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         rte_atomic64_clear(&soft_stats->tx_oversized);
 141 }
 142
 143 static void enic_init_soft_stats(struct enic *enic)
 144 {
 145         struct enic_soft_stats *soft_stats = &enic->soft_stats;
 146         rte_atomic64_init(&soft_stats->rx_nombuf);
 147         rte_atomic64_init(&soft_stats->rx_packet_errors);
 148         rte_atomic64_init(&soft_stats->tx_oversized);
 149         enic_clear_soft_stats(enic);
 150 }
 151
 152 void enic_dev_stats_clear(struct enic *enic)
 153 {
 154         if (vnic_dev_stats_clear(enic->vdev))
 155                 dev_err(enic, "Error in clearing stats\n");
 156         enic_clear_soft_stats(enic);
 157 }
 158
 159 void enic_dev_stats_get(struct enic *enic, struct rte_eth_stats *r_stats)
 160 {
 161         struct vnic_stats *stats;
 162         struct enic_soft_stats *soft_stats = &enic->soft_stats;
 163         int64_t rx_truncated;
 164         uint64_t rx_packet_errors;
 165
 166         if (vnic_dev_stats_dump(enic->vdev, &stats)) {
 167                 dev_err(enic, "Error in getting stats\n");
 168                 return;
 169         }
 170
 171         /* The number of truncated packets can only be calculated by
 172          * subtracting a hardware counter from error packets received by
 173          * the driver. Note: this causes transient inaccuracies in the
 174          * ipackets count. Also, the length of truncated packets are
 175          * counted in ibytes even though truncated packets are dropped
 176          * which can make ibytes be slightly higher than it should be.
 177          */
 178         rx_packet_errors = rte_atomic64_read(&soft_stats->rx_packet_errors);
 179         rx_truncated = rx_packet_errors - stats->rx.rx_errors;
 180
 181         r_stats->ipackets = stats->rx.rx_frames_ok - rx_truncated;
 182         r_stats->opackets = stats->tx.tx_frames_ok;
 183
 184         r_stats->ibytes = stats->rx.rx_bytes_ok;
 185         r_stats->obytes = stats->tx.tx_bytes_ok;
 186
 187         r_stats->ierrors = stats->rx.rx_errors + stats->rx.rx_drop;
 188         r_stats->oerrors = stats->tx.tx_errors
 189                            + rte_atomic64_read(&soft_stats->tx_oversized);
 190
 191         r_stats->imissed = stats->rx.rx_no_bufs + rx_truncated;
 192
 193         r_stats->rx_nombuf = rte_atomic64_read(&soft_stats->rx_nombuf);
 194 }
 195
 196 void enic_del_mac_address(struct enic *enic, int mac_index)
 197 {
 198         struct rte_eth_dev *eth_dev = enic->rte_dev;
 199         uint8_t *mac_addr = eth_dev->data->mac_addrs[mac_index].addr_bytes;
 200
 201         if (vnic_dev_del_addr(enic->vdev, mac_addr))
 202                 dev_err(enic, "del mac addr failed\n");
 203 }
 204
 205 void enic_set_mac_address(struct enic *enic, uint8_t *mac_addr)
 206 {
 207         int err;
 208
 209         if (!is_eth_addr_valid(mac_addr)) {
 210                 dev_err(enic, "invalid mac address\n");
 211                 return;
 212         }
 213
 214         err = vnic_dev_add_addr(enic->vdev, mac_addr);
 215         if (err) {
 216                 dev_err(enic, "add mac addr failed\n");
 217                 return;
 218         }
 219 }
 220
 221 static void
 222 enic_free_rq_buf(struct rte_mbuf **mbuf)
 223 {
 224         if (*mbuf == NULL)
 225                 return;
 226
 227         rte_pktmbuf_free(*mbuf);
 228         mbuf = NULL;
 229 }
 230
 231 void enic_init_vnic_resources(struct enic *enic)
 232 {
 233         unsigned int error_interrupt_enable = 1;
 234         unsigned int error_interrupt_offset = 0;
 235         unsigned int index = 0;
 236         unsigned int cq_idx;
 237         struct vnic_rq *data_rq;
 238
 239         for (index = 0; index < enic->rq_count; index++) {
 240                 cq_idx = enic_cq_rq(enic, enic_rte_rq_idx_to_sop_idx(index));
 241
 242                 vnic_rq_init(&enic->rq[enic_rte_rq_idx_to_sop_idx(index)],
 243                         cq_idx,
 244                         error_interrupt_enable,
 245                         error_interrupt_offset);
 246
 247                 data_rq = &enic->rq[enic_rte_rq_idx_to_data_idx(index)];
 248                 if (data_rq->in_use)
 249                         vnic_rq_init(data_rq,
 250                                      cq_idx,
 251                                      error_interrupt_enable,
 252                                      error_interrupt_offset);
 253
 254                 vnic_cq_init(&enic->cq[cq_idx],
 255                         0 /* flow_control_enable */,
 256                         1 /* color_enable */,
 257                         0 /* cq_head */,
 258                         0 /* cq_tail */,
 259                         1 /* cq_tail_color */,
 260                         0 /* interrupt_enable */,
 261                         1 /* cq_entry_enable */,
 262                         0 /* cq_message_enable */,
 263                         0 /* interrupt offset */,
 264                         0 /* cq_message_addr */);
 265         }
 266
 267         for (index = 0; index < enic->wq_count; index++) {
 268                 vnic_wq_init(&enic->wq[index],
 269                         enic_cq_wq(enic, index),
 270                         error_interrupt_enable,
 271                         error_interrupt_offset);
 272
 273                 cq_idx = enic_cq_wq(enic, index);
 274                 vnic_cq_init(&enic->cq[cq_idx],
 275                         0 /* flow_control_enable */,
 276                         1 /* color_enable */,
 277                         0 /* cq_head */,
 278                         0 /* cq_tail */,
 279                         1 /* cq_tail_color */,
 280                         0 /* interrupt_enable */,
 281                         0 /* cq_entry_enable */,
 282                         1 /* cq_message_enable */,
 283                         0 /* interrupt offset */,
 284                         (u64)enic->wq[index].cqmsg_rz->phys_addr);
 285         }
 286
 287         vnic_intr_init(&enic->intr,
