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_bus_pci.h>
  44 #include <rte_memzone.h>
  45 #include <rte_malloc.h>
  46 #include <rte_mbuf.h>
  47 #include <rte_string_fns.h>
  48 #include <rte_ethdev.h>
  49
  50 #include "enic_compat.h"
  51 #include "enic.h"
  52 #include "wq_enet_desc.h"
  53 #include "rq_enet_desc.h"
  54 #include "cq_enet_desc.h"
  55 #include "vnic_enet.h"
  56 #include "vnic_dev.h"
  57 #include "vnic_wq.h"
  58 #include "vnic_rq.h"
  59 #include "vnic_cq.h"
  60 #include "vnic_intr.h"
  61 #include "vnic_nic.h"
  62
  63 static inline int enic_is_sriov_vf(struct enic *enic)
  64 {
  65         return enic->pdev->id.device_id == PCI_DEVICE_ID_CISCO_VIC_ENET_VF;
  66 }
  67
  68 static int is_zero_addr(uint8_t *addr)
  69 {
  70         return !(addr[0] |  addr[1] | addr[2] | addr[3] | addr[4] | addr[5]);
  71 }
  72
  73 static int is_mcast_addr(uint8_t *addr)
  74 {
  75         return addr[0] & 1;
  76 }
  77
  78 static int is_eth_addr_valid(uint8_t *addr)
  79 {
  80         return !is_mcast_addr(addr) && !is_zero_addr(addr);
  81 }
  82
  83 static void
  84 enic_rxmbuf_queue_release(__rte_unused struct enic *enic, struct vnic_rq *rq)
  85 {
  86         uint16_t i;
  87
  88         if (!rq || !rq->mbuf_ring) {
  89                 dev_debug(enic, "Pointer to rq or mbuf_ring is NULL");
  90                 return;
  91         }
  92
  93         for (i = 0; i < rq->ring.desc_count; i++) {
  94                 if (rq->mbuf_ring[i]) {
  95                         rte_pktmbuf_free_seg(rq->mbuf_ring[i]);
  96                         rq->mbuf_ring[i] = NULL;
  97                 }
  98         }
  99 }
 100
 101 static void enic_free_wq_buf(struct vnic_wq_buf *buf)
 102 {
 103         struct rte_mbuf *mbuf = (struct rte_mbuf *)buf->mb;
 104
 105         rte_pktmbuf_free_seg(mbuf);
 106         buf->mb = NULL;
 107 }
 108
 109 static void enic_log_q_error(struct enic *enic)
 110 {
 111         unsigned int i;
 112         u32 error_status;
 113
 114         for (i = 0; i < enic->wq_count; i++) {
 115                 error_status = vnic_wq_error_status(&enic->wq[i]);
 116                 if (error_status)
 117                         dev_err(enic, "WQ[%d] error_status %d\n", i,
 118                                 error_status);
 119         }
 120
 121         for (i = 0; i < enic_vnic_rq_count(enic); i++) {
 122                 if (!enic->rq[i].in_use)
 123                         continue;
 124                 error_status = vnic_rq_error_status(&enic->rq[i]);
 125                 if (error_status)
 126                         dev_err(enic, "RQ[%d] error_status %d\n", i,
 127                                 error_status);
 128         }
 129 }
 130
 131 static void enic_clear_soft_stats(struct enic *enic)
 132 {
 133         struct enic_soft_stats *soft_stats = &enic->soft_stats;
 134         rte_atomic64_clear(&soft_stats->rx_nombuf);
 135         rte_atomic64_clear(&soft_stats->rx_packet_errors);
 136         rte_atomic64_clear(&soft_stats->tx_oversized);
 137 }
 138
 139 static void enic_init_soft_stats(struct enic *enic)
 140 {
 141         struct enic_soft_stats *soft_stats = &enic->soft_stats;
 142         rte_atomic64_init(&soft_stats->rx_nombuf);
 143         rte_atomic64_init(&soft_stats->rx_packet_errors);
 144         rte_atomic64_init(&soft_stats->tx_oversized);
 145         enic_clear_soft_stats(enic);
 146 }
 147
 148 void enic_dev_stats_clear(struct enic *enic)
 149 {
 150         if (vnic_dev_stats_clear(enic->vdev))
 151                 dev_err(enic, "Error in clearing stats\n");
 152         enic_clear_soft_stats(enic);
 153 }
 154
 155 int enic_dev_stats_get(struct enic *enic, struct rte_eth_stats *r_stats)
 156 {
 157         struct vnic_stats *stats;
 158         struct enic_soft_stats *soft_stats = &enic->soft_stats;
 159         int64_t rx_truncated;
 160         uint64_t rx_packet_errors;
 161         int ret = vnic_dev_stats_dump(enic->vdev, &stats);
 162
 163         if (ret) {
 164                 dev_err(enic, "Error in getting stats\n");
 165                 return ret;
 166         }
 167
 168         /* The number of truncated packets can only be calculated by
 169          * subtracting a hardware counter from error packets received by
 170          * the driver. Note: this causes transient inaccuracies in the
 171          * ipackets count. Also, the length of truncated packets are
 172          * counted in ibytes even though truncated packets are dropped
 173          * which can make ibytes be slightly higher than it should be.
