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                 error_status = vnic_rq_error_status(&enic->rq[i]);
 127                 if (error_status)
 128                         dev_err(enic, "RQ[%d] error_status %d\n", i,
 129                                 error_status);
 130         }
 131 }
 132
 133 static void enic_clear_soft_stats(struct enic *enic)
 134 {
 135         struct enic_soft_stats *soft_stats = &enic->soft_stats;
 136         rte_atomic64_clear(&soft_stats->rx_nombuf);
 137         rte_atomic64_clear(&soft_stats->rx_packet_errors);
 138 }
 139
 140 static void enic_init_soft_stats(struct enic *enic)
 141 {
 142         struct enic_soft_stats *soft_stats = &enic->soft_stats;
 143         rte_atomic64_init(&soft_stats->rx_nombuf);
 144         rte_atomic64_init(&soft_stats->rx_packet_errors);
 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 void 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
 162         if (vnic_dev_stats_dump(enic->vdev, &stats)) {
 163                 dev_err(enic, "Error in getting stats\n");
 164                 return;
 165         }
 166
 167         /* The number of truncated packets can only be calculated by
 168          * subtracting a hardware counter from error packets received by
 169          * the driver. Note: this causes transient inaccuracies in the
 170          * ipackets count. Also, the length of truncated packets are
 171          * counted in ibytes even though truncated packets are dropped
 172          * which can make ibytes be slightly higher than it should be.
 173          */
 174         rx_packet_errors = rte_atomic64_read(&soft_stats->rx_packet_errors);
 175         rx_truncated = rx_packet_errors - stats->rx.rx_errors -
 176                 stats->rx.rx_no_bufs;
 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
 187         r_stats->imissed = stats->rx.rx_no_bufs + rx_truncated;
 188
 189         r_stats->rx_nombuf = rte_atomic64_read(&soft_stats->rx_nombuf);
 190 }
 191
 192 void enic_del_mac_address(struct enic *enic)
 193 {
 194         if (vnic_dev_del_addr(enic->vdev, enic->mac_addr))
 195                 dev_err(enic, "del mac addr failed\n");
 196 }
 197
 198 void enic_set_mac_address(struct enic *enic, uint8_t *mac_addr)
 199 {
 200         int err;
 201
 202         if (!is_eth_addr_valid(mac_addr)) {
 203                 dev_err(enic, "invalid mac address\n");
 204                 return;
 205         }
 206
 207         err = vnic_dev_del_addr(enic->vdev, enic->mac_addr);
 208         if (err) {
 209                 dev_err(enic, "del mac addr failed\n");
 210                 return;
 211         }
 212
 213         ether_addr_copy((struct ether_addr *)mac_addr,
 214                 (struct ether_addr *)enic->mac_addr);
 215
 216         err = vnic_dev_add_addr(enic->vdev, mac_addr);
 217         if (err) {
 218                 dev_err(enic, "add mac addr failed\n");
 219                 return;
 220         }
 221 }
 222
 223 static void
 224 enic_free_rq_buf(struct rte_mbuf **mbuf)
 225 {
 226         if (*mbuf == NULL)
 227                 return;
 228
 229         rte_pktmbuf_free(*mbuf);
 230         mbuf = NULL;
 231 }
 232
 233 void enic_init_vnic_resources(struct enic *enic)
 234 {
 235         unsigned int error_interrupt_enable = 1;
 236         unsigned int error_interrupt_offset = 0;
 237         unsigned int index = 0;
 238         unsigned int cq_idx;
 239         struct vnic_rq *data_rq;
 240
 241         for (index = 0; index < enic->rq_count; index++) {
 242                 cq_idx = enic_cq_rq(enic, enic_sop_rq(index));
 243
 244                 vnic_rq_init(&enic->rq[enic_sop_rq(index)],
 245                         cq_idx,
 246                         error_interrupt_enable,
 247                         error_interrupt_offset);
 248
 249                 data_rq = &enic->rq[enic_data_rq(index)];
 250                 if (data_rq->in_use)
 251                         vnic_rq_init(data_rq,
 252                                      cq_idx,
 253                                      error_interrupt_enable,
 254                                      error_interrupt_offset);
 255
 256                 vnic_cq_init(&enic->cq[cq_idx],
 257                         0 /* flow_control_enable */,
 258                         1 /* color_enable */,
 259                         0 /* cq_head */,
 260                         0 /* cq_tail */,
 261                         1 /* cq_tail_color */,
 262                         0 /* interrupt_enable */,
 263                         1 /* cq_entry_enable */,
 264                         0 /* cq_message_enable */,
 265                         0 /* interrupt offset */,
 266                         0 /* cq_message_addr */);
 267         }
 268
 269         for (index = 0; index < enic->wq_count; index++) {
 270                 vnic_wq_init(&enic->wq[index],
 271                         enic_cq_wq(enic, index),
 272                         error_interrupt_enable,
 273                         error_interrupt_offset);
 274
