drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c

   1 /*******************************************************************************
   2
   3   Intel 82599 Virtual Function driver
   4   Copyright(c) 1999 - 2015 Intel Corporation.
   5
   6   This program is free software; you can redistribute it and/or modify it
   7   under the terms and conditions of the GNU General Public License,
   8   version 2, as published by the Free Software Foundation.
   9
  10   This program is distributed in the hope it will be useful, but WITHOUT
  11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13   more details.
  14
  15   You should have received a copy of the GNU General Public License along with
  16   this program; if not, see <http://www.gnu.org/licenses/>.
  17
  18   The full GNU General Public License is included in this distribution in
  19   the file called "COPYING".
  20
  21   Contact Information:
  22   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  23   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  24
  25 *******************************************************************************/
  26
  27 /******************************************************************************
  28  Copyright (c)2006 - 2007 Myricom, Inc. for some LRO specific code
  29 ******************************************************************************/
  30
  31 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  32
  33 #include <linux/types.h>
  34 #include <linux/bitops.h>
  35 #include <linux/module.h>
  36 #include <linux/pci.h>
  37 #include <linux/netdevice.h>
  38 #include <linux/vmalloc.h>
  39 #include <linux/string.h>
  40 #include <linux/in.h>
  41 #include <linux/ip.h>
  42 #include <linux/tcp.h>
  43 #include <linux/sctp.h>
  44 #include <linux/ipv6.h>
  45 #include <linux/slab.h>
  46 #include <net/checksum.h>
  47 #include <net/ip6_checksum.h>
  48 #include <linux/ethtool.h>
  49 #include <linux/if.h>
  50 #include <linux/if_vlan.h>
  51 #include <linux/prefetch.h>
  52
  53 #include "ixgbevf.h"
  54
  55 const char ixgbevf_driver_name[] = "ixgbevf";
  56 static const char ixgbevf_driver_string[] =
  57         "Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
  58
  59 #define DRV_VERSION "2.12.1-k"
  60 const char ixgbevf_driver_version[] = DRV_VERSION;
  61 static char ixgbevf_copyright[] =
  62         "Copyright (c) 2009 - 2012 Intel Corporation.";
  63
  64 static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
  65         [board_82599_vf] = &ixgbevf_82599_vf_info,
  66         [board_X540_vf]  = &ixgbevf_X540_vf_info,
  67         [board_X550_vf]  = &ixgbevf_X550_vf_info,
  68         [board_X550EM_x_vf] = &ixgbevf_X550EM_x_vf_info,
  69 };
  70
  71 /* ixgbevf_pci_tbl - PCI Device ID Table
  72  *
  73  * Wildcard entries (PCI_ANY_ID) should come last
  74  * Last entry must be all 0s
  75  *
  76  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
  77  *   Class, Class Mask, private data (not used) }
  78  */
  79 static const struct pci_device_id ixgbevf_pci_tbl[] = {
  80         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF), board_82599_vf },
  81         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF), board_X540_vf },
  82         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF), board_X550_vf },
  83         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF), board_X550EM_x_vf },
  84         /* required last entry */
  85         {0, }
  86 };
  87 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
  88
  89 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
  90 MODULE_DESCRIPTION("Intel(R) 10 Gigabit Virtual Function Network Driver");
  91 MODULE_LICENSE("GPL");
  92 MODULE_VERSION(DRV_VERSION);
  93
  94 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
  95 static int debug = -1;
  96 module_param(debug, int, 0);
  97 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  98
  99 static void ixgbevf_service_event_schedule(struct ixgbevf_adapter *adapter)
 100 {
 101         if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
 102             !test_bit(__IXGBEVF_REMOVING, &adapter->state) &&
 103             !test_and_set_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state))
 104                 schedule_work(&adapter->service_task);
 105 }
 106
 107 static void ixgbevf_service_event_complete(struct ixgbevf_adapter *adapter)
 108 {
 109         BUG_ON(!test_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state));
 110
 111         /* flush memory to make sure state is correct before next watchdog */
 112         smp_mb__before_atomic();
 113         clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
 114 }
 115
 116 /* forward decls */
 117 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter);
 118 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
 119 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
 120
 121 static void ixgbevf_remove_adapter(struct ixgbe_hw *hw)
 122 {
 123         struct ixgbevf_adapter *adapter = hw->back;
 124
 125         if (!hw->hw_addr)
 126                 return;
 127         hw->hw_addr = NULL;
 128         dev_err(&adapter->pdev->dev, "Adapter removed\n");
 129         if (test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
 130                 ixgbevf_service_event_schedule(adapter);
 131 }
 132
 133 static void ixgbevf_check_remove(struct ixgbe_hw *hw, u32 reg)
 134 {
 135         u32 value;
 136
 137         /* The following check not only optimizes a bit by not
 138          * performing a read on the status register when the
 139          * register just read was a status register read that
 140          * returned IXGBE_FAILED_READ_REG. It also blocks any
 141          * potential recursion.