 174          */
 175         rx_packet_errors = rte_atomic64_read(&soft_stats->rx_packet_errors);
 176         rx_truncated = rx_packet_errors - stats->rx.rx_errors;
 177
 178         r_stats->ipackets = stats->rx.rx_frames_ok - rx_truncated;
 179         r_stats->opackets = stats->tx.tx_frames_ok;
 180
 181         r_stats->ibytes = stats->rx.rx_bytes_ok;
 182         r_stats->obytes = stats->tx.tx_bytes_ok;
 183
 184         r_stats->ierrors = stats->rx.rx_errors + stats->rx.rx_drop;
 185         r_stats->oerrors = stats->tx.tx_errors
 186                            + rte_atomic64_read(&soft_stats->tx_oversized);
 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         return 0;
 192 }
 193
 194 void enic_del_mac_address(struct enic *enic, int mac_index)
 195 {
 196         struct rte_eth_dev *eth_dev = enic->rte_dev;
 197         uint8_t *mac_addr = eth_dev->data->mac_addrs[mac_index].addr_bytes;
 198
 199         if (vnic_dev_del_addr(enic->vdev, mac_addr))
 200                 dev_err(enic, "del mac addr failed\n");
 201 }
 202
 203 int enic_set_mac_address(struct enic *enic, uint8_t *mac_addr)
 204 {
 205         int err;
 206
 207         if (!is_eth_addr_valid(mac_addr)) {
 208                 dev_err(enic, "invalid mac address\n");
 209                 return -EINVAL;
 210         }
 211
 212         err = vnic_dev_add_addr(enic->vdev, mac_addr);
 213         if (err)
 214                 dev_err(enic, "add mac addr failed\n");
 215         return err;
 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->iova);
 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_iova
 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->iova;
 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                 return NULL;
 370         }
 371
 372         mze->rz = rz;
 373
 374         rte_spinlock_lock(&enic->memzone_list_lock);
 375         LIST_INSERT_HEAD(&enic->memzone_list, mze, entries);
 376         rte_spinlock_unlock(&enic->memzone_list_lock);
 377
 378         return vaddr;
 379 }
 380
 381 static void
 382 enic_free_consistent(void *priv,
 383                      __rte_unused size_t size,
 384                      void *vaddr,
 385                      dma_addr_t dma_handle)
 386 {
 387         struct enic_memzone_entry *mze;
 388         struct enic *enic = (struct enic *)priv;
 389
 390         rte_spinlock_lock(&enic->memzone_list_lock);
 391         LIST_FOREACH(mze, &enic->memzone_list, entries) {
 392                 if (mze->rz->addr == vaddr &&
 393                     mze->rz->iova == dma_handle)
 394                         break;
 395         }
 396         if (mze == NULL) {
 397                 rte_spinlock_unlock(&enic->memzone_list_lock);
 398                 dev_warning(enic,
 399                             "Tried to free memory, but couldn't find it in the memzone list\n");
 400                 return;
 401         }
 402         LIST_REMOVE(mze, entries);
 403         rte_spinlock_unlock(&enic->memzone_list_lock);
 404         rte_memzone_free(mze->rz);
 405         rte_free(mze);
 406 }
 407
 408 int enic_link_update(struct enic *enic)
 409 {
 410         struct rte_eth_dev *eth_dev = enic->rte_dev;
 411         int ret;
 412         int link_status = 0;
 413
 414         link_status = enic_get_link_status(enic);
 415         ret = (link_status == enic->link_status);
 416         enic->link_status = link_status;
 417         eth_dev->data->dev_link.link_status = link_status;
 418         eth_dev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
 419         eth_dev->data->dev_link.link_speed = vnic_dev_port_speed(enic->vdev);
 420         return ret;
 421 }
 422
 423 static void
 424 enic_intr_handler(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.offloads &
 633             DEV_RX_OFFLOAD_SCATTER) {
 634                 dev_info(enic, "Rq %u Scatter rx mode enabled\n", queue_idx);
 635                 /* ceil((mtu + ETHER_HDR_LEN + 4)/mbuf_size) */
 636                 mbufs_per_pkt = ((mtu + ETHER_HDR_LEN + 4) +
 637                                  (mbuf_size - 1)) / mbuf_size;
 638         } else {
 639                 dev_info(enic, "Scatter rx mode disabled\n");
 640                 mbufs_per_pkt = 1;
 641         }
 642
 643         if (mbufs_per_pkt > 1) {
 644                 dev_info(enic, "Rq %u Scatter rx mode in use\n", queue_idx);
 645                 rq_sop->data_queue_enable = 1;
 646                 rq_data->in_use = 1;
 647         } else {
 648                 dev_info(enic, "Rq %u Scatter rx mode not being used\n",