 275                 cq_idx = enic_cq_wq(enic, index);
 276                 vnic_cq_init(&enic->cq[cq_idx],
 277                         0 /* flow_control_enable */,
 278                         1 /* color_enable */,
 279                         0 /* cq_head */,
 280                         0 /* cq_tail */,
 281                         1 /* cq_tail_color */,
 282                         0 /* interrupt_enable */,
 283                         0 /* cq_entry_enable */,
 284                         1 /* cq_message_enable */,
 285                         0 /* interrupt offset */,
 286                         (u64)enic->wq[index].cqmsg_rz->phys_addr);
 287         }
 288
 289         vnic_intr_init(&enic->intr,
 290                 enic->config.intr_timer_usec,
 291                 enic->config.intr_timer_type,
 292                 /*mask_on_assertion*/1);
 293 }
 294
 295
 296 static int
 297 enic_alloc_rx_queue_mbufs(struct enic *enic, struct vnic_rq *rq)
 298 {
 299         struct rte_mbuf *mb;
 300         struct rq_enet_desc *rqd = rq->ring.descs;
 301         unsigned i;
 302         dma_addr_t dma_addr;
 303
 304         if (!rq->in_use)
 305                 return 0;
 306
 307         dev_debug(enic, "queue %u, allocating %u rx queue mbufs\n", rq->index,
 308                   rq->ring.desc_count);
 309
 310         for (i = 0; i < rq->ring.desc_count; i++, rqd++) {
 311                 mb = rte_mbuf_raw_alloc(rq->mp);
 312                 if (mb == NULL) {
 313                         dev_err(enic, "RX mbuf alloc failed queue_id=%u\n",
 314                         (unsigned)rq->index);
 315                         return -ENOMEM;
 316                 }
 317
 318                 dma_addr = (dma_addr_t)(mb->buf_physaddr
 319                            + RTE_PKTMBUF_HEADROOM);
 320                 rq_enet_desc_enc(rqd, dma_addr,
 321                                 (rq->is_sop ? RQ_ENET_TYPE_ONLY_SOP
 322                                 : RQ_ENET_TYPE_NOT_SOP),
 323                                 mb->buf_len - RTE_PKTMBUF_HEADROOM);
 324                 rq->mbuf_ring[i] = mb;
 325         }
 326
 327         /* make sure all prior writes are complete before doing the PIO write */
 328         rte_rmb();
 329
 330         /* Post all but the last buffer to VIC. */
 331         rq->posted_index = rq->ring.desc_count - 1;
 332
 333         rq->rx_nb_hold = 0;
 334
 335         dev_debug(enic, "port=%u, qidx=%u, Write %u posted idx, %u sw held\n",
 336                 enic->port_id, rq->index, rq->posted_index, rq->rx_nb_hold);
 337         iowrite32(rq->posted_index, &rq->ctrl->posted_index);
 338         iowrite32(0, &rq->ctrl->fetch_index);
 339         rte_rmb();
 340
 341         return 0;
 342
 343 }
 344
 345 static void *
 346 enic_alloc_consistent(void *priv, size_t size,
 347         dma_addr_t *dma_handle, u8 *name)
 348 {
 349         void *vaddr;
 350         const struct rte_memzone *rz;
 351         *dma_handle = 0;
 352         struct enic *enic = (struct enic *)priv;
 353         struct enic_memzone_entry *mze;
 354
 355         rz = rte_memzone_reserve_aligned((const char *)name,
 356                                          size, SOCKET_ID_ANY, 0, ENIC_ALIGN);
 357         if (!rz) {
 358                 pr_err("%s : Failed to allocate memory requested for %s\n",
 359                         __func__, name);
 360                 return NULL;
 361         }
 362
 363         vaddr = rz->addr;
 364         *dma_handle = (dma_addr_t)rz->phys_addr;
 365
 366         mze = rte_malloc("enic memzone entry",
 367                          sizeof(struct enic_memzone_entry), 0);
 368
 369         if (!mze) {
 370                 pr_err("%s : Failed to allocate memory for memzone list\n",
 371                        __func__);
 372                 rte_memzone_free(rz);
 373         }
 374
 375         mze->rz = rz;
 376
 377         rte_spinlock_lock(&enic->memzone_list_lock);
 378         LIST_INSERT_HEAD(&enic->memzone_list, mze, entries);
 379         rte_spinlock_unlock(&enic->memzone_list_lock);
 380
 381         return vaddr;
 382 }
 383
 384 static void
 385 enic_free_consistent(void *priv,
 386                      __rte_unused size_t size,
 387                      void *vaddr,
 388                      dma_addr_t dma_handle)
 389 {
 390         struct enic_memzone_entry *mze;
 391         struct enic *enic = (struct enic *)priv;
 392
 393         rte_spinlock_lock(&enic->memzone_list_lock);
 394         LIST_FOREACH(mze, &enic->memzone_list, entries) {
 395                 if (mze->rz->addr == vaddr &&
 396                     mze->rz->phys_addr == dma_handle)
 397                         break;
 398         }
 399         if (mze == NULL) {
 400                 rte_spinlock_unlock(&enic->memzone_list_lock);
 401                 dev_warning(enic,
 402                             "Tried to free memory, but couldn't find it in the memzone list\n");
 403                 return;
 404         }
 405         LIST_REMOVE(mze, entries);
 406         rte_spinlock_unlock(&enic->memzone_list_lock);
