drivers/net/enic/enic_main.c

   1 /*
   2  * Copyright 2008-2014 Cisco Systems, Inc.  All rights reserved.
   3  * Copyright 2007 Nuova Systems, Inc.  All rights reserved.
   4  *
   5  * Copyright (c) 2014, Cisco Systems, Inc.
   6  * All rights reserved.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  *
  12  * 1. Redistributions of source code must retain the above copyright
  13  * notice, this list of conditions and the following disclaimer.
  14  *
  15  * 2. Redistributions in binary form must reproduce the above copyright
  16  * notice, this list of conditions and the following disclaimer in
  17  * the documentation and/or other materials provided with the
  18  * distribution.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  21  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
  24  * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  25  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
  26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
  28  * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
  30  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  31  * POSSIBILITY OF SUCH DAMAGE.
  32  *
  33  */
  34
  35 #include <stdio.h>
  36
  37 #include <sys/stat.h>
  38 #include <sys/mman.h>
  39 #include <fcntl.h>
  40 #include <libgen.h>
  41
  42 #include <rte_pci.h>
  43 #include <rte_memzone.h>
  44 #include <rte_malloc.h>
  45 #include <rte_mbuf.h>
  46 #include <rte_string_fns.h>
  47 #include <rte_ethdev.h>
  48
  49 #include "enic_compat.h"
  50 #include "enic.h"
  51 #include "wq_enet_desc.h"
  52 #include "rq_enet_desc.h"
  53 #include "cq_enet_desc.h"
  54 #include "vnic_enet.h"
  55 #include "vnic_dev.h"
  56 #include "vnic_wq.h"
  57 #include "vnic_rq.h"
  58 #include "vnic_cq.h"
  59 #include "vnic_intr.h"
  60 #include "vnic_nic.h"
  61
  62 static inline int enic_is_sriov_vf(struct enic *enic)
  63 {
  64         return enic->pdev->id.device_id == PCI_DEVICE_ID_CISCO_VIC_ENET_VF;
  65 }
  66
  67 static int is_zero_addr(uint8_t *addr)
  68 {
  69         return !(addr[0] |  addr[1] | addr[2] | addr[3] | addr[4] | addr[5]);
  70 }
  71
  72 static int is_mcast_addr(uint8_t *addr)
  73 {
  74         return addr[0] & 1;
  75 }
  76
  77 static int is_eth_addr_valid(uint8_t *addr)
  78 {
  79         return !is_mcast_addr(addr) && !is_zero_addr(addr);
  80 }
  81
  82 static void
  83 enic_rxmbuf_queue_release(__rte_unused struct enic *enic, struct vnic_rq *rq)
  84 {
  85         uint16_t i;
  86
  87         if (!rq || !rq->mbuf_ring) {
  88                 dev_debug(enic, "Pointer to rq or mbuf_ring is NULL");
  89                 return;
  90         }
  91
  92         for (i = 0; i < rq->ring.desc_count; i++) {
  93                 if (rq->mbuf_ring[i]) {
  94                         rte_pktmbuf_free_seg(rq->mbuf_ring[i]);
  95                         rq->mbuf_ring[i] = NULL;
  96                 }
  97         }
  98 }
  99
 100 void enic_set_hdr_split_size(struct enic *enic, u16 split_hdr_size)
 101 {
 102         vnic_set_hdr_split_size(enic->vdev, split_hdr_size);
 103 }
 104
 105 static void enic_free_wq_buf(struct vnic_wq_buf *buf)
 106 {
 107         struct rte_mbuf *mbuf = (struct rte_mbuf *)buf->mb;
 108
 109         rte_pktmbuf_free_seg(mbuf);
 110         buf->mb = NULL;
 111 }
 112
 113 static void enic_log_q_error(struct enic *enic)
 114 {
 115         unsigned int i;
 116         u32 error_status;
 117
 118         for (i = 0; i < enic->wq_count; i++) {
 119                 error_status = vnic_wq_error_status(&enic->wq[i]);
 120                 if (error_status)
 121                         dev_err(enic, "WQ[%d] error_status %d\n", i,
 122                                 error_status);
 123         }
 124
 125         for (i = 0; i < enic_vnic_rq_count(enic); i++) {
 126                 if (!enic->rq[i].in_use)
 127                         continue;
 128                 error_status = vnic_rq_error_status(&enic->rq[i]);
 129                 if (error_status)
 130                         dev_err(enic, "RQ[%d] error_status %d\n", i,
 131                                 error_status);
 132         }
 133 }
 134
 135 static void enic_clear_soft_stats(struct enic *enic)
 136 {
 137         struct enic_soft_stats *soft_stats = &enic->soft_stats;
 138         rte_atomic64_clear(&soft_stats->rx_nombuf);
 139         rte_atomic64_clear(&soft_stats->rx_packet_errors);
 140 }
 141
 142 static void enic_init_soft_stats(struct enic *enic)
 143 {
 144         struct enic_soft_stats *soft_stats = &enic->soft_stats;
 145         rte_atomic64_init(&soft_stats->rx_nombuf);
 146         rte_atomic64_init(&soft_stats->rx_packet_errors);
 147         enic_clear_soft_stats(enic);
 148 }
 149
 150 void enic_dev_stats_clear(struct enic *enic)
 151 {
 152         if (vnic_dev_stats_clear(enic->vdev))
 153                 dev_err(enic, "Error in clearing stats\n");
 154         enic_clear_soft_stats(enic);
 155 }
 156
 157 void enic_dev_stats_get(struct enic *enic, struct rte_eth_stats *r_stats)
 158 {
 159         struct vnic_stats *stats;
 160         struct enic_soft_stats *soft_stats = &enic->soft_stats;
 161         int64_t rx_truncated;
 162         uint64_t rx_packet_errors;
 163
 164         if (vnic_dev_stats_dump(enic->vdev, &stats)) {
 165                 dev_err(enic, "Error in getting stats\n");
 166                 return;
 167         }
 168
 169         /* The number of truncated packets can only be calculated by
 170          * subtracting a hardware counter from error packets received by
 171          * the driver. Note: this causes transient inaccuracies in the
 172          * ipackets count. Also, the length of truncated packets are
 173          * counted in ibytes even though truncated packets are dropped
 174          * which can make ibytes be slightly higher than it should be.