 288                 enic->config.intr_timer_usec,
 289                 enic->config.intr_timer_type,
 290                 /*mask_on_assertion*/1);
 291 }
 292
 293
 294 static int
 295 enic_alloc_rx_queue_mbufs(struct enic *enic, struct vnic_rq *rq)
 296 {
 297         struct rte_mbuf *mb;
 298         struct rq_enet_desc *rqd = rq->ring.descs;
 299         unsigned i;
 300         dma_addr_t dma_addr;
 301
 302         if (!rq->in_use)
 303                 return 0;
 304
 305         dev_debug(enic, "queue %u, allocating %u rx queue mbufs\n", rq->index,
 306                   rq->ring.desc_count);
 307
 308         for (i = 0; i < rq->ring.desc_count; i++, rqd++) {
 309                 mb = rte_mbuf_raw_alloc(rq->mp);
 310                 if (mb == NULL) {
 311                         dev_err(enic, "RX mbuf alloc failed queue_id=%u\n",
 312                         (unsigned)rq->index);
 313                         return -ENOMEM;
 314                 }
 315
 316                 mb->data_off = RTE_PKTMBUF_HEADROOM;
 317                 dma_addr = (dma_addr_t)(mb->buf_physaddr
 318                            + RTE_PKTMBUF_HEADROOM);
 319                 rq_enet_desc_enc(rqd, dma_addr,
 320                                 (rq->is_sop ? RQ_ENET_TYPE_ONLY_SOP
 321                                 : RQ_ENET_TYPE_NOT_SOP),
 322                                 mb->buf_len - RTE_PKTMBUF_HEADROOM);
 323                 rq->mbuf_ring[i] = mb;
 324         }
 325
 326         /* make sure all prior writes are complete before doing the PIO write */
 327         rte_rmb();
 328
 329         /* Post all but the last buffer to VIC. */
 330         rq->posted_index = rq->ring.desc_count - 1;
 331
 332         rq->rx_nb_hold = 0;
 333
 334         dev_debug(enic, "port=%u, qidx=%u, Write %u posted idx, %u sw held\n",
 335                 enic->port_id, rq->index, rq->posted_index, rq->rx_nb_hold);
 336         iowrite32(rq->posted_index, &rq->ctrl->posted_index);
 337         iowrite32(0, &rq->ctrl->fetch_index);
 338         rte_rmb();
 339
 340         return 0;
 341
 342 }
 343
 344 static void *
 345 enic_alloc_consistent(void *priv, size_t size,
 346         dma_addr_t *dma_handle, u8 *name)
 347 {
 348         void *vaddr;
 349         const struct rte_memzone *rz;
 350         *dma_handle = 0;
 351         struct enic *enic = (struct enic *)priv;
 352         struct enic_memzone_entry *mze;
 353
 354         rz = rte_memzone_reserve_aligned((const char *)name,
 355                                          size, SOCKET_ID_ANY, 0, ENIC_ALIGN);
 356         if (!rz) {
 357                 pr_err("%s : Failed to allocate memory requested for %s\n",
 358                         __func__, name);
 359                 return NULL;
 360         }
 361
 362         vaddr = rz->addr;
 363         *dma_handle = (dma_addr_t)rz->phys_addr;
 364
 365         mze = rte_malloc("enic memzone entry",
 366                          sizeof(struct enic_memzone_entry), 0);
 367
 368         if (!mze) {
 369                 pr_err("%s : Failed to allocate memory for memzone list\n",
 370                        __func__);
 371                 rte_memzone_free(rz);
 372         }
 373
 374         mze->rz = rz;
 375
 376         rte_spinlock_lock(&enic->memzone_list_lock);
 377         LIST_INSERT_HEAD(&enic->memzone_list, mze, entries);
 378         rte_spinlock_unlock(&enic->memzone_list_lock);
 379
 380         return vaddr;
 381 }
 382
 383 static void
 384 enic_free_consistent(void *priv,
 385                      __rte_unused size_t size,
 386                      void *vaddr,
 387                      dma_addr_t dma_handle)
 388 {
 389         struct enic_memzone_entry *mze;
 390         struct enic *enic = (struct enic *)priv;
 391
 392         rte_spinlock_lock(&enic->memzone_list_lock);
 393         LIST_FOREACH(mze, &enic->memzone_list, entries) {
 394                 if (mze->rz->addr == vaddr &&
 395                     mze->rz->phys_addr == dma_handle)
 396                         break;
 397         }
 398         if (mze == NULL) {
 399                 rte_spinlock_unlock(&enic->memzone_list_lock);
 400                 dev_warning(enic,
 401                             "Tried to free memory, but couldn't find it in the memzone list\n");
 402                 return;
 403         }
 404         LIST_REMOVE(mze, entries);
 405         rte_spinlock_unlock(&enic->memzone_list_lock);
 406         rte_memzone_free(mze->rz);
 407         rte_free(mze);
 408 }
 409
 410 int enic_link_update(struct enic *enic)
 411 {
 412         struct rte_eth_dev *eth_dev = enic->rte_dev;
 413         int ret;
 414         int link_status = 0;
 415
 416         link_status = enic_get_link_status(enic);
 417         ret = (link_status == enic->link_status);
 418         enic->link_status = link_status;
 419         eth_dev->data->dev_link.link_status = link_status;
 420         eth_dev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
 421         eth_dev->data->dev_link.link_speed = vnic_dev_port_speed(enic->vdev);
 422         return ret;
 423 }
 424
 425 static void
 426 enic_intr_handler(void *arg)
 427 {
 428         struct rte_eth_dev *dev = (struct rte_eth_dev *)arg;
 429         struct enic *enic = pmd_priv(dev);
 430
 431         vnic_intr_return_all_credits(&enic->intr);
 432
 433         enic_link_update(enic);
 434         _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL);
 435         enic_log_q_error(enic);
 436 }
 437
 438 int enic_enable(struct enic *enic)
 439 {
 440         unsigned int index;
 441         int err;
 442         struct rte_eth_dev *eth_dev = enic->rte_dev;
 443
 444         eth_dev->data->dev_link.link_speed = vnic_dev_port_speed(enic->vdev);
 445         eth_dev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
 446
 447         /* vnic notification of link status has already been turned on in
 448          * enic_dev_init() which is called during probe time.  Here we are
 449          * just turning on interrupt vector 0 if needed.
