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_mempool_put(mbuf->pool, 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, 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_enable_wait(enic->vdev);
 460
 461         /* Register and enable error interrupt */
 462         rte_intr_callback_register(&(enic->pdev->intr_handle),
 463                 enic_intr_handler, (void *)enic->rte_dev);
 464
 465         rte_intr_enable(&(enic->pdev->intr_handle));
 466         vnic_intr_unmask(&enic->intr);
 467
 468         return 0;
 469 }
 470
 471 int enic_alloc_intr_resources(struct enic *enic)
 472 {
 473         int err;
 474
 475         dev_info(enic, "vNIC resources used:  "\
 476                 "wq %d rq %d cq %d intr %d\n",
 477                 enic->wq_count, enic_vnic_rq_count(enic),
 478                 enic->cq_count, enic->intr_count);
 479
 480         err = vnic_intr_alloc(enic->vdev, &enic->intr, 0);
 481         if (err)
 482                 enic_free_vnic_resources(enic);
 483
 484         return err;
 485 }
 486
 487 void enic_free_rq(void *rxq)
 488 {
 489         struct vnic_rq *rq_sop, *rq_data;
 490         struct enic *enic;
 491
 492         if (rxq == NULL)
 493                 return;
 494
 495         rq_sop = (struct vnic_rq *)rxq;
 496         enic = vnic_dev_priv(rq_sop->vdev);
 497         rq_data = &enic->rq[rq_sop->data_queue_idx];
 498
 499         enic_rxmbuf_queue_release(enic, rq_sop);
 500         if (rq_data->in_use)
 501                 enic_rxmbuf_queue_release(enic, rq_data);
 502
 503         rte_free(rq_sop->mbuf_ring);
 504         if (rq_data->in_use)
 505                 rte_free(rq_data->mbuf_ring);
 506
 507         rq_sop->mbuf_ring = NULL;
 508         rq_data->mbuf_ring = NULL;
 509
 510         vnic_rq_free(rq_sop);
 511         if (rq_data->in_use)
 512                 vnic_rq_free(rq_data);
 513
 514         vnic_cq_free(&enic->cq[rq_sop->index]);
 515 }
 516
 517 void enic_start_wq(struct enic *enic, uint16_t queue_idx)
 518 {
 519         vnic_wq_enable(&enic->wq[queue_idx]);
 520 }
 521
 522 int enic_stop_wq(struct enic *enic, uint16_t queue_idx)
 523 {
 524         return vnic_wq_disable(&enic->wq[queue_idx]);
 525 }
 526
 527 void enic_start_rq(struct enic *enic, uint16_t queue_idx)
 528 {
 529         struct vnic_rq *rq_sop = &enic->rq[enic_sop_rq(queue_idx)];
 530         struct vnic_rq *rq_data = &enic->rq[rq_sop->data_queue_idx];
 531
 532         if (rq_data->in_use)
 533                 vnic_rq_enable(rq_data);
 534         rte_mb();
 535         vnic_rq_enable(rq_sop);
 536
 537 }
 538
 539 int enic_stop_rq(struct enic *enic, uint16_t queue_idx)
 540 {
 541         int ret1 = 0, ret2 = 0;
 542
 543         struct vnic_rq *rq_sop = &enic->rq[enic_sop_rq(queue_idx)];
 544         struct vnic_rq *rq_data = &enic->rq[rq_sop->data_queue_idx];
 545
 546         ret2 = vnic_rq_disable(rq_sop);
 547         rte_mb();
 548         if (rq_data->in_use)
 549                 ret1 = vnic_rq_disable(rq_data);
 550
 551         if (ret2)
 552                 return ret2;
 553         else
 554                 return ret1;
 555 }
 556
 557 int enic_alloc_rq(struct enic *enic, uint16_t queue_idx,
 558         unsigned int socket_id, struct rte_mempool *mp,
 559         uint16_t nb_desc)
 560 {
 561         int rc;
 562         uint16_t sop_queue_idx = enic_sop_rq(queue_idx);
 563         uint16_t data_queue_idx = enic_data_rq(queue_idx);
 564         struct vnic_rq *rq_sop = &enic->rq[sop_queue_idx];
 565         struct vnic_rq *rq_data = &enic->rq[data_queue_idx];
 566         unsigned int mbuf_size, mbufs_per_pkt;
 567         unsigned int nb_sop_desc, nb_data_desc;
 568         uint16_t min_sop, max_sop, min_data, max_data;
 569
 570         rq_sop->is_sop = 1;
 571         rq_sop->data_queue_idx = data_queue_idx;
 572         rq_data->is_sop = 0;
 573         rq_data->data_queue_idx = 0;
 574         rq_sop->socket_id = socket_id;
 575         rq_sop->mp = mp;
 576         rq_data->socket_id = socket_id;
 577         rq_data->mp = mp;
 578         rq_sop->in_use = 1;
 579
 580         mbuf_size = (uint16_t)(rte_pktmbuf_data_room_size(mp) -
