drivers/net/vmxnet3/vmxnet3_rxtx.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2015 Intel Corporation
   3  */
   4
   5 #include <sys/queue.h>
   6
   7 #include <stdio.h>
   8 #include <stdlib.h>
   9 #include <string.h>
  10 #include <errno.h>
  11 #include <stdint.h>
  12 #include <stdarg.h>
  13 #include <unistd.h>
  14 #include <inttypes.h>
  15
  16 #include <rte_byteorder.h>
  17 #include <rte_common.h>
  18 #include <rte_cycles.h>
  19 #include <rte_log.h>
  20 #include <rte_debug.h>
  21 #include <rte_interrupts.h>
  22 #include <rte_pci.h>
  23 #include <rte_memory.h>
  24 #include <rte_memzone.h>
  25 #include <rte_launch.h>
  26 #include <rte_eal.h>
  27 #include <rte_per_lcore.h>
  28 #include <rte_lcore.h>
  29 #include <rte_atomic.h>
  30 #include <rte_branch_prediction.h>
  31 #include <rte_mempool.h>
  32 #include <rte_malloc.h>
  33 #include <rte_mbuf.h>
  34 #include <rte_ether.h>
  35 #include <rte_ethdev_driver.h>
  36 #include <rte_prefetch.h>
  37 #include <rte_ip.h>
  38 #include <rte_udp.h>
  39 #include <rte_tcp.h>
  40 #include <rte_sctp.h>
  41 #include <rte_string_fns.h>
  42 #include <rte_errno.h>
  43 #include <rte_net.h>
  44
  45 #include "base/vmxnet3_defs.h"
  46 #include "vmxnet3_ring.h"
  47
  48 #include "vmxnet3_logs.h"
  49 #include "vmxnet3_ethdev.h"
  50
  51 #define VMXNET3_TX_OFFLOAD_MASK ( \
  52                 PKT_TX_VLAN_PKT | \
  53                 PKT_TX_L4_MASK |  \
  54                 PKT_TX_TCP_SEG)
  55
  56 #define VMXNET3_TX_OFFLOAD_NOTSUP_MASK  \
  57         (PKT_TX_OFFLOAD_MASK ^ VMXNET3_TX_OFFLOAD_MASK)
  58
  59 static const uint32_t rxprod_reg[2] = {VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2};
  60
  61 static int vmxnet3_post_rx_bufs(vmxnet3_rx_queue_t*, uint8_t);
  62 static void vmxnet3_tq_tx_complete(vmxnet3_tx_queue_t *);
  63 #ifdef RTE_LIBRTE_VMXNET3_DEBUG_DRIVER_NOT_USED
  64 static void vmxnet3_rxq_dump(struct vmxnet3_rx_queue *);
  65 static void vmxnet3_txq_dump(struct vmxnet3_tx_queue *);
  66 #endif
  67
  68 #ifdef RTE_LIBRTE_VMXNET3_DEBUG_DRIVER_NOT_USED
  69 static void
  70 vmxnet3_rxq_dump(struct vmxnet3_rx_queue *rxq)
  71 {
  72         uint32_t avail = 0;
  73
  74         if (rxq == NULL)
  75                 return;
  76
  77         PMD_RX_LOG(DEBUG,
  78                    "RXQ: cmd0 base : %p cmd1 base : %p comp ring base : %p.",
  79                    rxq->cmd_ring[0].base, rxq->cmd_ring[1].base, rxq->comp_ring.base);
  80         PMD_RX_LOG(DEBUG,
  81                    "RXQ: cmd0 basePA : 0x%lx cmd1 basePA : 0x%lx comp ring basePA : 0x%lx.",
  82                    (unsigned long)rxq->cmd_ring[0].basePA,
  83                    (unsigned long)rxq->cmd_ring[1].basePA,
  84                    (unsigned long)rxq->comp_ring.basePA);
  85
  86         avail = vmxnet3_cmd_ring_desc_avail(&rxq->cmd_ring[0]);
  87         PMD_RX_LOG(DEBUG,
  88                    "RXQ:cmd0: size=%u; free=%u; next2proc=%u; queued=%u",
  89                    (uint32_t)rxq->cmd_ring[0].size, avail,
  90                    rxq->comp_ring.next2proc,
  91                    rxq->cmd_ring[0].size - avail);
  92
  93         avail = vmxnet3_cmd_ring_desc_avail(&rxq->cmd_ring[1]);
  94         PMD_RX_LOG(DEBUG, "RXQ:cmd1 size=%u; free=%u; next2proc=%u; queued=%u",
  95                    (uint32_t)rxq->cmd_ring[1].size, avail, rxq->comp_ring.next2proc,
  96                    rxq->cmd_ring[1].size - avail);
  97
  98 }
  99
 100 static void
 101 vmxnet3_txq_dump(struct vmxnet3_tx_queue *txq)
 102 {
 103         uint32_t avail = 0;
 104
 105         if (txq == NULL)
 106                 return;
 107
 108         PMD_TX_LOG(DEBUG, "TXQ: cmd base : %p comp ring base : %p data ring base : %p.",
 109                    txq->cmd_ring.base, txq->comp_ring.base, txq->data_ring.base);
 110         PMD_TX_LOG(DEBUG, "TXQ: cmd basePA : 0x%lx comp ring basePA : 0x%lx data ring basePA : 0x%lx.",
 111                    (unsigned long)txq->cmd_ring.basePA,
 112                    (unsigned long)txq->comp_ring.basePA,
 113                    (unsigned long)txq->data_ring.basePA);
 114
 115         avail = vmxnet3_cmd_ring_desc_avail(&txq->cmd_ring);
 116         PMD_TX_LOG(DEBUG, "TXQ: size=%u; free=%u; next2proc=%u; queued=%u",
 117                    (uint32_t)txq->cmd_ring.size, avail,
 118                    txq->comp_ring.next2proc, txq->cmd_ring.size - avail);
 119 }
 120 #endif
 121
 122 static void
 123 vmxnet3_tx_cmd_ring_release_mbufs(vmxnet3_cmd_ring_t *ring)
 124 {
 125         while (ring->next2comp != ring->next2fill) {
 126                 /* No need to worry about desc ownership, device is quiesced by now. */
 127                 vmxnet3_buf_info_t *buf_info = ring->buf_info + ring->next2comp;
 128
 129                 if (buf_info->m) {
 130                         rte_pktmbuf_free(buf_info->m);
 131                         buf_info->m = NULL;
 132                         buf_info->bufPA = 0;
 133                         buf_info->len = 0;
 134                 }
 135                 vmxnet3_cmd_ring_adv_next2comp(ring);
 136         }
 137 }
 138
 139 static void
 140 vmxnet3_rx_cmd_ring_release_mbufs(vmxnet3_cmd_ring_t *ring)
 141 {
 142         uint32_t i;
 143
 144         for (i = 0; i < ring->size; i++) {
 145                 /* No need to worry about desc ownership, device is quiesced by now. */
 146                 vmxnet3_buf_info_t *buf_info = &ring->buf_info[i];
 147
 148                 if (buf_info->m) {
 149                         rte_pktmbuf_free_seg(buf_info->m);
 150                         buf_info->m = NULL;
 151                         buf_info->bufPA = 0;
 152                         buf_info->len = 0;
 153                 }
 154                 vmxnet3_cmd_ring_adv_next2comp(ring);
 155         }
 156 }
 157
 158 static void
 159 vmxnet3_cmd_ring_release(vmxnet3_cmd_ring_t *ring)
 160 {
 161         rte_free(ring->buf_info);
 162         ring->buf_info = NULL;
 163 }
 164
 165 void
 166 vmxnet3_dev_tx_queue_release(void *txq)
 167 {
 168         vmxnet3_tx_queue_t *tq = txq;
 169
 170         if (tq != NULL) {
 171                 /* Release mbufs */
 172                 vmxnet3_tx_cmd_ring_release_mbufs(&tq->cmd_ring);
 173                 /* Release the cmd_ring */
 174                 vmxnet3_cmd_ring_release(&tq->cmd_ring);
 175                 /* Release the memzone */
 176                 rte_memzone_free(tq->mz);
 177                 /* Release the queue */
