drivers/net/ice/ice_dcf_ethdev.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2020 Intel Corporation
   3  */
   4
   5 #include <errno.h>
   6 #include <stdbool.h>
   7 #include <sys/types.h>
   8 #include <sys/ioctl.h>
   9 #include <unistd.h>
  10
  11 #include <rte_interrupts.h>
  12 #include <rte_debug.h>
  13 #include <rte_pci.h>
  14 #include <rte_atomic.h>
  15 #include <rte_eal.h>
  16 #include <rte_ether.h>
  17 #include <rte_ethdev_pci.h>
  18 #include <rte_kvargs.h>
  19 #include <rte_malloc.h>
  20 #include <rte_memzone.h>
  21 #include <rte_dev.h>
  22
  23 #include <iavf_devids.h>
  24
  25 #include "ice_generic_flow.h"
  26 #include "ice_dcf_ethdev.h"
  27 #include "ice_rxtx.h"
  28
  29 static uint16_t
  30 ice_dcf_recv_pkts(__rte_unused void *rx_queue,
  31                   __rte_unused struct rte_mbuf **bufs,
  32                   __rte_unused uint16_t nb_pkts)
  33 {
  34         return 0;
  35 }
  36
  37 static uint16_t
  38 ice_dcf_xmit_pkts(__rte_unused void *tx_queue,
  39                   __rte_unused struct rte_mbuf **bufs,
  40                   __rte_unused uint16_t nb_pkts)
  41 {
  42         return 0;
  43 }
  44
  45 static int
  46 ice_dcf_init_rxq(struct rte_eth_dev *dev, struct ice_rx_queue *rxq)
  47 {
  48         struct ice_dcf_adapter *dcf_ad = dev->data->dev_private;
  49         struct rte_eth_dev_data *dev_data = dev->data;
  50         struct iavf_hw *hw = &dcf_ad->real_hw.avf;
  51         uint16_t buf_size, max_pkt_len, len;
  52
  53         buf_size = rte_pktmbuf_data_room_size(rxq->mp) - RTE_PKTMBUF_HEADROOM;
  54         rxq->rx_hdr_len = 0;
  55         rxq->rx_buf_len = RTE_ALIGN(buf_size, (1 << ICE_RLAN_CTX_DBUF_S));
  56         len = ICE_SUPPORT_CHAIN_NUM * rxq->rx_buf_len;
  57         max_pkt_len = RTE_MIN(len, dev->data->dev_conf.rxmode.max_rx_pkt_len);
  58
  59         /* Check if the jumbo frame and maximum packet length are set
  60          * correctly.
  61          */
  62         if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_JUMBO_FRAME) {
  63                 if (max_pkt_len <= ICE_ETH_MAX_LEN ||
  64                     max_pkt_len > ICE_FRAME_SIZE_MAX) {
  65                         PMD_DRV_LOG(ERR, "maximum packet length must be "
  66                                     "larger than %u and smaller than %u, "
  67                                     "as jumbo frame is enabled",
  68                                     (uint32_t)ICE_ETH_MAX_LEN,
  69                                     (uint32_t)ICE_FRAME_SIZE_MAX);
  70                         return -EINVAL;
  71                 }
  72         } else {
  73                 if (max_pkt_len < RTE_ETHER_MIN_LEN ||
  74                     max_pkt_len > ICE_ETH_MAX_LEN) {
  75                         PMD_DRV_LOG(ERR, "maximum packet length must be "
  76                                     "larger than %u and smaller than %u, "
  77                                     "as jumbo frame is disabled",
  78                                     (uint32_t)RTE_ETHER_MIN_LEN,
  79                                     (uint32_t)ICE_ETH_MAX_LEN);
  80                         return -EINVAL;
  81                 }
  82         }
  83
  84         rxq->max_pkt_len = max_pkt_len;
  85         if ((dev_data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_SCATTER) ||
  86             (rxq->max_pkt_len + 2 * ICE_VLAN_TAG_SIZE) > buf_size) {
  87                 dev_data->scattered_rx = 1;
  88         }
  89         rxq->qrx_tail = hw->hw_addr + IAVF_QRX_TAIL1(rxq->queue_id);
  90         IAVF_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
  91         IAVF_WRITE_FLUSH(hw);
  92
  93         return 0;
  94 }
  95
  96 static int
  97 ice_dcf_init_rx_queues(struct rte_eth_dev *dev)
  98 {
  99         struct ice_rx_queue **rxq =
 100                 (struct ice_rx_queue **)dev->data->rx_queues;
 101         int i, ret;
 102
 103         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 104                 if (!rxq[i] || !rxq[i]->q_set)
 105                         continue;
 106                 ret = ice_dcf_init_rxq(dev, rxq[i]);
 107                 if (ret)
 108                         return ret;
 109         }
 110
 111         ice_set_rx_function(dev);
 112         ice_set_tx_function(dev);
 113
 114         return 0;
 115 }
 116
 117 #define IAVF_MISC_VEC_ID                RTE_INTR_VEC_ZERO_OFFSET
 118 #define IAVF_RX_VEC_START               RTE_INTR_VEC_RXTX_OFFSET
 119
 120 #define IAVF_ITR_INDEX_DEFAULT          0
 121 #define IAVF_QUEUE_ITR_INTERVAL_DEFAULT 32 /* 32 us */
 122 #define IAVF_QUEUE_ITR_INTERVAL_MAX     8160 /* 8160 us */
 123
 124 static inline uint16_t
 125 iavf_calc_itr_interval(int16_t interval)
 126 {
 127         if (interval < 0 || interval > IAVF_QUEUE_ITR_INTERVAL_MAX)
 128                 interval = IAVF_QUEUE_ITR_INTERVAL_DEFAULT;
 129
 130         /* Convert to hardware count, as writing each 1 represents 2 us */
 131         return interval / 2;
 132 }
 133
 134 static int
 135 ice_dcf_config_rx_queues_irqs(struct rte_eth_dev *dev,
 136                                      struct rte_intr_handle *intr_handle)
 137 {
 138         struct ice_dcf_adapter *adapter = dev->data->dev_private;
 139         struct ice_dcf_hw *hw = &adapter->real_hw;
 140         uint16_t interval, i;
 141         int vec;
 142