 407         rte_memzone_free(mze->rz);
 408         rte_free(mze);
 409 }
 410
 411 static void
 412 enic_intr_handler(__rte_unused struct rte_intr_handle *handle,
 413         void *arg)
 414 {
 415         struct enic *enic = pmd_priv((struct rte_eth_dev *)arg);
 416
 417         vnic_intr_return_all_credits(&enic->intr);
 418
 419         enic_log_q_error(enic);
 420 }
 421
 422 int enic_enable(struct enic *enic)
 423 {
 424         unsigned int index;
 425         int err;
 426         struct rte_eth_dev *eth_dev = enic->rte_dev;
 427
 428         eth_dev->data->dev_link.link_speed = vnic_dev_port_speed(enic->vdev);
 429         eth_dev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
 430         vnic_dev_notify_set(enic->vdev, -1); /* No Intr for notify */
 431
 432         if (enic_clsf_init(enic))
 433                 dev_warning(enic, "Init of hash table for clsf failed."\
 434                         "Flow director feature will not work\n");
 435
 436         for (index = 0; index < enic->rq_count; index++) {
 437                 err = enic_alloc_rx_queue_mbufs(enic,
 438                         &enic->rq[enic_sop_rq(index)]);
 439                 if (err) {
 440                         dev_err(enic, "Failed to alloc sop RX queue mbufs\n");
 441                         return err;
 442                 }
 443                 err = enic_alloc_rx_queue_mbufs(enic,
 444                         &enic->rq[enic_data_rq(index)]);
 445                 if (err) {
 446                         /* release the allocated mbufs for the sop rq*/
 447                         enic_rxmbuf_queue_release(enic,
 448                                 &enic->rq[enic_sop_rq(index)]);
 449
 450                         dev_err(enic, "Failed to alloc data RX queue mbufs\n");
 451                         return err;
 452                 }
 453         }
 454
 455         for (index = 0; index < enic->wq_count; index++)
 456                 enic_start_wq(enic, index);
 457         for (index = 0; index < enic->rq_count; index++)
 458                 enic_start_rq(enic, index);
 459
 460         vnic_dev_add_addr(enic->vdev, enic->mac_addr);
 461
 462         vnic_dev_enable_wait(enic->vdev);
 463
 464         /* Register and enable error interrupt */
 465         rte_intr_callback_register(&(enic->pdev->intr_handle),
 466                 enic_intr_handler, (void *)enic->rte_dev);
 467
 468         rte_intr_enable(&(enic->pdev->intr_handle));
 469         vnic_intr_unmask(&enic->intr);
 470
 471         return 0;
 472 }
 473
 474 int enic_alloc_intr_resources(struct enic *enic)
 475 {
 476         int err;
 477
 478         dev_info(enic, "vNIC resources used:  "\
 479                 "wq %d rq %d cq %d intr %d\n",
 480                 enic->wq_count, enic_vnic_rq_count(enic),
 481                 enic->cq_count, enic->intr_count);
 482
 483         err = vnic_intr_alloc(enic->vdev, &enic->intr, 0);
 484         if (err)
 485                 enic_free_vnic_resources(enic);
 486
 487         return err;
 488 }
 489
 490 void enic_free_rq(void *rxq)
 491 {
 492         struct vnic_rq *rq_sop, *rq_data;
 493         struct enic *enic;
 494
 495         if (rxq == NULL)
 496                 return;
 497
 498         rq_sop = (struct vnic_rq *)rxq;
 499         enic = vnic_dev_priv(rq_sop->vdev);
 500         rq_data = &enic->rq[rq_sop->data_queue_idx];
 501
 502         enic_rxmbuf_queue_release(enic, rq_sop);
 503         if (rq_data->in_use)
 504                 enic_rxmbuf_queue_release(enic, rq_data);
 505
 506         rte_free(rq_sop->mbuf_ring);
 507         if (rq_data->in_use)
 508                 rte_free(rq_data->mbuf_ring);
 509
 510         rq_sop->mbuf_ring = NULL;
 511         rq_data->mbuf_ring = NULL;
 512
 513         vnic_rq_free(rq_sop);
 514         if (rq_data->in_use)
 515                 vnic_rq_free(rq_data);
 516
 517         vnic_cq_free(&enic->cq[rq_sop->index]);
 518 }
 519
 520 void enic_start_wq(struct enic *enic, uint16_t queue_idx)
 521 {
 522         struct rte_eth_dev *eth_dev = enic->rte_dev;
 523         vnic_wq_enable(&enic->wq[queue_idx]);
 524         eth_dev->data->tx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STARTED;
 525 }
 526
 527 int enic_stop_wq(struct enic *enic, uint16_t queue_idx)
 528 {
 529         struct rte_eth_dev *eth_dev = enic->rte_dev;
 530         int ret;
 531
 532         ret = vnic_wq_disable(&enic->wq[queue_idx]);
 533         if (ret)
 534                 return ret;
 535
 536         eth_dev->data->tx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STOPPED;
 537         return 0;
 538 }
 539
 540 void enic_start_rq(struct enic *enic, uint16_t queue_idx)
 541 {
 542         struct vnic_rq *rq_sop = &enic->rq[enic_sop_rq(queue_idx)];
 543         struct vnic_rq *rq_data = &enic->rq[rq_sop->data_queue_idx];
 544         struct rte_eth_dev *eth_dev = enic->rte_dev;
 545
 546         if (rq_data->in_use)
 547                 vnic_rq_enable(rq_data);
 548         rte_mb();
 549         vnic_rq_enable(rq_sop);