 581                                RTE_PKTMBUF_HEADROOM);
 582
 583         if (enic->rte_dev->data->dev_conf.rxmode.enable_scatter) {
 584                 dev_info(enic, "Scatter rx mode enabled\n");
 585                 /* ceil((mtu + ETHER_HDR_LEN + 4)/mbuf_size) */
 586                 mbufs_per_pkt = ((enic->config.mtu + ETHER_HDR_LEN + 4) +
 587                                  (mbuf_size - 1)) / mbuf_size;
 588         } else {
 589                 dev_info(enic, "Scatter rx mode disabled\n");
 590                 mbufs_per_pkt = 1;
 591         }
 592
 593         if (mbufs_per_pkt > 1) {
 594                 dev_info(enic, "Scatter rx mode in use\n");
 595                 rq_data->in_use = 1;
 596         } else {
 597                 dev_info(enic, "Scatter rx mode not being used\n");
 598                 rq_data->in_use = 0;
 599         }
 600
 601         /* number of descriptors have to be a multiple of 32 */
 602         nb_sop_desc = (nb_desc / mbufs_per_pkt) & ~0x1F;
 603         nb_data_desc = (nb_desc - nb_sop_desc) & ~0x1F;
 604
 605         rq_sop->max_mbufs_per_pkt = mbufs_per_pkt;
 606         rq_data->max_mbufs_per_pkt = mbufs_per_pkt;
 607
 608         if (mbufs_per_pkt > 1) {
 609                 min_sop = 64;
 610                 max_sop = ((enic->config.rq_desc_count /
 611                             (mbufs_per_pkt - 1)) & ~0x1F);
 612                 min_data = min_sop * (mbufs_per_pkt - 1);
 613                 max_data = enic->config.rq_desc_count;
 614         } else {
 615                 min_sop = 64;
 616                 max_sop = enic->config.rq_desc_count;
 617                 min_data = 0;
 618                 max_data = 0;
 619         }
 620
 621         if (nb_desc < (min_sop + min_data)) {
 622                 dev_warning(enic,
 623                             "Number of rx descs too low, adjusting to minimum\n");
 624                 nb_sop_desc = min_sop;
 625                 nb_data_desc = min_data;
 626         } else if (nb_desc > (max_sop + max_data)) {
 627                 dev_warning(enic,
 628                             "Number of rx_descs too high, adjusting to maximum\n");
 629                 nb_sop_desc = max_sop;
 630                 nb_data_desc = max_data;
 631         }
 632         if (mbufs_per_pkt > 1) {
 633                 dev_info(enic, "For mtu %d and mbuf size %d valid rx descriptor range is %d to %d\n",
 634                          enic->config.mtu, mbuf_size, min_sop + min_data,
 635                          max_sop + max_data);
 636         }
 637         dev_info(enic, "Using %d rx descriptors (sop %d, data %d)\n",
 638                  nb_sop_desc + nb_data_desc, nb_sop_desc, nb_data_desc);
 639
 640         /* Allocate sop queue resources */
 641         rc = vnic_rq_alloc(enic->vdev, rq_sop, sop_queue_idx,
 642                 nb_sop_desc, sizeof(struct rq_enet_desc));
 643         if (rc) {
 644                 dev_err(enic, "error in allocation of sop rq\n");
 645                 goto err_exit;
 646         }
 647         nb_sop_desc = rq_sop->ring.desc_count;
 648
 649         if (rq_data->in_use) {
 650                 /* Allocate data queue resources */
 651                 rc = vnic_rq_alloc(enic->vdev, rq_data, data_queue_idx,
 652                                    nb_data_desc,
 653                                    sizeof(struct rq_enet_desc));
 654                 if (rc) {
 655                         dev_err(enic, "error in allocation of data rq\n");
 656                         goto err_free_rq_sop;
 657                 }
 658                 nb_data_desc = rq_data->ring.desc_count;
 659         }
 660         rc = vnic_cq_alloc(enic->vdev, &enic->cq[queue_idx], queue_idx,
 661                            socket_id, nb_sop_desc + nb_data_desc,
 662                            sizeof(struct cq_enet_rq_desc));
 663         if (rc) {
 664                 dev_err(enic, "error in allocation of cq for rq\n");
 665                 goto err_free_rq_data;
 666         }
 667
 668         /* Allocate the mbuf rings */
 669         rq_sop->mbuf_ring = (struct rte_mbuf **)
 670                 rte_zmalloc_socket("rq->mbuf_ring",