 649                          queue_idx);
 650                 rq_sop->data_queue_enable = 0;
 651                 rq_data->in_use = 0;
 652         }
 653
 654         /* number of descriptors have to be a multiple of 32 */
 655         nb_sop_desc = (nb_desc / mbufs_per_pkt) & ~0x1F;
 656         nb_data_desc = (nb_desc - nb_sop_desc) & ~0x1F;
 657
 658         rq_sop->max_mbufs_per_pkt = mbufs_per_pkt;
 659         rq_data->max_mbufs_per_pkt = mbufs_per_pkt;
 660
 661         if (mbufs_per_pkt > 1) {
 662                 min_sop = 64;
 663                 max_sop = ((enic->config.rq_desc_count /
 664                             (mbufs_per_pkt - 1)) & ~0x1F);
 665                 min_data = min_sop * (mbufs_per_pkt - 1);
 666                 max_data = enic->config.rq_desc_count;
 667         } else {
 668                 min_sop = 64;
 669                 max_sop = enic->config.rq_desc_count;
 670                 min_data = 0;
 671                 max_data = 0;
 672         }
 673
 674         if (nb_desc < (min_sop + min_data)) {
 675                 dev_warning(enic,
 676                             "Number of rx descs too low, adjusting to minimum\n");
 677                 nb_sop_desc = min_sop;
 678                 nb_data_desc = min_data;
 679         } else if (nb_desc > (max_sop + max_data)) {
 680                 dev_warning(enic,
 681                             "Number of rx_descs too high, adjusting to maximum\n");
 682                 nb_sop_desc = max_sop;
 683                 nb_data_desc = max_data;
 684         }
 685         if (mbufs_per_pkt > 1) {
 686                 dev_info(enic, "For mtu %d and mbuf size %d valid rx descriptor range is %d to %d\n",
 687                          mtu, mbuf_size, min_sop + min_data,
 688                          max_sop + max_data);
 689         }
 690         dev_info(enic, "Using %d rx descriptors (sop %d, data %d)\n",
 691                  nb_sop_desc + nb_data_desc, nb_sop_desc, nb_data_desc);
 692
 693         /* Allocate sop queue resources */
 694         rc = vnic_rq_alloc(enic->vdev, rq_sop, sop_queue_idx,
 695                 nb_sop_desc, sizeof(struct rq_enet_desc));
 696         if (rc) {
 697                 dev_err(enic, "error in allocation of sop rq\n");
 698                 goto err_exit;
 699         }
 700         nb_sop_desc = rq_sop->ring.desc_count;
 701
 702         if (rq_data->in_use) {
 703                 /* Allocate data queue resources */
 704                 rc = vnic_rq_alloc(enic->vdev, rq_data, data_queue_idx,
 705                                    nb_data_desc,
 706                                    sizeof(struct rq_enet_desc));
 707                 if (rc) {
 708                         dev_err(enic, "error in allocation of data rq\n");
 709                         goto err_free_rq_sop;
 710                 }
 711                 nb_data_desc = rq_data->ring.desc_count;
 712         }
 713         rc = vnic_cq_alloc(enic->vdev, &enic->cq[queue_idx], queue_idx,
 714                            socket_id, nb_sop_desc + nb_data_desc,
 715                            sizeof(struct cq_enet_rq_desc));
 716         if (rc) {
 717                 dev_err(enic, "error in allocation of cq for rq\n");
 718                 goto err_free_rq_data;
 719         }
 720
 721         /* Allocate the mbuf rings */
 722         rq_sop->mbuf_ring = (struct rte_mbuf **)
 723                 rte_zmalloc_socket("rq->mbuf_ring",
 724                                    sizeof(struct rte_mbuf *) * nb_sop_desc,
 725                                    RTE_CACHE_LINE_SIZE, rq_sop->socket_id);
 726         if (rq_sop->mbuf_ring == NULL)
 727                 goto err_free_cq;
 728
 729         if (rq_data->in_use) {
 730                 rq_data->mbuf_ring = (struct rte_mbuf **)
 731                         rte_zmalloc_socket("rq->mbuf_ring",
 732                                 sizeof(struct rte_mbuf *) * nb_data_desc,
 733                                 RTE_CACHE_LINE_SIZE, rq_sop->socket_id);
 734                 if (rq_data->mbuf_ring == NULL)
 735                         goto err_free_sop_mbuf;
 736         }
 737
 738         rq_sop->tot_nb_desc = nb_desc; /* squirl away for MTU update function */
 739
 740         return 0;
 741
 742 err_free_sop_mbuf:
 743         rte_free(rq_sop->mbuf_ring);
 744 err_free_cq:
 745         /* cleanup on error */
 746         vnic_cq_free(&enic->cq[queue_idx]);
 747 err_free_rq_data:
 748         if (rq_data->in_use)
 749                 vnic_rq_free(rq_data);