 142          */
 143         if (reg == IXGBE_VFSTATUS) {
 144                 ixgbevf_remove_adapter(hw);
 145                 return;
 146         }
 147         value = ixgbevf_read_reg(hw, IXGBE_VFSTATUS);
 148         if (value == IXGBE_FAILED_READ_REG)
 149                 ixgbevf_remove_adapter(hw);
 150 }
 151
 152 u32 ixgbevf_read_reg(struct ixgbe_hw *hw, u32 reg)
 153 {
 154         u8 __iomem *reg_addr = ACCESS_ONCE(hw->hw_addr);
 155         u32 value;
 156
 157         if (IXGBE_REMOVED(reg_addr))
 158                 return IXGBE_FAILED_READ_REG;
 159         value = readl(reg_addr + reg);
 160         if (unlikely(value == IXGBE_FAILED_READ_REG))
 161                 ixgbevf_check_remove(hw, reg);
 162         return value;
 163 }
 164
 165 /**
 166  * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
 167  * @adapter: pointer to adapter struct
 168  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
 169  * @queue: queue to map the corresponding interrupt to
 170  * @msix_vector: the vector to map to the corresponding queue
 171  **/
 172 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
 173                              u8 queue, u8 msix_vector)
 174 {
 175         u32 ivar, index;
 176         struct ixgbe_hw *hw = &adapter->hw;
 177
 178         if (direction == -1) {
 179                 /* other causes */
 180                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 181                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
 182                 ivar &= ~0xFF;
 183                 ivar |= msix_vector;
 184                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
 185         } else {
 186                 /* Tx or Rx causes */
 187                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 188                 index = ((16 * (queue & 1)) + (8 * direction));
 189                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
 190                 ivar &= ~(0xFF << index);
 191                 ivar |= (msix_vector << index);
 192                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
 193         }
 194 }
 195
 196 static void ixgbevf_unmap_and_free_tx_resource(struct ixgbevf_ring *tx_ring,
 197                                         struct ixgbevf_tx_buffer *tx_buffer)
 198 {
 199         if (tx_buffer->skb) {
 200                 dev_kfree_skb_any(tx_buffer->skb);
 201                 if (dma_unmap_len(tx_buffer, len))
 202                         dma_unmap_single(tx_ring->dev,
 203                                          dma_unmap_addr(tx_buffer, dma),
 204                                          dma_unmap_len(tx_buffer, len),
 205                                          DMA_TO_DEVICE);
 206         } else if (dma_unmap_len(tx_buffer, len)) {
 207                 dma_unmap_page(tx_ring->dev,
 208                                dma_unmap_addr(tx_buffer, dma),
 209                                dma_unmap_len(tx_buffer, len),
 210                                DMA_TO_DEVICE);
 211         }
 212         tx_buffer->next_to_watch = NULL;
 213         tx_buffer->skb = NULL;
 214         dma_unmap_len_set(tx_buffer, len, 0);
 215         /* tx_buffer must be completely set up in the transmit path */
 216 }
 217
 218 static u64 ixgbevf_get_tx_completed(struct ixgbevf_ring *ring)
 219 {
 220         return ring->stats.packets;
 221 }
 222
 223 static u32 ixgbevf_get_tx_pending(struct ixgbevf_ring *ring)
 224 {
 225         struct ixgbevf_adapter *adapter = netdev_priv(ring->netdev);
 226         struct ixgbe_hw *hw = &adapter->hw;
 227
 228         u32 head = IXGBE_READ_REG(hw, IXGBE_VFTDH(ring->reg_idx));
 229         u32 tail = IXGBE_READ_REG(hw, IXGBE_VFTDT(ring->reg_idx));
 230
 231         if (head != tail)
 232                 return (head < tail) ?
 233                         tail - head : (tail + ring->count - head);
 234
 235         return 0;
 236 }
 237
 238 static inline bool ixgbevf_check_tx_hang(struct ixgbevf_ring *tx_ring)
 239 {
 240         u32 tx_done = ixgbevf_get_tx_completed(tx_ring);
 241         u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
 242         u32 tx_pending = ixgbevf_get_tx_pending(tx_ring);
 243
 244         clear_check_for_tx_hang(tx_ring);
 245
 246         /* Check for a hung queue, but be thorough. This verifies
 247          * that a transmit has been completed since the previous
 248          * check AND there is at least one packet pending. The
 249          * ARMED bit is set to indicate a potential hang.