 671                                    sizeof(struct rte_mbuf *) * nb_sop_desc,
 672                                    RTE_CACHE_LINE_SIZE, rq_sop->socket_id);
 673         if (rq_sop->mbuf_ring == NULL)
 674                 goto err_free_cq;
 675
 676         if (rq_data->in_use) {
 677                 rq_data->mbuf_ring = (struct rte_mbuf **)
 678                         rte_zmalloc_socket("rq->mbuf_ring",
 679                                 sizeof(struct rte_mbuf *) * nb_data_desc,
 680                                 RTE_CACHE_LINE_SIZE, rq_sop->socket_id);
 681                 if (rq_data->mbuf_ring == NULL)
 682                         goto err_free_sop_mbuf;
 683         }
 684
 685         return 0;
 686
 687 err_free_sop_mbuf:
 688         rte_free(rq_sop->mbuf_ring);
 689 err_free_cq:
 690         /* cleanup on error */
 691         vnic_cq_free(&enic->cq[queue_idx]);
 692 err_free_rq_data:
 693         if (rq_data->in_use)
 694                 vnic_rq_free(rq_data);
 695 err_free_rq_sop:
 696         vnic_rq_free(rq_sop);
 697 err_exit:
 698         return -ENOMEM;
 699 }
 700
 701 void enic_free_wq(void *txq)
 702 {
 703         struct vnic_wq *wq;
 704         struct enic *enic;
 705
 706         if (txq == NULL)
 707                 return;
 708
 709         wq = (struct vnic_wq *)txq;
 710         enic = vnic_dev_priv(wq->vdev);
 711         rte_memzone_free(wq->cqmsg_rz);
 712         vnic_wq_free(wq);
 713         vnic_cq_free(&enic->cq[enic->rq_count + wq->index]);
 714 }
 715
 716 int enic_alloc_wq(struct enic *enic, uint16_t queue_idx,
 717         unsigned int socket_id, uint16_t nb_desc)
 718 {
 719         int err;
 720         struct vnic_wq *wq = &enic->wq[queue_idx];
 721         unsigned int cq_index = enic_cq_wq(enic, queue_idx);
 722         char name[NAME_MAX];
 723         static int instance;
 724
 725         wq->socket_id = socket_id;
 726         if (nb_desc) {
 727                 if (nb_desc > enic->config.wq_desc_count) {
 728                         dev_warning(enic,
 729                                 "WQ %d - number of tx desc in cmd line (%d)"\
 730                                 "is greater than that in the UCSM/CIMC adapter"\
 731                                 "policy.  Applying the value in the adapter "\
 732                                 "policy (%d)\n",
 733                                 queue_idx, nb_desc, enic->config.wq_desc_count);
 734                 } else if (nb_desc != enic->config.wq_desc_count) {
 735                         enic->config.wq_desc_count = nb_desc;
 736                         dev_info(enic,
 737                                 "TX Queues - effective number of descs:%d\n",
 738                                 nb_desc);
 739                 }
 740         }
 741
 742         /* Allocate queue resources */
 743         err = vnic_wq_alloc(enic->vdev, &enic->wq[queue_idx], queue_idx,
 744                 enic->config.wq_desc_count,
 745                 sizeof(struct wq_enet_desc));
 746         if (err) {
 747                 dev_err(enic, "error in allocation of wq\n");
 748                 return err;
 749         }
 750
 751         err = vnic_cq_alloc(enic->vdev, &enic->cq[cq_index], cq_index,
 752                 socket_id, enic->config.wq_desc_count,
 753                 sizeof(struct cq_enet_wq_desc));
 754         if (err) {
 755                 vnic_wq_free(wq);
 756                 dev_err(enic, "error in allocation of cq for wq\n");
 757         }
 758
 759         /* setup up CQ message */
 760         snprintf((char *)name, sizeof(name),
 761                  "vnic_cqmsg-%s-%d-%d", enic->bdf_name, queue_idx,
 762                 instance++);
 763
 764         wq->cqmsg_rz = rte_memzone_reserve_aligned((const char *)name,
 765                                                    sizeof(uint32_t),
 766                                                    SOCKET_ID_ANY, 0,
 767                                                    ENIC_ALIGN);
 768         if (!wq->cqmsg_rz)
 769                 return -ENOMEM;
 770
 771         return err;
 772 }
 773
 774 int enic_disable(struct enic *enic)
 775 {
 776         unsigned int i;
 777         int err;