 750 err_free_rq_sop:
 751         vnic_rq_free(rq_sop);
 752 err_exit:
 753         return -ENOMEM;
 754 }
 755
 756 void enic_free_wq(void *txq)
 757 {
 758         struct vnic_wq *wq;
 759         struct enic *enic;
 760
 761         if (txq == NULL)
 762                 return;
 763
 764         wq = (struct vnic_wq *)txq;
 765         enic = vnic_dev_priv(wq->vdev);
 766         rte_memzone_free(wq->cqmsg_rz);
 767         vnic_wq_free(wq);
 768         vnic_cq_free(&enic->cq[enic->rq_count + wq->index]);
 769 }
 770
 771 int enic_alloc_wq(struct enic *enic, uint16_t queue_idx,
 772         unsigned int socket_id, uint16_t nb_desc)
 773 {
 774         int err;
 775         struct vnic_wq *wq = &enic->wq[queue_idx];
 776         unsigned int cq_index = enic_cq_wq(enic, queue_idx);
 777         char name[NAME_MAX];
 778         static int instance;
 779
 780         wq->socket_id = socket_id;
 781         if (nb_desc) {
 782                 if (nb_desc > enic->config.wq_desc_count) {
 783                         dev_warning(enic,
 784                                 "WQ %d - number of tx desc in cmd line (%d)"\
 785                                 "is greater than that in the UCSM/CIMC adapter"\
 786                                 "policy.  Applying the value in the adapter "\
 787                                 "policy (%d)\n",
 788                                 queue_idx, nb_desc, enic->config.wq_desc_count);
 789                 } else if (nb_desc != enic->config.wq_desc_count) {
 790                         enic->config.wq_desc_count = nb_desc;
 791                         dev_info(enic,
 792                                 "TX Queues - effective number of descs:%d\n",
 793                                 nb_desc);
 794                 }
 795         }
 796
 797         /* Allocate queue resources */
 798         err = vnic_wq_alloc(enic->vdev, &enic->wq[queue_idx], queue_idx,
 799                 enic->config.wq_desc_count,
 800                 sizeof(struct wq_enet_desc));
 801         if (err) {
 802                 dev_err(enic, "error in allocation of wq\n");
 803                 return err;
 804         }
 805
 806         err = vnic_cq_alloc(enic->vdev, &enic->cq[cq_index], cq_index,
 807                 socket_id, enic->config.wq_desc_count,
 808                 sizeof(struct cq_enet_wq_desc));
 809         if (err) {
 810                 vnic_wq_free(wq);
 811                 dev_err(enic, "error in allocation of cq for wq\n");
 812         }
 813
 814         /* setup up CQ message */
 815         snprintf((char *)name, sizeof(name),
 816                  "vnic_cqmsg-%s-%d-%d", enic->bdf_name, queue_idx,
 817                 instance++);
 818
 819         wq->cqmsg_rz = rte_memzone_reserve_aligned((const char *)name,
 820                                                    sizeof(uint32_t),
 821                                                    SOCKET_ID_ANY, 0,
 822                                                    ENIC_ALIGN);
 823         if (!wq->cqmsg_rz)
 824                 return -ENOMEM;
 825
 826         return err;
 827 }
 828
 829 int enic_disable(struct enic *enic)
 830 {
 831         unsigned int i;
 832         int err;
 833
 834         vnic_intr_mask(&enic->intr);
 835         (void)vnic_intr_masked(&enic->intr); /* flush write */
 836         rte_intr_disable(&enic->pdev->intr_handle);
 837         rte_intr_callback_unregister(&enic->pdev->intr_handle,
 838                                      enic_intr_handler,
 839                                      (void *)enic->rte_dev);
 840
 841         vnic_dev_disable(enic->vdev);
 842
 843         enic_clsf_destroy(enic);
 844
 845         if (!enic_is_sriov_vf(enic))
 846                 vnic_dev_del_addr(enic->vdev, enic->mac_addr);
 847
 848         for (i = 0; i < enic->wq_count; i++) {
 849                 err = vnic_wq_disable(&enic->wq[i]);
 850                 if (err)
 851                         return err;
 852         }
 853         for (i = 0; i < enic_vnic_rq_count(enic); i++) {
 854                 if (enic->rq[i].in_use) {
 855                         err = vnic_rq_disable(&enic->rq[i]);
 856                         if (err)
 857                                 return err;
 858                 }
 859         }
 860
 861         /* If we were using interrupts, set the interrupt vector to -1
 862          * to disable interrupts.  We are not disabling link notifcations,
 863          * though, as we want the polling of link status to continue working.