 250          */
 251         if ((tx_done_old == tx_done) && tx_pending) {
 252                 /* make sure it is true for two checks in a row */
 253                 return test_and_set_bit(__IXGBEVF_HANG_CHECK_ARMED,
 254                                         &tx_ring->state);
 255         }
 256         /* reset the countdown */
 257         clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &tx_ring->state);
 258
 259         /* update completed stats and continue */
 260         tx_ring->tx_stats.tx_done_old = tx_done;
 261
 262         return false;
 263 }
 264
 265 static void ixgbevf_tx_timeout_reset(struct ixgbevf_adapter *adapter)
 266 {
 267         /* Do the reset outside of interrupt context */
 268         if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
 269                 adapter->flags |= IXGBEVF_FLAG_RESET_REQUESTED;
 270                 ixgbevf_service_event_schedule(adapter);
 271         }
 272 }
 273
 274 /**
 275  * ixgbevf_tx_timeout - Respond to a Tx Hang
 276  * @netdev: network interface device structure
 277  **/
 278 static void ixgbevf_tx_timeout(struct net_device *netdev)
 279 {
 280         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
 281
 282         ixgbevf_tx_timeout_reset(adapter);
 283 }
 284
 285 /**
 286  * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
 287  * @q_vector: board private structure
 288  * @tx_ring: tx ring to clean
 289  **/
 290 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
 291                                  struct ixgbevf_ring *tx_ring)
 292 {
 293         struct ixgbevf_adapter *adapter = q_vector->adapter;
 294         struct ixgbevf_tx_buffer *tx_buffer;
 295         union ixgbe_adv_tx_desc *tx_desc;
 296         unsigned int total_bytes = 0, total_packets = 0;
 297         unsigned int budget = tx_ring->count / 2;
 298         unsigned int i = tx_ring->next_to_clean;
 299
 300         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
 301                 return true;
 302
 303         tx_buffer = &tx_ring->tx_buffer_info[i];
 304         tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
 305         i -= tx_ring->count;
 306
 307         do {
 308                 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
 309
 310                 /* if next_to_watch is not set then there is no work pending */
 311                 if (!eop_desc)
 312                         break;
 313
 314                 /* prevent any other reads prior to eop_desc */
 315                 smp_rmb();
 316
 317                 /* if DD is not set pending work has not been completed */
 318                 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
 319                         break;
 320
 321                 /* clear next_to_watch to prevent false hangs */
 322                 tx_buffer->next_to_watch = NULL;
 323
 324                 /* update the statistics for this packet */
 325                 total_bytes += tx_buffer->bytecount;
 326                 total_packets += tx_buffer->gso_segs;
 327
 328                 /* free the skb */
 329                 dev_kfree_skb_any(tx_buffer->skb);
 330
 331                 /* unmap skb header data */
 332                 dma_unmap_single(tx_ring->dev,
 333                                  dma_unmap_addr(tx_buffer, dma),
 334                                  dma_unmap_len(tx_buffer, len),
 335                                  DMA_TO_DEVICE);
 336
 337                 /* clear tx_buffer data */
 338                 tx_buffer->skb = NULL;
 339                 dma_unmap_len_set(tx_buffer, len, 0);
 340
 341                 /* unmap remaining buffers */
 342                 while (tx_desc != eop_desc) {
 343                         tx_buffer++;
 344                         tx_desc++;
 345                         i++;
 346                         if (unlikely(!i)) {
 347                                 i -= tx_ring->count;
 348                                 tx_buffer = tx_ring->tx_buffer_info;
 349                                 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
 350                         }
 351
 352                         /* unmap any remaining paged data */
 353                         if (dma_unmap_len(tx_buffer, len)) {
 354                                 dma_unmap_page(tx_ring->dev,
 355                                                dma_unmap_addr(tx_buffer, dma),
 356                                                dma_unmap_len(tx_buffer, len),
 357                                                DMA_TO_DEVICE);
 358                                 dma_unmap_len_set(tx_buffer, len, 0);
 359                         }
 360                 }
 361
 362                 /* move us one more past the eop_desc for start of next pkt */
 363                 tx_buffer++;
 364                 tx_desc++;
 365                 i++;
 366                 if (unlikely(!i)) {
 367                         i -= tx_ring->count;
 368                         tx_buffer = tx_ring->tx_buffer_info;
 369                         tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
 370                 }
 371
 372                 /* issue prefetch for next Tx descriptor */
 373                 prefetch(tx_desc);
 374
 375                 /* update budget accounting */
 376                 budget--;
 377         } while (likely(budget));
 378
 379         i += tx_ring->count;
 380         tx_ring->next_to_clean = i;
 381         u64_stats_update_begin(&tx_ring->syncp);
 382         tx_ring->stats.bytes += total_bytes;
 383         tx_ring->stats.packets += total_packets;
 384         u64_stats_update_end(&tx_ring->syncp);
 385         q_vector->tx.total_bytes += total_bytes;