 450          */
 451         if (eth_dev->data->dev_conf.intr_conf.lsc)
 452                 vnic_dev_notify_set(enic->vdev, 0);
 453
 454         if (enic_clsf_init(enic))
 455                 dev_warning(enic, "Init of hash table for clsf failed."\
 456                         "Flow director feature will not work\n");
 457
 458         for (index = 0; index < enic->rq_count; index++) {
 459                 err = enic_alloc_rx_queue_mbufs(enic,
 460                         &enic->rq[enic_rte_rq_idx_to_sop_idx(index)]);
 461                 if (err) {
 462                         dev_err(enic, "Failed to alloc sop RX queue mbufs\n");
 463                         return err;
 464                 }
 465                 err = enic_alloc_rx_queue_mbufs(enic,
 466                         &enic->rq[enic_rte_rq_idx_to_data_idx(index)]);
 467                 if (err) {
 468                         /* release the allocated mbufs for the sop rq*/
 469                         enic_rxmbuf_queue_release(enic,
 470                                 &enic->rq[enic_rte_rq_idx_to_sop_idx(index)]);
 471
 472                         dev_err(enic, "Failed to alloc data RX queue mbufs\n");
 473                         return err;
 474                 }
 475         }
 476
 477         for (index = 0; index < enic->wq_count; index++)
 478                 enic_start_wq(enic, index);
 479         for (index = 0; index < enic->rq_count; index++)
 480                 enic_start_rq(enic, index);
 481
 482         vnic_dev_add_addr(enic->vdev, enic->mac_addr);
 483
 484         vnic_dev_enable_wait(enic->vdev);
 485
 486         /* Register and enable error interrupt */
 487         rte_intr_callback_register(&(enic->pdev->intr_handle),
 488                 enic_intr_handler, (void *)enic->rte_dev);
 489
 490         rte_intr_enable(&(enic->pdev->intr_handle));
 491         vnic_intr_unmask(&enic->intr);
 492
 493         return 0;
 494 }
 495
 496 int enic_alloc_intr_resources(struct enic *enic)
 497 {
 498         int err;
 499
 500         dev_info(enic, "vNIC resources used:  "\
 501                 "wq %d rq %d cq %d intr %d\n",
 502                 enic->wq_count, enic_vnic_rq_count(enic),
 503                 enic->cq_count, enic->intr_count);
 504
 505         err = vnic_intr_alloc(enic->vdev, &enic->intr, 0);
 506         if (err)
 507                 enic_free_vnic_resources(enic);
 508
 509         return err;
 510 }
 511
 512 void enic_free_rq(void *rxq)
 513 {
 514         struct vnic_rq *rq_sop, *rq_data;
 515         struct enic *enic;
 516
 517         if (rxq == NULL)
 518                 return;
 519
 520         rq_sop = (struct vnic_rq *)rxq;
 521         enic = vnic_dev_priv(rq_sop->vdev);
 522         rq_data = &enic->rq[rq_sop->data_queue_idx];
 523
 524         enic_rxmbuf_queue_release(enic, rq_sop);
 525         if (rq_data->in_use)
 526                 enic_rxmbuf_queue_release(enic, rq_data);
 527
 528         rte_free(rq_sop->mbuf_ring);
 529         if (rq_data->in_use)
 530                 rte_free(rq_data->mbuf_ring);
 531
 532         rq_sop->mbuf_ring = NULL;
 533         rq_data->mbuf_ring = NULL;
 534
 535         vnic_rq_free(rq_sop);
 536         if (rq_data->in_use)
 537                 vnic_rq_free(rq_data);
 538
 539         vnic_cq_free(&enic->cq[enic_sop_rq_idx_to_cq_idx(rq_sop->index)]);
 540
 541         rq_sop->in_use = 0;
 542         rq_data->in_use = 0;
 543 }
 544
 545 void enic_start_wq(struct enic *enic, uint16_t queue_idx)
 546 {
 547         struct rte_eth_dev *eth_dev = enic->rte_dev;
 548         vnic_wq_enable(&enic->wq[queue_idx]);
 549         eth_dev->data->tx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STARTED;
 550 }
 551
 552 int enic_stop_wq(struct enic *enic, uint16_t queue_idx)
 553 {
 554         struct rte_eth_dev *eth_dev = enic->rte_dev;
 555         int ret;
 556
 557         ret = vnic_wq_disable(&enic->wq[queue_idx]);
 558         if (ret)
 559                 return ret;
 560
 561         eth_dev->data->tx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STOPPED;
 562         return 0;
 563 }
 564
 565 void enic_start_rq(struct enic *enic, uint16_t queue_idx)
 566 {
 567         struct vnic_rq *rq_sop;
 568         struct vnic_rq *rq_data;
 569         rq_sop = &enic->rq[enic_rte_rq_idx_to_sop_idx(queue_idx)];
 570         rq_data = &enic->rq[rq_sop->data_queue_idx];
 571         struct rte_eth_dev *eth_dev = enic->rte_dev;
 572
 573         if (rq_data->in_use)
 574                 vnic_rq_enable(rq_data);
 575         rte_mb();
 576         vnic_rq_enable(rq_sop);
 577         eth_dev->data->rx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STARTED;
 578 }
 579
 580 int enic_stop_rq(struct enic *enic, uint16_t queue_idx)
 581 {
 582         int ret1 = 0, ret2 = 0;
 583         struct rte_eth_dev *eth_dev = enic->rte_dev;
 584         struct vnic_rq *rq_sop;
 585         struct vnic_rq *rq_data;
 586         rq_sop = &enic->rq[enic_rte_rq_idx_to_sop_idx(queue_idx)];
 587         rq_data = &enic->rq[rq_sop->data_queue_idx];
 588
 589         ret2 = vnic_rq_disable(rq_sop);
 590         rte_mb();
 591         if (rq_data->in_use)
 592                 ret1 = vnic_rq_disable(rq_data);
 593
 594         if (ret2)
 595                 return ret2;
 596         else if (ret1)
 597                 return ret1;
 598
 599         eth_dev->data->rx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STOPPED;
 600         return 0;
 601 }
 602
 603 int enic_alloc_rq(struct enic *enic, uint16_t queue_idx,
 604         unsigned int socket_id, struct rte_mempool *mp,