 778
 779         vnic_intr_mask(&enic->intr);
 780         (void)vnic_intr_masked(&enic->intr); /* flush write */
 781
 782         vnic_dev_disable(enic->vdev);
 783
 784         enic_clsf_destroy(enic);
 785
 786         if (!enic_is_sriov_vf(enic))
 787                 vnic_dev_del_addr(enic->vdev, enic->mac_addr);
 788
 789         for (i = 0; i < enic->wq_count; i++) {
 790                 err = vnic_wq_disable(&enic->wq[i]);
 791                 if (err)
 792                         return err;
 793         }
 794         for (i = 0; i < enic_vnic_rq_count(enic); i++) {
 795                 if (enic->rq[i].in_use) {
 796                         err = vnic_rq_disable(&enic->rq[i]);
 797                         if (err)
 798                                 return err;
 799                 }
 800         }
 801
 802         vnic_dev_set_reset_flag(enic->vdev, 1);
 803         vnic_dev_notify_unset(enic->vdev);
 804
 805         for (i = 0; i < enic->wq_count; i++)
 806                 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
 807
 808         for (i = 0; i < enic_vnic_rq_count(enic); i++)
 809                 if (enic->rq[i].in_use)
 810                         vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
 811         for (i = 0; i < enic->cq_count; i++)
 812                 vnic_cq_clean(&enic->cq[i]);
 813         vnic_intr_clean(&enic->intr);
 814
 815         return 0;
 816 }
 817
 818 static int enic_dev_wait(struct vnic_dev *vdev,
 819         int (*start)(struct vnic_dev *, int),
 820         int (*finished)(struct vnic_dev *, int *),
 821         int arg)
 822 {
 823         int done;
 824         int err;
 825         int i;
 826
 827         err = start(vdev, arg);
 828         if (err)
 829                 return err;
 830
 831         /* Wait for func to complete...2 seconds max */
 832         for (i = 0; i < 2000; i++) {
 833                 err = finished(vdev, &done);
 834                 if (err)
 835                         return err;
 836                 if (done)
 837                         return 0;
 838                 usleep(1000);
 839         }
 840         return -ETIMEDOUT;
 841 }
 842
 843 static int enic_dev_open(struct enic *enic)
 844 {
 845         int err;
 846
 847         err = enic_dev_wait(enic->vdev, vnic_dev_open,
 848                 vnic_dev_open_done, 0);
 849         if (err)
 850                 dev_err(enic_get_dev(enic),
 851                         "vNIC device open failed, err %d\n", err);
 852
 853         return err;
 854 }
 855
 856 static int enic_set_rsskey(struct enic *enic)
 857 {
 858         dma_addr_t rss_key_buf_pa;
 859         union vnic_rss_key *rss_key_buf_va = NULL;
 860         static union vnic_rss_key rss_key = {
 861                 .key = {
 862                         [0] = {.b = {85, 67, 83, 97, 119, 101, 115, 111, 109, 101}},
 863                         [1] = {.b = {80, 65, 76, 79, 117, 110, 105, 113, 117, 101}},
 864                         [2] = {.b = {76, 73, 78, 85, 88, 114, 111, 99, 107, 115}},
 865                         [3] = {.b = {69, 78, 73, 67, 105, 115, 99, 111, 111, 108}},
 866                 }
 867         };
 868         int err;
 869         u8 name[NAME_MAX];
 870
 871         snprintf((char *)name, NAME_MAX, "rss_key-%s", enic->bdf_name);
 872         rss_key_buf_va = enic_alloc_consistent(enic, sizeof(union vnic_rss_key),
 873                 &rss_key_buf_pa, name);
 874         if (!rss_key_buf_va)
 875                 return -ENOMEM;
 876
 877         rte_memcpy(rss_key_buf_va, &rss_key, sizeof(union vnic_rss_key));
 878
 879         err = enic_set_rss_key(enic,
 880                 rss_key_buf_pa,
 881                 sizeof(union vnic_rss_key));
 882
 883         enic_free_consistent(enic, sizeof(union vnic_rss_key),
 884                 rss_key_buf_va, rss_key_buf_pa);
 885
 886         return err;
 887 }
 888
 889 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
 890 {
 891         dma_addr_t rss_cpu_buf_pa;
 892         union vnic_rss_cpu *rss_cpu_buf_va = NULL;
 893         int i;
 894         int err;
 895         u8 name[NAME_MAX];
 896
