function does a fake walk through over completion queue
2219  * to calculate number of BDs used by HW.
2220  * At the end, it restores the state of completion queue.
2221  */
2222 static uint16_t
2223 qede_parse_fp_cqe(struct qede_rx_queue *rxq)
2224 {
2225         uint16_t hw_comp_cons, sw_comp_cons, bd_count = 0;
2226         union eth_rx_cqe *cqe, *orig_cqe = NULL;
2227
2228         hw_comp_cons = rte_le_to_cpu_16(*rxq->hw_cons_ptr);
2229         sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
2230
2231         if (hw_comp_cons == sw_comp_cons)
2232                 return 0;
2233
2234         /* Get the CQE from the completion ring */
2235         cqe = (union eth_rx_cqe *)ecore_chain_consume(&rxq->rx_comp_ring);
2236         orig_cqe = cqe;
2237
2238         while (sw_comp_cons != hw_comp_cons) {
2239                 switch (cqe->fast_path_regular.type) {
2240                 case ETH_RX_CQE_TYPE_REGULAR:
2241                         bd_count += cqe->fast_path_regular.bd_num;
2242                         break;
2243                 case ETH_RX_CQE_TYPE_TPA_END:
2244                         bd_count += cqe->fast_path_tpa_end.num_of_bds;
2245                         break;
2246                 default:
2247                         break;
2248                 }
2249
2250                 cqe =
2251                 (union eth_rx_cqe *)ecore_chain_consume(&rxq->rx_comp_ring);
2252                 sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
2253         }
2254
2255         /* revert comp_ring to original state */
2256         ecore_chain_set_cons(&rxq->rx_comp_ring, sw_comp_cons, orig_cqe);
2257
2258         return bd_count;
2259 }
2260
2261 int
2262 qede_rx_descriptor_status(void *p_rxq, uint16_t offset)
2263 {
2264         uint16_t hw_bd_cons, sw_bd_cons, sw_bd_prod;
2265         uint16_t produced, consumed;
2266         struct qede_rx_queue *rxq = p_rxq;
2267
2268         if (offset > rxq->nb_rx_desc)
2269                 return -EINVAL;
2270
2271         sw_bd_cons = ecore_chain_get_cons_idx(&rxq->rx_bd_ring);
2272         sw_bd_prod = ecore_chain_get_prod_idx(&rxq->rx_bd_ring);
2273
2274         /* find BDs used by HW from completion queue elements */
2275         hw_bd_cons = sw_bd_cons + qede_parse_fp_cqe(rxq);
2276
2277         if (hw_bd_cons < sw_bd_cons)
2278                 /* wraparound case */
2279                 consumed = (0xffff - sw_bd_cons) + hw_bd_cons;
2280         else
2281                 consumed = hw_bd_cons - sw_bd_cons;
2282
2283         if (offset <= consumed)
2284                 return RTE_ETH_RX_DESC_DONE;
2285
2286         if (sw_bd_prod < sw_bd_cons)
2287                 /* wraparound case */
2288                 produced = (0xffff - sw_bd_cons) + sw_bd_prod;
2289         else
2290                 produced = sw_bd_prod - sw_bd_cons;
2291
2292         if (offset <= produced)
2293                 return RTE_ETH_RX_DESC_AVAIL;
2294
2295         return RTE_ETH_RX_DESC_UNAVAIL;
2296 }