 605         uint16_t nb_desc, uint16_t free_thresh)
 606 {
 607         int rc;
 608         uint16_t sop_queue_idx = enic_rte_rq_idx_to_sop_idx(queue_idx);
 609         uint16_t data_queue_idx = enic_rte_rq_idx_to_data_idx(queue_idx);
 610         struct vnic_rq *rq_sop = &enic->rq[sop_queue_idx];
 611         struct vnic_rq *rq_data = &enic->rq[data_queue_idx];
 612         unsigned int mbuf_size, mbufs_per_pkt;
 613         unsigned int nb_sop_desc, nb_data_desc;
 614         uint16_t min_sop, max_sop, min_data, max_data;
 615         uint16_t mtu = enic->rte_dev->data->mtu;
 616
 617         rq_sop->is_sop = 1;
 618         rq_sop->data_queue_idx = data_queue_idx;
 619         rq_data->is_sop = 0;
 620         rq_data->data_queue_idx = 0;
 621         rq_sop->socket_id = socket_id;
 622         rq_sop->mp = mp;
 623         rq_data->socket_id = socket_id;
 624         rq_data->mp = mp;
 625         rq_sop->in_use = 1;
 626         rq_sop->rx_free_thresh = free_thresh;
 627         rq_data->rx_free_thresh = free_thresh;
 628         dev_debug(enic, "Set queue_id:%u free thresh:%u\n", queue_idx,
 629                   free_thresh);
 630
 631         mbuf_size = (uint16_t)(rte_pktmbuf_data_room_size(mp) -
 632                                RTE_PKTMBUF_HEADROOM);
 633
 634         if (enic->rte_dev->data->dev_conf.rxmode.enable_scatter) {
 635                 dev_info(enic, "Rq %u Scatter rx mode enabled\n", queue_idx);
 636                 /* ceil((mtu + ETHER_HDR_LEN + 4)/mbuf_size) */
 637                 mbufs_per_pkt = ((mtu + ETHER_HDR_LEN + 4) +
 638                                  (mbuf_size - 1)) / mbuf_size;
 639         } else {
 640                 dev_info(enic, "Scatter rx mode disabled\n");
 641                 mbufs_per_pkt = 1;
 642         }
 643
 644         if (mbufs_per_pkt > 1) {
 645                 dev_info(enic, "Rq %u Scatter rx mode in use\n", queue_idx);
 646                 rq_sop->data_queue_enable = 1;
 647                 rq_data->in_use = 1;
 648         } else {
 649                 dev_info(enic, "Rq %u Scatter rx mode not being used\n",
 650                          queue_idx);
 651                 rq_sop->data_queue_enable = 0;
 652                 rq_data->in_use = 0;
 653         }
 654
 655         /* number of descriptors have to be a multiple of 32 */
 656         nb_sop_desc = (nb_desc / mbufs_per_pkt) & ~0x1F;
 657         nb_data_desc = (nb_desc - nb_sop_desc) & ~0x1F;
 658
 659         rq_sop->max_mbufs_per_pkt = mbufs_per_pkt;
 660         rq_data->max_mbufs_per_pkt = mbufs_per_pkt;
 661
 662         if (mbufs_per_pkt > 1) {
 663                 min_sop = 64;
 664                 max_sop = ((enic->config.rq_desc_count /
 665                             (mbufs_per_pkt - 1)) & ~0x1F);
 666                 min_data = min_sop * (mbufs_per_pkt - 1);
 667                 max_data = enic->config.rq_desc_count;
 668         } else {
 669                 min_sop = 64;
 670                 max_sop = enic->config.rq_desc_count;
 671                 min_data = 0;
 672                 max_data = 0;
 673         }
 674
 675         if (nb_desc < (min_sop + min_data)) {
 676                 dev_warning(enic,
 677                             "Number of rx descs too low, adjusting to minimum\n");
 678                 nb_sop_desc = min_sop;
 679                 nb_data_desc = min_data;
 680         } else if (nb_desc > (max_sop + max_data)) {
 681                 dev_warning(enic,
 682                             "Number of rx_descs too high, adjusting to maximum\n");
 683                 nb_sop_desc = max_sop;
 684                 nb_data_desc = max_data;
 685         }
 686         if (mbufs_per_pkt > 1) {
 687                 dev_info(enic, "For mtu %d and mbuf size %d valid rx descriptor range is %d to %d\n",
 688                          mtu, mbuf_size, min_sop + min_data,
 689                          max_sop + max_data);
 690         }
 691         dev_info(enic, "Using %d rx descriptors (sop %d, data %d)\n",
 692                  nb_sop_desc + nb_data_desc, nb_sop_desc, nb_data_desc);
 693
 694         /* Allocate sop queue resources */
 695         rc = vnic_rq_alloc(enic->vdev, rq_sop, sop_queue_idx,
 696                 nb_sop_desc, sizeof(struct rq_enet_desc));
 697         if (rc) {
 698                 dev_err(enic, "error in allocation of sop rq\n");
 699                 goto err_exit;
 700         }
 701         nb_sop_desc = rq_sop->ring.desc_count;
 702
 703         if (rq_data->in_use) {
 704                 /* Allocate data queue resources */
 705                 rc = vnic_rq_alloc(enic->vdev, rq_data, data_queue_idx,
 706                                    nb_data_desc,
 707                                    sizeof(struct rq_enet_desc));
 708                 if (rc) {
 709                         dev_err(enic, "error in allocation of data rq\n");
 710                         goto err_free_rq_sop;
 711                 }
 712                 nb_data_desc = rq_data->ring.desc_count;
 713         }
 714         rc = vnic_cq_alloc(enic->vdev, &enic->cq[queue_idx], queue_idx,
 715                            socket_id, nb_sop_desc + nb_data_desc,
 716                            sizeof(struct cq_enet_rq_desc));
 717         if (rc) {
 718                 dev_err(enic, "error in allocation of cq for rq\n");
 719                 goto err_free_rq_data;
 720         }
 721
 722         /* Allocate the mbuf rings */
 723         rq_sop->mbuf_ring = (struct rte_mbuf **)
 724                 rte_zmalloc_socket("rq->mbuf_ring",