 178                 rte_free(tq);
 179         }
 180 }
 181
 182 void
 183 vmxnet3_dev_rx_queue_release(void *rxq)
 184 {
 185         int i;
 186         vmxnet3_rx_queue_t *rq = rxq;
 187
 188         if (rq != NULL) {
 189                 /* Release mbufs */
 190                 for (i = 0; i < VMXNET3_RX_CMDRING_SIZE; i++)
 191                         vmxnet3_rx_cmd_ring_release_mbufs(&rq->cmd_ring[i]);
 192
 193                 /* Release both the cmd_rings */
 194                 for (i = 0; i < VMXNET3_RX_CMDRING_SIZE; i++)
 195                         vmxnet3_cmd_ring_release(&rq->cmd_ring[i]);
 196
 197                 /* Release the memzone */
 198                 rte_memzone_free(rq->mz);
 199
 200                 /* Release the queue */
 201                 rte_free(rq);
 202         }
 203 }
 204
 205 static void
 206 vmxnet3_dev_tx_queue_reset(void *txq)
 207 {
 208         vmxnet3_tx_queue_t *tq = txq;
 209         struct vmxnet3_cmd_ring *ring = &tq->cmd_ring;
 210         struct vmxnet3_comp_ring *comp_ring = &tq->comp_ring;
 211         struct vmxnet3_data_ring *data_ring = &tq->data_ring;
 212         int size;
 213
 214         if (tq != NULL) {
 215                 /* Release the cmd_ring mbufs */
 216                 vmxnet3_tx_cmd_ring_release_mbufs(&tq->cmd_ring);
 217         }
 218
 219         /* Tx vmxnet rings structure initialization*/
 220         ring->next2fill = 0;
 221         ring->next2comp = 0;
 222         ring->gen = VMXNET3_INIT_GEN;
 223         comp_ring->next2proc = 0;
 224         comp_ring->gen = VMXNET3_INIT_GEN;
 225
 226         size = sizeof(struct Vmxnet3_TxDesc) * ring->size;
 227         size += sizeof(struct Vmxnet3_TxCompDesc) * comp_ring->size;
 228         size += tq->txdata_desc_size * data_ring->size;
 229
 230         memset(ring->base, 0, size);
 231 }
 232
 233 static void
 234 vmxnet3_dev_rx_queue_reset(void *rxq)
 235 {
 236         int i;
 237         vmxnet3_rx_queue_t *rq = rxq;
 238         struct vmxnet3_hw *hw = rq->hw;
 239         struct vmxnet3_cmd_ring *ring0, *ring1;
 240         struct vmxnet3_comp_ring *comp_ring;
 241         struct vmxnet3_rx_data_ring *data_ring = &rq->data_ring;
 242         int size;
 243
 244         /* Release both the cmd_rings mbufs */
 245         for (i = 0; i < VMXNET3_RX_CMDRING_SIZE; i++)
 246                 vmxnet3_rx_cmd_ring_release_mbufs(&rq->cmd_ring[i]);
 247
 248         ring0 = &rq->cmd_ring[0];
 249         ring1 = &rq->cmd_ring[1];
 250         comp_ring = &rq->comp_ring;
 251
 252         /* Rx vmxnet rings structure initialization */
 253         ring0->next2fill = 0;
 254         ring1->next2fill = 0;
 255         ring0->next2comp = 0;
 256         ring1->next2comp = 0;
 257         ring0->gen = VMXNET3_INIT_GEN;
 258         ring1->gen = VMXNET3_INIT_GEN;
 259         comp_ring->next2proc = 0;
 260         comp_ring->gen = VMXNET3_INIT_GEN;
 261
 262         size = sizeof(struct Vmxnet3_RxDesc) * (ring0->size + ring1->size);
 263         size += sizeof(struct Vmxnet3_RxCompDesc) * comp_ring->size;
 264         if (VMXNET3_VERSION_GE_3(hw) && rq->data_desc_size)
 265                 size += rq->data_desc_size * data_ring->size;
 266
 267         memset(ring0->base, 0, size);
 268 }
 269
 270 void
 271 vmxnet3_dev_clear_queues(struct rte_eth_dev *dev)
 272 {
 273         unsigned i;
 274
 275         PMD_INIT_FUNC_TRACE();
 276
 277         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 278                 struct vmxnet3_tx_queue *txq = dev->data->tx_queues[i];
 279
 280                 if (txq != NULL) {
 281                         txq->stopped = TRUE;
 282                         vmxnet3_dev_tx_queue_reset(txq);
 283                 }
 284         }
 285
 286         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 287                 struct vmxnet3_rx_queue *rxq = dev->data->rx_queues[i];
 288
 289                 if (rxq != NULL) {
 290                         rxq->stopped = TRUE;
 291                         vmxnet3_dev_rx_queue_reset(rxq);
 292                 }
 293         }
 294 }
 295
 296 static int
 297 vmxnet3_unmap_pkt(uint16_t eop_idx, vmxnet3_tx_queue_t *txq)
 298 {
 299         int completed = 0;
 300         struct rte_mbuf *mbuf;
 301
 302         /* Release cmd_ring descriptor and free mbuf */
 303         RTE_ASSERT(txq->cmd_ring.base[eop_idx].txd.eop == 1);
 304
 305         mbuf = txq->cmd_ring.buf_info[eop_idx].m;
 306         if (mbuf == NULL)
 307                 rte_panic("EOP desc does not point to a valid mbuf");
 308         rte_pktmbuf_free(mbuf);
 309
 310         txq->cmd_ring.buf_info[eop_idx].m = NULL;
 311
 312         while (txq->cmd_ring.next2comp != eop_idx) {
 313                 /* no out-of-order completion */
 314                 RTE_ASSERT(txq->cmd_ring.base[txq->cmd_ring.next2comp].txd.cq == 0);
 315                 vmxnet3_cmd_ring_adv_next2comp(&txq->cmd_ring);
 316                 completed++;
 317         }
 318
 319         /* Mark the txd for which tcd was generated as completed */
 320         vmxnet3_cmd_ring_adv_next2comp(&txq->cmd_ring);
 321
 322         return completed + 1;
 323 }
 324
 325 static void
 326 vmxnet3_tq_tx_complete(vmxnet3_tx_queue_t *txq)
 327 {
 328         int completed = 0;
 329         vmxnet3_comp_ring_t *comp_ring = &txq->comp_ring;
 330         struct Vmxnet3_TxCompDesc *tcd = (struct Vmxnet3_TxCompDesc *)
 331                 (comp_ring->base + comp_ring->next2proc);
 332
 333         while (tcd->gen == comp_ring->gen) {
 334                 completed += vmxnet3_unmap_pkt(tcd->txdIdx, txq);
 335
 336                 vmxnet3_comp_ring_adv_next2proc(comp_ring);
 337                 tcd = (struct Vmxnet3_TxCompDesc *)(comp_ring->base +
 338                                                     comp_ring->next2proc);
 339         }
 340
 341         PMD_TX_LOG(DEBUG, "Processed %d tx comps & command descs.", completed);
 342 }
 343
 344 uint16_t
 345 vmxnet3_prep_pkts(__rte_unused void *tx_queue, struct rte_mbuf **tx_pkts,
 346         uint16_t nb_pkts)
 347 {
 348         int32_t ret;
 349         uint32_t i;
 350         uint64_t ol_flags;
 351         struct rte_mbuf *m;
 352
 353         for (i = 0; i != nb_pkts; i++) {
 354                 m = tx_pkts[i];
 355                 ol_flags = m->ol_flags;
 356
 357                 /* Non-TSO packet cannot occupy more than
 358                  * VMXNET3_MAX_TXD_PER_PKT TX descriptors.