 175          */
 176         rx_packet_errors = rte_atomic64_read(&soft_stats->rx_packet_errors);
 177         rx_truncated = rx_packet_errors - stats->rx.rx_errors;
 178
 179         r_stats->ipackets = stats->rx.rx_frames_ok - rx_truncated;
 180         r_stats->opackets = stats->tx.tx_frames_ok;
 181
 182         r_stats->ibytes = stats->rx.rx_bytes_ok;
 183         r_stats->obytes = stats->tx.tx_bytes_ok;
 184
 185         r_stats->ierrors = stats->rx.rx_errors + stats->rx.rx_drop;
 186         r_stats->oerrors = stats->tx.tx_errors;
 187
 188         r_stats->imissed = stats->rx.rx_no_bufs + rx_truncated;
 189
 190         r_stats->rx_nombuf = rte_atomic64_read(&soft_stats->rx_nombuf);
 191 }
 192
 193 void enic_del_mac_address(struct enic *enic, int mac_index)
 194 {
 195         struct rte_eth_dev *eth_dev = enic->rte_dev;
 196         uint8_t *mac_addr = eth_dev->data->mac_addrs[mac_index].addr_bytes;
 197
 198         if (vnic_dev_del_addr(enic->vdev, mac_addr))
 199                 dev_err(enic, "del mac addr failed\n");
 200 }
 201
 202 void enic_set_mac_address(struct enic *enic, uint8_t *mac_addr)
 203 {
 204         int err;
 205
 206         if (!is_eth_addr_valid(mac_addr)) {
 207                 dev_err(enic, "invalid mac address\n");
 208                 return;
 209         }
 210
 211         err = vnic_dev_add_addr(enic->vdev, mac_addr);
 212         if (err) {
 213                 dev_err(enic, "add mac addr failed\n");
 214                 return;
 215         }
 216 }
 217
 218 static void
 219 enic_free_rq_buf(struct rte_mbuf **mbuf)
 220 {
 221         if (*mbuf == NULL)
 222                 return;
 223
 224         rte_pktmbuf_free(*mbuf);
 225         mbuf = NULL;
 226 }
 227
 228 void enic_init_vnic_resources(struct enic *enic)
 229 {
 230         unsigned int error_interrupt_enable = 1;
 231         unsigned int error_interrupt_offset = 0;
 232         unsigned int index = 0;
 233         unsigned int cq_idx;
 234         struct vnic_rq *data_rq;
 235
 236         for (index = 0; index < enic->rq_count; index++) {
 237                 cq_idx = enic_cq_rq(enic, enic_rte_rq_idx_to_sop_idx(index));
 238
 239                 vnic_rq_init(&enic->rq[enic_rte_rq_idx_to_sop_idx(index)],
 240                         cq_idx,
 241                         error_interrupt_enable,
 242                         error_interrupt_offset);
 243
 244                 data_rq = &enic->rq[enic_rte_rq_idx_to_data_idx(index)];
 245                 if (data_rq->in_use)
 246                         vnic_rq_init(data_rq,
 247                                      cq_idx,
 248                                      error_interrupt_enable,
 249                                      error_interrupt_offset);
 250
 251                 vnic_cq_init(&enic->cq[cq_idx],
 252                         0 /* flow_control_enable */,
 253                         1 /* color_enable */,
 254                         0 /* cq_head */,
 255                         0 /* cq_tail */,
 256                         1 /* cq_tail_color */,
 257                         0 /* interrupt_enable */,
 258                         1 /* cq_entry_enable */,
 259                         0 /* cq_message_enable */,
 260                         0 /* interrupt offset */,
 261                         0 /* cq_message_addr */);
 262         }
 263
 264         for (index = 0; index < enic->wq_count; index++) {
 265                 vnic_wq_init(&enic->wq[index],
 266                         enic_cq_wq(enic, index),
 267                         error_interrupt_enable,
 268                         error_interrupt_offset);
 269
 270                 cq_idx = enic_cq_wq(enic, index);
 271                 vnic_cq_init(&enic->cq[cq_idx],
 272                         0 /* flow_control_enable */,
 273                         1 /* color_enable */,
 274                         0 /* cq_head */,
 275                         0 /* cq_tail */,
 276                         1 /* cq_tail_color */,
 277                         0 /* interrupt_enable */,
 278                         0 /* cq_entry_enable */,
 279                         1 /* cq_message_enable */,
 280                         0 /* interrupt offset */,
 281                         (u64)enic->wq[index].cqmsg_rz->phys_addr);
 282         }
 283
 284         vnic_intr_init(&enic->intr,
 285                 enic->config.intr_timer_usec,
 286                 enic->config.intr_timer_type,
 287                 /*mask_on_assertion*/1);
 288 }
 289
 290
 291 static int
 292 enic_alloc_rx_queue_mbufs(struct enic *enic, struct vnic_rq *rq)
 293 {
 294         struct rte_mbuf *mb;
 295         struct rq_enet_desc *rqd = rq->ring.descs;
 296         unsigned i;
 297         dma_addr_t dma_addr;
 298
 299         if (!rq->in_use)
 300                 return 0;
 301
 302         dev_debug(enic, "queue %u, allocating %u rx queue mbufs\n", rq->index,
 303                   rq->ring.desc_count);
 304
 305         for (i = 0; i < rq->ring.desc_count; i++, rqd++) {
 306                 mb = rte_mbuf_raw_alloc(rq->mp);
 307                 if (mb == NULL) {
 308                         dev_err(enic, "RX mbuf alloc failed queue_id=%u\n",
 309                         (unsigned)rq->index);
 310                         return -ENOMEM;
 311                 }
 312
 313                 mb->data_off = RTE_PKTMBUF_HEADROOM;
 314                 dma_addr = (dma_addr_t)(mb->buf_physaddr
 315                            + RTE_PKTMBUF_HEADROOM);
 316                 rq_enet_desc_enc(rqd, dma_addr,
 317                                 (rq->is_sop ? RQ_ENET_TYPE_ONLY_SOP
 318                                 : RQ_ENET_TYPE_NOT_SOP),
 319                                 mb->buf_len - RTE_PKTMBUF_HEADROOM);
 320                 rq->mbuf_ring[i] = mb;
 321         }
 322
 323         /* make sure all prior writes are complete before doing the PIO write */
 324         rte_rmb();
 325
 326         /* Post all but the last buffer to VIC. */
 327         rq->posted_index = rq->ring.desc_count - 1;
 328
 329         rq->rx_nb_hold = 0;
 330
 331         dev_debug(enic, "port=%u, qidx=%u, Write %u posted idx, %u sw held\n",
 332                 enic->port_id, rq->index, rq->posted_index, rq->rx_nb_hold);
 333         iowrite32(rq->posted_index, &rq->ctrl->posted_index);
 334         iowrite32(0, &rq->ctrl->fetch_index);
 335         rte_rmb();
 336
 337         return 0;
 338
 339 }
 340
 341 static void *
 342 enic_alloc_consistent(void *priv, size_t size,
 343         dma_addr_t *dma_handle, u8 *name)
 344 {
 345         void *vaddr;
 346         const struct rte_memzone *rz;
 347         *dma_handle = 0;
 348         struct enic *enic = (struct enic *)priv;
 349         struct enic_memzone_entry *mze;
 350
 351         rz = rte_memzone_reserve_aligned((const char *)name,