 897         snprintf((char *)name, NAME_MAX, "rss_cpu-%s", enic->bdf_name);
 898         rss_cpu_buf_va = enic_alloc_consistent(enic, sizeof(union vnic_rss_cpu),
 899                 &rss_cpu_buf_pa, name);
 900         if (!rss_cpu_buf_va)
 901                 return -ENOMEM;
 902
 903         for (i = 0; i < (1 << rss_hash_bits); i++)
 904                 (*rss_cpu_buf_va).cpu[i / 4].b[i % 4] =
 905                         enic_sop_rq(i % enic->rq_count);
 906
 907         err = enic_set_rss_cpu(enic,
 908                 rss_cpu_buf_pa,
 909                 sizeof(union vnic_rss_cpu));
 910
 911         enic_free_consistent(enic, sizeof(union vnic_rss_cpu),
 912                 rss_cpu_buf_va, rss_cpu_buf_pa);
 913
 914         return err;
 915 }
 916
 917 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
 918         u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
 919 {
 920         const u8 tso_ipid_split_en = 0;
 921         int err;
 922
 923         /* Enable VLAN tag stripping */
 924
 925         err = enic_set_nic_cfg(enic,
 926                 rss_default_cpu, rss_hash_type,
 927                 rss_hash_bits, rss_base_cpu,
 928                 rss_enable, tso_ipid_split_en,
 929                 enic->ig_vlan_strip_en);
 930
 931         return err;
 932 }
 933
 934 int enic_set_rss_nic_cfg(struct enic *enic)
 935 {
 936         const u8 rss_default_cpu = 0;
 937         const u8 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
 938             NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
 939             NIC_CFG_RSS_HASH_TYPE_IPV6 |
 940             NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
 941         const u8 rss_hash_bits = 7;
 942         const u8 rss_base_cpu = 0;
 943         u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
 944
 945         if (rss_enable) {
 946                 if (!enic_set_rsskey(enic)) {
 947                         if (enic_set_rsscpu(enic, rss_hash_bits)) {
 948                                 rss_enable = 0;
 949                                 dev_warning(enic, "RSS disabled, "\
 950                                         "Failed to set RSS cpu indirection table.");
 951                         }
 952                 } else {
 953                         rss_enable = 0;
 954                         dev_warning(enic,
 955                                 "RSS disabled, Failed to set RSS key.\n");
 956                 }
 957         }
 958
 959         return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
 960                 rss_hash_bits, rss_base_cpu, rss_enable);
 961 }
 962
 963 int enic_setup_finish(struct enic *enic)
 964 {
 965         int ret;
 966
 967         enic_init_soft_stats(enic);
 968
 969         ret = enic_set_rss_nic_cfg(enic);
 970         if (ret) {
 971                 dev_err(enic, "Failed to config nic, aborting.\n");
 972                 return -1;
 973         }
 974
 975         vnic_dev_add_addr(enic->vdev, enic->mac_addr);
 976
 977         /* Default conf */
 978         vnic_dev_packet_filter(enic->vdev,
 979                 1 /* directed  */,
 980                 1 /* multicast */,
 981                 1 /* broadcast */,
 982                 0 /* promisc   */,
 983                 1 /* allmulti  */);
 984
 985         enic->promisc = 0;
 986         enic->allmulti = 1;
 987
 988         return 0;
 989 }
 990
 991 void enic_add_packet_filter(struct enic *enic)
 992 {
 993         /* Args -> directed, multicast, broadcast, promisc, allmulti */
 994         vnic_dev_packet_filter(enic->vdev, 1, 1, 1,
 995                 enic->promisc, enic->allmulti);
 996 }
 997
 998 int enic_get_link_status(struct enic *enic)
 999 {
1000         return vnic_dev_link_status(enic->vdev);
1001 }
1002
1003 static void enic_dev_deinit(struct enic *enic)
1004 {
1005         struct rte_eth_dev *eth_dev = enic->rte_dev;
1006
1007         rte_free(eth_dev->data->mac_addrs);
1008 }
1009
1010
1011 int enic_set_vnic_res(struct enic *enic)
1012 {
1013         struct rte_eth_dev *eth_dev = enic->rte_dev;
1014         int rc = 0;
1015
1016         /* With Rx scatter support, two RQs are now used per RQ used by
1017          * the application.