 725                                    sizeof(struct rte_mbuf *) * nb_sop_desc,
 726                                    RTE_CACHE_LINE_SIZE, rq_sop->socket_id);
 727         if (rq_sop->mbuf_ring == NULL)
 728                 goto err_free_cq;
 729
 730         if (rq_data->in_use) {
 731                 rq_data->mbuf_ring = (struct rte_mbuf **)
 732                         rte_zmalloc_socket("rq->mbuf_ring",
 733                                 sizeof(struct rte_mbuf *) * nb_data_desc,
 734                                 RTE_CACHE_LINE_SIZE, rq_sop->socket_id);
 735                 if (rq_data->mbuf_ring == NULL)
 736                         goto err_free_sop_mbuf;
 737         }
 738
 739         rq_sop->tot_nb_desc = nb_desc; /* squirl away for MTU update function */
 740
 741         return 0;
 742
 743 err_free_sop_mbuf:
 744         rte_free(rq_sop->mbuf_ring);
 745 err_free_cq:
 746         /* cleanup on error */
 747         vnic_cq_free(&enic->cq[queue_idx]);
 748 err_free_rq_data:
 749         if (rq_data->in_use)
 750                 vnic_rq_free(rq_data);
 751 err_free_rq_sop:
 752         vnic_rq_free(rq_sop);
 753 err_exit:
 754         return -ENOMEM;
 755 }
 756
 757 void enic_free_wq(void *txq)
 758 {
 759         struct vnic_wq *wq;
 760         struct enic *enic;
 761
 762         if (txq == NULL)
 763                 return;
 764
 765         wq = (struct vnic_wq *)txq;
 766         enic = vnic_dev_priv(wq->vdev);
 767         rte_memzone_free(wq->cqmsg_rz);
 768         vnic_wq_free(wq);
 769         vnic_cq_free(&enic->cq[enic->rq_count + wq->index]);
 770 }
 771
 772 int enic_alloc_wq(struct enic *enic, uint16_t queue_idx,
 773         unsigned int socket_id, uint16_t nb_desc)
 774 {
 775         int err;
 776         struct vnic_wq *wq = &enic->wq[queue_idx];
 777         unsigned int cq_index = enic_cq_wq(enic, queue_idx);
 778         char name[NAME_MAX];
 779         static int instance;
 780
 781         wq->socket_id = socket_id;
 782         if (nb_desc) {
 783                 if (nb_desc > enic->config.wq_desc_count) {
 784                         dev_warning(enic,
 785                                 "WQ %d - number of tx desc in cmd line (%d)"\
 786                                 "is greater than that in the UCSM/CIMC adapter"\
 787                                 "policy.  Applying the value in the adapter "\
 788                                 "policy (%d)\n",
 789                                 queue_idx, nb_desc, enic->config.wq_desc_count);
 790                 } else if (nb_desc != enic->config.wq_desc_count) {
 791                         enic->config.wq_desc_count = nb_desc;
 792                         dev_info(enic,
 793                                 "TX Queues - effective number of descs:%d\n",
 794                                 nb_desc);
 795                 }
 796         }
 797
 798         /* Allocate queue resources */
 799         err = vnic_wq_alloc(enic->vdev, &enic->wq[queue_idx], queue_idx,
 800                 enic->config.wq_desc_count,
 801                 sizeof(struct wq_enet_desc));
 802         if (err) {
 803                 dev_err(enic, "error in allocation of wq\n");
 804                 return err;
 805         }
 806
 807         err = vnic_cq_alloc(enic->vdev, &enic->cq[cq_index], cq_index,
 808                 socket_id, enic->config.wq_desc_count,
 809                 sizeof(struct cq_enet_wq_desc));
 810         if (err) {
 811                 vnic_wq_free(wq);
 812                 dev_err(enic, "error in allocation of cq for wq\n");
 813         }
 814
 815         /* setup up CQ message */
 816         snprintf((char *)name, sizeof(name),
 817                  "vnic_cqmsg-%s-%d-%d", enic->bdf_name, queue_idx,
 818                 instance++);
 819
 820         wq->cqmsg_rz = rte_memzone_reserve_aligned((const char *)name,
 821                                                    sizeof(uint32_t),
 822                                                    SOCKET_ID_ANY, 0,
 823                                                    ENIC_ALIGN);
 824         if (!wq->cqmsg_rz)
 825                 return -ENOMEM;
 826
 827         return err;
 828 }
 829
 830 int enic_disable(struct enic *enic)
 831 {
 832         unsigned int i;
 833         int err;
 834
 835         vnic_intr_mask(&enic->intr);
 836         (void)vnic_intr_masked(&enic->intr); /* flush write */
 837         rte_intr_disable(&enic->pdev->intr_handle);
 838         rte_intr_callback_unregister(&enic->pdev->intr_handle,
 839                                      enic_intr_handler,
 840                                      (void *)enic->rte_dev);
 841
 842         vnic_dev_disable(enic->vdev);
 843
 844         enic_clsf_destroy(enic);
 845
 846         if (!enic_is_sriov_vf(enic))
 847                 vnic_dev_del_addr(enic->vdev, enic->mac_addr);
 848
 849         for (i = 0; i < enic->wq_count; i++) {
 850                 err = vnic_wq_disable(&enic->wq[i]);
 851                 if (err)
 852                         return err;
 853         }
 854         for (i = 0; i < enic_vnic_rq_count(enic); i++) {
 855                 if (enic->rq[i].in_use) {
 856                         err = vnic_rq_disable(&enic->rq[i]);
 857                         if (err)
 858                                 return err;
 859                 }
 860         }
 861
 862         /* If we were using interrupts, set the interrupt vector to -1
 863          * to disable interrupts.  We are not disabling link notifcations,
 864          * though, as we want the polling of link status to continue working.
