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