 864          */
 865         if (enic->rte_dev->data->dev_conf.intr_conf.lsc)
 866                 vnic_dev_notify_set(enic->vdev, -1);
 867
 868         vnic_dev_set_reset_flag(enic->vdev, 1);
 869
 870         for (i = 0; i < enic->wq_count; i++)
 871                 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
 872
 873         for (i = 0; i < enic_vnic_rq_count(enic); i++)
 874                 if (enic->rq[i].in_use)
 875                         vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
 876         for (i = 0; i < enic->cq_count; i++)
 877                 vnic_cq_clean(&enic->cq[i]);
 878         vnic_intr_clean(&enic->intr);
 879
 880         return 0;
 881 }
 882
 883 static int enic_dev_wait(struct vnic_dev *vdev,
 884         int (*start)(struct vnic_dev *, int),
 885         int (*finished)(struct vnic_dev *, int *),
 886         int arg)
 887 {
 888         int done;
 889         int err;
 890         int i;
 891
 892         err = start(vdev, arg);
 893         if (err)
 894                 return err;
 895
 896         /* Wait for func to complete...2 seconds max */
 897         for (i = 0; i < 2000; i++) {
 898                 err = finished(vdev, &done);
 899                 if (err)
 900                         return err;
 901                 if (done)
 902                         return 0;
 903                 usleep(1000);
 904         }
 905         return -ETIMEDOUT;
 906 }
 907
 908 static int enic_dev_open(struct enic *enic)
 909 {
 910         int err;
 911
 912         err = enic_dev_wait(enic->vdev, vnic_dev_open,
 913                 vnic_dev_open_done, 0);
 914         if (err)
 915                 dev_err(enic_get_dev(enic),
 916                         "vNIC device open failed, err %d\n", err);
 917
 918         return err;
 919 }
 920
 921 static int enic_set_rsskey(struct enic *enic)
 922 {
 923         dma_addr_t rss_key_buf_pa;
 924         union vnic_rss_key *rss_key_buf_va = NULL;
 925         static union vnic_rss_key rss_key = {
 926                 .key = {
 927                         [0] = {.b = {85, 67, 83, 97, 119, 101, 115, 111, 109, 101}},
 928                         [1] = {.b = {80, 65, 76, 79, 117, 110, 105, 113, 117, 101}},
 929                         [2] = {.b = {76, 73, 78, 85, 88, 114, 111, 99, 107, 115}},
 930                         [3] = {.b = {69, 78, 73, 67, 105, 115, 99, 111, 111, 108}},
 931                 }
 932         };
 933         int err;
 934         u8 name[NAME_MAX];
 935
 936         snprintf((char *)name, NAME_MAX, "rss_key-%s", enic->bdf_name);
 937         rss_key_buf_va = enic_alloc_consistent(enic, sizeof(union vnic_rss_key),
 938                 &rss_key_buf_pa, name);
 939         if (!rss_key_buf_va)
 940                 return -ENOMEM;
 941
 942         rte_memcpy(rss_key_buf_va, &rss_key, sizeof(union vnic_rss_key));
 943
 944         err = enic_set_rss_key(enic,
 945                 rss_key_buf_pa,
 946                 sizeof(union vnic_rss_key));
 947
 948         enic_free_consistent(enic, sizeof(union vnic_rss_key),
 949                 rss_key_buf_va, rss_key_buf_pa);
 950
 951         return err;
 952 }
 953
 954 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
 955 {
 956         dma_addr_t rss_cpu_buf_pa;
 957         union vnic_rss_cpu *rss_cpu_buf_va = NULL;
 958         int i;
 959         int err;
 960         u8 name[NAME_MAX];
 961
 962         snprintf((char *)name, NAME_MAX, "rss_cpu-%s", enic->bdf_name);
 963         rss_cpu_buf_va = enic_alloc_consistent(enic, sizeof(union vnic_rss_cpu),
 964                 &rss_cpu_buf_pa, name);
 965         if (!rss_cpu_buf_va)
 966                 return -ENOMEM;
 967
 968         for (i = 0; i < (1 << rss_hash_bits); i++)
 969                 (*rss_cpu_buf_va).cpu[i / 4].b[i % 4] =
 970                         enic_rte_rq_idx_to_sop_idx(i % enic->rq_count);
 971
 972         err = enic_set_rss_cpu(enic,
 973                 rss_cpu_buf_pa,
 974                 sizeof(union vnic_rss_cpu));
 975
 976         enic_free_consistent(enic, sizeof(union vnic_rss_cpu),
 977                 rss_cpu_buf_va, rss_cpu_buf_pa);
 978
 979         return err;
 980 }
 981
 982 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
 983         u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
 984 {
 985         const u8 tso_ipid_split_en = 0;
 986         int err;
 987
 988         /* Enable VLAN tag stripping */
 989
 990         err = enic_set_nic_cfg(enic,
 991                 rss_default_cpu, rss_hash_type,
 992                 rss_hash_bits, rss_base_cpu,
 993                 rss_enable, tso_ipid_split_en,
 994                 enic->ig_vlan_strip_en);
 995
 996         return err;
 997 }
 998
 999 int enic_set_rss_nic_cfg(struct enic *enic)
1000 {
1001         const u8 rss_default_cpu = 0;
1002         const u8 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
1003             NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
1004             NIC_CFG_RSS_HASH_TYPE_IPV6 |
1005             NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
1006         const u8 rss_hash_bits = 7;
1007         const u8 rss_base_cpu = 0;
1008         u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
1009
1010         if (rss_enable) {