1018          */
1019         if (enic->conf_rq_count < eth_dev->data->nb_rx_queues) {
1020                 dev_err(dev, "Not enough Receive queues. Requested:%u which uses %d RQs on VIC, Configured:%u\n",
1021                         eth_dev->data->nb_rx_queues,
1022                         eth_dev->data->nb_rx_queues * 2, enic->conf_rq_count);
1023                 rc = -EINVAL;
1024         }
1025         if (enic->conf_wq_count < eth_dev->data->nb_tx_queues) {
1026                 dev_err(dev, "Not enough Transmit queues. Requested:%u, Configured:%u\n",
1027                         eth_dev->data->nb_tx_queues, enic->conf_wq_count);
1028                 rc = -EINVAL;
1029         }
1030
1031         if (enic->conf_cq_count < (eth_dev->data->nb_rx_queues +
1032                                    eth_dev->data->nb_tx_queues)) {
1033                 dev_err(dev, "Not enough Completion queues. Required:%u, Configured:%u\n",
1034                         (eth_dev->data->nb_rx_queues +
1035                          eth_dev->data->nb_tx_queues), enic->conf_cq_count);
1036                 rc = -EINVAL;
1037         }
1038
1039         if (rc == 0) {
1040                 enic->rq_count = eth_dev->data->nb_rx_queues;
1041                 enic->wq_count = eth_dev->data->nb_tx_queues;
1042                 enic->cq_count = enic->rq_count + enic->wq_count;
1043         }
1044
1045         return rc;
1046 }
1047
1048 /* The Cisco NIC can send and receive packets up to a max packet size
1049  * determined by the NIC type and firmware. There is also an MTU
1050  * configured into the NIC via the CIMC/UCSM management interface
1051  * which can be overridden by this function (up to the max packet size).
1052  * Depending on the network setup, doing so may cause packet drops
1053  * and unexpected behavior.
1054  */
1055 int enic_set_mtu(struct enic *enic, uint16_t new_mtu)
1056 {
1057         uint16_t old_mtu;       /* previous setting */
1058         uint16_t config_mtu;    /* Value configured into NIC via CIMC/UCSM */
1059         struct rte_eth_dev *eth_dev = enic->rte_dev;
1060
1061         old_mtu = eth_dev->data->mtu;
1062         config_mtu = enic->config.mtu;
1063
1064         /* only works with Rx scatter disabled */
1065         if (enic->rte_dev->data->dev_conf.rxmode.enable_scatter)
1066                 return -ENOTSUP;
1067
1068         if (new_mtu > enic->max_mtu) {
1069                 dev_err(enic,
1070                         "MTU not updated: requested (%u) greater than max (%u)\n",
1071                         new_mtu, enic->max_mtu);
1072                 return -EINVAL;
1073         }
1074         if (new_mtu < ENIC_MIN_MTU) {
1075                 dev_info(enic,
1076                         "MTU not updated: requested (%u) less than min (%u)\n",
1077                         new_mtu, ENIC_MIN_MTU);
1078                 return -EINVAL;
1079         }
1080         if (new_mtu > config_mtu)
1081                 dev_warning(enic,
1082                         "MTU (%u) is greater than value configured in NIC (%u)\n",
1083                         new_mtu, config_mtu);
1084
1085         /* update the mtu */
1086         eth_dev->data->mtu = new_mtu;
1087
1088         dev_info(enic, "MTU changed from %u to %u\n",  old_mtu, new_mtu);
1089         return 0;
1090 }
1091
1092 static int enic_dev_init(struct enic *enic)
1093 {
1094         int err;
1095         struct rte_eth_dev *eth_dev = enic->rte_dev;
1096
1097         vnic_dev_intr_coal_timer_info_default(enic->vdev);
1098
1099         /* Get vNIC configuration
1100         */