 359                  */
 360                 if ((ol_flags & PKT_TX_TCP_SEG) == 0 &&
 361                                 m->nb_segs > VMXNET3_MAX_TXD_PER_PKT) {
 362                         rte_errno = -EINVAL;
 363                         return i;
 364                 }
 365
 366                 /* check that only supported TX offloads are requested. */
 367                 if ((ol_flags & VMXNET3_TX_OFFLOAD_NOTSUP_MASK) != 0 ||
 368                                 (ol_flags & PKT_TX_L4_MASK) ==
 369                                 PKT_TX_SCTP_CKSUM) {
 370                         rte_errno = -ENOTSUP;
 371                         return i;
 372                 }
 373
 374 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
 375                 ret = rte_validate_tx_offload(m);
 376                 if (ret != 0) {
 377                         rte_errno = ret;
 378                         return i;
 379                 }
 380 #endif
 381                 ret = rte_net_intel_cksum_prepare(m);
 382                 if (ret != 0) {
 383                         rte_errno = ret;
 384                         return i;
 385                 }
 386         }
 387
 388         return i;
 389 }
 390
 391 uint16_t
 392 vmxnet3_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
 393                   uint16_t nb_pkts)
 394 {
 395         uint16_t nb_tx;
 396         vmxnet3_tx_queue_t *txq = tx_queue;
 397         struct vmxnet3_hw *hw = txq->hw;
 398         Vmxnet3_TxQueueCtrl *txq_ctrl = &txq->shared->ctrl;
 399         uint32_t deferred = rte_le_to_cpu_32(txq_ctrl->txNumDeferred);
 400
 401         if (unlikely(txq->stopped)) {
 402                 PMD_TX_LOG(DEBUG, "Tx queue is stopped.");
 403                 return 0;
 404         }
 405
 406         /* Free up the comp_descriptors aggressively */
 407         vmxnet3_tq_tx_complete(txq);
 408
 409         nb_tx = 0;
 410         while (nb_tx < nb_pkts) {
 411                 Vmxnet3_GenericDesc *gdesc;
 412                 vmxnet3_buf_info_t *tbi;
 413                 uint32_t first2fill, avail, dw2;
 414                 struct rte_mbuf *txm = tx_pkts[nb_tx];
 415                 struct rte_mbuf *m_seg = txm;
 416                 int copy_size = 0;
 417                 bool tso = (txm->ol_flags & PKT_TX_TCP_SEG) != 0;
 418                 /* # of descriptors needed for a packet. */
 419                 unsigned count = txm->nb_segs;
 420
 421                 avail = vmxnet3_cmd_ring_desc_avail(&txq->cmd_ring);
 422                 if (count > avail) {
 423                         /* Is command ring full? */
 424                         if (unlikely(avail == 0)) {
 425                                 PMD_TX_LOG(DEBUG, "No free ring descriptors");
 426                                 txq->stats.tx_ring_full++;
 427                                 txq->stats.drop_total += (nb_pkts - nb_tx);
 428                                 break;
 429                         }
 430
 431                         /* Command ring is not full but cannot handle the
 432                          * multi-segmented packet. Let's try the next packet
 433                          * in this case.
 434                          */
 435                         PMD_TX_LOG(DEBUG, "Running out of ring descriptors "
 436                                    "(avail %d needed %d)", avail, count);
 437                         txq->stats.drop_total++;
 438                         if (tso)
 439                                 txq->stats.drop_tso++;
 440                         rte_pktmbuf_free(txm);
 441                         nb_tx++;
 442                         continue;
 443                 }
 444
 445                 /* Drop non-TSO packet that is excessively fragmented */
 446                 if (unlikely(!tso && count > VMXNET3_MAX_TXD_PER_PKT)) {
 447                         PMD_TX_LOG(ERR, "Non-TSO packet cannot occupy more than %d tx "
 448                                    "descriptors. Packet dropped.", VMXNET3_MAX_TXD_PER_PKT);
 449                         txq->stats.drop_too_many_segs++;
 450                         txq->stats.drop_total++;
 451                         rte_pktmbuf_free(txm);
 452                         nb_tx++;
 453                         continue;
 454                 }
 455
 456                 if (txm->nb_segs == 1 &&
 457                     rte_pktmbuf_pkt_len(txm) <= txq->txdata_desc_size) {
 458                         struct Vmxnet3_TxDataDesc *tdd;
 459
 460                         tdd = (struct Vmxnet3_TxDataDesc *)
 461                                 ((uint8 *)txq->data_ring.base +
 462                                  txq->cmd_ring.next2fill *
 463                                  txq->txdata_desc_size);
 464                         copy_size = rte_pktmbuf_pkt_len(txm);
 465                         rte_memcpy(tdd->data, rte_pktmbuf_mtod(txm, char *), copy_size);
 466                 }
 467
 468                 /* use the previous gen bit for the SOP desc */
 469                 dw2 = (txq->cmd_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT;
 470                 first2fill = txq->cmd_ring.next2fill;
 471                 do {
 472                         /* Remember the transmit buffer for cleanup */
 473                         tbi = txq->cmd_ring.buf_info + txq->cmd_ring.next2fill;
 474
 475                         /* NB: the following assumes that VMXNET3 maximum
 476                          * transmit buffer size (16K) is greater than
 477                          * maximum size of mbuf segment size.