1011                 if (!enic_set_rsskey(enic)) {
1012                         if (enic_set_rsscpu(enic, rss_hash_bits)) {
1013                                 rss_enable = 0;
1014                                 dev_warning(enic, "RSS disabled, "\
1015                                         "Failed to set RSS cpu indirection table.");
1016                         }
1017                 } else {
1018                         rss_enable = 0;
1019                         dev_warning(enic,
1020                                 "RSS disabled, Failed to set RSS key.\n");
1021                 }
1022         }
1023
1024         return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
1025                 rss_hash_bits, rss_base_cpu, rss_enable);
1026 }
1027
1028 int enic_setup_finish(struct enic *enic)
1029 {
1030         int ret;
1031
1032         enic_init_soft_stats(enic);
1033
1034         ret = enic_set_rss_nic_cfg(enic);
1035         if (ret) {
1036                 dev_err(enic, "Failed to config nic, aborting.\n");
1037                 return -1;
1038         }
1039
1040         /* Default conf */
1041         vnic_dev_packet_filter(enic->vdev,
1042                 1 /* directed  */,
1043                 1 /* multicast */,
1044                 1 /* broadcast */,
1045                 0 /* promisc   */,
1046                 1 /* allmulti  */);
1047
1048         enic->promisc = 0;
1049         enic->allmulti = 1;
1050
1051         return 0;
1052 }
1053
1054 void enic_add_packet_filter(struct enic *enic)
1055 {
1056         /* Args -> directed, multicast, broadcast, promisc, allmulti */
1057         vnic_dev_packet_filter(enic->vdev, 1, 1, 1,
1058                 enic->promisc, enic->allmulti);
1059 }
1060
1061 int enic_get_link_status(struct enic *enic)
1062 {
1063         return vnic_dev_link_status(enic->vdev);
1064 }
1065
1066 static void enic_dev_deinit(struct enic *enic)
1067 {
1068         struct rte_eth_dev *eth_dev = enic->rte_dev;
1069
1070         /* stop link status checking */
1071         vnic_dev_notify_unset(enic->vdev);
1072
1073         rte_free(eth_dev->data->mac_addrs);
1074 }
1075
1076
1077 int enic_set_vnic_res(struct enic *enic)
1078 {
1079         struct rte_eth_dev *eth_dev = enic->rte_dev;
1080         int rc = 0;
1081
1082         /* With Rx scatter support, two RQs are now used per RQ used by
1083          * the application.
1084          */
1085         if (enic->conf_rq_count < eth_dev->data->nb_rx_queues) {
1086                 dev_err(dev, "Not enough Receive queues. Requested:%u which uses %d RQs on VIC, Configured:%u\n",
1087                         eth_dev->data->nb_rx_queues,
1088                         eth_dev->data->nb_rx_queues * 2, enic->conf_rq_count);
1089                 rc = -EINVAL;
1090         }
1091         if (enic->conf_wq_count < eth_dev->data->nb_tx_queues) {
1092                 dev_err(dev, "Not enough Transmit queues. Requested:%u, Configured:%u\n",
1093                         eth_dev->data->nb_tx_queues, enic->conf_wq_count);
1094                 rc = -EINVAL;
1095         }
1096
1097         if (enic->conf_cq_count < (eth_dev->data->nb_rx_queues +
1098                                    eth_dev->data->nb_tx_queues)) {
1099                 dev_err(dev, "Not enough Completion queues. Required:%u, Configured:%u\n",
1100                         (eth_dev->data->nb_rx_queues +
1101                          eth_dev->data->nb_tx_queues), enic->conf_cq_count);
1102                 rc = -EINVAL;
1103         }
1104
1105         if (rc == 0) {
1106                 enic->rq_count = eth_dev->data->nb_rx_queues;
1107                 enic->wq_count = eth_dev->data->nb_tx_queues;
1108                 enic->cq_count = enic->rq_count + enic->wq_count;
1109         }
1110
1111         return rc;
1112 }
1113
1114 /* Initialize the completion queue for an RQ */
1115 static int
1116 enic_reinit_rq(struct enic *enic, unsigned int rq_idx)
1117 {
1118         struct vnic_rq *sop_rq, *data_rq;
1119         unsigned int cq_idx;
1120         int rc = 0;
1121
1122         sop_rq = &enic->rq[enic_rte_rq_idx_to_sop_idx(rq_idx)];
1123         data_rq = &enic->rq[enic_rte_rq_idx_to_data_idx(rq_idx)];
1124         cq_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 (rte_eal_process_type() != RTE_PROC_PRIMARY)
1185                 return -E_RTE_SECONDARY;
1186
1187         if (new_mtu > enic->max_mtu) {
1188                 dev_err(enic,
1189                         "MTU not updated: requested (%u) greater than max (%u)\n",
1190                         new_mtu, enic->max_mtu);
1191                 return -EINVAL;
1192         }
1193         if (new_mtu < ENIC_MIN_MTU) {
1194                 dev_info(enic,
1195                         "MTU not updated: requested (%u) less than min (%u)\n",
1196                         new_mtu, ENIC_MIN_MTU);
1197                 return -EINVAL;
1198         }
1199         if (new_mtu > config_mtu)
1200                 dev_warning(enic,
1201                         "MTU (%u) is greater than value configured in NIC (%u)\n",
1202                         new_mtu, config_mtu);
1203
1204         /* The easy case is when scatter is disabled. However if the MTU
1205          * becomes greater than the mbuf data size, packet drops will ensue.