 143         if (rte_intr_cap_multiple(intr_handle) &&
 144             dev->data->dev_conf.intr_conf.rxq) {
 145                 if (rte_intr_efd_enable(intr_handle, dev->data->nb_rx_queues))
 146                         return -1;
 147         }
 148
 149         if (rte_intr_dp_is_en(intr_handle) && !intr_handle->intr_vec) {
 150                 intr_handle->intr_vec =
 151                         rte_zmalloc("intr_vec",
 152                                     dev->data->nb_rx_queues * sizeof(int), 0);
 153                 if (!intr_handle->intr_vec) {
 154                         PMD_DRV_LOG(ERR, "Failed to allocate %d rx intr_vec",
 155                                     dev->data->nb_rx_queues);
 156                         return -1;
 157                 }
 158         }
 159
 160         if (!dev->data->dev_conf.intr_conf.rxq ||
 161             !rte_intr_dp_is_en(intr_handle)) {
 162                 /* Rx interrupt disabled, Map interrupt only for writeback */
 163                 hw->nb_msix = 1;
 164                 if (hw->vf_res->vf_cap_flags &
 165                     VIRTCHNL_VF_OFFLOAD_WB_ON_ITR) {
 166                         /* If WB_ON_ITR supports, enable it */
 167                         hw->msix_base = IAVF_RX_VEC_START;
 168                         IAVF_WRITE_REG(&hw->avf,
 169                                        IAVF_VFINT_DYN_CTLN1(hw->msix_base - 1),
 170                                        IAVF_VFINT_DYN_CTLN1_ITR_INDX_MASK |
 171                                        IAVF_VFINT_DYN_CTLN1_WB_ON_ITR_MASK);
 172                 } else {
 173                         /* If no WB_ON_ITR offload flags, need to set
 174                          * interrupt for descriptor write back.
 175                          */
 176                         hw->msix_base = IAVF_MISC_VEC_ID;
 177
 178                         /* set ITR to max */
 179                         interval =
 180                         iavf_calc_itr_interval(IAVF_QUEUE_ITR_INTERVAL_MAX);
 181                         IAVF_WRITE_REG(&hw->avf, IAVF_VFINT_DYN_CTL01,
 182                                        IAVF_VFINT_DYN_CTL01_INTENA_MASK |
 183                                        (IAVF_ITR_INDEX_DEFAULT <<
 184                                         IAVF_VFINT_DYN_CTL01_ITR_INDX_SHIFT) |
 185                                        (interval <<
 186                                         IAVF_VFINT_DYN_CTL01_INTERVAL_SHIFT));
 187                 }
 188                 IAVF_WRITE_FLUSH(&hw->avf);
 189                 /* map all queues to the same interrupt */
 190                 for (i = 0; i < dev->data->nb_rx_queues; i++)
 191                         hw->rxq_map[hw->msix_base] |= 1 << i;
 192         } else {
 193                 if (!rte_intr_allow_others(intr_handle)) {
 194                         hw->nb_msix = 1;
 195                         hw->msix_base = IAVF_MISC_VEC_ID;
 196                         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 197                                 hw->rxq_map[hw->msix_base] |= 1 << i;
 198                                 intr_handle->intr_vec[i] = IAVF_MISC_VEC_ID;
 199                         }
 200                         PMD_DRV_LOG(DEBUG,
 201                                     "vector %u are mapping to all Rx queues",
 202                                     hw->msix_base);
 203                 } else {
 204                         /* If Rx interrupt is reuquired, and we can use
 205                          * multi interrupts, then the vec is from 1
 206                          */
 207                         hw->nb_msix = RTE_MIN(hw->vf_res->max_vectors,
 208                                               intr_handle->nb_efd);
 209                         hw->msix_base = IAVF_MISC_VEC_ID;
 210                         vec = IAVF_MISC_VEC_ID;
 211                         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 212                                 hw->rxq_map[vec] |= 1 << i;
 213                                 intr_handle->intr_vec[i] = vec++;
 214                                 if (vec >= hw->nb_msix)
 215                                         vec = IAVF_RX_VEC_START;
 216                         }
 217                         PMD_DRV_LOG(DEBUG,
 218                                     "%u vectors are mapping to %u Rx queues",
 219                                     hw->nb_msix, dev->data->nb_rx_queues);
 220                 }
 221         }
 222
 223         if (ice_dcf_config_irq_map(hw)) {
 224                 PMD_DRV_LOG(ERR, "config interrupt mapping failed");
 225                 return -1;
 226         }
 227         return 0;
 228 }
 229
 230 static int
 231 alloc_rxq_mbufs(struct ice_rx_queue *rxq)
 232 {
 233         volatile union ice_rx_flex_desc *rxd;
 234         struct rte_mbuf *mbuf = NULL;
 235         uint64_t dma_addr;
 236         uint16_t i;
 237
 238         for (i = 0; i < rxq->nb_rx_desc; i++) {
 239                 mbuf = rte_mbuf_raw_alloc(rxq->mp);
 240                 if (unlikely(!mbuf)) {
 241                         PMD_DRV_LOG(ERR, "Failed to allocate mbuf for RX");
 242                         return -ENOMEM;
 243                 }
 244
 245                 rte_mbuf_refcnt_set(mbuf, 1);
 246                 mbuf->next = NULL;
 247                 mbuf->data_off = RTE_PKTMBUF_HEADROOM;
 248                 mbuf->nb_segs = 1;