 352                                          size, SOCKET_ID_ANY, 0, ENIC_ALIGN);
 353         if (!rz) {
 354                 pr_err("%s : Failed to allocate memory requested for %s\n",
 355                         __func__, name);
 356                 return NULL;
 357         }
 358
 359         vaddr = rz->addr;
 360         *dma_handle = (dma_addr_t)rz->phys_addr;
 361
 362         mze = rte_malloc("enic memzone entry",
 363                          sizeof(struct enic_memzone_entry), 0);
 364
 365         if (!mze) {
 366                 pr_err("%s : Failed to allocate memory for memzone list\n",
 367                        __func__);
 368                 rte_memzone_free(rz);
 369         }
 370
 371         mze->rz = rz;
 372
 373         rte_spinlock_lock(&enic->memzone_list_lock);
 374         LIST_INSERT_HEAD(&enic->memzone_list, mze, entries);
 375         rte_spinlock_unlock(&enic->memzone_list_lock);
 376
 377         return vaddr;
 378 }
 379
 380 static void
 381 enic_free_consistent(void *priv,
 382                      __rte_unused size_t size,
 383                      void *vaddr,
 384                      dma_addr_t dma_handle)
 385 {
 386         struct enic_memzone_entry *mze;
 387         struct enic *enic = (struct enic *)priv;
 388
 389         rte_spinlock_lock(&enic->memzone_list_lock);
 390         LIST_FOREACH(mze, &enic->memzone_list, entries) {
 391                 if (mze->rz->addr == vaddr &&
 392                     mze->rz->phys_addr == dma_handle)
 393                         break;
 394         }
 395         if (mze == NULL) {
 396                 rte_spinlock_unlock(&enic->memzone_list_lock);
 397                 dev_warning(enic,
 398                             "Tried to free memory, but couldn't find it in the memzone list\n");
 399                 return;
 400         }
 401         LIST_REMOVE(mze, entries);
 402         rte_spinlock_unlock(&enic->memzone_list_lock);
 403         rte_memzone_free(mze->rz);
 404         rte_free(mze);
 405 }
 406
 407 int enic_link_update(struct enic *enic)
 408 {
 409         struct rte_eth_dev *eth_dev = enic->rte_dev;
 410         int ret;
 411         int link_status = 0;
 412
 413         link_status = enic_get_link_status(enic);
 414         ret = (link_status == enic->link_status);
 415         enic->link_status = link_status;
 416         eth_dev->data->dev_link.link_status = link_status;
 417         eth_dev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
 418         eth_dev->data->dev_link.link_speed = vnic_dev_port_speed(enic->vdev);
 419         return ret;
 420 }
 421
 422 static void
 423 enic_intr_handler(__rte_unused struct rte_intr_handle *handle,
 424         void *arg)
 425 {
 426         struct rte_eth_dev *dev = (struct rte_eth_dev *)arg;
 427         struct enic *enic = pmd_priv(dev);
 428
 429         vnic_intr_return_all_credits(&enic->intr);
 430
 431         enic_link_update(enic);
 432         _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL);
 433         enic_log_q_error(enic);
 434 }
 435
 436 int enic_enable(struct enic *enic)
 437 {
 438         unsigned int index;
 439         int err;
 440         struct rte_eth_dev *eth_dev = enic->rte_dev;
 441
 442         eth_dev->data->dev_link.link_speed = vnic_dev_port_speed(enic->vdev);
 443         eth_dev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
 444
 445         /* vnic notification of link status has already been turned on in
 446          * enic_dev_init() which is called during probe time.  Here we are
 447          * just turning on interrupt vector 0 if needed.
 448          */
 449         if (eth_dev->data->dev_conf.intr_conf.lsc)
 450                 vnic_dev_notify_set(enic->vdev, 0);
 451
 452         if (enic_clsf_init(enic))
 453                 dev_warning(enic, "Init of hash table for clsf failed."\
 454                         "Flow director feature will not work\n");
 455
 456         for (index = 0; index < enic->rq_count; index++) {
 457                 err = enic_alloc_rx_queue_mbufs(enic,
 458                         &enic->rq[enic_rte_rq_idx_to_sop_idx(index)]);
 459                 if (err) {
 460                         dev_err(enic, "Failed to alloc sop RX queue mbufs\n");
 461                         return err;
 462                 }
 463                 err = enic_alloc_rx_queue_mbufs(enic,
 464                         &enic->rq[enic_rte_rq_idx_to_data_idx(index)]);
 465                 if (err) {
 466                         /* release the allocated mbufs for the sop rq*/
 467                         enic_rxmbuf_queue_release(enic,
 468                                 &enic->rq[enic_rte_rq_idx_to_sop_idx(index)]);
 469
 470                         dev_err(enic, "Failed to alloc data RX queue mbufs\n");
 471                         return err;
 472                 }
 473         }
 474
 475         for (index = 0; index < enic->wq_count; index++)
 476                 enic_start_wq(enic, index);
 477         for (index = 0; index < enic->rq_count; index++)
 478                 enic_start_rq(enic, index);
 479
 480         vnic_dev_add_addr(enic->vdev, enic->mac_addr);
 481
 482         vnic_dev_enable_wait(enic->vdev);
 483
 484         /* Register and enable error interrupt */
 485         rte_intr_callback_register(&(enic->pdev->intr_handle),
 486                 enic_intr_handler, (void *)enic->rte_dev);
 487
 488         rte_intr_enable(&(enic->pdev->intr_handle));
 489         vnic_intr_unmask(&enic->intr);
 490
 491         return 0;
 492 }
 493
 494 int enic_alloc_intr_resources(struct enic *enic)
 495 {
 496         int err;
 497
 498         dev_info(enic, "vNIC resources used:  "\
 499                 "wq %d rq %d cq %d intr %d\n",
 500                 enic->wq_count, enic_vnic_rq_count(enic),
 501                 enic->cq_count, enic->intr_count);
 502
 503         err = vnic_intr_alloc(enic->vdev, &enic->intr, 0);
 504         if (err)
 505                 enic_free_vnic_resources(enic);
 506
 507         return err;
 508 }
 509
 510 void enic_free_rq(void *rxq)
 511 {
 512         struct vnic_rq *rq_sop, *rq_data;
 513         struct enic *enic;
 514
 515         if (rxq == NULL)
 516                 return;
 517
 518         rq_sop = (struct vnic_rq *)rxq;
 519         enic = vnic_dev_priv(rq_sop->vdev);
 520         rq_data = &enic->rq[rq_sop->data_queue_idx];
 521
 522         enic_rxmbuf_queue_release(enic, rq_sop);
 523         if (rq_data->in_use)
 524                 enic_rxmbuf_queue_release(enic, rq_data);
 525
 526         rte_free(rq_sop->mbuf_ring);
 527         if (rq_data->in_use)
 528                 rte_free(rq_data->mbuf_ring);
 529
 530         rq_sop->mbuf_ring = NULL;
 531         rq_data->mbuf_ring = NULL;
 532