1101         err = enic_get_vnic_config(enic);
1102         if (err) {
1103                 dev_err(dev, "Get vNIC configuration failed, aborting\n");
1104                 return err;
1105         }
1106
1107         eth_dev->data->mac_addrs = rte_zmalloc("enic_mac_addr", ETH_ALEN, 0);
1108         if (!eth_dev->data->mac_addrs) {
1109                 dev_err(enic, "mac addr storage alloc failed, aborting.\n");
1110                 return -1;
1111         }
1112         ether_addr_copy((struct ether_addr *) enic->mac_addr,
1113                 &eth_dev->data->mac_addrs[0]);
1114
1115
1116         /* Get available resource counts
1117         */
1118         enic_get_res_counts(enic);
1119
1120         vnic_dev_set_reset_flag(enic->vdev, 0);
1121
1122         return 0;
1123
1124 }
1125
1126 int enic_probe(struct enic *enic)
1127 {
1128         struct rte_pci_device *pdev = enic->pdev;
1129         int err = -1;
1130
1131         dev_debug(enic, " Initializing ENIC PMD version %s\n", DRV_VERSION);
1132
1133         enic->bar0.vaddr = (void *)pdev->mem_resource[0].addr;
1134         enic->bar0.len = pdev->mem_resource[0].len;
1135
1136         /* Register vNIC device */
1137         enic->vdev = vnic_dev_register(NULL, enic, enic->pdev, &enic->bar0, 1);
1138         if (!enic->vdev) {
1139                 dev_err(enic, "vNIC registration failed, aborting\n");
1140                 goto err_out;
1141         }
1142
1143         LIST_INIT(&enic->memzone_list);
1144         rte_spinlock_init(&enic->memzone_list_lock);
1145
1146         vnic_register_cbacks(enic->vdev,
1147                 enic_alloc_consistent,
1148                 enic_free_consistent);
1149
1150         /* Issue device open to get device in known state */
1151         err = enic_dev_open(enic);
1152         if (err) {
1153                 dev_err(enic, "vNIC dev open failed, aborting\n");
1154                 goto err_out_unregister;
1155         }
1156
1157         /* Set ingress vlan rewrite mode before vnic initialization */
1158         err = vnic_dev_set_ig_vlan_rewrite_mode(enic->vdev,
1159                 IG_VLAN_REWRITE_MODE_PASS_THRU);
1160         if (err) {
1161                 dev_err(enic,
1162                         "Failed to set ingress vlan rewrite mode, aborting.\n");
1163                 goto err_out_dev_close;
1164         }
1165
1166         /* Issue device init to initialize the vnic-to-switch link.
1167          * We'll start with carrier off and wait for link UP
1168          * notification later to turn on carrier.  We don't need
1169          * to wait here for the vnic-to-switch link initialization
1170          * to complete; link UP notification is the indication that
1171          * the process is complete.
1172          */
1173
1174         err = vnic_dev_init(enic->vdev, 0);
1175         if (err) {
1176                 dev_err(enic, "vNIC dev init failed, aborting\n");
1177                 goto err_out_dev_close;
1178         }
1179
1180         err = enic_dev_init(enic);
1181         if (err) {
1182                 dev_err(enic, "Device initialization failed, aborting\n");
1183                 goto err_out_dev_close;
1184         }
1185
1186         return 0;
1187
1188 err_out_dev_close:
1189         vnic_dev_close(enic->vdev);
1190 err_out_unregister:
1191         vnic_dev_unregister(enic->vdev);
1192 err_out:
1193         return err;
1194 }
1195
1196 void enic_remove(struct enic *enic)
1197 {
1198         enic_dev_deinit(enic);
1199         vnic_dev_close(enic->vdev);
1200         vnic_dev_unregister(enic->vdev);
1201 }