 249                 mbuf->port = rxq->port_id;
 250
 251                 dma_addr =
 252                         rte_cpu_to_le_64(rte_mbuf_data_iova_default(mbuf));
 253
 254                 rxd = &rxq->rx_ring[i];
 255                 rxd->read.pkt_addr = dma_addr;
 256                 rxd->read.hdr_addr = 0;
 257 #ifndef RTE_LIBRTE_ICE_16BYTE_RX_DESC
 258                 rxd->read.rsvd1 = 0;
 259                 rxd->read.rsvd2 = 0;
 260 #endif
 261
 262                 rxq->sw_ring[i].mbuf = (void *)mbuf;
 263         }
 264
 265         return 0;
 266 }
 267
 268 static int
 269 ice_dcf_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 270 {
 271         struct ice_dcf_adapter *ad = dev->data->dev_private;
 272         struct iavf_hw *hw = &ad->real_hw.avf;
 273         struct ice_rx_queue *rxq;
 274         int err = 0;
 275
 276         if (rx_queue_id >= dev->data->nb_rx_queues)
 277                 return -EINVAL;
 278
 279         rxq = dev->data->rx_queues[rx_queue_id];
 280
 281         err = alloc_rxq_mbufs(rxq);
 282         if (err) {
 283                 PMD_DRV_LOG(ERR, "Failed to allocate RX queue mbuf");
 284                 return err;
 285         }
 286
 287         rte_wmb();
 288
 289         /* Init the RX tail register. */
 290         IAVF_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
 291         IAVF_WRITE_FLUSH(hw);
 292
 293         /* Ready to switch the queue on */
 294         err = ice_dcf_switch_queue(&ad->real_hw, rx_queue_id, true, true);
 295         if (err) {
 296                 PMD_DRV_LOG(ERR, "Failed to switch RX queue %u on",
 297                             rx_queue_id);
 298                 return err;
 299         }
 300
 301         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
 302
 303         return 0;
 304 }
 305
 306 static inline void
 307 reset_rx_queue(struct ice_rx_queue *rxq)
 308 {
 309         uint16_t len;
 310         uint32_t i;
 311
 312         if (!rxq)
 313                 return;
 314
 315         len = rxq->nb_rx_desc + ICE_RX_MAX_BURST;
 316
 317         for (i = 0; i < len * sizeof(union ice_rx_flex_desc); i++)
 318                 ((volatile char *)rxq->rx_ring)[i] = 0;
 319
 320         memset(&rxq->fake_mbuf, 0x0, sizeof(rxq->fake_mbuf));
 321
 322         for (i = 0; i < ICE_RX_MAX_BURST; i++)
 323                 rxq->sw_ring[rxq->nb_rx_desc + i].mbuf = &rxq->fake_mbuf;
 324
 325         /* for rx bulk */
 326         rxq->rx_nb_avail = 0;
 327         rxq->rx_next_avail = 0;
 328         rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
 329
 330         rxq->rx_tail = 0;
 331         rxq->nb_rx_hold = 0;
 332         rxq->pkt_first_seg = NULL;
 333         rxq->pkt_last_seg = NULL;
 334 }
 335
 336 static inline void
 337 reset_tx_queue(struct ice_tx_queue *txq)
 338 {
 339         struct ice_tx_entry *txe;
 340         uint32_t i, size;
 341         uint16_t prev;
 342
 343         if (!txq) {
 344                 PMD_DRV_LOG(DEBUG, "Pointer to txq is NULL");
 345                 return;
 346         }
 347
 348         txe = txq->sw_ring;
 349         size = sizeof(struct ice_tx_desc) * txq->nb_tx_desc;
 350         for (i = 0; i < size; i++)
 351                 ((volatile char *)txq->tx_ring)[i] = 0;
 352
 353         prev = (uint16_t)(txq->nb_tx_desc - 1);
 354         for (i = 0; i < txq->nb_tx_desc; i++) {
 355                 txq->tx_ring[i].cmd_type_offset_bsz =
 356                         rte_cpu_to_le_64(IAVF_TX_DESC_DTYPE_DESC_DONE);
 357                 txe[i].mbuf =  NULL;
 358                 txe[i].last_id = i;
 359                 txe[prev].next_id = i;
 360                 prev = i;
 361         }
 362
 363         txq->tx_tail = 0;
 364         txq->nb_tx_used = 0;
 365
 366         txq->last_desc_cleaned = txq->nb_tx_desc - 1;
 367         txq->nb_tx_free = txq->nb_tx_desc - 1;
 368
 369         txq->tx_next_dd = txq->tx_rs_thresh - 1;
 370         txq->tx_next_rs = txq->tx_rs_thresh - 1;
 371 }
 372
 373 static int
 374 ice_dcf_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 375 {
 376         struct ice_dcf_adapter *ad = dev->data->dev_private;
 377         struct ice_dcf_hw *hw = &ad->real_hw;
 378         struct ice_rx_queue *rxq;
 379         int err;
 380
 381         if (rx_queue_id >= dev->data->nb_rx_queues)
 382                 return -EINVAL;
 383
 384         err = ice_dcf_switch_queue(hw, rx_queue_id, true, false);
 385         if (err) {
 386                 PMD_DRV_LOG(ERR, "Failed to switch RX queue %u off",
 387                             rx_queue_id);
 388                 return err;
 389         }
 390
 391         rxq = dev->data->rx_queues[rx_queue_id];
 392         rxq->rx_rel_mbufs(rxq);
 393         reset_rx_queue(rxq);
 394         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
 395
 396         return 0;
 397 }
 398
 399 static int
 400 ice_dcf_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 401 {
 402         struct ice_dcf_adapter *ad = dev->data->dev_private;
 403         struct iavf_hw *hw = &ad->real_hw.avf;
 404         struct ice_tx_queue *txq;
 405         int err = 0;
 406
 407         if (tx_queue_id >= dev->data->nb_tx_queues)
 408                 return -EINVAL;
 409
 410         txq = dev->data->tx_queues[tx_queue_id];
 411
 412         /* Init the RX tail register. */