 533         vnic_rq_free(rq_sop);
 534         if (rq_data->in_use)
 535                 vnic_rq_free(rq_data);
 536
 537         vnic_cq_free(&enic->cq[enic_sop_rq_idx_to_cq_idx(rq_sop->index)]);
 538
 539         rq_sop->in_use = 0;
 540         rq_data->in_use = 0;
 541 }
 542
 543 void enic_start_wq(struct enic *enic, uint16_t queue_idx)
 544 {
 545         struct rte_eth_dev *eth_dev = enic->rte_dev;
 546         vnic_wq_enable(&enic->wq[queue_idx]);
 547         eth_dev->data->tx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STARTED;
 548 }
 549
 550 int enic_stop_wq(struct enic *enic, uint16_t queue_idx)
 551 {
 552         struct rte_eth_dev *eth_dev = enic->rte_dev;
 553         int ret;
 554
 555         ret = vnic_wq_disable(&enic->wq[queue_idx]);
 556         if (ret)
 557                 return ret;
 558
 559         eth_dev->data->tx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STOPPED;
 560         return 0;
 561 }
 562
 563 void enic_start_rq(struct enic *enic, uint16_t queue_idx)
 564 {
 565         struct vnic_rq *rq_sop;
 566         struct vnic_rq *rq_data;
 567         rq_sop = &enic->rq[enic_rte_rq_idx_to_sop_idx(queue_idx)];
 568         rq_data = &enic->rq[rq_sop->data_queue_idx];
 569         struct rte_eth_dev *eth_dev = enic->rte_dev;
 570
 571         if (rq_data->in_use)
 572                 vnic_rq_enable(rq_data);
 573         rte_mb();
 574         vnic_rq_enable(rq_sop);
 575         eth_dev->data->rx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STARTED;
 576 }
 577
 578 int enic_stop_rq(struct enic *enic, uint16_t queue_idx)
 579 {
 580         int ret1 = 0, ret2 = 0;
 581         struct rte_eth_dev *eth_dev = enic->rte_dev;
 582         struct vnic_rq *rq_sop;
 583         struct vnic_rq *rq_data;
 584         rq_sop = &enic->rq[enic_rte_rq_idx_to_sop_idx(queue_idx)];
 585         rq_data = &enic->rq[rq_sop->data_queue_idx];
 586
 587         ret2 = vnic_rq_disable(rq_sop);
 588         rte_mb();
 589         if (rq_data->in_use)
 590                 ret1 = vnic_rq_disable(rq_data);
 591
 592         if (ret2)
 593                 return ret2;
 594         else if (ret1)
 595                 return ret1;
 596
 597         eth_dev->data->rx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STOPPED;
 598         return 0;
 599 }
 600
 601 int enic_alloc_rq(struct enic *enic, uint16_t queue_idx,
 602         unsigned int socket_id, struct rte_mempool *mp,
 603         uint16_t nb_desc, uint16_t free_thresh)
 604 {
 605         int rc;
 606         uint16_t sop_queue_idx = enic_rte_rq_idx_to_sop_idx(queue_idx);
 607         uint16_t data_queue_idx = enic_rte_rq_idx_to_data_idx(queue_idx);
 608         struct vnic_rq *rq_sop = &enic->rq[sop_queue_idx];
 609         struct vnic_rq *rq_data = &enic->rq[data_queue_idx];
 610         unsigned int mbuf_size, mbufs_per_pkt;
 611         unsigned int nb_sop_desc, nb_data_desc;
 612         uint16_t min_sop, max_sop, min_data, max_data;
 613         uint16_t mtu = enic->rte_dev->data->mtu;
 614
 615         rq_sop->is_sop = 1;
 616         rq_sop->data_queue_idx = data_queue_idx;
 617         rq_data->is_sop = 0;
 618         rq_data->data_queue_idx = 0;
 619         rq_sop->socket_id = socket_id;
 620         rq_sop->mp = mp;
 621         rq_data->socket_id = socket_id;
 622         rq_data->mp = mp;
 623         rq_sop->in_use = 1;
 624         rq_sop->rx_free_thresh = free_thresh;
 625         rq_data->rx_free_thresh = free_thresh;
 626         dev_debug(enic, "Set queue_id:%u free thresh:%u\n", queue_idx,
 627                   free_thresh);
 628
 629         mbuf_size = (uint16_t)(rte_pktmbuf_data_room_size(mp) -
 630                                RTE_PKTMBUF_HEADROOM);
 631
 632         if (enic->rte_dev->data->dev_conf.rxmode.enable_scatter) {
 633                 dev_info(enic, "Rq %u Scatter rx mode enabled\n", queue_idx);
 634                 /* ceil((mtu + ETHER_HDR_LEN + 4)/mbuf_size) */
 635                 mbufs_per_pkt = ((mtu + ETHER_HDR_LEN + 4) +
 636                                  (mbuf_size - 1)) / mbuf_size;
 637         } else {
 638                 dev_info(enic, "Scatter rx mode disabled\n");
 639                 mbufs_per_pkt = 1;
 640         }
 641
 642         if (mbufs_per_pkt > 1) {
 643                 dev_info(enic, "Rq %u Scatter rx mode in use\n", queue_idx);
 644                 rq_sop->data_queue_enable = 1;
 645                 rq_data->in_use = 1;
 646         } else {
 647                 dev_info(enic, "Rq %u Scatter rx mode not being used\n",
 648                          queue_idx);
 649                 rq_sop->data_queue_enable = 0;
 650                 rq_data->in_use = 0;
 651         }
 652
 653         /* number of descriptors have to be a multiple of 32 */
 654         nb_sop_desc = (nb_desc / mbufs_per_pkt) & ~0x1F;
 655         nb_data_desc = (nb_desc - nb_sop_desc) & ~0x1F;
 656
 657         rq_sop->max_mbufs_per_pkt = mbufs_per_pkt;
 658         rq_data->max_mbufs_per_pkt = mbufs_per_pkt;
 659
 660         if (mbufs_per_pkt > 1) {
 661                 min_sop = 64;
 662                 max_sop = ((enic->config.rq_desc_count /
 663                             (mbufs_per_pkt - 1)) & ~0x1F);
 664                 min_data = min_sop * (mbufs_per_pkt - 1);
 665                 max_data = enic->config.rq_desc_count;
 666         } else {
 667                 min_sop = 64;
 668                 max_sop = enic->config.rq_desc_count;
 669                 min_data = 0;
 670                 max_data = 0;
 671         }
 672
 673         if (nb_desc < (min_sop + min_data)) {
 674                 dev_warning(enic,
 675                             "Number of rx descs too low, adjusting to minimum\n");
 676                 nb_sop_desc = min_sop;
 677                 nb_data_desc = min_data;
 678         } else if (nb_desc > (max_sop + max_data)) {
 679                 dev_warning(enic,
 680                             "Number of rx_descs too high, adjusting to maximum\n");
 681                 nb_sop_desc = max_sop;
 682                 nb_data_desc = max_data;
 683         }
 684         if (mbufs_per_pkt > 1) {
 685                 dev_info(enic, "For mtu %d and mbuf size %d valid rx descriptor range is %d to %d\n",
 686                          mtu, mbuf_size, min_sop + min_data,
 687                          max_sop + max_data);
 688         }
 689         dev_info(enic, "Using %d rx descriptors (sop %d, data %d)\n",
 690                  nb_sop_desc + nb_data_desc, nb_sop_desc, nb_data_desc);
 691
 692         /* Allocate sop queue resources */
 693         rc = vnic_rq_alloc(enic->vdev, rq_sop, sop_queue_idx,
 694                 nb_sop_desc, sizeof(struct rq_enet_desc));
 695         if (rc) {
 696                 dev_err(enic, "error in allocation of sop rq\n");