 478                          */
 479                         gdesc = txq->cmd_ring.base + txq->cmd_ring.next2fill;
 480                         if (copy_size) {
 481                                 uint64 offset =
 482                                         (uint64)txq->cmd_ring.next2fill *
 483                                                         txq->txdata_desc_size;
 484                                 gdesc->txd.addr =
 485                                         rte_cpu_to_le_64(txq->data_ring.basePA +
 486                                                          offset);
 487                         } else {
 488                                 gdesc->txd.addr = rte_mbuf_data_iova(m_seg);
 489                         }
 490
 491                         gdesc->dword[2] = dw2 | m_seg->data_len;
 492                         gdesc->dword[3] = 0;
 493
 494                         /* move to the next2fill descriptor */
 495                         vmxnet3_cmd_ring_adv_next2fill(&txq->cmd_ring);
 496
 497                         /* use the right gen for non-SOP desc */
 498                         dw2 = txq->cmd_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 499                 } while ((m_seg = m_seg->next) != NULL);
 500
 501                 /* set the last buf_info for the pkt */
 502                 tbi->m = txm;
 503                 /* Update the EOP descriptor */
 504                 gdesc->dword[3] |= VMXNET3_TXD_EOP | VMXNET3_TXD_CQ;
 505
 506                 /* Add VLAN tag if present */
 507                 gdesc = txq->cmd_ring.base + first2fill;
 508                 if (txm->ol_flags & PKT_TX_VLAN_PKT) {
 509                         gdesc->txd.ti = 1;
 510                         gdesc->txd.tci = txm->vlan_tci;
 511                 }
 512
 513                 if (tso) {
 514                         uint16_t mss = txm->tso_segsz;
 515
 516                         RTE_ASSERT(mss > 0);
 517
 518                         gdesc->txd.hlen = txm->l2_len + txm->l3_len + txm->l4_len;
 519                         gdesc->txd.om = VMXNET3_OM_TSO;
 520                         gdesc->txd.msscof = mss;
 521
 522                         deferred += (rte_pktmbuf_pkt_len(txm) - gdesc->txd.hlen + mss - 1) / mss;
 523                 } else if (txm->ol_flags & PKT_TX_L4_MASK) {
 524                         gdesc->txd.om = VMXNET3_OM_CSUM;
 525                         gdesc->txd.hlen = txm->l2_len + txm->l3_len;
 526
 527                         switch (txm->ol_flags & PKT_TX_L4_MASK) {
 528                         case PKT_TX_TCP_CKSUM:
 529                                 gdesc->txd.msscof = gdesc->txd.hlen + offsetof(struct tcp_hdr, cksum);
 530                                 break;
 531                         case PKT_TX_UDP_CKSUM:
 532                                 gdesc->txd.msscof = gdesc->txd.hlen + offsetof(struct udp_hdr, dgram_cksum);
 533                                 break;
 534                         default:
 535                                 PMD_TX_LOG(WARNING, "requested cksum offload not supported %#llx",
 536                                            txm->ol_flags & PKT_TX_L4_MASK);
 537                                 abort();
 538                         }
 539                         deferred++;
 540                 } else {
 541                         gdesc->txd.hlen = 0;
 542                         gdesc->txd.om = VMXNET3_OM_NONE;
 543                         gdesc->txd.msscof = 0;
 544                         deferred++;
 545                 }
 546
 547                 /* flip the GEN bit on the SOP */
 548                 rte_compiler_barrier();
 549                 gdesc->dword[2] ^= VMXNET3_TXD_GEN;
 550
 551                 txq_ctrl->txNumDeferred = rte_cpu_to_le_32(deferred);
 552                 nb_tx++;
 553         }
 554
 555         PMD_TX_LOG(DEBUG, "vmxnet3 txThreshold: %u", rte_le_to_cpu_32(txq_ctrl->txThreshold));
 556
 557         if (deferred >= rte_le_to_cpu_32(txq_ctrl->txThreshold)) {
 558                 txq_ctrl->txNumDeferred = 0;
 559                 /* Notify vSwitch that packets are available. */
 560                 VMXNET3_WRITE_BAR0_REG(hw, (VMXNET3_REG_TXPROD + txq->queue_id * VMXNET3_REG_ALIGN),
 561                                        txq->cmd_ring.next2fill);
 562         }
 563
 564         return nb_tx;
 565 }
 566
 567 static inline void
 568 vmxnet3_renew_desc(vmxnet3_rx_queue_t *rxq, uint8_t ring_id,
 569                    struct rte_mbuf *mbuf)
 570 {
 571         uint32_t val;
 572         struct vmxnet3_cmd_ring *ring = &rxq->cmd_ring[ring_id];
 573         struct Vmxnet3_RxDesc *rxd =
 574                 (struct Vmxnet3_RxDesc *)(ring->base + ring->next2fill);
 575         vmxnet3_buf_info_t *buf_info = &ring->buf_info[ring->next2fill];
 576
 577         if (ring_id == 0) {
 578                 /* Usually: One HEAD type buf per packet
 579                  * val = (ring->next2fill % rxq->hw->bufs_per_pkt) ?
 580                  * VMXNET3_RXD_BTYPE_BODY : VMXNET3_RXD_BTYPE_HEAD;
 581                  */
 582
 583                 /* We use single packet buffer so all heads here */
 584                 val = VMXNET3_RXD_BTYPE_HEAD;
 585         } else {
 586                 /* All BODY type buffers for 2nd ring */
 587                 val = VMXNET3_RXD_BTYPE_BODY;
 588         }
 589
 590         /*
 591          * Load mbuf pointer into buf_info[ring_size]
 592          * buf_info structure is equivalent to cookie for virtio-virtqueue
 593          */
 594         buf_info->m = mbuf;
 595         buf_info->len = (uint16_t)(mbuf->buf_len - RTE_PKTMBUF_HEADROOM);
 596         buf_info->bufPA = rte_mbuf_data_iova_default(mbuf);
 597
 598         /* Load Rx Descriptor with the buffer's GPA */
 599         rxd->addr = buf_info->bufPA;
 600
 601         /* After this point rxd->addr MUST not be NULL */
 602         rxd->btype = val;
 603         rxd->len = buf_info->len;
 604         /* Flip gen bit at the end to change ownership */
 605         rxd->gen = ring->gen;
 606
 607         vmxnet3_cmd_ring_adv_next2fill(ring);
 608 }
 609 /*
 610  *  Allocates mbufs and clusters. Post rx descriptors with buffer details
 611  *  so that device can receive packets in those buffers.
 612  *  Ring layout:
 613  *      Among the two rings, 1st ring contains buffers of type 0 and type 1.
 614  *      bufs_per_pkt is set such that for non-LRO cases all the buffers required
 615  *      by a frame will fit in 1st ring (1st buf of type0 and rest of type1).
 616  *      2nd ring contains buffers of type 1 alone. Second ring mostly be used
 617  *      only for LRO.
 618  */
 619 static int
 620 vmxnet3_post_rx_bufs(vmxnet3_rx_queue_t *rxq, uint8_t ring_id)
 621 {
 622         int err = 0;
 623         uint32_t i = 0;
 624         struct vmxnet3_cmd_ring *ring = &rxq->cmd_ring[ring_id];
 625
 626         while (vmxnet3_cmd_ring_desc_avail(ring) > 0) {
 627                 struct rte_mbuf *mbuf;
 628
 629                 /* Allocate blank mbuf for the current Rx Descriptor */
 630                 mbuf = rte_mbuf_raw_alloc(rxq->mp);
 631                 if (unlikely(mbuf == NULL)) {
 632                         PMD_RX_LOG(ERR, "Error allocating mbuf");
 633                         rxq->stats.rx_buf_alloc_failure++;
 634                         err = ENOMEM;
 635                         break;
 636                 }
 637
 638                 vmxnet3_renew_desc(rxq, ring_id, mbuf);
 639                 i++;
 640         }
 641
 642         /* Return error only if no buffers are posted at present */