 413         txq->qtx_tail = hw->hw_addr + IAVF_QTX_TAIL1(tx_queue_id);
 414         IAVF_PCI_REG_WRITE(txq->qtx_tail, 0);
 415         IAVF_WRITE_FLUSH(hw);
 416
 417         /* Ready to switch the queue on */
 418         err = ice_dcf_switch_queue(&ad->real_hw, tx_queue_id, false, true);
 419
 420         if (err) {
 421                 PMD_DRV_LOG(ERR, "Failed to switch TX queue %u on",
 422                             tx_queue_id);
 423                 return err;
 424         }
 425
 426         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
 427
 428         return 0;
 429 }
 430
 431 static int
 432 ice_dcf_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 433 {
 434         struct ice_dcf_adapter *ad = dev->data->dev_private;
 435         struct ice_dcf_hw *hw = &ad->real_hw;
 436         struct ice_tx_queue *txq;
 437         int err;
 438
 439         if (tx_queue_id >= dev->data->nb_tx_queues)
 440                 return -EINVAL;
 441
 442         err = ice_dcf_switch_queue(hw, tx_queue_id, false, false);
 443         if (err) {
 444                 PMD_DRV_LOG(ERR, "Failed to switch TX queue %u off",
 445                             tx_queue_id);
 446                 return err;
 447         }
 448
 449         txq = dev->data->tx_queues[tx_queue_id];
 450         txq->tx_rel_mbufs(txq);
 451         reset_tx_queue(txq);
 452         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
 453
 454         return 0;
 455 }
 456
 457 static int
 458 ice_dcf_start_queues(struct rte_eth_dev *dev)
 459 {
 460         struct ice_rx_queue *rxq;
 461         struct ice_tx_queue *txq;
 462         int nb_rxq = 0;
 463         int nb_txq, i;
 464
 465         for (nb_txq = 0; nb_txq < dev->data->nb_tx_queues; nb_txq++) {
 466                 txq = dev->data->tx_queues[nb_txq];
 467                 if (txq->tx_deferred_start)
 468                         continue;
 469                 if (ice_dcf_tx_queue_start(dev, nb_txq) != 0) {
 470                         PMD_DRV_LOG(ERR, "Fail to start queue %u", nb_txq);
 471                         goto tx_err;
 472                 }
 473         }
 474
 475         for (nb_rxq = 0; nb_rxq < dev->data->nb_rx_queues; nb_rxq++) {
 476                 rxq = dev->data->rx_queues[nb_rxq];
 477                 if (rxq->rx_deferred_start)
 478                         continue;
 479                 if (ice_dcf_rx_queue_start(dev, nb_rxq) != 0) {
 480                         PMD_DRV_LOG(ERR, "Fail to start queue %u", nb_rxq);
 481                         goto rx_err;
 482                 }
 483         }
 484
 485         return 0;
 486
 487         /* stop the started queues if failed to start all queues */
 488 rx_err:
 489         for (i = 0; i < nb_rxq; i++)
 490                 ice_dcf_rx_queue_stop(dev, i);
 491 tx_err:
 492         for (i = 0; i < nb_txq; i++)
 493                 ice_dcf_tx_queue_stop(dev, i);
 494
 495         return -1;
 496 }
 497
 498 static int
 499 ice_dcf_dev_start(struct rte_eth_dev *dev)
 500 {
 501         struct ice_dcf_adapter *dcf_ad = dev->data->dev_private;
 502         struct rte_intr_handle *intr_handle = dev->intr_handle;
 503         struct ice_adapter *ad = &dcf_ad->parent;
 504         struct ice_dcf_hw *hw = &dcf_ad->real_hw;
 505         int ret;
 506
 507         ad->pf.adapter_stopped = 0;
 508
 509         hw->num_queue_pairs = RTE_MAX(dev->data->nb_rx_queues,
 510                                       dev->data->nb_tx_queues);
 511
 512         ret = ice_dcf_init_rx_queues(dev);
 513         if (ret) {
 514                 PMD_DRV_LOG(ERR, "Fail to init queues");
 515                 return ret;
 516         }
 517
 518         if (hw->vf_res->vf_cap_flags & VIRTCHNL_VF_OFFLOAD_RSS_PF) {
 519                 ret = ice_dcf_init_rss(hw);
 520                 if (ret) {
 521                         PMD_DRV_LOG(ERR, "Failed to configure RSS");
 522                         return ret;
 523                 }
 524         }
 525
 526         ret = ice_dcf_configure_queues(hw);
 527         if (ret) {
 528                 PMD_DRV_LOG(ERR, "Fail to config queues");
 529                 return ret;
 530         }
 531
 532         ret = ice_dcf_config_rx_queues_irqs(dev, intr_handle);
 533         if (ret) {
 534                 PMD_DRV_LOG(ERR, "Fail to config rx queues' irqs");
 535                 return ret;
 536         }
 537
 538         if (dev->data->dev_conf.intr_conf.rxq != 0) {
 539                 rte_intr_disable(intr_handle);
 540                 rte_intr_enable(intr_handle);
 541         }
 542
 543         ret = ice_dcf_start_queues(dev);
 544         if (ret) {
 545                 PMD_DRV_LOG(ERR, "Failed to enable queues");
 546                 return ret;
 547         }
 548
 549         ret = ice_dcf_add_del_all_mac_addr(hw, true);
 550         if (ret) {
 551                 PMD_DRV_LOG(ERR, "Failed to add mac addr");
 552                 return ret;
 553         }
 554
 555         dev->data->dev_link.link_status = ETH_LINK_UP;
 556
 557         return 0;
 558 }
 559
 560 static void
 561 ice_dcf_stop_queues(struct rte_eth_dev *dev)
 562 {
 563         struct ice_dcf_adapter *ad = dev->data->dev_private;
 564         struct ice_dcf_hw *hw = &ad->real_hw;
 565         struct ice_rx_queue *rxq;
 566         struct ice_tx_queue *txq;
 567         int ret, i;
 568
 569         /* Stop All queues */