 550         eth_dev->data->rx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STARTED;
 551 }
 552
 553 int enic_stop_rq(struct enic *enic, uint16_t queue_idx)
 554 {
 555         int ret1 = 0, ret2 = 0;
 556         struct rte_eth_dev *eth_dev = enic->rte_dev;
 557         struct vnic_rq *rq_sop = &enic->rq[enic_sop_rq(queue_idx)];
 558         struct vnic_rq *rq_data = &enic->rq[rq_sop->data_queue_idx];
 559
 560         ret2 = vnic_rq_disable(rq_sop);
 561         rte_mb();
 562         if (rq_data->in_use)
 563                 ret1 = vnic_rq_disable(rq_data);
 564
 565         if (ret2)
 566                 return ret2;
 567         else if (ret1)
 568                 return ret1;
 569
 570         eth_dev->data->rx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STOPPED;
 571         return 0;
 572 }
 573
 574 int enic_alloc_rq(struct enic *enic, uint16_t queue_idx,
 575         unsigned int socket_id, struct rte_mempool *mp,
 576         uint16_t nb_desc)
 577 {
 578         int rc;
 579         uint16_t sop_queue_idx = enic_sop_rq(queue_idx);
 580         uint16_t data_queue_idx = enic_data_rq(queue_idx);
 581         struct vnic_rq *rq_sop = &enic->rq[sop_queue_idx];
 582         struct vnic_rq *rq_data = &enic->rq[data_queue_idx];
 583         unsigned int mbuf_size, mbufs_per_pkt;
 584         unsigned int nb_sop_desc, nb_data_desc;
 585         uint16_t min_sop, max_sop, min_data, max_data;
 586
 587         rq_sop->is_sop = 1;
 588         rq_sop->data_queue_idx = data_queue_idx;
 589         rq_data->is_sop = 0;
 590         rq_data->data_queue_idx = 0;
 591         rq_sop->socket_id = socket_id;
 592         rq_sop->mp = mp;
 593         rq_data->socket_id = socket_id;
 594         rq_data->mp = mp;
 595         rq_sop->in_use = 1;
 596
 597         mbuf_size = (uint16_t)(rte_pktmbuf_data_room_size(mp) -
 598                                RTE_PKTMBUF_HEADROOM);
 599
 600         if (enic->rte_dev->data->dev_conf.rxmode.enable_scatter) {
 601                 dev_info(enic, "Scatter rx mode enabled\n");
 602                 /* ceil((mtu + ETHER_HDR_LEN + 4)/mbuf_size) */
 603                 mbufs_per_pkt = ((enic->config.mtu + ETHER_HDR_LEN + 4) +
 604                                  (mbuf_size - 1)) / mbuf_size;
 605         } else {
 606                 dev_info(enic, "Scatter rx mode disabled\n");
 607                 mbufs_per_pkt = 1;
 608         }
 609
 610         if (mbufs_per_pkt > 1) {
 611                 dev_info(enic, "Scatter rx mode in use\n");
 612                 rq_data->in_use = 1;
 613         } else {
 614                 dev_info(enic, "Scatter rx mode not being used\n");
 615                 rq_data->in_use = 0;
 616         }
 617
 618         /* number of descriptors have to be a multiple of 32 */
 619         nb_sop_desc = (nb_desc / mbufs_per_pkt) & ~0x1F;
 620         nb_data_desc = (nb_desc - nb_sop_desc) & ~0x1F;
 621
 622         rq_sop->max_mbufs_per_pkt = mbufs_per_pkt;
 623         rq_data->max_mbufs_per_pkt = mbufs_per_pkt;
 624
 625         if (mbufs_per_pkt > 1) {
 626                 min_sop = 64;
 627                 max_sop = ((enic->config.rq_desc_count /
 628                             (mbufs_per_pkt - 1)) & ~0x1F);
 629                 min_data = min_sop * (mbufs_per_pkt - 1);
 630                 max_data = enic->config.rq_desc_count;
 631         } else {
 632                 min_sop = 64;
 633                 max_sop = enic->config.rq_desc_count;
 634                 min_data = 0;
 635                 max_data = 0;
 636         }
 637
 638         if (nb_desc < (min_sop + min_data)) {
 639                 dev_warning(enic,
 640                             "Number of rx descs too low, adjusting to minimum\n");
 641                 nb_sop_desc = min_sop;
 642                 nb_data_desc = min_data;
 643         } else if (nb_desc > (max_sop + max_data)) {
 644                 dev_warning(enic,
 645                             "Number of rx_descs too high, adjusting to maximum\n");
 646                 nb_sop_desc = max_sop;
 647                 nb_data_desc = max_data;
 648         }
 649         if (mbufs_per_pkt > 1) {
 650                 dev_info(enic, "For mtu %d and mbuf size %d valid rx descriptor range is %d to %d\n",
 651                          enic->config.mtu, mbuf_size, min_sop + min_data,
 652                          max_sop + max_data);
 653         }
 654         dev_info(enic, "Using %d rx descriptors (sop %d, data %d)\n",
 655                  nb_sop_desc + nb_data_desc, nb_sop_desc, nb_data_desc);
 656
 657         /* Allocate sop queue resources */
 658         rc = vnic_rq_alloc(enic->vdev, rq_sop, sop_queue_idx,
 659                 nb_sop_desc, sizeof(struct rq_enet_desc));
 660         if (rc) {
 661                 dev_err(enic, "error in allocation of sop rq\n");
 662                 goto err_exit;
 663         }
 664         nb_sop_desc = rq_sop->ring.desc_count;
 665
 666         if (rq_data->in_use) {
 667                 /* Allocate data queue resources */
 668                 rc = vnic_rq_alloc(enic->vdev, rq_data, data_queue_idx,