 697                 goto err_exit;
 698         }
 699         nb_sop_desc = rq_sop->ring.desc_count;
 700
 701         if (rq_data->in_use) {
 702                 /* Allocate data queue resources */
 703                 rc = vnic_rq_alloc(enic->vdev, rq_data, data_queue_idx,
 704                                    nb_data_desc,
 705                                    sizeof(struct rq_enet_desc));
 706                 if (rc) {
 707                         dev_err(enic, "error in allocation of data rq\n");
 708                         goto err_free_rq_sop;
 709                 }
 710                 nb_data_desc = rq_data->ring.desc_count;
 711         }
 712         rc = vnic_cq_alloc(enic->vdev, &enic->cq[queue_idx], queue_idx,
 713                            socket_id, nb_sop_desc + nb_data_desc,
 714                            sizeof(struct cq_enet_rq_desc));
 715         if (rc) {
 716                 dev_err(enic, "error in allocation of cq for rq\n");
 717                 goto err_free_rq_data;
 718         }
 719
 720         /* Allocate the mbuf rings */
 721         rq_sop->mbuf_ring = (struct rte_mbuf **)
 722                 rte_zmalloc_socket("rq->mbuf_ring",
 723                                    sizeof(struct rte_mbuf *) * nb_sop_desc,
 724                                    RTE_CACHE_LINE_SIZE, rq_sop->socket_id);
 725         if (rq_sop->mbuf_ring == NULL)
 726                 goto err_free_cq;
 727
 728         if (rq_data->in_use) {
 729                 rq_data->mbuf_ring = (struct rte_mbuf **)
 730                         rte_zmalloc_socket("rq->mbuf_ring",
 731                                 sizeof(struct rte_mbuf *) * nb_data_desc,
 732                                 RTE_CACHE_LINE_SIZE, rq_sop->socket_id);
 733                 if (rq_data->mbuf_ring == NULL)
 734                         goto err_free_sop_mbuf;
 735         }
 736
 737         rq_sop->tot_nb_desc = nb_desc; /* squirl away for MTU update function */
 738
 739         return 0;
 740
 741 err_free_sop_mbuf:
 742         rte_free(rq_sop->mbuf_ring);
 743 err_free_cq:
 744         /* cleanup on error */
 745         vnic_cq_free(&enic->cq[queue_idx]);
 746 err_free_rq_data:
 747         if (rq_data->in_use)
 748                 vnic_rq_free(rq_data);
 749 err_free_rq_sop:
 750         vnic_rq_free(rq_sop);
 751 err_exit:
 752         return -ENOMEM;
 753 }
 754
 755 void enic_free_wq(void *txq)
 756 {
 757         struct vnic_wq *wq;
 758         struct enic *enic;
 759
 760         if (txq == NULL)
 761                 return;
 762
 763         wq = (struct vnic_wq *)txq;
 764         enic = vnic_dev_priv(wq->vdev);
 765         rte_memzone_free(wq->cqmsg_rz);
 766         vnic_wq_free(wq);
 767         vnic_cq_free(&enic->cq[enic->rq_count + wq->index]);
 768 }
 769
 770 int enic_alloc_wq(struct enic *enic, uint16_t queue_idx,
 771         unsigned int socket_id, uint16_t nb_desc)
 772 {
 773         int err;
 774         struct vnic_wq *wq = &enic->wq[queue_idx];
 775         unsigned int cq_index = enic_cq_wq(enic, queue_idx);
 776         char name[NAME_MAX];
 777         static int instance;
 778
 779         wq->socket_id = socket_id;
 780         if (nb_desc) {
 781                 if (nb_desc > enic->config.wq_desc_count) {
 782                         dev_warning(enic,
 783                                 "WQ %d - number of tx desc in cmd line (%d)"\
 784                                 "is greater than that in the UCSM/CIMC adapter"\
 785                                 "policy.  Applying the value in the adapter "\
 786                                 "policy (%d)\n",
 787                                 queue_idx, nb_desc, enic->config.wq_desc_count);
 788                 } else if (nb_desc != enic->config.wq_desc_count) {
 789                         enic->config.wq_desc_count = nb_desc;
 790                         dev_info(enic,
 791                                 "TX Queues - effective number of descs:%d\n",
 792                                 nb_desc);
 793                 }
 794         }
 795
 796         /* Allocate queue resources */
 797         err = vnic_wq_alloc(enic->vdev, &enic->wq[queue_idx], queue_idx,
 798                 enic->config.wq_desc_count,
 799                 sizeof(struct wq_enet_desc));
 800         if (err) {
 801                 dev_err(enic, "error in allocation of wq\n");
 802                 return err;
 803         }
 804
 805         err = vnic_cq_alloc(enic->vdev, &enic->cq[cq_index], cq_index,
 806                 socket_id, enic->config.wq_desc_count,
 807                 sizeof(struct cq_enet_wq_desc));
 808         if (err) {
 809                 vnic_wq_free(wq);
 810                 dev_err(enic, "error in allocation of cq for wq\n");
 811         }
 812
 813         /* setup up CQ message */
 814         snprintf((char *)name, sizeof(name),
 815                  "vnic_cqmsg-%s-%d-%d", enic->bdf_name, queue_idx,
 816                 instance++);
 817
 818         wq->cqmsg_rz = rte_memzone_reserve_aligned((const char *)name,
 819                                                    sizeof(uint32_t),
 820                                                    SOCKET_ID_ANY, 0,
 821                                                    ENIC_ALIGN);
 822         if (!wq->cqmsg_rz)
 823                 return -ENOMEM;
 824
 825         return err;
 826 }
 827
 828 int enic_disable(struct enic *enic)
 829 {
 830         unsigned int i;
 831         int err;
 832
 833         vnic_intr_mask(&enic->intr);
 834         (void)vnic_intr_masked(&enic->intr); /* flush write */
 835         rte_intr_disable(&enic->pdev->intr_handle);
 836         rte_intr_callback_unregister(&enic->pdev->intr_handle,
 837                                      enic_intr_handler,
 838                                      (void *)enic->rte_dev);
 839
 840         vnic_dev_disable(enic->vdev);
 841
 842         enic_clsf_destroy(enic);
 843
 844         if (!enic_is_sriov_vf(enic))
 845                 vnic_dev_del_addr(enic->vdev, enic->mac_addr);
 846
 847         for (i = 0; i < enic->wq_count; i++) {
 848                 err = vnic_wq_disable(&enic->wq[i]);
 849                 if (err)
 850                         return err;
 851         }
 852         for (i = 0; i < enic_vnic_rq_count(enic); i++) {
 853                 if (enic->rq[i].in_use) {
 854                         err = vnic_rq_disable(&enic->rq[i]);
 855                         if (err)
 856                                 return err;
 857                 }
 858         }
 859
 860         /* If we were using interrupts, set the interrupt vector to -1
 861          * to disable interrupts.  We are not disabling link notifcations,
 862          * though, as we want the polling of link status to continue working.