 643         if (vmxnet3_cmd_ring_desc_avail(ring) >= (ring->size - 1))
 644                 return -err;
 645         else
 646                 return i;
 647 }
 648
 649 /* MSS not provided by vmxnet3, guess one with available information */
 650 static uint16_t
 651 vmxnet3_guess_mss(struct vmxnet3_hw *hw, const Vmxnet3_RxCompDesc *rcd,
 652                 struct rte_mbuf *rxm)
 653 {
 654         uint32_t hlen, slen;
 655         struct ipv4_hdr *ipv4_hdr;
 656         struct ipv6_hdr *ipv6_hdr;
 657         struct tcp_hdr *tcp_hdr;
 658         char *ptr;
 659
 660         RTE_ASSERT(rcd->tcp);
 661
 662         ptr = rte_pktmbuf_mtod(rxm, char *);
 663         slen = rte_pktmbuf_data_len(rxm);
 664         hlen = sizeof(struct ether_hdr);
 665
 666         if (rcd->v4) {
 667                 if (unlikely(slen < hlen + sizeof(struct ipv4_hdr)))
 668                         return hw->mtu - sizeof(struct ipv4_hdr)
 669                                         - sizeof(struct tcp_hdr);
 670
 671                 ipv4_hdr = (struct ipv4_hdr *)(ptr + hlen);
 672                 hlen += (ipv4_hdr->version_ihl & IPV4_HDR_IHL_MASK) *
 673                                 IPV4_IHL_MULTIPLIER;
 674         } else if (rcd->v6) {
 675                 if (unlikely(slen < hlen + sizeof(struct ipv6_hdr)))
 676                         return hw->mtu - sizeof(struct ipv6_hdr) -
 677                                         sizeof(struct tcp_hdr);
 678
 679                 ipv6_hdr = (struct ipv6_hdr *)(ptr + hlen);
 680                 hlen += sizeof(struct ipv6_hdr);
 681                 if (unlikely(ipv6_hdr->proto != IPPROTO_TCP)) {
 682                         int frag;
 683
 684                         rte_net_skip_ip6_ext(ipv6_hdr->proto, rxm,
 685                                         &hlen, &frag);
 686                 }
 687         }
 688
 689         if (unlikely(slen < hlen + sizeof(struct tcp_hdr)))
 690                 return hw->mtu - hlen - sizeof(struct tcp_hdr) +
 691                                 sizeof(struct ether_hdr);
 692
 693         tcp_hdr = (struct tcp_hdr *)(ptr + hlen);
 694         hlen += (tcp_hdr->data_off & 0xf0) >> 2;
 695
 696         if (rxm->udata64 > 1)
 697                 return (rte_pktmbuf_pkt_len(rxm) - hlen +
 698                                 rxm->udata64 - 1) / rxm->udata64;
 699         else
 700                 return hw->mtu - hlen + sizeof(struct ether_hdr);
 701 }
 702
 703 /* Receive side checksum and other offloads */
 704 static inline void
 705 vmxnet3_rx_offload(struct vmxnet3_hw *hw, const Vmxnet3_RxCompDesc *rcd,
 706                 struct rte_mbuf *rxm, const uint8_t sop)
 707 {
 708         uint64_t ol_flags = rxm->ol_flags;
 709         uint32_t packet_type = rxm->packet_type;
 710
 711         /* Offloads set in sop */
 712         if (sop) {
 713                 /* Set packet type */
 714                 packet_type |= RTE_PTYPE_L2_ETHER;
 715
 716                 /* Check large packet receive */
 717                 if (VMXNET3_VERSION_GE_2(hw) &&
 718                     rcd->type == VMXNET3_CDTYPE_RXCOMP_LRO) {
 719                         const Vmxnet3_RxCompDescExt *rcde =
 720                                         (const Vmxnet3_RxCompDescExt *)rcd;
 721
 722                         rxm->tso_segsz = rcde->mss;
 723                         rxm->udata64 = rcde->segCnt;
 724                         ol_flags |= PKT_RX_LRO;
 725                 }
 726         } else { /* Offloads set in eop */
 727                 /* Check for RSS */
 728                 if (rcd->rssType != VMXNET3_RCD_RSS_TYPE_NONE) {
 729                         ol_flags |= PKT_RX_RSS_HASH;
 730                         rxm->hash.rss = rcd->rssHash;
 731                 }
 732
 733                 /* Check for hardware stripped VLAN tag */
 734                 if (rcd->ts) {
 735                         ol_flags |= (PKT_RX_VLAN | PKT_RX_VLAN_STRIPPED);
 736                         rxm->vlan_tci = rte_le_to_cpu_16((uint16_t)rcd->tci);
 737                 }
 738
 739                 /* Check packet type, checksum errors, etc. */
 740                 if (rcd->cnc) {
 741                         ol_flags |= PKT_RX_L4_CKSUM_UNKNOWN;
 742                 } else {
 743                         if (rcd->v4) {
 744                                 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
 745
 746                                 if (rcd->ipc)
 747                                         ol_flags |= PKT_RX_IP_CKSUM_GOOD;
 748                                 else
 749                                         ol_flags |= PKT_RX_IP_CKSUM_BAD;
 750
 751                                 if (rcd->tuc) {
 752                                         ol_flags |= PKT_RX_L4_CKSUM_GOOD;
 753                                         if (rcd->tcp)
 754                                                 packet_type |= RTE_PTYPE_L4_TCP;
 755                                         else
 756                                                 packet_type |= RTE_PTYPE_L4_UDP;
 757                                 } else {
 758                                         if (rcd->tcp) {
 759                                                 packet_type |= RTE_PTYPE_L4_TCP;
 760                                                 ol_flags |= PKT_RX_L4_CKSUM_BAD;
 761                                         } else if (rcd->udp) {
 762                                                 packet_type |= RTE_PTYPE_L4_UDP;
 763                                                 ol_flags |= PKT_RX_L4_CKSUM_BAD;
 764                                         }
 765                                 }
 766                         } else if (rcd->v6) {
 767                                 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
 768
 769                                 if (rcd->tuc) {
 770                                         ol_flags |= PKT_RX_L4_CKSUM_GOOD;
 771                                         if (rcd->tcp)
 772                                                 packet_type |= RTE_PTYPE_L4_TCP;
 773                                         else
 774                                                 packet_type |= RTE_PTYPE_L4_UDP;
 775                                 } else {
 776                                         if (rcd->tcp) {
 777                                                 packet_type |= RTE_PTYPE_L4_TCP;
 778                                                 ol_flags |= PKT_RX_L4_CKSUM_BAD;
 779                                         } else if (rcd->udp) {
 780                                                 packet_type |= RTE_PTYPE_L4_UDP;
 781                                                 ol_flags |= PKT_RX_L4_CKSUM_BAD;
 782                                         }
 783                                 }
 784                         } else {
 785                                 packet_type |= RTE_PTYPE_UNKNOWN;
 786                         }
 787
 788                         /* Old variants of vmxnet3 do not provide MSS */
 789                         if ((ol_flags & PKT_RX_LRO) && rxm->tso_segsz == 0)
 790                                 rxm->tso_segsz = vmxnet3_guess_mss(hw,
 791                                                 rcd, rxm);
 792                 }
 793         }
 794
 795         rxm->ol_flags = ol_flags;
 796         rxm->packet_type = packet_type;
 797 }
 798
 799 /*
 800  * Process the Rx Completion Ring of given vmxnet3_rx_queue
 801  * for nb_pkts burst and return the number of packets received