 865          */
 866         if (enic->rte_dev->data->dev_conf.intr_conf.lsc)
 867                 vnic_dev_notify_set(enic->vdev, -1);
 868
 869         vnic_dev_set_reset_flag(enic->vdev, 1);
 870
 871         for (i = 0; i < enic->wq_count; i++)
 872                 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
 873
 874         for (i = 0; i < enic_vnic_rq_count(enic); i++)
 875                 if (enic->rq[i].in_use)
 876                         vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
 877         for (i = 0; i < enic->cq_count; i++)
 878                 vnic_cq_clean(&enic->cq[i]);
 879         vnic_intr_clean(&enic->intr);
 880
 881         return 0;
 882 }
 883
 884 static int enic_dev_wait(struct vnic_dev *vdev,
 885         int (*start)(struct vnic_dev *, int),
 886         int (*finished)(struct vnic_dev *, int *),
 887         int arg)
 888 {
 889         int done;
 890         int err;
 891         int i;
 892
 893         err = start(vdev, arg);
 894         if (err)
 895                 return err;
 896
 897         /* Wait for func to complete...2 seconds max */
 898         for (i = 0; i < 2000; i++) {
 899                 err = finished(vdev, &done);
 900                 if (err)
 901                         return err;
 902                 if (done)
 903                         return 0;
 904                 usleep(1000);
 905         }
 906         return -ETIMEDOUT;
 907 }
 908
 909 static int enic_dev_open(struct enic *enic)
 910 {
 911         int err;
 912
 913         err = enic_dev_wait(enic->vdev, vnic_dev_open,
 914                 vnic_dev_open_done, 0);
 915         if (err)
 916                 dev_err(enic_get_dev(enic),
 917                         "vNIC device open failed, err %d\n", err);
 918
 919         return err;
 920 }
 921
 922 static int enic_set_rsskey(struct enic *enic)
 923 {
 924         dma_addr_t rss_key_buf_pa;
 925         union vnic_rss_key *rss_key_buf_va = NULL;
 926         static union vnic_rss_key rss_key = {
 927                 .key = {
 928                         [0] = {.b = {85, 67, 83, 97, 119, 101, 115, 111, 109, 101}},
 929                         [1] = {.b = {80, 65, 76, 79, 117, 110, 105, 113, 117, 101}},
 930                         [2] = {.b = {76, 73, 78, 85, 88, 114, 111, 99, 107, 115}},
 931                         [3] = {.b = {69, 78, 73, 67, 105, 115, 99, 111, 111, 108}},
 932                 }
 933         };
 934         int err;
 935         u8 name[NAME_MAX];
 936
 937         snprintf((char *)name, NAME_MAX, "rss_key-%s", enic->bdf_name);
 938         rss_key_buf_va = enic_alloc_consistent(enic, sizeof(union vnic_rss_key),
 939                 &rss_key_buf_pa, name);
 940         if (!rss_key_buf_va)
 941                 return -ENOMEM;
 942
 943         rte_memcpy(rss_key_buf_va, &rss_key, sizeof(union vnic_rss_key));
 944
 945         err = enic_set_rss_key(enic,
 946                 rss_key_buf_pa,
 947                 sizeof(union vnic_rss_key));
 948
 949         enic_free_consistent(enic, sizeof(union vnic_rss_key),
 950                 rss_key_buf_va, rss_key_buf_pa);
 951
 952         return err;
 953 }
 954
 955 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
 956 {
 957         dma_addr_t rss_cpu_buf_pa;
 958         union vnic_rss_cpu *rss_cpu_buf_va = NULL;
 959         int i;
 960         int err;
 961         u8 name[NAME_MAX];
 962
 963         snprintf((char *)name, NAME_MAX, "rss_cpu-%s", enic->bdf_name);
 964         rss_cpu_buf_va = enic_alloc_consistent(enic, sizeof(union vnic_rss_cpu),
 965                 &rss_cpu_buf_pa, name);
 966         if (!rss_cpu_buf_va)
 967                 return -ENOMEM;
 968
 969         for (i = 0; i < (1 << rss_hash_bits); i++)
 970                 (*rss_cpu_buf_va).cpu[i / 4].b[i % 4] =
 971                         enic_rte_rq_idx_to_sop_idx(i % enic->rq_count);
 972
 973         err = enic_set_rss_cpu(enic,
 974                 rss_cpu_buf_pa,
 975                 sizeof(union vnic_rss_cpu));
 976
 977         enic_free_consistent(enic, sizeof(union vnic_rss_cpu),
 978                 rss_cpu_buf_va, rss_cpu_buf_pa);
 979
 980         return err;
 981 }
 982
 983 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
 984         u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
 985 {
 986         const u8 tso_ipid_split_en = 0;
 987         int err;
 988
 989         /* Enable VLAN tag stripping */
 990
 991         err = enic_set_nic_cfg(enic,
 992                 rss_default_cpu, rss_hash_type,
 993                 rss_hash_bits, rss_base_cpu,
 994                 rss_enable, tso_ipid_split_en,
 995                 enic->ig_vlan_strip_en);
 996
 997         return err;
 998 }
 999
1000 int enic_set_rss_nic_cfg(struct enic *enic)
1001 {
1002         const u8 rss_default_cpu = 0;
1003         const u8 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
1004             NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
1005             NIC_CFG_RSS_HASH_TYPE_IPV6 |
1006             NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
1007         const u8 rss_hash_bits = 7;
1008         const u8 rss_base_cpu = 0;
1009         u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
1010
1011         if (rss_enable) {
1012                 if (!enic_set_rsskey(enic)) {
1013                         if (enic_set_rsscpu(enic, rss_hash_bits)) {
1014                                 rss_enable = 0;
1015                                 dev_warning(enic, "RSS disabled, "\
1016                                         "Failed to set RSS cpu indirection table.");
1017                         }
1018                 } else {
1019                         rss_enable = 0;
1020                         dev_warning(enic,
1021                                 "RSS disabled, Failed to set RSS key.\n");
1022                 }
1023         }
1024
1025         return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
1026                 rss_hash_bits, rss_base_cpu, rss_enable);
1027 }
1028
1029 int enic_setup_finish(struct enic *enic)
1030 {
1031         int ret;
1032
1033         enic_init_soft_stats(enic);
1034
1035         ret = enic_set_rss_nic_cfg(enic);
1036         if (ret) {
1037                 dev_err(enic, "Failed to config nic, aborting.\n");
1038                 return -1;
1039         }
1040
1041         /* Default conf */
1042         vnic_dev_packet_filter(enic->vdev,
1043                 1 /* directed  */,
1044                 1 /* multicast */,
1045                 1 /* broadcast */,
1046                 0 /* promisc   */,
1047                 1 /* allmulti  */);
1048
1049         enic->promisc = 0;
1050         enic->allmulti = 1;
1051
1052         return 0;
1053 }
1054
1055 void enic_add_packet_filter(struct enic *enic)
1056 {
1057         /* Args -> directed, multicast, broadcast, promisc, allmulti */
1058         vnic_dev_packet_filter(enic->vdev, 1, 1, 1,
1059                 enic->promisc, enic->allmulti);
1060 }
1061
1062 int enic_get_link_status(struct enic *enic)
1063 {
1064         return vnic_dev_link_status(enic->vdev);
1065 }
1066
1067 static void enic_dev_deinit(struct enic *enic)
1068 {
1069         struct rte_eth_dev *eth_dev = enic->rte_dev;
1070
1071         /* stop link status checking */
1072         vnic_dev_notify_unset(enic->vdev);
1073
1074         rte_free(eth_dev->data->mac_addrs);
1075 }
1076
1077
1078 int enic_set_vnic_res(struct enic *enic)
1079 {
1080         struct rte_eth_dev *eth_dev = enic->rte_dev;
1081         int rc = 0;
1082
1083         /* With Rx scatter support, two RQs are now used per RQ used by
1084          * the application.