 863          */
 864         if (enic->rte_dev->data->dev_conf.intr_conf.lsc)
 865                 vnic_dev_notify_set(enic->vdev, -1);
 866
 867         vnic_dev_set_reset_flag(enic->vdev, 1);
 868
 869         for (i = 0; i < enic->wq_count; i++)
 870                 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
 871
 872         for (i = 0; i < enic_vnic_rq_count(enic); i++)
 873                 if (enic->rq[i].in_use)
 874                         vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
 875         for (i = 0; i < enic->cq_count; i++)
 876                 vnic_cq_clean(&enic->cq[i]);
 877         vnic_intr_clean(&enic->intr);
 878
 879         return 0;
 880 }
 881
 882 static int enic_dev_wait(struct vnic_dev *vdev,
 883         int (*start)(struct vnic_dev *, int),
 884         int (*finished)(struct vnic_dev *, int *),
 885         int arg)
 886 {
 887         int done;
 888         int err;
 889         int i;
 890
 891         err = start(vdev, arg);
 892         if (err)
 893                 return err;
 894
 895         /* Wait for func to complete...2 seconds max */
 896         for (i = 0; i < 2000; i++) {
 897                 err = finished(vdev, &done);
 898                 if (err)
 899                         return err;
 900                 if (done)
 901                         return 0;
 902                 usleep(1000);
 903         }
 904         return -ETIMEDOUT;
 905 }
 906
 907 static int enic_dev_open(struct enic *enic)
 908 {
 909         int err;
 910
 911         err = enic_dev_wait(enic->vdev, vnic_dev_open,
 912                 vnic_dev_open_done, 0);
 913         if (err)
 914                 dev_err(enic_get_dev(enic),
 915                         "vNIC device open failed, err %d\n", err);
 916
 917         return err;
 918 }
 919
 920 static int enic_set_rsskey(struct enic *enic)
 921 {
 922         dma_addr_t rss_key_buf_pa;
 923         union vnic_rss_key *rss_key_buf_va = NULL;
 924         static union vnic_rss_key rss_key = {
 925                 .key = {
 926                         [0] = {.b = {85, 67, 83, 97, 119, 101, 115, 111, 109, 101}},
 927                         [1] = {.b = {80, 65, 76, 79, 117, 110, 105, 113, 117, 101}},
 928                         [2] = {.b = {76, 73, 78, 85, 88, 114, 111, 99, 107, 115}},
 929                         [3] = {.b = {69, 78, 73, 67, 105, 115, 99, 111, 111, 108}},
 930                 }
 931         };
 932         int err;
 933         u8 name[NAME_MAX];
 934
 935         snprintf((char *)name, NAME_MAX, "rss_key-%s", enic->bdf_name);
 936         rss_key_buf_va = enic_alloc_consistent(enic, sizeof(union vnic_rss_key),
 937                 &rss_key_buf_pa, name);
 938         if (!rss_key_buf_va)
 939                 return -ENOMEM;
 940
 941         rte_memcpy(rss_key_buf_va, &rss_key, sizeof(union vnic_rss_key));
 942
 943         err = enic_set_rss_key(enic,
 944                 rss_key_buf_pa,
 945                 sizeof(union vnic_rss_key));
 946
 947         enic_free_consistent(enic, sizeof(union vnic_rss_key),
 948                 rss_key_buf_va, rss_key_buf_pa);
 949
 950         return err;
 951 }
 952
 953 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
 954 {
 955         dma_addr_t rss_cpu_buf_pa;
 956         union vnic_rss_cpu *rss_cpu_buf_va = NULL;
 957         int i;
 958         int err;
 959         u8 name[NAME_MAX];
 960
 961         snprintf((char *)name, NAME_MAX, "rss_cpu-%s", enic->bdf_name);
 962         rss_cpu_buf_va = enic_alloc_consistent(enic, sizeof(union vnic_rss_cpu),
 963                 &rss_cpu_buf_pa, name);
 964         if (!rss_cpu_buf_va)
 965                 return -ENOMEM;
 966
 967         for (i = 0; i < (1 << rss_hash_bits); i++)
 968                 (*rss_cpu_buf_va).cpu[i / 4].b[i % 4] =
 969                         enic_rte_rq_idx_to_sop_idx(i % enic->rq_count);
 970
 971         err = enic_set_rss_cpu(enic,
 972                 rss_cpu_buf_pa,
 973                 sizeof(union vnic_rss_cpu));
 974
 975         enic_free_consistent(enic, sizeof(union vnic_rss_cpu),
 976                 rss_cpu_buf_va, rss_cpu_buf_pa);
 977
 978         return err;
 979 }
 980
 981 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
 982         u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
 983 {
 984         const u8 tso_ipid_split_en = 0;
 985         int err;
 986
 987         /* Enable VLAN tag stripping */
 988
 989         err = enic_set_nic_cfg(enic,
 990                 rss_default_cpu, rss_hash_type,
 991                 rss_hash_bits, rss_base_cpu,
 992                 rss_enable, tso_ipid_split_en,
 993                 enic->ig_vlan_strip_en);
 994
 995         return err;
 996 }
 997
 998 int enic_set_rss_nic_cfg(struct enic *enic)
 999 {
1000         const u8 rss_default_cpu = 0;
1001         const u8 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
1002             NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
1003             NIC_CFG_RSS_HASH_TYPE_IPV6 |
1004             NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
1005         const u8 rss_hash_bits = 7;
1006         const u8 rss_base_cpu = 0;
1007         u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
1008
1009         if (rss_enable) {
1010                 if (!enic_set_rsskey(enic)) {
1011                         if (enic_set_rsscpu(enic, rss_hash_bits)) {
1012                                 rss_enable = 0;
1013                                 dev_warning(enic, "RSS disabled, "\
1014                                         "Failed to set RSS cpu indirection table.");
1015                         }
1016                 } else {
1017                         rss_enable = 0;
1018                         dev_warning(enic,
1019                                 "RSS disabled, Failed to set RSS key.\n");
1020                 }
1021         }
1022
1023         return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
1024                 rss_hash_bits, rss_base_cpu, rss_enable);
1025 }
1026
1027 int enic_setup_finish(struct enic *enic)
1028 {
1029         int ret;
1030
1031         enic_init_soft_stats(enic);
1032
1033         ret = enic_set_rss_nic_cfg(enic);
1034         if (ret) {
1035                 dev_err(enic, "Failed to config nic, aborting.\n");
1036                 return -1;
1037         }
1038
1039         /* Default conf */
1040         vnic_dev_packet_filter(enic->vdev,
1041                 1 /* directed  */,
1042                 1 /* multicast */,
1043                 1 /* broadcast */,
1044                 0 /* promisc   */,
1045                 1 /* allmulti  */);
1046
1047         enic->promisc = 0;
1048         enic->allmulti = 1;
1049
1050         return 0;
1051 }
1052
1053 void enic_add_packet_filter(struct enic *enic)
1054 {
1055         /* Args -> directed, multicast, broadcast, promisc, allmulti */
1056         vnic_dev_packet_filter(enic->vdev, 1, 1, 1,
1057                 enic->promisc, enic->allmulti);
1058 }
1059
1060 int enic_get_link_status(struct enic *enic)
1061 {
1062         return vnic_dev_link_status(enic->vdev);
1063 }
1064
1065 static void enic_dev_deinit(struct enic *enic)
1066 {
1067         struct rte_eth_dev *eth_dev = enic->rte_dev;
1068
1069         /* stop link status checking */
1070         vnic_dev_notify_unset(enic->vdev);
1071
1072         rte_free(eth_dev->data->mac_addrs);
1073 }
1074
1075
1076 int enic_set_vnic_res(struct enic *enic)
1077 {
1078         struct rte_eth_dev *eth_dev = enic->rte_dev;
1079         int rc = 0;
1080
1081         /* With Rx scatter support, two RQs are now used per RQ used by
1082          * the application.