 669                                    nb_data_desc,
 670                                    sizeof(struct rq_enet_desc));
 671                 if (rc) {
 672                         dev_err(enic, "error in allocation of data rq\n");
 673                         goto err_free_rq_sop;
 674                 }
 675                 nb_data_desc = rq_data->ring.desc_count;
 676         }
 677         rc = vnic_cq_alloc(enic->vdev, &enic->cq[queue_idx], queue_idx,
 678                            socket_id, nb_sop_desc + nb_data_desc,
 679                            sizeof(struct cq_enet_rq_desc));
 680         if (rc) {
 681                 dev_err(enic, "error in allocation of cq for rq\n");
 682                 goto err_free_rq_data;
 683         }
 684
 685         /* Allocate the mbuf rings */
 686         rq_sop->mbuf_ring = (struct rte_mbuf **)
 687                 rte_zmalloc_socket("rq->mbuf_ring",
 688                                    sizeof(struct rte_mbuf *) * nb_sop_desc,
 689                                    RTE_CACHE_LINE_SIZE, rq_sop->socket_id);
 690         if (rq_sop->mbuf_ring == NULL)
 691                 goto err_free_cq;
 692
 693         if (rq_data->in_use) {
 694                 rq_data->mbuf_ring = (struct rte_mbuf **)
 695                         rte_zmalloc_socket("rq->mbuf_ring",
 696                                 sizeof(struct rte_mbuf *) * nb_data_desc,
 697                                 RTE_CACHE_LINE_SIZE, rq_sop->socket_id);
 698                 if (rq_data->mbuf_ring == NULL)
 699                         goto err_free_sop_mbuf;
 700         }
 701
 702         return 0;
 703
 704 err_free_sop_mbuf:
 705         rte_free(rq_sop->mbuf_ring);
 706 err_free_cq:
 707         /* cleanup on error */
 708         vnic_cq_free(&enic->cq[queue_idx]);
 709 err_free_rq_data:
 710         if (rq_data->in_use)
 711                 vnic_rq_free(rq_data);
 712 err_free_rq_sop:
 713         vnic_rq_free(rq_sop);
 714 err_exit:
 715         return -ENOMEM;
 716 }
 717
 718 void enic_free_wq(void *txq)
 719 {
 720         struct vnic_wq *wq;
 721         struct enic *enic;
 722
 723         if (txq == NULL)
 724                 return;
 725
 726         wq = (struct vnic_wq *)txq;
 727         enic = vnic_dev_priv(wq->vdev);
 728         rte_memzone_free(wq->cqmsg_rz);
 729         vnic_wq_free(wq);
 730         vnic_cq_free(&enic->cq[enic->rq_count + wq->index]);
 731 }
 732
 733 int enic_alloc_wq(struct enic *enic, uint16_t queue_idx,
 734         unsigned int socket_id, uint16_t nb_desc)
 735 {
 736         int err;
 737         struct vnic_wq *wq = &enic->wq[queue_idx];
 738         unsigned int cq_index = enic_cq_wq(enic, queue_idx);
 739         char name[NAME_MAX];
 740         static int instance;
 741
 742         wq->socket_id = socket_id;
 743         if (nb_desc) {
 744                 if (nb_desc > enic->config.wq_desc_count) {
 745                         dev_warning(enic,
 746                                 "WQ %d - number of tx desc in cmd line (%d)"\
 747                                 "is greater than that in the UCSM/CIMC adapter"\
 748                                 "policy.  Applying the value in the adapter "\
 749                                 "policy (%d)\n",
 750                                 queue_idx, nb_desc, enic->config.wq_desc_count);
 751                 } else if (nb_desc != enic->config.wq_desc_count) {
 752                         enic->config.wq_desc_count = nb_desc;
 753                         dev_info(enic,
 754                                 "TX Queues - effective number of descs:%d\n",
 755                                 nb_desc);
 756                 }
 757         }
 758
 759         /* Allocate queue resources */
 760         err = vnic_wq_alloc(enic->vdev, &enic->wq[queue_idx], queue_idx,
 761                 enic->config.wq_desc_count,
 762                 sizeof(struct wq_enet_desc));
 763         if (err) {
 764                 dev_err(enic, "error in allocation of wq\n");
 765                 return err;
 766         }
 767
 768         err = vnic_cq_alloc(enic->vdev, &enic->cq[cq_index], cq_index,
 769                 socket_id, enic->config.wq_desc_count,
 770                 sizeof(struct cq_enet_wq_desc));
 771         if (err) {
 772                 vnic_wq_free(wq);
 773                 dev_err(enic, "error in allocation of cq for wq\n");
 774         }
 775
 776         /* setup up CQ message */
 777         snprintf((char *)name, sizeof(name),
 778                  "vnic_cqmsg-%s-%d-%d", enic->bdf_name, queue_idx,
 779                 instance++);
 780
 781         wq->cqmsg_rz = rte_memzone_reserve_aligned((const char *)name,
 782                                                    sizeof(uint32_t),
 783                                                    SOCKET_ID_ANY, 0,
 784                                                    ENIC_ALIGN);
 785         if (!wq->cqmsg_rz)
 786                 return -ENOMEM;
 787
 788         return err;
 789 }
 790
 791 int enic_disable(struct enic *enic)
 792 {