 570         ret = ice_dcf_disable_queues(hw);
 571         if (ret)
 572                 PMD_DRV_LOG(WARNING, "Fail to stop queues");
 573
 574         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 575                 txq = dev->data->tx_queues[i];
 576                 if (!txq)
 577                         continue;
 578                 txq->tx_rel_mbufs(txq);
 579                 reset_tx_queue(txq);
 580                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 581         }
 582         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 583                 rxq = dev->data->rx_queues[i];
 584                 if (!rxq)
 585                         continue;
 586                 rxq->rx_rel_mbufs(rxq);
 587                 reset_rx_queue(rxq);
 588                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 589         }
 590 }
 591
 592 static int
 593 ice_dcf_dev_stop(struct rte_eth_dev *dev)
 594 {
 595         struct ice_dcf_adapter *dcf_ad = dev->data->dev_private;
 596         struct rte_intr_handle *intr_handle = dev->intr_handle;
 597         struct ice_adapter *ad = &dcf_ad->parent;
 598
 599         if (ad->pf.adapter_stopped == 1) {
 600                 PMD_DRV_LOG(DEBUG, "Port is already stopped");
 601                 return 0;
 602         }
 603
 604         /* Stop the VF representors for this device */
 605         ice_dcf_vf_repr_stop_all(dcf_ad);
 606
 607         ice_dcf_stop_queues(dev);
 608
 609         rte_intr_efd_disable(intr_handle);
 610         if (intr_handle->intr_vec) {
 611                 rte_free(intr_handle->intr_vec);
 612                 intr_handle->intr_vec = NULL;
 613         }
 614
 615         ice_dcf_add_del_all_mac_addr(&dcf_ad->real_hw, false);
 616         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 617         ad->pf.adapter_stopped = 1;
 618
 619         return 0;
 620 }
 621
 622 static int
 623 ice_dcf_dev_configure(struct rte_eth_dev *dev)
 624 {
 625         struct ice_dcf_adapter *dcf_ad = dev->data->dev_private;
 626         struct ice_adapter *ad = &dcf_ad->parent;
 627
 628         ad->rx_bulk_alloc_allowed = true;
 629         ad->tx_simple_allowed = true;
 630
 631         if (dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG)
 632                 dev->data->dev_conf.rxmode.offloads |= DEV_RX_OFFLOAD_RSS_HASH;
 633
 634         return 0;
 635 }
 636
 637 static int
 638 ice_dcf_dev_info_get(struct rte_eth_dev *dev,
 639                      struct rte_eth_dev_info *dev_info)
 640 {
 641         struct ice_dcf_adapter *adapter = dev->data->dev_private;
 642         struct ice_dcf_hw *hw = &adapter->real_hw;
 643
 644         dev_info->max_mac_addrs = 1;
 645         dev_info->max_rx_queues = hw->vsi_res->num_queue_pairs;
 646         dev_info->max_tx_queues = hw->vsi_res->num_queue_pairs;
 647         dev_info->min_rx_bufsize = ICE_BUF_SIZE_MIN;
 648         dev_info->max_rx_pktlen = ICE_FRAME_SIZE_MAX;
 649         dev_info->hash_key_size = hw->vf_res->rss_key_size;
 650         dev_info->reta_size = hw->vf_res->rss_lut_size;
 651         dev_info->flow_type_rss_offloads = ICE_RSS_OFFLOAD_ALL;
 652
 653         dev_info->rx_offload_capa =
 654                 DEV_RX_OFFLOAD_VLAN_STRIP |
 655                 DEV_RX_OFFLOAD_IPV4_CKSUM |
 656                 DEV_RX_OFFLOAD_UDP_CKSUM |
 657                 DEV_RX_OFFLOAD_TCP_CKSUM |
 658                 DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM |
 659                 DEV_RX_OFFLOAD_SCATTER |
 660                 DEV_RX_OFFLOAD_JUMBO_FRAME |
 661                 DEV_RX_OFFLOAD_VLAN_FILTER |
 662                 DEV_RX_OFFLOAD_RSS_HASH;
 663         dev_info->tx_offload_capa =
 664                 DEV_TX_OFFLOAD_VLAN_INSERT |
 665                 DEV_TX_OFFLOAD_IPV4_CKSUM |
 666                 DEV_TX_OFFLOAD_UDP_CKSUM |
 667                 DEV_TX_OFFLOAD_TCP_CKSUM |
 668                 DEV_TX_OFFLOAD_SCTP_CKSUM |
 669                 DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM |
 670                 DEV_TX_OFFLOAD_TCP_TSO |
 671                 DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
 672                 DEV_TX_OFFLOAD_GRE_TNL_TSO |
 673                 DEV_TX_OFFLOAD_IPIP_TNL_TSO |
 674                 DEV_TX_OFFLOAD_GENEVE_TNL_TSO |
 675                 DEV_TX_OFFLOAD_MULTI_SEGS;
 676
 677         dev_info->default_rxconf = (struct rte_eth_rxconf) {
 678                 .rx_thresh = {
 679                         .pthresh = ICE_DEFAULT_RX_PTHRESH,
 680                         .hthresh = ICE_DEFAULT_RX_HTHRESH,
 681                         .wthresh = ICE_DEFAULT_RX_WTHRESH,
 682                 },
 683                 .rx_free_thresh = ICE_DEFAULT_RX_FREE_THRESH,
 684                 .rx_drop_en = 0,
 685                 .offloads = 0,
 686         };
 687
 688         dev_info->default_txconf = (struct rte_eth_txconf) {
 689                 .tx_thresh = {
 690                         .pthresh = ICE_DEFAULT_TX_PTHRESH,
 691                         .hthresh = ICE_DEFAULT_TX_HTHRESH,
 692                         .wthresh = ICE_DEFAULT_TX_WTHRESH,
 693                 },
 694                 .tx_free_thresh = ICE_DEFAULT_TX_FREE_THRESH,
 695                 .tx_rs_thresh = ICE_DEFAULT_TX_RSBIT_THRESH,
 696                 .offloads = 0,
 697         };
 698
 699         dev_info->rx_desc_lim = (struct rte_eth_desc_lim) {