1083          */
1084         if (enic->conf_rq_count < eth_dev->data->nb_rx_queues) {
1085                 dev_err(dev, "Not enough Receive queues. Requested:%u which uses %d RQs on VIC, Configured:%u\n",
1086                         eth_dev->data->nb_rx_queues,
1087                         eth_dev->data->nb_rx_queues * 2, enic->conf_rq_count);
1088                 rc = -EINVAL;
1089         }
1090         if (enic->conf_wq_count < eth_dev->data->nb_tx_queues) {
1091                 dev_err(dev, "Not enough Transmit queues. Requested:%u, Configured:%u\n",
1092                         eth_dev->data->nb_tx_queues, enic->conf_wq_count);
1093                 rc = -EINVAL;
1094         }
1095
1096         if (enic->conf_cq_count < (eth_dev->data->nb_rx_queues +
1097                                    eth_dev->data->nb_tx_queues)) {
1098                 dev_err(dev, "Not enough Completion queues. Required:%u, Configured:%u\n",
1099                         (eth_dev->data->nb_rx_queues +
1100                          eth_dev->data->nb_tx_queues), enic->conf_cq_count);
1101                 rc = -EINVAL;
1102         }
1103
1104         if (rc == 0) {
1105                 enic->rq_count = eth_dev->data->nb_rx_queues;
1106                 enic->wq_count = eth_dev->data->nb_tx_queues;
1107                 enic->cq_count = enic->rq_count + enic->wq_count;
1108         }
1109
1110         return rc;
1111 }
1112
1113 /* Initialize the completion queue for an RQ */
1114 static int
1115 enic_reinit_rq(struct enic *enic, unsigned int rq_idx)
1116 {
1117         struct vnic_rq *sop_rq, *data_rq;
1118         unsigned int cq_idx = enic_cq_rq(enic, rq_idx);
1119         int rc = 0;
1120
1121         sop_rq = &enic->rq[enic_rte_rq_idx_to_sop_idx(rq_idx)];
1122         data_rq = &enic->rq[enic_rte_rq_idx_to_data_idx(rq_idx)];
1123
1124         vnic_cq_clean(&enic->cq[cq_idx]);
1125         vnic_cq_init(&enic->cq[cq_idx],
1126                      0 /* flow_control_enable */,
1127                      1 /* color_enable */,
1128                      0 /* cq_head */,
1129                      0 /* cq_tail */,
1130                      1 /* cq_tail_color */,
1131                      0 /* interrupt_enable */,
1132                      1 /* cq_entry_enable */,
1133                      0 /* cq_message_enable */,
1134                      0 /* interrupt offset */,
1135                      0 /* cq_message_addr */);
1136
1137
1138         vnic_rq_init_start(sop_rq, enic_cq_rq(enic,
1139                            enic_rte_rq_idx_to_sop_idx(rq_idx)), 0,
1140                            sop_rq->ring.desc_count - 1, 1, 0);
1141         if (data_rq->in_use) {
1142                 vnic_rq_init_start(data_rq,
1143                                    enic_cq_rq(enic,
1144                                    enic_rte_rq_idx_to_data_idx(rq_idx)), 0,
1145                                    data_rq->ring.desc_count - 1, 1, 0);
1146         }
1147
1148         rc = enic_alloc_rx_queue_mbufs(enic, sop_rq);
1149         if (rc)
1150                 return rc;
1151
1152         if (data_rq->in_use) {
1153                 rc = enic_alloc_rx_queue_mbufs(enic, data_rq);
1154                 if (rc) {
1155                         enic_rxmbuf_queue_release(enic, sop_rq);
1156                         return rc;
1157                 }
1158         }
1159
1160         return 0;
1161 }
1162
1163 /* The Cisco NIC can send and receive packets up to a max packet size
1164  * determined by the NIC type and firmware. There is also an MTU
1165  * configured into the NIC via the CIMC/UCSM management interface
1166  * which can be overridden by this function (up to the max packet size).
1167  * Depending on the network setup, doing so may cause packet drops
1168  * and unexpected behavior.
1169  */
1170 int enic_set_mtu(struct enic *enic, uint16_t new_mtu)
1171 {
1172         unsigned int rq_idx;
1173         struct vnic_rq *rq;
1174         int rc = 0;
1175         uint16_t old_mtu;       /* previous setting */
1176         uint16_t config_mtu;    /* Value configured into NIC via CIMC/UCSM */
1177         struct rte_eth_dev *eth_dev = enic->rte_dev;
1178
1179         old_mtu = eth_dev->data->mtu;
1180         config_mtu = enic->config.mtu;
1181
1182         if (new_mtu > enic->max_mtu) {
1183                 dev_err(enic,
1184                         "MTU not updated: requested (%u) greater than max (%u)\n",
1185                         new_mtu, enic->max_mtu);
1186                 return -EINVAL;
1187         }
1188         if (new_mtu < ENIC_MIN_MTU) {
1189                 dev_info(enic,
1190                         "MTU not updated: requested (%u) less than min (%u)\n",
1191                         new_mtu, ENIC_MIN_MTU);
1192                 return -EINVAL;
1193         }
1194         if (new_mtu > config_mtu)
1195                 dev_warning(enic,
1196                         "MTU (%u) is greater than value configured in NIC (%u)\n",
1197                         new_mtu, config_mtu);
1198
1199         /* The easy case is when scatter is disabled. However if the MTU
1200          * becomes greater than the mbuf data size, packet drops will ensue.
1201          */
1202         if (!enic->rte_dev->data->dev_conf.rxmode.enable_scatter) {
1203                 eth_dev->data->mtu = new_mtu;
1204                 goto set_mtu_done;
1205         }
1206
1207         /* Rx scatter is enabled so reconfigure RQ's on the fly. The point is to
1208          * change Rx scatter mode if necessary for better performance. I.e. if
1209          * MTU was greater than the mbuf size and now it's less, scatter Rx
1210          * doesn't have to be used and vice versa.