1206          */
1207         if (!(enic->rte_dev->data->dev_conf.rxmode.offloads &
1208               DEV_RX_OFFLOAD_SCATTER)) {
1209                 eth_dev->data->mtu = new_mtu;
1210                 goto set_mtu_done;
1211         }
1212
1213         /* Rx scatter is enabled so reconfigure RQ's on the fly. The point is to
1214          * change Rx scatter mode if necessary for better performance. I.e. if
1215          * MTU was greater than the mbuf size and now it's less, scatter Rx
1216          * doesn't have to be used and vice versa.
1217           */
1218         rte_spinlock_lock(&enic->mtu_lock);
1219
1220         /* Stop traffic on all RQs */
1221         for (rq_idx = 0; rq_idx < enic->rq_count * 2; rq_idx++) {
1222                 rq = &enic->rq[rq_idx];
1223                 if (rq->is_sop && rq->in_use) {
1224                         rc = enic_stop_rq(enic,
1225                                           enic_sop_rq_idx_to_rte_idx(rq_idx));
1226                         if (rc) {
1227                                 dev_err(enic, "Failed to stop Rq %u\n", rq_idx);
1228                                 goto set_mtu_done;
1229                         }
1230                 }
1231         }
1232
1233         /* replace Rx function with a no-op to avoid getting stale pkts */
1234         eth_dev->rx_pkt_burst = enic_dummy_recv_pkts;
1235         rte_mb();
1236
1237         /* Allow time for threads to exit the real Rx function. */
1238         usleep(100000);
1239
1240         /* now it is safe to reconfigure the RQs */
1241
1242         /* update the mtu */
1243         eth_dev->data->mtu = new_mtu;
1244
1245         /* free and reallocate RQs with the new MTU */
1246         for (rq_idx = 0; rq_idx < enic->rq_count; rq_idx++) {
1247                 rq = &enic->rq[enic_rte_rq_idx_to_sop_idx(rq_idx)];
1248
1249                 enic_free_rq(rq);
1250                 rc = enic_alloc_rq(enic, rq_idx, rq->socket_id, rq->mp,
1251                                    rq->tot_nb_desc, rq->rx_free_thresh);
1252                 if (rc) {
1253                         dev_err(enic,
1254                                 "Fatal MTU alloc error- No traffic will pass\n");
1255                         goto set_mtu_done;
1256                 }
1257
1258                 rc = enic_reinit_rq(enic, rq_idx);
1259                 if (rc) {
1260                         dev_err(enic,
1261                                 "Fatal MTU RQ reinit- No traffic will pass\n");
1262                         goto set_mtu_done;
1263                 }
1264         }
1265
1266         /* put back the real receive function */
1267         rte_mb();
1268         eth_dev->rx_pkt_burst = enic_recv_pkts;
1269         rte_mb();
1270
1271         /* restart Rx traffic */
1272         for (rq_idx = 0; rq_idx < enic->rq_count; rq_idx++) {
1273                 rq = &enic->rq[enic_rte_rq_idx_to_sop_idx(rq_idx)];
1274                 if (rq->is_sop && rq->in_use)
1275                         enic_start_rq(enic, rq_idx);
1276         }
1277
1278 set_mtu_done:
1279         dev_info(enic, "MTU changed from %u to %u\n",  old_mtu, new_mtu);
1280         rte_spinlock_unlock(&enic->mtu_lock);
1281         return rc;
1282 }
1283
1284 static int enic_dev_init(struct enic *enic)
1285 {
1286         int err;
1287         struct rte_eth_dev *eth_dev = enic->rte_dev;
1288
1289         vnic_dev_intr_coal_timer_info_default(enic->vdev);
1290
1291         /* Get vNIC configuration
1292         */
1293         err = enic_get_vnic_config(enic);
1294         if (err) {
1295                 dev_err(dev, "Get vNIC configuration failed, aborting\n");
1296                 return err;
1297         }
1298
1299         /* Get available resource counts */
1300         enic_get_res_counts(enic);
1301         if (enic->conf_rq_count == 1) {
1302                 dev_err(enic, "Running with only 1 RQ configured in the vNIC is not supported.\n");
1303                 dev_err(enic, "Please configure 2 RQs in the vNIC for each Rx queue used by DPDK.\n");