 700                 .nb_max = ICE_MAX_RING_DESC,
 701                 .nb_min = ICE_MIN_RING_DESC,
 702                 .nb_align = ICE_ALIGN_RING_DESC,
 703         };
 704
 705         dev_info->tx_desc_lim = (struct rte_eth_desc_lim) {
 706                 .nb_max = ICE_MAX_RING_DESC,
 707                 .nb_min = ICE_MIN_RING_DESC,
 708                 .nb_align = ICE_ALIGN_RING_DESC,
 709         };
 710
 711         return 0;
 712 }
 713
 714 static int
 715 ice_dcf_dev_promiscuous_enable(__rte_unused struct rte_eth_dev *dev)
 716 {
 717         return 0;
 718 }
 719
 720 static int
 721 ice_dcf_dev_promiscuous_disable(__rte_unused struct rte_eth_dev *dev)
 722 {
 723         return 0;
 724 }
 725
 726 static int
 727 ice_dcf_dev_allmulticast_enable(__rte_unused struct rte_eth_dev *dev)
 728 {
 729         return 0;
 730 }
 731
 732 static int
 733 ice_dcf_dev_allmulticast_disable(__rte_unused struct rte_eth_dev *dev)
 734 {
 735         return 0;
 736 }
 737
 738 static int
 739 ice_dcf_dev_filter_ctrl(struct rte_eth_dev *dev,
 740                         enum rte_filter_type filter_type,
 741                         enum rte_filter_op filter_op,
 742                         void *arg)
 743 {
 744         int ret = 0;
 745
 746         if (!dev)
 747                 return -EINVAL;
 748
 749         switch (filter_type) {
 750         case RTE_ETH_FILTER_GENERIC:
 751                 if (filter_op != RTE_ETH_FILTER_GET)
 752                         return -EINVAL;
 753                 *(const void **)arg = &ice_flow_ops;
 754                 break;
 755
 756         default:
 757                 PMD_DRV_LOG(WARNING, "Filter type (%d) not supported",
 758                             filter_type);
 759                 ret = -EINVAL;
 760                 break;
 761         }
 762
 763         return ret;
 764 }
 765
 766 #define ICE_DCF_32_BIT_WIDTH (CHAR_BIT * 4)
 767 #define ICE_DCF_48_BIT_WIDTH (CHAR_BIT * 6)
 768 #define ICE_DCF_48_BIT_MASK  RTE_LEN2MASK(ICE_DCF_48_BIT_WIDTH, uint64_t)
 769
 770 static void
 771 ice_dcf_stat_update_48(uint64_t *offset, uint64_t *stat)
 772 {
 773         if (*stat >= *offset)
 774                 *stat = *stat - *offset;
 775         else
 776                 *stat = (uint64_t)((*stat +
 777                         ((uint64_t)1 << ICE_DCF_48_BIT_WIDTH)) - *offset);
 778
 779         *stat &= ICE_DCF_48_BIT_MASK;
 780 }
 781
 782 static void
 783 ice_dcf_stat_update_32(uint64_t *offset, uint64_t *stat)
 784 {
 785         if (*stat >= *offset)
 786                 *stat = (uint64_t)(*stat - *offset);
 787         else
 788                 *stat = (uint64_t)((*stat +
 789                         ((uint64_t)1 << ICE_DCF_32_BIT_WIDTH)) - *offset);
 790 }
 791
 792 static void
 793 ice_dcf_update_stats(struct virtchnl_eth_stats *oes,
 794                      struct virtchnl_eth_stats *nes)
 795 {
 796         ice_dcf_stat_update_48(&oes->rx_bytes, &nes->rx_bytes);
 797         ice_dcf_stat_update_48(&oes->rx_unicast, &nes->rx_unicast);
 798         ice_dcf_stat_update_48(&oes->rx_multicast, &nes->rx_multicast);
 799         ice_dcf_stat_update_48(&oes->rx_broadcast, &nes->rx_broadcast);
 800         ice_dcf_stat_update_32(&oes->rx_discards, &nes->rx_discards);
 801         ice_dcf_stat_update_48(&oes->tx_bytes, &nes->tx_bytes);
 802         ice_dcf_stat_update_48(&oes->tx_unicast, &nes->tx_unicast);
 803         ice_dcf_stat_update_48(&oes->tx_multicast, &nes->tx_multicast);
 804         ice_dcf_stat_update_48(&oes->tx_broadcast, &nes->tx_broadcast);
 805         ice_dcf_stat_update_32(&oes->tx_errors, &nes->tx_errors);
 806         ice_dcf_stat_update_32(&oes->tx_discards, &nes->tx_discards);
 807 }
 808
 809
 810 static int
 811 ice_dcf_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 812 {
 813         struct ice_dcf_adapter *ad = dev->data->dev_private;
 814         struct ice_dcf_hw *hw = &ad->real_hw;
 815         struct virtchnl_eth_stats pstats;
 816         int ret;
 817
 818         ret = ice_dcf_query_stats(hw, &pstats);
 819         if (ret == 0) {
 820                 ice_dcf_update_stats(&hw->eth_stats_offset, &pstats);
 821                 stats->ipackets = pstats.rx_unicast + pstats.rx_multicast +
 822                                 pstats.rx_broadcast - pstats.rx_discards;
 823                 stats->opackets = pstats.tx_broadcast + pstats.tx_multicast +
 824                                                 pstats.tx_unicast;
 825                 stats->imissed = pstats.rx_discards;
 826                 stats->oerrors = pstats.tx_errors + pstats.tx_discards;
 827                 stats->ibytes = pstats.rx_bytes;
 828                 stats->ibytes -= stats->ipackets * RTE_ETHER_CRC_LEN;
 829                 stats->obytes = pstats.tx_bytes;
 830         } else {
 831                 PMD_DRV_LOG(ERR, "Get statistics failed");
 832         }
 833         return ret;
 834 }
 835
 836 static int
 837 ice_dcf_stats_reset(struct rte_eth_dev *dev)
 838 {
 839         struct ice_dcf_adapter *ad = dev->data->dev_private;
 840         struct ice_dcf_hw *hw = &ad->real_hw;
 841         struct virtchnl_eth_stats pstats;
 842         int ret;
 843
 844         /* read stat values to clear hardware registers */
 845         ret = ice_dcf_query_stats(hw, &pstats);
 846         if (ret != 0)
 847                 return ret;
 848