1211           */
1212         rte_spinlock_lock(&enic->mtu_lock);
1213
1214         /* Stop traffic on all RQs */
1215         for (rq_idx = 0; rq_idx < enic->rq_count * 2; rq_idx++) {
1216                 rq = &enic->rq[rq_idx];
1217                 if (rq->is_sop && rq->in_use) {
1218                         rc = enic_stop_rq(enic,
1219                                           enic_sop_rq_idx_to_rte_idx(rq_idx));
1220                         if (rc) {
1221                                 dev_err(enic, "Failed to stop Rq %u\n", rq_idx);
1222                                 goto set_mtu_done;
1223                         }
1224                 }
1225         }
1226
1227         /* replace Rx funciton with a no-op to avoid getting stale pkts */
1228         eth_dev->rx_pkt_burst = enic_dummy_recv_pkts;
1229         rte_mb();
1230
1231         /* Allow time for threads to exit the real Rx function. */
1232         usleep(100000);
1233
1234         /* now it is safe to reconfigure the RQs */
1235
1236         /* update the mtu */
1237         eth_dev->data->mtu = new_mtu;
1238
1239         /* free and reallocate RQs with the new MTU */
1240         for (rq_idx = 0; rq_idx < enic->rq_count; rq_idx++) {
1241                 rq = &enic->rq[enic_rte_rq_idx_to_sop_idx(rq_idx)];
1242
1243                 enic_free_rq(rq);
1244                 rc = enic_alloc_rq(enic, rq_idx, rq->socket_id, rq->mp,
1245                                    rq->tot_nb_desc, rq->rx_free_thresh);
1246                 if (rc) {
1247                         dev_err(enic,
1248                                 "Fatal MTU alloc error- No traffic will pass\n");
1249                         goto set_mtu_done;
1250                 }
1251
1252                 rc = enic_reinit_rq(enic, rq_idx);
1253                 if (rc) {
1254                         dev_err(enic,
1255                                 "Fatal MTU RQ reinit- No traffic will pass\n");
1256                         goto set_mtu_done;
1257                 }
1258         }
1259
1260         /* put back the real receive function */
1261         rte_mb();
1262         eth_dev->rx_pkt_burst = enic_recv_pkts;
1263         rte_mb();
1264
1265         /* restart Rx traffic */
1266         for (rq_idx = 0; rq_idx < enic->rq_count; rq_idx++) {
1267                 rq = &enic->rq[enic_rte_rq_idx_to_sop_idx(rq_idx)];
1268                 if (rq->is_sop && rq->in_use)
1269                         enic_start_rq(enic, rq_idx);
1270         }
1271
1272 set_mtu_done:
1273         dev_info(enic, "MTU changed from %u to %u\n",  old_mtu, new_mtu);
1274         rte_spinlock_unlock(&enic->mtu_lock);
1275         return rc;
1276 }
1277
1278 static int enic_dev_init(struct enic *enic)
1279 {
1280         int err;
1281         struct rte_eth_dev *eth_dev = enic->rte_dev;
1282
1283         vnic_dev_intr_coal_timer_info_default(enic->vdev);
1284
1285         /* Get vNIC configuration
1286         */
1287         err = enic_get_vnic_config(enic);
1288         if (err) {
1289                 dev_err(dev, "Get vNIC configuration failed, aborting\n");
1290                 return err;
1291         }
1292
1293         /* Get available resource counts */
1294         enic_get_res_counts(enic);
1295         if (enic->conf_rq_count == 1) {
1296                 dev_err(enic, "Running with only 1 RQ configured in the vNIC is not supported.\n");
1297                 dev_err(enic, "Please configure 2 RQs in the vNIC for each Rx queue used by DPDK.\n");
1298                 dev_err(enic, "See the ENIC PMD guide for more information.\n");
1299                 return -EINVAL;
1300         }
1301
1302         /* Get the supported filters */
1303         enic_fdir_info(enic);
1304
1305         eth_dev->data->mac_addrs = rte_zmalloc("enic_mac_addr", ETH_ALEN
1306                                                 * ENIC_MAX_MAC_ADDR, 0);
1307         if (!eth_dev->data->mac_addrs) {
1308                 dev_err(enic, "mac addr storage alloc failed, aborting.\n");
1309                 return -1;
1310         }
1311         ether_addr_copy((struct ether_addr *) enic->mac_addr,
1312                         eth_dev->data->mac_addrs);
1313
1314         vnic_dev_set_reset_flag(enic->vdev, 0);
1315
1316         /* set up link status checking */
1317         vnic_dev_notify_set(enic->vdev, -1); /* No Intr for notify */
1318
1319         return 0;
1320
1321 }
1322
1323 int enic_probe(struct enic *enic)
1324 {
1325         struct rte_pci_device *pdev = enic->pdev;
1326         int err = -1;
1327
1328         dev_debug(enic, " Initializing ENIC PMD\n");
1329
1330         enic->bar0.vaddr = (void *)pdev->mem_resource[0].addr;
1331         enic->bar0.len = pdev->mem_resource[0].len;
1332
1333         /* Register vNIC device */
1334         enic->vdev = vnic_dev_register(NULL, enic, enic->pdev, &enic->bar0, 1);
1335         if (!enic->vdev) {
1336                 dev_err(enic, "vNIC registration failed, aborting\n");
1337                 goto err_out;
1338         }
1339
1340         LIST_INIT(&enic->memzone_list);
1341         rte_spinlock_init(&enic->memzone_list_lock);
1342
1343         vnic_register_cbacks(enic->vdev,
1344                 enic_alloc_consistent,
1345                 enic_free_consistent);
1346
1347         /* Issue device open to get device in known state */
1348         err = enic_dev_open(enic);
1349         if (err) {
1350                 dev_err(enic, "vNIC dev open failed, aborting\n");
1351                 goto err_out_unregister;
1352         }
1353
1354         /* Set ingress vlan rewrite mode before vnic initialization */
1355         err = vnic_dev_set_ig_vlan_rewrite_mode(enic->vdev,
1356                 IG_VLAN_REWRITE_MODE_PASS_THRU);
1357         if (err) {
1358                 dev_err(enic,
1359                         "Failed to set ingress vlan rewrite mode, aborting.\n");
1360                 goto err_out_dev_close;
1361         }
1362
1363         /* Issue device init to initialize the vnic-to-switch link.
1364          * We'll start with carrier off and wait for link UP
1365          * notification later to turn on carrier.  We don't need
1366          * to wait here for the vnic-to-switch link initialization
1367          * to complete; link UP notification is the indication that
1368          * the process is complete.
1369          */
1370
1371         err = vnic_dev_init(enic->vdev, 0);
1372         if (err) {
1373                 dev_err(enic, "vNIC dev init failed, aborting\n");
1374                 goto err_out_dev_close;
1375         }
1376
1377         err = enic_dev_init(enic);
1378         if (err) {
1379                 dev_err(enic, "Device initialization failed, aborting\n");
1380                 goto err_out_dev_close;
1381         }
1382
1383         return 0;
1384
1385 err_out_dev_close:
1386         vnic_dev_close(enic->vdev);
1387 err_out_unregister:
1388         vnic_dev_unregister(enic->vdev);
1389 err_out:
1390         return err;
1391 }
1392
1393 void enic_remove(struct enic *enic)
1394 {
1395         enic_dev_deinit(enic);
1396         vnic_dev_close(enic->vdev);
1397         vnic_dev_unregister(enic->vdev);
1398 }