1304                 dev_err(enic, "See the ENIC PMD guide for more information.\n");
1305                 return -EINVAL;
1306         }
1307
1308         /* Get the supported filters */
1309         enic_fdir_info(enic);
1310
1311         eth_dev->data->mac_addrs = rte_zmalloc("enic_mac_addr", ETH_ALEN
1312                                                 * ENIC_MAX_MAC_ADDR, 0);
1313         if (!eth_dev->data->mac_addrs) {
1314                 dev_err(enic, "mac addr storage alloc failed, aborting.\n");
1315                 return -1;
1316         }
1317         ether_addr_copy((struct ether_addr *) enic->mac_addr,
1318                         eth_dev->data->mac_addrs);
1319
1320         vnic_dev_set_reset_flag(enic->vdev, 0);
1321
1322         LIST_INIT(&enic->flows);
1323         rte_spinlock_init(&enic->flows_lock);
1324
1325         /* set up link status checking */
1326         vnic_dev_notify_set(enic->vdev, -1); /* No Intr for notify */
1327
1328         return 0;
1329
1330 }
1331
1332 int enic_probe(struct enic *enic)
1333 {
1334         struct rte_pci_device *pdev = enic->pdev;
1335         int err = -1;
1336
1337         dev_debug(enic, " Initializing ENIC PMD\n");
1338
1339         /* if this is a secondary process the hardware is already initialized */
1340         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1341                 return 0;
1342
1343         enic->bar0.vaddr = (void *)pdev->mem_resource[0].addr;
1344         enic->bar0.len = pdev->mem_resource[0].len;
1345
1346         /* Register vNIC device */
1347         enic->vdev = vnic_dev_register(NULL, enic, enic->pdev, &enic->bar0, 1);
1348         if (!enic->vdev) {
1349                 dev_err(enic, "vNIC registration failed, aborting\n");
1350                 goto err_out;
1351         }
1352
1353         LIST_INIT(&enic->memzone_list);
1354         rte_spinlock_init(&enic->memzone_list_lock);
1355
1356         vnic_register_cbacks(enic->vdev,
1357                 enic_alloc_consistent,
1358                 enic_free_consistent);
1359
1360         /* Issue device open to get device in known state */
1361         err = enic_dev_open(enic);
1362         if (err) {
1363                 dev_err(enic, "vNIC dev open failed, aborting\n");
1364                 goto err_out_unregister;
1365         }
1366
1367         /* Set ingress vlan rewrite mode before vnic initialization */
1368         err = vnic_dev_set_ig_vlan_rewrite_mode(enic->vdev,
1369                 IG_VLAN_REWRITE_MODE_PASS_THRU);
1370         if (err) {
1371                 dev_err(enic,
1372                         "Failed to set ingress vlan rewrite mode, aborting.\n");
1373                 goto err_out_dev_close;
1374         }
1375
1376         /* Issue device init to initialize the vnic-to-switch link.
1377          * We'll start with carrier off and wait for link UP
1378          * notification later to turn on carrier.  We don't need
1379          * to wait here for the vnic-to-switch link initialization
1380          * to complete; link UP notification is the indication that
1381          * the process is complete.
1382          */
1383
1384         err = vnic_dev_init(enic->vdev, 0);
1385         if (err) {
1386                 dev_err(enic, "vNIC dev init failed, aborting\n");
1387                 goto err_out_dev_close;
1388         }
1389
1390         err = enic_dev_init(enic);
1391         if (err) {
1392                 dev_err(enic, "Device initialization failed, aborting\n");
1393                 goto err_out_dev_close;
1394         }
1395
1396         return 0;
1397
1398 err_out_dev_close:
1399         vnic_dev_close(enic->vdev);
1400 err_out_unregister:
1401         vnic_dev_unregister(enic->vdev);
1402 err_out:
1403         return err;
1404 }
1405
1406 void enic_remove(struct enic *enic)
1407 {
1408         enic_dev_deinit(enic);
1409         vnic_dev_close(enic->vdev);
1410         vnic_dev_unregister(enic->vdev);
1411 }