1085          */
1086         if (enic->conf_rq_count < eth_dev->data->nb_rx_queues) {
1087                 dev_err(dev, "Not enough Receive queues. Requested:%u which uses %d RQs on VIC, Configured:%u\n",
1088                         eth_dev->data->nb_rx_queues,
1089                         eth_dev->data->nb_rx_queues * 2, enic->conf_rq_count);
1090                 rc = -EINVAL;
1091         }
1092         if (enic->conf_wq_count < eth_dev->data->nb_tx_queues) {
1093                 dev_err(dev, "Not enough Transmit queues. Requested:%u, Configured:%u\n",
1094                         eth_dev->data->nb_tx_queues, enic->conf_wq_count);
1095                 rc = -EINVAL;
1096         }
1097
1098         if (enic->conf_cq_count < (eth_dev->data->nb_rx_queues +
1099                                    eth_dev->data->nb_tx_queues)) {
1100                 dev_err(dev, "Not enough Completion queues. Required:%u, Configured:%u\n",
1101                         (eth_dev->data->nb_rx_queues +
1102                          eth_dev->data->nb_tx_queues), enic->conf_cq_count);
1103                 rc = -EINVAL;
1104         }
1105
1106         if (rc == 0) {
1107                 enic->rq_count = eth_dev->data->nb_rx_queues;
1108                 enic->wq_count = eth_dev->data->nb_tx_queues;
1109                 enic->cq_count = enic->rq_count + enic->wq_count;
1110         }
1111
1112         return rc;
1113 }
1114
1115 /* Initialize the completion queue for an RQ */
1116 static int
1117 enic_reinit_rq(struct enic *enic, unsigned int rq_idx)
1118 {
1119         struct vnic_rq *sop_rq, *data_rq;
1120         unsigned int cq_idx = enic_cq_rq(enic, rq_idx);
1121         int rc = 0;
1122
1123         sop_rq = &enic->rq[enic_rte_rq_idx_to_sop_idx(rq_idx)];
1124         data_rq = &enic->rq[enic_rte_rq_idx_to_data_idx(rq_idx)];
1125
1126         vnic_cq_clean(&enic->cq[cq_idx]);
1127         vnic_cq_init(&enic->cq[cq_idx],
1128                      0 /* flow_control_enable */,
1129                      1 /* color_enable */,
1130                      0 /* cq_head */,
1131                      0 /* cq_tail */,
1132                      1 /* cq_tail_color */,
1133                      0 /* interrupt_enable */,
1134                      1 /* cq_entry_enable */,
1135                      0 /* cq_message_enable */,
1136                      0 /* interrupt offset */,
1137                      0 /* cq_message_addr */);
1138
1139
1140         vnic_rq_init_start(sop_rq, enic_cq_rq(enic,
1141                            enic_rte_rq_idx_to_sop_idx(rq_idx)), 0,
1142                            sop_rq->ring.desc_count - 1, 1, 0);
1143         if (data_rq->in_use) {
1144                 vnic_rq_init_start(data_rq,
1145                                    enic_cq_rq(enic,
1146                                    enic_rte_rq_idx_to_data_idx(rq_idx)), 0,
1147                                    data_rq->ring.desc_count - 1, 1, 0);
1148         }
1149
1150         rc = enic_alloc_rx_queue_mbufs(enic, sop_rq);
1151         if (rc)
1152                 return rc;
1153
1154         if (data_rq->in_use) {
1155                 rc = enic_alloc_rx_queue_mbufs(enic, data_rq);
1156                 if (rc) {
1157                         enic_rxmbuf_queue_release(enic, sop_rq);
1158                         return rc;
1159                 }
1160         }
1161
1162         return 0;
1163 }
1164
1165 /* The Cisco NIC can send and receive packets up to a max packet size
1166  * determined by the NIC type and firmware. There is also an MTU
1167  * configured into the NIC via the CIMC/UCSM management interface
1168  * which can be overridden by this function (up to the max packet size).
1169  * Depending on the network setup, doing so may cause packet drops
1170  * and unexpected behavior.
1171  */
1172 int enic_set_mtu(struct enic *enic, uint16_t new_mtu)
1173 {
1174         unsigned int rq_idx;
1175         struct vnic_rq *rq;
1176         int rc = 0;
1177         uint16_t old_mtu;       /* previous setting */
1178         uint16_t config_mtu;    /* Value configured into NIC via CIMC/UCSM */
1179         struct rte_eth_dev *eth_dev = enic->rte_dev;
1180
1181         old_mtu = eth_dev->data->mtu;
1182         config_mtu = enic->config.mtu;
1183
1184         if (new_mtu > enic->max_mtu) {
1185                 dev_err(enic,
1186                         "MTU not updated: requested (%u) greater than max (%u)\n",
1187                         new_mtu, enic->max_mtu);
1188                 return -EINVAL;
1189         }
1190         if (new_mtu < ENIC_MIN_MTU) {
1191                 dev_info(enic,
1192                         "MTU not updated: requested (%u) less than min (%u)\n",
1193                         new_mtu, ENIC_MIN_MTU);
1194                 return -EINVAL;
1195         }
1196         if (new_mtu > config_mtu)
1197                 dev_warning(enic,
1198                         "MTU (%u) is greater than value configured in NIC (%u)\n",
1199                         new_mtu, config_mtu);
1200
1201         /* The easy case is when scatter is disabled. However if the MTU
1202          * becomes greater than the mbuf data size, packet drops will ensue.
1203          */
1204         if (!enic->rte_dev->data->dev_conf.rxmode.enable_scatter) {
1205                 eth_dev->data->mtu = new_mtu;
1206                 goto set_mtu_done;
1207         }
1208
1209         /* Rx scatter is enabled so reconfigure RQ's on the fly. The point is to
1210          * change Rx scatter mode if necessary for better performance. I.e. if
1211          * MTU was greater than the mbuf size and now it's less, scatter Rx
1212          * doesn't have to be used and vice versa.