 849         /* set stats offset base on current values */
 850         hw->eth_stats_offset = pstats;
 851
 852         return 0;
 853 }
 854
 855 static void
 856 ice_dcf_free_repr_info(struct ice_dcf_adapter *dcf_adapter)
 857 {
 858         if (dcf_adapter->repr_infos) {
 859                 rte_free(dcf_adapter->repr_infos);
 860                 dcf_adapter->repr_infos = NULL;
 861         }
 862 }
 863
 864 static int
 865 ice_dcf_init_repr_info(struct ice_dcf_adapter *dcf_adapter)
 866 {
 867         dcf_adapter->repr_infos =
 868                         rte_calloc("ice_dcf_rep_info",
 869                                    dcf_adapter->real_hw.num_vfs,
 870                                    sizeof(dcf_adapter->repr_infos[0]), 0);
 871         if (!dcf_adapter->repr_infos) {
 872                 PMD_DRV_LOG(ERR, "Failed to alloc memory for VF representors\n");
 873                 return -ENOMEM;
 874         }
 875
 876         return 0;
 877 }
 878
 879 static int
 880 ice_dcf_dev_close(struct rte_eth_dev *dev)
 881 {
 882         struct ice_dcf_adapter *adapter = dev->data->dev_private;
 883
 884         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 885                 return 0;
 886
 887         ice_dcf_free_repr_info(adapter);
 888         ice_dcf_uninit_parent_adapter(dev);
 889         ice_dcf_uninit_hw(dev, &adapter->real_hw);
 890
 891         return 0;
 892 }
 893
 894 static int
 895 ice_dcf_link_update(__rte_unused struct rte_eth_dev *dev,
 896                     __rte_unused int wait_to_complete)
 897 {
 898         return 0;
 899 }
 900
 901 static const struct eth_dev_ops ice_dcf_eth_dev_ops = {
 902         .dev_start               = ice_dcf_dev_start,
 903         .dev_stop                = ice_dcf_dev_stop,
 904         .dev_close               = ice_dcf_dev_close,
 905         .dev_configure           = ice_dcf_dev_configure,
 906         .dev_infos_get           = ice_dcf_dev_info_get,
 907         .rx_queue_setup          = ice_rx_queue_setup,
 908         .tx_queue_setup          = ice_tx_queue_setup,
 909         .rx_queue_release        = ice_rx_queue_release,
 910         .tx_queue_release        = ice_tx_queue_release,
 911         .rx_queue_start          = ice_dcf_rx_queue_start,
 912         .tx_queue_start          = ice_dcf_tx_queue_start,
 913         .rx_queue_stop           = ice_dcf_rx_queue_stop,
 914         .tx_queue_stop           = ice_dcf_tx_queue_stop,
 915         .link_update             = ice_dcf_link_update,
 916         .stats_get               = ice_dcf_stats_get,
 917         .stats_reset             = ice_dcf_stats_reset,
 918         .promiscuous_enable      = ice_dcf_dev_promiscuous_enable,
 919         .promiscuous_disable     = ice_dcf_dev_promiscuous_disable,
 920         .allmulticast_enable     = ice_dcf_dev_allmulticast_enable,
 921         .allmulticast_disable    = ice_dcf_dev_allmulticast_disable,
 922         .filter_ctrl             = ice_dcf_dev_filter_ctrl,
 923 };
 924
 925 static int
 926 ice_dcf_dev_init(struct rte_eth_dev *eth_dev)
 927 {
 928         struct ice_dcf_adapter *adapter = eth_dev->data->dev_private;
 929
 930         eth_dev->dev_ops = &ice_dcf_eth_dev_ops;
 931         eth_dev->rx_pkt_burst = ice_dcf_recv_pkts;
 932         eth_dev->tx_pkt_burst = ice_dcf_xmit_pkts;
 933
 934         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 935                 return 0;
 936
 937         eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
 938
 939         adapter->real_hw.vc_event_msg_cb = ice_dcf_handle_pf_event_msg;
 940         if (ice_dcf_init_hw(eth_dev, &adapter->real_hw) != 0) {
 941                 PMD_INIT_LOG(ERR, "Failed to init DCF hardware");
 942                 return -1;
 943         }
 944
 945         if (ice_dcf_init_parent_adapter(eth_dev) != 0) {
 946                 PMD_INIT_LOG(ERR, "Failed to init DCF parent adapter");
 947                 ice_dcf_uninit_hw(eth_dev, &adapter->real_hw);
 948                 return -1;
 949         }
 950
 951         return 0;
 952 }
 953
 954 static int
 955 ice_dcf_dev_uninit(struct rte_eth_dev *eth_dev)
 956 {
 957         ice_dcf_dev_close(eth_dev);
 958
 959         return 0;
 960 }
 961
 962 static int
 963 ice_dcf_cap_check_handler(__rte_unused const char *key,
 964                           const char *value, __rte_unused void *opaque)
 965 {
 966         if (strcmp(value, "dcf"))
 967                 return -1;
 968
 969         return 0;
 970 }
 971
 972 static int
 973 ice_dcf_cap_selected(struct rte_devargs *devargs)
 974 {
 975         struct rte_kvargs *kvlist;
 976         const char *key = "cap";
 977         int ret = 0;
 978
 979         if (devargs == NULL)
 980                 return 0;
 981
 982         kvlist = rte_kvargs_parse(devargs->args, NULL);
 983         if (kvlist == NULL)
 984                 return 0;
 985
 986         if (!rte_kvargs_count(kvlist, key))
 987                 goto exit;
 988
 989         /* dcf capability selected when there's a key-value pair: cap=dcf */
 990         if (rte_kvargs_process(kvlist, key,
 991                                ice_dcf_cap_check_handler, NULL) < 0)
 992                 goto exit;
 993
 994         ret = 1;
 995
 996 exit:
 997         rte_kvargs_free(kvlist);
 998         return ret;
 999 }
1000
1001 static int
1002 eth_ice_dcf_pci_probe(__rte_unused struct rte_pci_driver *pci_drv,