 793         unsigned int i;
 794         int err;
 795
 796         vnic_intr_mask(&enic->intr);
 797         (void)vnic_intr_masked(&enic->intr); /* flush write */
 798
 799         vnic_dev_disable(enic->vdev);
 800
 801         enic_clsf_destroy(enic);
 802
 803         if (!enic_is_sriov_vf(enic))
 804                 vnic_dev_del_addr(enic->vdev, enic->mac_addr);
 805
 806         for (i = 0; i < enic->wq_count; i++) {
 807                 err = vnic_wq_disable(&enic->wq[i]);
 808                 if (err)
 809                         return err;
 810         }
 811         for (i = 0; i < enic_vnic_rq_count(enic); i++) {
 812                 if (enic->rq[i].in_use) {
 813                         err = vnic_rq_disable(&enic->rq[i]);
 814                         if (err)
 815                                 return err;
 816                 }
 817         }
 818
 819         vnic_dev_set_reset_flag(enic->vdev, 1);
 820         vnic_dev_notify_unset(enic->vdev);
 821
 822         for (i = 0; i < enic->wq_count; i++)
 823                 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
 824
 825         for (i = 0; i < enic_vnic_rq_count(enic); i++)
 826                 if (enic->rq[i].in_use)
 827                         vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
 828         for (i = 0; i < enic->cq_count; i++)
 829                 vnic_cq_clean(&enic->cq[i]);
 830         vnic_intr_clean(&enic->intr);
 831
 832         return 0;
 833 }
 834
 835 static int enic_dev_wait(struct vnic_dev *vdev,
 836         int (*start)(struct vnic_dev *, int),
 837         int (*finished)(struct vnic_dev *, int *),
 838         int arg)
 839 {
 840         int done;
 841         int err;
 842         int i;
 843
 844         err = start(vdev, arg);
 845         if (err)
 846                 return err;
 847
 848         /* Wait for func to complete...2 seconds max */
 849         for (i = 0; i < 2000; i++) {
 850                 err = finished(vdev, &done);
 851                 if (err)
 852                         return err;
 853                 if (done)
 854                         return 0;
 855                 usleep(1000);
 856         }
 857         return -ETIMEDOUT;
 858 }
 859
 860 static int enic_dev_open(struct enic *enic)
 861 {
 862         int err;
 863
 864         err = enic_dev_wait(enic->vdev, vnic_dev_open,
 865                 vnic_dev_open_done, 0);
 866         if (err)
 867                 dev_err(enic_get_dev(enic),
 868                         "vNIC device open failed, err %d\n", err);
 869
 870         return err;
 871 }
 872
 873 static int enic_set_rsskey(struct enic *enic)
 874 {
 875         dma_addr_t rss_key_buf_pa;
 876         union vnic_rss_key *rss_key_buf_va = NULL;
 877         static union vnic_rss_key rss_key = {
 878                 .key = {
 879                         [0] = {.b = {85, 67, 83, 97, 119, 101, 115, 111, 109, 101}},
 880                         [1] = {.b = {80, 65, 76, 79, 117, 110, 105, 113, 117, 101}},
 881                         [2] = {.b = {76, 73, 78, 85, 88, 114, 111, 99, 107, 115}},
 882                         [3] = {.b = {69, 78, 73, 67, 105, 115, 99, 111, 111, 108}},
 883                 }
 884         };
 885         int err;
 886         u8 name[NAME_MAX];
 887
 888         snprintf((char *)name, NAME_MAX, "rss_key-%s", enic->bdf_name);
 889         rss_key_buf_va = enic_alloc_consistent(enic, sizeof(union vnic_rss_key),
 890                 &rss_key_buf_pa, name);
 891         if (!rss_key_buf_va)
 892                 return -ENOMEM;
 893
 894         rte_memcpy(rss_key_buf_va, &rss_key, sizeof(union vnic_rss_key));
 895
 896         err = enic_set_rss_key(enic,
 897                 rss_key_buf_pa,
 898                 sizeof(union vnic_rss_key));
 899
 900         enic_free_consistent(enic, sizeof(union vnic_rss_key),
 901                 rss_key_buf_va, rss_key_buf_pa);
 902
 903         return err;
 904 }
 905
 906 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
 907 {
 908         dma_addr_t rss_cpu_buf_pa;
 909         union vnic_rss_cpu *rss_cpu_buf_va = NULL;
 910         int i;
 911         int err;
 912         u8 name[NAME_MAX];
 913
 914         snprintf((char *)name, NAME_MAX, "rss_cpu-%s", enic->bdf_name);
 915         rss_cpu_buf_va = enic_alloc_consistent(enic, sizeof(union vnic_rss_cpu),
 916                 &rss_cpu_buf_pa, name);
 917         if (!rss_cpu_buf_va)
 918                 return -ENOMEM;
 919
 920         for (i = 0; i < (1 << rss_hash_bits); i++)
 921                 (*rss_cpu_buf_va).cpu[i / 4].b[i % 4] =
 922                         enic_sop_rq(i % enic->rq_count);
 923
 924         err = enic_set_rss_cpu(enic,
 925                 rss_cpu_buf_pa,
 926                 sizeof(union vnic_rss_cpu));
 927
 928         enic_free_consistent(enic, sizeof(union vnic_rss_cpu),
 929                 rss_cpu_buf_va, rss_cpu_buf_pa);
 930
 931         return err;
 932 }
 933
 934 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
 935         u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