 802  */
 803 uint16_t
 804 vmxnet3_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
 805 {
 806         uint16_t nb_rx;
 807         uint32_t nb_rxd, idx;
 808         uint8_t ring_idx;
 809         vmxnet3_rx_queue_t *rxq;
 810         Vmxnet3_RxCompDesc *rcd;
 811         vmxnet3_buf_info_t *rbi;
 812         Vmxnet3_RxDesc *rxd;
 813         struct rte_mbuf *rxm = NULL;
 814         struct vmxnet3_hw *hw;
 815
 816         nb_rx = 0;
 817         ring_idx = 0;
 818         nb_rxd = 0;
 819         idx = 0;
 820
 821         rxq = rx_queue;
 822         hw = rxq->hw;
 823
 824         rcd = &rxq->comp_ring.base[rxq->comp_ring.next2proc].rcd;
 825
 826         if (unlikely(rxq->stopped)) {
 827                 PMD_RX_LOG(DEBUG, "Rx queue is stopped.");
 828                 return 0;
 829         }
 830
 831         while (rcd->gen == rxq->comp_ring.gen) {
 832                 struct rte_mbuf *newm;
 833
 834                 if (nb_rx >= nb_pkts)
 835                         break;
 836
 837                 newm = rte_mbuf_raw_alloc(rxq->mp);
 838                 if (unlikely(newm == NULL)) {
 839                         PMD_RX_LOG(ERR, "Error allocating mbuf");
 840                         rxq->stats.rx_buf_alloc_failure++;
 841                         break;
 842                 }
 843
 844                 idx = rcd->rxdIdx;
 845                 ring_idx = vmxnet3_get_ring_idx(hw, rcd->rqID);
 846                 rxd = (Vmxnet3_RxDesc *)rxq->cmd_ring[ring_idx].base + idx;
 847                 RTE_SET_USED(rxd); /* used only for assert when enabled */
 848                 rbi = rxq->cmd_ring[ring_idx].buf_info + idx;
 849
 850                 PMD_RX_LOG(DEBUG, "rxd idx: %d ring idx: %d.", idx, ring_idx);
 851
 852                 RTE_ASSERT(rcd->len <= rxd->len);
 853                 RTE_ASSERT(rbi->m);
 854
 855                 /* Get the packet buffer pointer from buf_info */
 856                 rxm = rbi->m;
 857
 858                 /* Clear descriptor associated buf_info to be reused */
 859                 rbi->m = NULL;
 860                 rbi->bufPA = 0;
 861
 862                 /* Update the index that we received a packet */
 863                 rxq->cmd_ring[ring_idx].next2comp = idx;
 864
 865                 /* For RCD with EOP set, check if there is frame error */
 866                 if (unlikely(rcd->eop && rcd->err)) {
 867                         rxq->stats.drop_total++;
 868                         rxq->stats.drop_err++;
 869
 870                         if (!rcd->fcs) {
 871                                 rxq->stats.drop_fcs++;
 872                                 PMD_RX_LOG(ERR, "Recv packet dropped due to frame err.");
 873                         }
 874                         PMD_RX_LOG(ERR, "Error in received packet rcd#:%d rxd:%d",
 875                                    (int)(rcd - (struct Vmxnet3_RxCompDesc *)
 876                                          rxq->comp_ring.base), rcd->rxdIdx);
 877                         rte_pktmbuf_free_seg(rxm);
 878                         if (rxq->start_seg) {
 879                                 struct rte_mbuf *start = rxq->start_seg;
 880
 881                                 rxq->start_seg = NULL;
 882                                 rte_pktmbuf_free(start);
 883                         }
 884                         goto rcd_done;
 885                 }
 886
 887                 /* Initialize newly received packet buffer */
 888                 rxm->port = rxq->port_id;
 889                 rxm->nb_segs = 1;
 890                 rxm->next = NULL;
 891                 rxm->pkt_len = (uint16_t)rcd->len;
 892                 rxm->data_len = (uint16_t)rcd->len;
 893                 rxm->data_off = RTE_PKTMBUF_HEADROOM;
 894                 rxm->ol_flags = 0;
 895                 rxm->vlan_tci = 0;
 896                 rxm->packet_type = 0;
 897
 898                 /*
 899                  * If this is the first buffer of the received packet,
 900                  * set the pointer to the first mbuf of the packet
 901                  * Otherwise, update the total length and the number of segments
 902                  * of the current scattered packet, and update the pointer to
 903                  * the last mbuf of the current packet.
 904                  */
 905                 if (rcd->sop) {
 906                         RTE_ASSERT(rxd->btype == VMXNET3_RXD_BTYPE_HEAD);
 907
 908                         if (unlikely(rcd->len == 0)) {
 909                                 RTE_ASSERT(rcd->eop);
 910
 911                                 PMD_RX_LOG(DEBUG,
 912                                            "Rx buf was skipped. rxring[%d][%d])",
 913                                            ring_idx, idx);
 914                                 rte_pktmbuf_free_seg(rxm);
 915                                 goto rcd_done;
 916                         }
 917
 918                         if (vmxnet3_rx_data_ring(hw, rcd->rqID)) {
 919                                 uint8_t *rdd = rxq->data_ring.base +
 920                                         idx * rxq->data_desc_size;
 921
 922                                 RTE_ASSERT(VMXNET3_VERSION_GE_3(hw));
 923                                 rte_memcpy(rte_pktmbuf_mtod(rxm, char *),
 924                                            rdd, rcd->len);
 925                         }
 926
 927                         rxq->start_seg = rxm;
 928                         vmxnet3_rx_offload(hw, rcd, rxm, 1);
 929                 } else {
 930                         struct rte_mbuf *start = rxq->start_seg;
 931
 932                         RTE_ASSERT(rxd->btype == VMXNET3_RXD_BTYPE_BODY);
 933
 934                         start->pkt_len += rxm->data_len;
 935                         start->nb_segs++;
 936
 937                         rxq->last_seg->next = rxm;
 938                 }
 939                 rxq->last_seg = rxm;
 940
 941                 if (rcd->eop) {
 942                         struct rte_mbuf *start = rxq->start_seg;
 943
 944                         vmxnet3_rx_offload(hw, rcd, start, 0);
 945                         rx_pkts[nb_rx++] = start;
 946                         rxq->start_seg = NULL;
 947                 }
 948
 949 rcd_done:
 950                 rxq->cmd_ring[ring_idx].next2comp = idx;
 951                 VMXNET3_INC_RING_IDX_ONLY(rxq->cmd_ring[ring_idx].next2comp,
 952                                           rxq->cmd_ring[ring_idx].size);
 953
 954                 /* It's time to renew descriptors */
 955                 vmxnet3_renew_desc(rxq, ring_idx, newm);
 956                 if (unlikely(rxq->shared->ctrl.updateRxProd)) {
 957                         VMXNET3_WRITE_BAR0_REG(hw, rxprod_reg[ring_idx] + (rxq->queue_id * VMXNET3_REG_ALIGN),
 958                                                rxq->cmd_ring[ring_idx].next2fill);
 959                 }
 960
 961                 /* Advance to the next descriptor in comp_ring */
 962                 vmxnet3_comp_ring_adv_next2proc(&rxq->comp_ring);
 963
 964                 rcd = &rxq->comp_ring.base[rxq->comp_ring.next2proc].rcd;
 965                 nb_rxd++;
 966                 if (nb_rxd > rxq->cmd_ring[0].size) {
 967                         PMD_RX_LOG(ERR, "Used up quota of receiving packets,"
 968                                    " relinquish control.");
 969                         break;
 970                 }
 971         }
 972
 973         if (unlikely(nb_rxd == 0)) {
 974                 uint32_t avail;