1003                       struct rte_pci_device *pci_dev)
1004 {
1005         struct rte_eth_devargs eth_da = { .nb_representor_ports = 0 };
1006         struct ice_dcf_vf_repr_param repr_param;
1007         char repr_name[RTE_ETH_NAME_MAX_LEN];
1008         struct ice_dcf_adapter *dcf_adapter;
1009         struct rte_eth_dev *dcf_ethdev;
1010         uint16_t dcf_vsi_id;
1011         int i, ret;
1012
1013         if (!ice_dcf_cap_selected(pci_dev->device.devargs))
1014                 return 1;
1015
1016         ret = rte_eth_devargs_parse(pci_dev->device.devargs->args, &eth_da);
1017         if (ret)
1018                 return ret;
1019
1020         ret = rte_eth_dev_pci_generic_probe(pci_dev,
1021                                             sizeof(struct ice_dcf_adapter),
1022                                             ice_dcf_dev_init);
1023         if (ret || !eth_da.nb_representor_ports)
1024                 return ret;
1025
1026         dcf_ethdev = rte_eth_dev_allocated(pci_dev->device.name);
1027         if (dcf_ethdev == NULL)
1028                 return -ENODEV;
1029
1030         dcf_adapter = dcf_ethdev->data->dev_private;
1031         ret = ice_dcf_init_repr_info(dcf_adapter);
1032         if (ret)
1033                 return ret;
1034
1035         if (eth_da.nb_representor_ports > dcf_adapter->real_hw.num_vfs ||
1036             eth_da.nb_representor_ports >= RTE_MAX_ETHPORTS) {
1037                 PMD_DRV_LOG(ERR, "the number of port representors is too large: %u",
1038                             eth_da.nb_representor_ports);
1039                 ice_dcf_free_repr_info(dcf_adapter);
1040                 return -EINVAL;
1041         }
1042
1043         dcf_vsi_id = dcf_adapter->real_hw.vsi_id | VIRTCHNL_DCF_VF_VSI_VALID;
1044
1045         repr_param.dcf_eth_dev = dcf_ethdev;
1046         repr_param.switch_domain_id = 0;
1047
1048         for (i = 0; i < eth_da.nb_representor_ports; i++) {
1049                 uint16_t vf_id = eth_da.representor_ports[i];
1050                 struct rte_eth_dev *vf_rep_eth_dev;
1051
1052                 if (vf_id >= dcf_adapter->real_hw.num_vfs) {
1053                         PMD_DRV_LOG(ERR, "VF ID %u is out of range (0 ~ %u)",
1054                                     vf_id, dcf_adapter->real_hw.num_vfs - 1);
1055                         ret = -EINVAL;
1056                         break;
1057                 }
1058
1059                 if (dcf_adapter->real_hw.vf_vsi_map[vf_id] == dcf_vsi_id) {
1060                         PMD_DRV_LOG(ERR, "VF ID %u is DCF's ID.\n", vf_id);
1061                         ret = -EINVAL;
1062                         break;
1063                 }
1064
1065                 repr_param.vf_id = vf_id;
1066                 snprintf(repr_name, sizeof(repr_name), "net_%s_representor_%u",
1067                          pci_dev->device.name, vf_id);
1068                 ret = rte_eth_dev_create(&pci_dev->device, repr_name,
1069                                          sizeof(struct ice_dcf_vf_repr),
1070                                          NULL, NULL, ice_dcf_vf_repr_init,
1071                                          &repr_param);
1072                 if (ret) {
1073                         PMD_DRV_LOG(ERR, "failed to create DCF VF representor %s",
1074                                     repr_name);
1075                         break;
1076                 }
1077
1078                 vf_rep_eth_dev = rte_eth_dev_allocated(repr_name);
1079                 if (!vf_rep_eth_dev) {
1080                         PMD_DRV_LOG(ERR,
1081                                     "Failed to find the ethdev for DCF VF representor: %s",
1082                                     repr_name);
1083                         ret = -ENODEV;
1084                         break;
1085                 }
1086
1087                 dcf_adapter->repr_infos[vf_id].vf_rep_eth_dev = vf_rep_eth_dev;
1088                 dcf_adapter->num_reprs++;
1089         }
1090
1091         return ret;
1092 }
1093
1094 static int
1095 eth_ice_dcf_pci_remove(struct rte_pci_device *pci_dev)
1096 {
1097         struct rte_eth_dev *eth_dev;
1098
1099         eth_dev = rte_eth_dev_allocated(pci_dev->device.name);
1100         if (!eth_dev)
1101                 return 0;
1102
1103         if (eth_dev->data->dev_flags & RTE_ETH_DEV_REPRESENTOR)
1104                 return rte_eth_dev_pci_generic_remove(pci_dev,
1105                                                       ice_dcf_vf_repr_uninit);
1106         else
1107                 return rte_eth_dev_pci_generic_remove(pci_dev,
1108                                                       ice_dcf_dev_uninit);
1109 }
1110
1111 static const struct rte_pci_id pci_id_ice_dcf_map[] = {
1112         { RTE_PCI_DEVICE(IAVF_INTEL_VENDOR_ID, IAVF_DEV_ID_ADAPTIVE_VF) },
1113         { .vendor_id = 0, /* sentinel */ },
1114 };
1115
1116 static struct rte_pci_driver rte_ice_dcf_pmd = {
1117         .id_table = pci_id_ice_dcf_map,
1118         .drv_flags = RTE_PCI_DRV_NEED_MAPPING,
1119         .probe = eth_ice_dcf_pci_probe,
1120         .remove = eth_ice_dcf_pci_remove,
1121 };
1122
1123 RTE_PMD_REGISTER_PCI(net_ice_dcf, rte_ice_dcf_pmd);
1124 RTE_PMD_REGISTER_PCI_TABLE(net_ice_dcf, pci_id_ice_dcf_map);
1125 RTE_PMD_REGISTER_KMOD_DEP(net_ice_dcf, "* igb_uio | vfio-pci");
1126 RTE_PMD_REGISTER_PARAM_STRING(net_ice_dcf, "cap=dcf");