 936 {
 937         const u8 tso_ipid_split_en = 0;
 938         int err;
 939
 940         /* Enable VLAN tag stripping */
 941
 942         err = enic_set_nic_cfg(enic,
 943                 rss_default_cpu, rss_hash_type,
 944                 rss_hash_bits, rss_base_cpu,
 945                 rss_enable, tso_ipid_split_en,
 946                 enic->ig_vlan_strip_en);
 947
 948         return err;
 949 }
 950
 951 int enic_set_rss_nic_cfg(struct enic *enic)
 952 {
 953         const u8 rss_default_cpu = 0;
 954         const u8 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
 955             NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
 956             NIC_CFG_RSS_HASH_TYPE_IPV6 |
 957             NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
 958         const u8 rss_hash_bits = 7;
 959         const u8 rss_base_cpu = 0;
 960         u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
 961
 962         if (rss_enable) {
 963                 if (!enic_set_rsskey(enic)) {
 964                         if (enic_set_rsscpu(enic, rss_hash_bits)) {
 965                                 rss_enable = 0;
 966                                 dev_warning(enic, "RSS disabled, "\
 967                                         "Failed to set RSS cpu indirection table.");
 968                         }
 969                 } else {
 970                         rss_enable = 0;
 971                         dev_warning(enic,
 972                                 "RSS disabled, Failed to set RSS key.\n");
 973                 }
 974         }
 975
 976         return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
 977                 rss_hash_bits, rss_base_cpu, rss_enable);
 978 }
 979
 980 int enic_setup_finish(struct enic *enic)
 981 {
 982         int ret;
 983
 984         enic_init_soft_stats(enic);
 985
 986         ret = enic_set_rss_nic_cfg(enic);
 987         if (ret) {
 988                 dev_err(enic, "Failed to config nic, aborting.\n");
 989                 return -1;
 990         }
 991
 992         /* Default conf */
 993         vnic_dev_packet_filter(enic->vdev,
 994                 1 /* directed  */,
 995                 1 /* multicast */,
 996                 1 /* broadcast */,
 997                 0 /* promisc   */,
 998                 1 /* allmulti  */);
 999
1000         enic->promisc = 0;
1001         enic->allmulti = 1;
1002
1003         return 0;
1004 }
1005
1006 void enic_add_packet_filter(struct enic *enic)
1007 {
1008         /* Args -> directed, multicast, broadcast, promisc, allmulti */
1009         vnic_dev_packet_filter(enic->vdev, 1, 1, 1,
1010                 enic->promisc, enic->allmulti);
1011 }
1012
1013 int enic_get_link_status(struct enic *enic)
1014 {
1015         return vnic_dev_link_status(enic->vdev);
1016 }
1017
1018 static void enic_dev_deinit(struct enic *enic)
1019 {
1020         struct rte_eth_dev *eth_dev = enic->rte_dev;
1021
1022         rte_free(eth_dev->data->mac_addrs);
1023 }
1024
1025
1026 int enic_set_vnic_res(struct enic *enic)
1027 {
1028         struct rte_eth_dev *eth_dev = enic->rte_dev;
1029         int rc = 0;
1030
1031         /* With Rx scatter support, two RQs are now used per RQ used by
1032          * the application.
1033          */
1034         if (enic->conf_rq_count < eth_dev->data->nb_rx_queues) {
1035                 dev_err(dev, "Not enough Receive queues. Requested:%u which uses %d RQs on VIC, Configured:%u\n",
1036                         eth_dev->data->nb_rx_queues,
1037                         eth_dev->data->nb_rx_queues * 2, enic->conf_rq_count);
1038                 rc = -EINVAL;
1039         }
1040         if (enic->conf_wq_count < eth_dev->data->nb_tx_queues) {
1041                 dev_err(dev, "Not enough Transmit queues. Requested:%u, Configured:%u\n",
1042                         eth_dev->data->nb_tx_queues, enic->conf_wq_count);
1043                 rc = -EINVAL;
1044         }
1045
1046         if (enic->conf_cq_count < (eth_dev->data->nb_rx_queues +
1047                                    eth_dev->data->nb_tx_queues)) {
1048                 dev_err(dev, "Not enough Completion queues. Required:%u, Configured:%u\n",
1049                         (eth_dev->data->nb_rx_queues +
1050                          eth_dev->data->nb_tx_queues), enic->conf_cq_count);
1051                 rc = -EINVAL;
1052         }
1053
1054         if (rc == 0) {
1055                 enic->rq_count = eth_dev->data->nb_rx_queues;
1056                 enic->wq_count = eth_dev->data->nb_tx_queues;
1057                 enic->cq_count = enic->rq_count + enic->wq_count;
1058         }
1059
1060         return rc;
1061 }
1062
1063 /* The Cisco NIC can send and receive packets up to a max packet size
1064  * determined by the NIC type and firmware. There is also an MTU
1065  * configured into the NIC via the CIMC/UCSM management interface
1066  * which can be overridden by this function (up to the max packet size).
1067  * Depending on the network setup, doing so may cause packet drops
1068  * and unexpected behavior.