 975                 for (ring_idx = 0; ring_idx < VMXNET3_RX_CMDRING_SIZE; ring_idx++) {
 976                         avail = vmxnet3_cmd_ring_desc_avail(&rxq->cmd_ring[ring_idx]);
 977                         if (unlikely(avail > 0)) {
 978                                 /* try to alloc new buf and renew descriptors */
 979                                 vmxnet3_post_rx_bufs(rxq, ring_idx);
 980                         }
 981                 }
 982                 if (unlikely(rxq->shared->ctrl.updateRxProd)) {
 983                         for (ring_idx = 0; ring_idx < VMXNET3_RX_CMDRING_SIZE; ring_idx++) {
 984                                 VMXNET3_WRITE_BAR0_REG(hw, rxprod_reg[ring_idx] + (rxq->queue_id * VMXNET3_REG_ALIGN),
 985                                                        rxq->cmd_ring[ring_idx].next2fill);
 986                         }
 987                 }
 988         }
 989
 990         return nb_rx;
 991 }
 992
 993 int
 994 vmxnet3_dev_tx_queue_setup(struct rte_eth_dev *dev,
 995                            uint16_t queue_idx,
 996                            uint16_t nb_desc,
 997                            unsigned int socket_id,
 998                            const struct rte_eth_txconf *tx_conf)
 999 {
1000         struct vmxnet3_hw *hw = dev->data->dev_private;
1001         const struct rte_memzone *mz;
1002         struct vmxnet3_tx_queue *txq;
1003         struct vmxnet3_cmd_ring *ring;
1004         struct vmxnet3_comp_ring *comp_ring;
1005         struct vmxnet3_data_ring *data_ring;
1006         int size;
1007
1008         PMD_INIT_FUNC_TRACE();
1009
1010         if ((tx_conf->txq_flags & ETH_TXQ_FLAGS_NOXSUMSCTP) !=
1011             ETH_TXQ_FLAGS_NOXSUMSCTP) {
1012                 PMD_INIT_LOG(ERR, "SCTP checksum offload not supported");
1013                 return -EINVAL;
1014         }
1015
1016         txq = rte_zmalloc("ethdev_tx_queue", sizeof(struct vmxnet3_tx_queue),
1017                           RTE_CACHE_LINE_SIZE);
1018         if (txq == NULL) {
1019                 PMD_INIT_LOG(ERR, "Can not allocate tx queue structure");
1020                 return -ENOMEM;
1021         }
1022
1023         txq->queue_id = queue_idx;
1024         txq->port_id = dev->data->port_id;
1025         txq->shared = NULL; /* set in vmxnet3_setup_driver_shared() */
1026         txq->hw = hw;
1027         txq->qid = queue_idx;
1028         txq->stopped = TRUE;
1029         txq->txdata_desc_size = hw->txdata_desc_size;
1030
1031         ring = &txq->cmd_ring;
1032         comp_ring = &txq->comp_ring;
1033         data_ring = &txq->data_ring;
1034
1035         /* Tx vmxnet ring length should be between 512-4096 */
1036         if (nb_desc < VMXNET3_DEF_TX_RING_SIZE) {
1037                 PMD_INIT_LOG(ERR, "VMXNET3 Tx Ring Size Min: %u",
1038                              VMXNET3_DEF_TX_RING_SIZE);
1039                 return -EINVAL;
1040         } else if (nb_desc > VMXNET3_TX_RING_MAX_SIZE) {
1041                 PMD_INIT_LOG(ERR, "VMXNET3 Tx Ring Size Max: %u",
1042                              VMXNET3_TX_RING_MAX_SIZE);
1043                 return -EINVAL;
1044         } else {
1045                 ring->size = nb_desc;
1046                 ring->size &= ~VMXNET3_RING_SIZE_MASK;
1047         }
1048         comp_ring->size = data_ring->size = ring->size;
1049
1050         /* Tx vmxnet rings structure initialization*/
1051         ring->next2fill = 0;
1052         ring->next2comp = 0;
1053         ring->gen = VMXNET3_INIT_GEN;
1054         comp_ring->next2proc = 0;
1055         comp_ring->gen = VMXNET3_INIT_GEN;
1056
1057         size = sizeof(struct Vmxnet3_TxDesc) * ring->size;
1058         size += sizeof(struct Vmxnet3_TxCompDesc) * comp_ring->size;
1059         size += txq->txdata_desc_size * data_ring->size;
1060
1061         mz = rte_eth_dma_zone_reserve(dev, "txdesc", queue_idx, size,
1062                                       VMXNET3_RING_BA_ALIGN, socket_id);
1063         if (mz == NULL) {
1064                 PMD_INIT_LOG(ERR, "ERROR: Creating queue descriptors zone");
1065                 return -ENOMEM;
1066         }
1067         txq->mz = mz;
1068         memset(mz->addr, 0, mz->len);
1069
1070         /* cmd_ring initialization */
1071         ring->base = mz->addr;
1072         ring->basePA = mz->iova;
1073
1074         /* comp_ring initialization */
1075         comp_ring->base = ring->base + ring->size;
1076         comp_ring->basePA = ring->basePA +
1077                 (sizeof(struct Vmxnet3_TxDesc) * ring->size);
1078
1079         /* data_ring initialization */
1080         data_ring->base = (Vmxnet3_TxDataDesc *)(comp_ring->base + comp_ring->size);
1081         data_ring->basePA = comp_ring->basePA +
1082                         (sizeof(struct Vmxnet3_TxCompDesc) * comp_ring->size);
1083
1084         /* cmd_ring0 buf_info allocation */
1085         ring->buf_info = rte_zmalloc("tx_ring_buf_info",
1086                                      ring->size * sizeof(vmxnet3_buf_info_t), RTE_CACHE_LINE_SIZE);
1087         if (ring->buf_info == NULL) {
1088                 PMD_INIT_LOG(ERR, "ERROR: Creating tx_buf_info structure");
1089                 return -ENOMEM;
1090         }
1091
1092         /* Update the data portion with txq */
1093         dev->data->tx_queues[queue_idx] = txq;
1094
1095         return 0;
1096 }
1097
1098 int
1099 vmxnet3_dev_rx_queue_setup(struct rte_eth_dev *dev,
1100                            uint16_t queue_idx,
1101                            uint16_t nb_desc,
1102                            unsigned int socket_id,
1103                            __rte_unused const struct rte_eth_rxconf *rx_conf,
1104                            struct rte_mempool *mp)
1105 {
1106         const struct rte_memzone *mz;
1107         struct vmxnet3_rx_queue *rxq;
1108         struct vmxnet3_hw *hw = dev->data->dev_private;
1109         struct vmxnet3_cmd_ring *ring0, *ring1, *ring;
1110         struct vmxnet3_comp_ring *comp_ring;
1111         struct vmxnet3_rx_data_ring *data_ring;
1112         int size;
1113         uint8_t i;
1114         char mem_name[32];
1115
1116         PMD_INIT_FUNC_TRACE();
1117
1118         rxq = rte_zmalloc("ethdev_rx_queue", sizeof(struct vmxnet3_rx_queue),
1119                           RTE_CACHE_LINE_SIZE);
1120         if (rxq == NULL) {
1121                 PMD_INIT_LOG(ERR, "Can not allocate rx queue structure");
1122                 return -ENOMEM;
1123         }
1124
1125         rxq->mp = mp;
1126         rxq->queue_id = queue_idx;
1127         rxq->port_id = dev->data->port_id;
1128         rxq->shared = NULL; /* set in vmxnet3_setup_driver_shared() */
1129         rxq->hw = hw;
1130         rxq->qid1 = queue_idx;
1131         rxq->qid2 = queue_idx + hw->num_rx_queues;
1132         rxq->data_ring_qid = queue_idx + 2 * hw->num_rx_queues;
1133         rxq->data_desc_size = hw->rxdata_desc_size;
1134         rxq->stopped = TRUE;
1135
1136         ring0 = &rxq->cmd_ring[0];
1137         ring1 = &rxq->cmd_ring[1];
1138         comp_ring = &rxq->comp_ring;
1139         data_ring = &rxq->data_ring;
1140
1141         /* Rx vmxnet rings length should be between 256-4096 */
1142         if (nb_desc < VMXNET3_DEF_RX_RING_SIZE) {
1143                 PMD_INIT_LOG(ERR, "VMXNET3 Rx Ring Size Min: 256");
1144                 return -EINVAL;
1145         } else if (nb_desc > VMXNET3_RX_RING_MAX_SIZE) {
1146                 PMD_INIT_LOG(ERR, "VMXNET3 Rx Ring Size Max: 4096");
1147                 return -EINVAL;
1148         } else {
1149                 ring0->size = nb_desc;