1213           */
1214         rte_spinlock_lock(&enic->mtu_lock);
1215
1216         /* Stop traffic on all RQs */
1217         for (rq_idx = 0; rq_idx < enic->rq_count * 2; rq_idx++) {
1218                 rq = &enic->rq[rq_idx];
1219                 if (rq->is_sop && rq->in_use) {
1220                         rc = enic_stop_rq(enic,
1221                                           enic_sop_rq_idx_to_rte_idx(rq_idx));
1222                         if (rc) {
1223                                 dev_err(enic, "Failed to stop Rq %u\n", rq_idx);
1224                                 goto set_mtu_done;
1225                         }
1226                 }
1227         }
1228
1229         /* replace Rx funciton with a no-op to avoid getting stale pkts */
1230         eth_dev->rx_pkt_burst = enic_dummy_recv_pkts;
1231         rte_mb();
1232
1233         /* Allow time for threads to exit the real Rx function. */
1234         usleep(100000);
1235
1236         /* now it is safe to reconfigure the RQs */
1237
1238         /* update the mtu */
1239         eth_dev->data->mtu = new_mtu;
1240
1241         /* free and reallocate RQs with the new MTU */
1242         for (rq_idx = 0; rq_idx < enic->rq_count; rq_idx++) {
1243                 rq = &enic->rq[enic_rte_rq_idx_to_sop_idx(rq_idx)];
1244
1245                 enic_free_rq(rq);
1246                 rc = enic_alloc_rq(enic, rq_idx, rq->socket_id, rq->mp,
1247                                    rq->tot_nb_desc, rq->rx_free_thresh);
1248                 if (rc) {
1249                         dev_err(enic,
1250                                 "Fatal MTU alloc error- No traffic will pass\n");
1251                         goto set_mtu_done;
1252                 }
1253
1254                 rc = enic_reinit_rq(enic, rq_idx);
1255                 if (rc) {
1256                         dev_err(enic,
1257                                 "Fatal MTU RQ reinit- No traffic will pass\n");
1258                         goto set_mtu_done;
1259                 }
1260         }
1261
1262         /* put back the real receive function */
1263         rte_mb();
1264         eth_dev->rx_pkt_burst = enic_recv_pkts;
1265         rte_mb();
1266
1267         /* restart Rx traffic */
1268         for (rq_idx = 0; rq_idx < enic->rq_count; rq_idx++) {
1269                 rq = &enic->rq[enic_rte_rq_idx_to_sop_idx(rq_idx)];
1270                 if (rq->is_sop && rq->in_use)
1271                         enic_start_rq(enic, rq_idx);
1272         }
1273
1274 set_mtu_done:
1275         dev_info(enic, "MTU changed from %u to %u\n",  old_mtu, new_mtu);
1276         rte_spinlock_unlock(&enic->mtu_lock);
1277         return rc;
1278 }
1279
1280 static int enic_dev_init(struct enic *enic)
1281 {
1282         int err;
1283         struct rte_eth_dev *eth_dev = enic->rte_dev;
1284
1285         vnic_dev_intr_coal_timer_info_default(enic->vdev);
1286
1287         /* Get vNIC configuration
1288         */
1289         err = enic_get_vnic_config(enic);
1290         if (err) {
1291                 dev_err(dev, "Get vNIC configuration failed, aborting\n");
1292                 return err;
1293         }
1294
1295         /* Get available resource counts */
1296         enic_get_res_counts(enic);
1297         if (enic->conf_rq_count == 1) {
1298                 dev_err(enic, "Running with only 1 RQ configured in the vNIC is not supported.\n");
1299                 dev_err(enic, "Please configure 2 RQs in the vNIC for each Rx queue used by DPDK.\n");
1300                 dev_err(enic, "See the ENIC PMD guide for more information.\n");
1301                 return -EINVAL;
1302         }
1303
1304         /* Get the supported filters */
1305         enic_fdir_info(enic);
1306
1307         eth_dev->data->mac_addrs = rte_zmalloc("enic_mac_addr", ETH_ALEN
1308                                                 * ENIC_MAX_MAC_ADDR, 0);
1309         if (!eth_dev->data->mac_addrs) {
1310                 dev_err(enic, "mac addr storage alloc failed, aborting.\n");
1311                 return -1;
1312         }
1313         ether_addr_copy((struct ether_addr *) enic->mac_addr,
1314                         eth_dev->data->mac_addrs);
1315
1316         vnic_dev_set_reset_flag(enic->vdev, 0);
1317
1318         /* set up link status checking */
1319         vnic_dev_notify_set(enic->vdev, -1); /* No Intr for notify */
1320
1321         return 0;
1322
1323 }
1324
1325 int enic_probe(struct enic *enic)
1326 {
1327         struct rte_pci_device *pdev = enic->pdev;
1328         int err = -1;
1329
1330         dev_debug(enic, " Initializing ENIC PMD\n");
1331
1332         enic->bar0.vaddr = (void *)pdev->mem_resource[0].addr;
1333         enic->bar0.len = pdev->mem_resource[0].len;
1334
1335         /* Register vNIC device */
1336         enic->vdev = vnic_dev_register(NULL, enic, enic->pdev, &enic->bar0, 1);
1337         if (!enic->vdev) {
1338                 dev_err(enic, "vNIC registration failed, aborting\n");
1339                 goto err_out;
1340         }
1341
1342         LIST_INIT(&enic->memzone_list);
1343         rte_spinlock_init(&enic->memzone_list_lock);
1344
1345         vnic_register_cbacks(enic->vdev,
1346                 enic_alloc_consistent,
1347                 enic_free_consistent);
1348
1349         /* Issue device open to get device in known state */
1350         err = enic_dev_open(enic);
1351         if (err) {
1352                 dev_err(enic, "vNIC dev open failed, aborting\n");
1353                 goto err_out_unregister;
1354         }
1355
1356         /* Set ingress vlan rewrite mode before vnic initialization */
1357         err = vnic_dev_set_ig_vlan_rewrite_mode(enic->vdev,
1358                 IG_VLAN_REWRITE_MODE_PASS_THRU);
1359         if (err) {
1360                 dev_err(enic,
1361                         "Failed to set ingress vlan rewrite mode, aborting.\n");
1362                 goto err_out_dev_close;
1363         }
1364
1365         /* Issue device init to initialize the vnic-to-switch link.
1366          * We'll start with carrier off and wait for link UP
1367          * notification later to turn on carrier.  We don't need
1368          * to wait here for the vnic-to-switch link initialization
1369          * to complete; link UP notification is the indication that
1370          * the process is complete.
1371          */
1372
1373         err = vnic_dev_init(enic->vdev, 0);
1374         if (err) {
1375                 dev_err(enic, "vNIC dev init failed, aborting\n");
1376                 goto err_out_dev_close;
1377         }
1378
1379         err = enic_dev_init(enic);
1380         if (err) {
1381                 dev_err(enic, "Device initialization failed, aborting\n");
1382                 goto err_out_dev_close;
1383         }
1384
1385         return 0;
1386
1387 err_out_dev_close:
1388         vnic_dev_close(enic->vdev);
1389 err_out_unregister:
1390         vnic_dev_unregister(enic->vdev);
1391 err_out:
1392         return err;
1393 }
1394
1395 void enic_remove(struct enic *enic)
1396 {
1397         enic_dev_deinit(enic);
1398         vnic_dev_close(enic->vdev);
1399         vnic_dev_unregister(enic->vdev);
1400 }