1069  */
1070 int enic_set_mtu(struct enic *enic, uint16_t new_mtu)
1071 {
1072         uint16_t old_mtu;       /* previous setting */
1073         uint16_t config_mtu;    /* Value configured into NIC via CIMC/UCSM */
1074         struct rte_eth_dev *eth_dev = enic->rte_dev;
1075
1076         old_mtu = eth_dev->data->mtu;
1077         config_mtu = enic->config.mtu;
1078
1079         /* only works with Rx scatter disabled */
1080         if (enic->rte_dev->data->dev_conf.rxmode.enable_scatter)
1081                 return -ENOTSUP;
1082
1083         if (new_mtu > enic->max_mtu) {
1084                 dev_err(enic,
1085                         "MTU not updated: requested (%u) greater than max (%u)\n",
1086                         new_mtu, enic->max_mtu);
1087                 return -EINVAL;
1088         }
1089         if (new_mtu < ENIC_MIN_MTU) {
1090                 dev_info(enic,
1091                         "MTU not updated: requested (%u) less than min (%u)\n",
1092                         new_mtu, ENIC_MIN_MTU);
1093                 return -EINVAL;
1094         }
1095         if (new_mtu > config_mtu)
1096                 dev_warning(enic,
1097                         "MTU (%u) is greater than value configured in NIC (%u)\n",
1098                         new_mtu, config_mtu);
1099
1100         /* update the mtu */
1101         eth_dev->data->mtu = new_mtu;
1102
1103         dev_info(enic, "MTU changed from %u to %u\n",  old_mtu, new_mtu);
1104         return 0;
1105 }
1106
1107 static int enic_dev_init(struct enic *enic)
1108 {
1109         int err;
1110         struct rte_eth_dev *eth_dev = enic->rte_dev;
1111
1112         vnic_dev_intr_coal_timer_info_default(enic->vdev);
1113
1114         /* Get vNIC configuration
1115         */
1116         err = enic_get_vnic_config(enic);
1117         if (err) {
1118                 dev_err(dev, "Get vNIC configuration failed, aborting\n");
1119                 return err;
1120         }
1121
1122         eth_dev->data->mac_addrs = rte_zmalloc("enic_mac_addr", ETH_ALEN, 0);
1123         if (!eth_dev->data->mac_addrs) {
1124                 dev_err(enic, "mac addr storage alloc failed, aborting.\n");
1125                 return -1;
1126         }
1127         ether_addr_copy((struct ether_addr *) enic->mac_addr,
1128                 &eth_dev->data->mac_addrs[0]);
1129
1130
1131         /* Get available resource counts
1132         */
1133         enic_get_res_counts(enic);
1134
1135         vnic_dev_set_reset_flag(enic->vdev, 0);
1136
1137         return 0;
1138
1139 }
1140
1141 int enic_probe(struct enic *enic)
1142 {
1143         struct rte_pci_device *pdev = enic->pdev;
1144         int err = -1;
1145
1146         dev_debug(enic, " Initializing ENIC PMD\n");
1147
1148         enic->bar0.vaddr = (void *)pdev->mem_resource[0].addr;
1149         enic->bar0.len = pdev->mem_resource[0].len;
1150
1151         /* Register vNIC device */
1152         enic->vdev = vnic_dev_register(NULL, enic, enic->pdev, &enic->bar0, 1);
1153         if (!enic->vdev) {
1154                 dev_err(enic, "vNIC registration failed, aborting\n");
1155                 goto err_out;
1156         }
1157
1158         LIST_INIT(&enic->memzone_list);
1159         rte_spinlock_init(&enic->memzone_list_lock);
1160
1161         vnic_register_cbacks(enic->vdev,
1162                 enic_alloc_consistent,
1163                 enic_free_consistent);
1164
1165         /* Issue device open to get device in known state */
1166         err = enic_dev_open(enic);
1167         if (err) {
1168                 dev_err(enic, "vNIC dev open failed, aborting\n");
1169                 goto err_out_unregister;
1170         }
1171
1172         /* Set ingress vlan rewrite mode before vnic initialization */
1173         err = vnic_dev_set_ig_vlan_rewrite_mode(enic->vdev,
1174                 IG_VLAN_REWRITE_MODE_PASS_THRU);
1175         if (err) {
1176                 dev_err(enic,
1177                         "Failed to set ingress vlan rewrite mode, aborting.\n");
1178                 goto err_out_dev_close;
1179         }
1180
1181         /* Issue device init to initialize the vnic-to-switch link.
1182          * We'll start with carrier off and wait for link UP
1183          * notification later to turn on carrier.  We don't need
1184          * to wait here for the vnic-to-switch link initialization
1185          * to complete; link UP notification is the indication that
1186          * the process is complete.
1187          */
1188
1189         err = vnic_dev_init(enic->vdev, 0);
1190         if (err) {
1191                 dev_err(enic, "vNIC dev init failed, aborting\n");
1192                 goto err_out_dev_close;
1193         }
1194
1195         err = enic_dev_init(enic);
1196         if (err) {
1197                 dev_err(enic, "Device initialization failed, aborting\n");
1198                 goto err_out_dev_close;
1199         }
1200
1201         return 0;
1202
1203 err_out_dev_close:
1204         vnic_dev_close(enic->vdev);
1205 err_out_unregister:
1206         vnic_dev_unregister(enic->vdev);
1207 err_out:
1208         return err;
1209 }
1210
1211 void enic_remove(struct enic *enic)
1212 {
1213         enic_dev_deinit(enic);
1214         vnic_dev_close(enic->vdev);
1215         vnic_dev_unregister(enic->vdev);
1216 }