1150                 ring0->size &= ~VMXNET3_RING_SIZE_MASK;
1151                 ring1->size = ring0->size;
1152         }
1153
1154         comp_ring->size = ring0->size + ring1->size;
1155         data_ring->size = ring0->size;
1156
1157         /* Rx vmxnet rings structure initialization */
1158         ring0->next2fill = 0;
1159         ring1->next2fill = 0;
1160         ring0->next2comp = 0;
1161         ring1->next2comp = 0;
1162         ring0->gen = VMXNET3_INIT_GEN;
1163         ring1->gen = VMXNET3_INIT_GEN;
1164         comp_ring->next2proc = 0;
1165         comp_ring->gen = VMXNET3_INIT_GEN;
1166
1167         size = sizeof(struct Vmxnet3_RxDesc) * (ring0->size + ring1->size);
1168         size += sizeof(struct Vmxnet3_RxCompDesc) * comp_ring->size;
1169         if (VMXNET3_VERSION_GE_3(hw) && rxq->data_desc_size)
1170                 size += rxq->data_desc_size * data_ring->size;
1171
1172         mz = rte_eth_dma_zone_reserve(dev, "rxdesc", queue_idx, size,
1173                                       VMXNET3_RING_BA_ALIGN, socket_id);
1174         if (mz == NULL) {
1175                 PMD_INIT_LOG(ERR, "ERROR: Creating queue descriptors zone");
1176                 return -ENOMEM;
1177         }
1178         rxq->mz = mz;
1179         memset(mz->addr, 0, mz->len);
1180
1181         /* cmd_ring0 initialization */
1182         ring0->base = mz->addr;
1183         ring0->basePA = mz->iova;
1184
1185         /* cmd_ring1 initialization */
1186         ring1->base = ring0->base + ring0->size;
1187         ring1->basePA = ring0->basePA + sizeof(struct Vmxnet3_RxDesc) * ring0->size;
1188
1189         /* comp_ring initialization */
1190         comp_ring->base = ring1->base + ring1->size;
1191         comp_ring->basePA = ring1->basePA + sizeof(struct Vmxnet3_RxDesc) *
1192                 ring1->size;
1193
1194         /* data_ring initialization */
1195         if (VMXNET3_VERSION_GE_3(hw) && rxq->data_desc_size) {
1196                 data_ring->base =
1197                         (uint8_t *)(comp_ring->base + comp_ring->size);
1198                 data_ring->basePA = comp_ring->basePA +
1199                         sizeof(struct Vmxnet3_RxCompDesc) * comp_ring->size;
1200         }
1201
1202         /* cmd_ring0-cmd_ring1 buf_info allocation */
1203         for (i = 0; i < VMXNET3_RX_CMDRING_SIZE; i++) {
1204
1205                 ring = &rxq->cmd_ring[i];
1206                 ring->rid = i;
1207                 snprintf(mem_name, sizeof(mem_name), "rx_ring_%d_buf_info", i);
1208
1209                 ring->buf_info = rte_zmalloc(mem_name,
1210                                              ring->size * sizeof(vmxnet3_buf_info_t),
1211                                              RTE_CACHE_LINE_SIZE);
1212                 if (ring->buf_info == NULL) {
1213                         PMD_INIT_LOG(ERR, "ERROR: Creating rx_buf_info structure");
1214                         return -ENOMEM;
1215                 }
1216         }
1217
1218         /* Update the data portion with rxq */
1219         dev->data->rx_queues[queue_idx] = rxq;
1220
1221         return 0;
1222 }
1223
1224 /*
1225  * Initializes Receive Unit
1226  * Load mbufs in rx queue in advance
1227  */
1228 int
1229 vmxnet3_dev_rxtx_init(struct rte_eth_dev *dev)
1230 {
1231         struct vmxnet3_hw *hw = dev->data->dev_private;
1232
1233         int i, ret;
1234         uint8_t j;
1235
1236         PMD_INIT_FUNC_TRACE();
1237
1238         for (i = 0; i < hw->num_rx_queues; i++) {
1239                 vmxnet3_rx_queue_t *rxq = dev->data->rx_queues[i];
1240
1241                 for (j = 0; j < VMXNET3_RX_CMDRING_SIZE; j++) {
1242                         /* Passing 0 as alloc_num will allocate full ring */
1243                         ret = vmxnet3_post_rx_bufs(rxq, j);
1244                         if (ret <= 0) {
1245                                 PMD_INIT_LOG(ERR,
1246                                              "ERROR: Posting Rxq: %d buffers ring: %d",
1247                                              i, j);
1248                                 return -ret;
1249                         }
1250                         /*
1251                          * Updating device with the index:next2fill to fill the
1252                          * mbufs for coming packets.
1253                          */
1254                         if (unlikely(rxq->shared->ctrl.updateRxProd)) {
1255                                 VMXNET3_WRITE_BAR0_REG(hw, rxprod_reg[j] + (rxq->queue_id * VMXNET3_REG_ALIGN),
1256                                                        rxq->cmd_ring[j].next2fill);
1257                         }
1258                 }
1259                 rxq->stopped = FALSE;
1260                 rxq->start_seg = NULL;
1261         }
1262
1263         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1264                 struct vmxnet3_tx_queue *txq = dev->data->tx_queues[i];
1265
1266                 txq->stopped = FALSE;
1267         }
1268
1269         return 0;
1270 }
1271
1272 static uint8_t rss_intel_key[40] = {
1273         0x6D, 0x5A, 0x56, 0xDA, 0x25, 0x5B, 0x0E, 0xC2,
1274         0x41, 0x67, 0x25, 0x3D, 0x43, 0xA3, 0x8F, 0xB0,
1275         0xD0, 0xCA, 0x2B, 0xCB, 0xAE, 0x7B, 0x30, 0xB4,
1276         0x77, 0xCB, 0x2D, 0xA3, 0x80, 0x30, 0xF2, 0x0C,
1277         0x6A, 0x42, 0xB7, 0x3B, 0xBE, 0xAC, 0x01, 0xFA,
1278 };
1279
1280 /*
1281  * Configure RSS feature
1282  */
1283 int
1284 vmxnet3_rss_configure(struct rte_eth_dev *dev)
1285 {
1286         struct vmxnet3_hw *hw = dev->data->dev_private;
1287         struct VMXNET3_RSSConf *dev_rss_conf;
1288         struct rte_eth_rss_conf *port_rss_conf;
1289         uint64_t rss_hf;
1290         uint8_t i, j;
1291
1292         PMD_INIT_FUNC_TRACE();
1293
1294         dev_rss_conf = hw->rss_conf;
1295         port_rss_conf = &dev->data->dev_conf.rx_adv_conf.rss_conf;
1296
1297         /* loading hashFunc */
1298         dev_rss_conf->hashFunc = VMXNET3_RSS_HASH_FUNC_TOEPLITZ;
1299         /* loading hashKeySize */
1300         dev_rss_conf->hashKeySize = VMXNET3_RSS_MAX_KEY_SIZE;
1301         /* loading indTableSize: Must not exceed VMXNET3_RSS_MAX_IND_TABLE_SIZE (128)*/
1302         dev_rss_conf->indTableSize = (uint16_t)(hw->num_rx_queues * 4);
1303
1304         if (port_rss_conf->rss_key == NULL) {
1305                 /* Default hash key */
1306                 port_rss_conf->rss_key = rss_intel_key;
1307         }
1308
1309         /* loading hashKey */
1310         memcpy(&dev_rss_conf->hashKey[0], port_rss_conf->rss_key,
1311                dev_rss_conf->hashKeySize);
1312
1313         /* loading indTable */
1314         for (i = 0, j = 0; i < dev_rss_conf->indTableSize; i++, j++) {
1315                 if (j == dev->data->nb_rx_queues)
1316                         j = 0;
1317                 dev_rss_conf->indTable[i] = j;
1318         }
1319
1320         /* loading hashType */
1321         dev_rss_conf->hashType = 0;
1322         rss_hf = port_rss_conf->rss_hf & VMXNET3_RSS_OFFLOAD_ALL;
1323         if (rss_hf & ETH_RSS_IPV4)
1324                 dev_rss_conf->hashType |= VMXNET3_RSS_HASH_TYPE_IPV4;
1325         if (rss_hf & ETH_RSS_NONFRAG_IPV4_TCP)
1326                 dev_rss_conf->hashType |= VMXNET3_RSS_HASH_TYPE_TCP_IPV4;
1327         if (rss_hf & ETH_RSS_IPV6)
1328                 dev_rss_conf->hashType |= VMXNET3_RSS_HASH_TYPE_IPV6;
1329         if (rss_hf & ETH_RSS_NONFRAG_IPV6_TCP)
1330                 dev_rss_conf->hashType |= VMXNET3_RSS_HASH_TYPE_TCP_IPV6;
1331
1332         return VMXNET3_SUCCESS;
1333 }