 386         q_vector->tx.total_packets += total_packets;
 387
 388         if (check_for_tx_hang(tx_ring) && ixgbevf_check_tx_hang(tx_ring)) {
 389                 struct ixgbe_hw *hw = &adapter->hw;
 390                 union ixgbe_adv_tx_desc *eop_desc;
 391
 392                 eop_desc = tx_ring->tx_buffer_info[i].next_to_watch;
 393
 394                 pr_err("Detected Tx Unit Hang\n"
 395                        "  Tx Queue             <%d>\n"
 396                        "  TDH, TDT             <%x>, <%x>\n"
 397                        "  next_to_use          <%x>\n"
 398                        "  next_to_clean        <%x>\n"
 399                        "tx_buffer_info[next_to_clean]\n"
 400                        "  next_to_watch        <%p>\n"
 401                        "  eop_desc->wb.status  <%x>\n"
 402                        "  time_stamp           <%lx>\n"
 403                        "  jiffies              <%lx>\n",
 404                        tx_ring->queue_index,
 405                        IXGBE_READ_REG(hw, IXGBE_VFTDH(tx_ring->reg_idx)),
 406                        IXGBE_READ_REG(hw, IXGBE_VFTDT(tx_ring->reg_idx)),
 407                        tx_ring->next_to_use, i,
 408                        eop_desc, (eop_desc ? eop_desc->wb.status : 0),
 409                        tx_ring->tx_buffer_info[i].time_stamp, jiffies);
 410
 411                 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
 412
 413                 /* schedule immediate reset if we believe we hung */
 414                 ixgbevf_tx_timeout_reset(adapter);
 415
 416                 return true;
 417         }
 418
 419 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
 420         if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
 421                      (ixgbevf_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
 422                 /* Make sure that anybody stopping the queue after this
 423                  * sees the new next_to_clean.
 424                  */
 425                 smp_mb();
 426
 427                 if (__netif_subqueue_stopped(tx_ring->netdev,
 428                                              tx_ring->queue_index) &&
 429                     !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
 430                         netif_wake_subqueue(tx_ring->netdev,
 431                                             tx_ring->queue_index);
 432                         ++tx_ring->tx_stats.restart_queue;
 433                 }
 434         }
 435
 436         return !!budget;
 437 }
 438
 439 /**
 440  * ixgbevf_rx_skb - Helper function to determine proper Rx method
 441  * @q_vector: structure containing interrupt and ring information
 442  * @skb: packet to send up
 443  **/
 444 static void ixgbevf_rx_skb(struct ixgbevf_q_vector *q_vector,
 445                            struct sk_buff *skb)
 446 {
 447 #ifdef CONFIG_NET_RX_BUSY_POLL
 448         skb_mark_napi_id(skb, &q_vector->napi);
 449
 450         if (ixgbevf_qv_busy_polling(q_vector)) {
 451                 netif_receive_skb(skb);
 452                 /* exit early if we busy polled */
 453                 return;
 454         }
 455 #endif /* CONFIG_NET_RX_BUSY_POLL */
 456
 457         napi_gro_receive(&q_vector->napi, skb);
 458 }
 459
 460 #define IXGBE_RSS_L4_TYPES_MASK \
 461         ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
 462          (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
 463          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
 464          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
 465
 466 static inline void ixgbevf_rx_hash(struct ixgbevf_ring *ring,
 467                                    union ixgbe_adv_rx_desc *rx_desc,
 468                                    struct sk_buff *skb)
 469 {
 470         u16 rss_type;
 471
 472         if (!(ring->netdev->features & NETIF_F_RXHASH))
 473                 return;
 474
 475         rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
 476                    IXGBE_RXDADV_RSSTYPE_MASK;
 477
 478         if (!rss_type)
 479                 return;
 480
 481         skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
 482                      (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
 483                      PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
 484 }
 485
 486 /**
 487  * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
 488  * @ring: structure containig ring specific data
 489  * @rx_desc: current Rx descriptor being processed
 490  * @skb: skb currently being received and modified
 491  **/
 492 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
 493                                        union ixgbe_adv_rx_desc *rx_desc,
 494                                        struct sk_buff *skb)
 495 {
 496         skb_checksum_none_assert(skb);
 497
 498         /* Rx csum disabled */
 499         if (!(ring->netdev->features & NETIF_F_RXCSUM))
 500                 return;
 501
 502         /* if IP and error */
 503         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
 504             ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
 505                 ring->rx_stats.csum_err++;
 506                 return;
 507         }
 508
 509         if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
 510                 return;
 511
 512         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
 513                 ring->rx_stats.csum_err++;
 514                 return;
 515         }
 516
 517         /* It must be a TCP or UDP packet with a valid checksum */
 518         skb->ip_summed = CHECKSUM_UNNECESSARY;
 519 }
 520
 521 /**
 522  * ixgbevf_process_skb_fields - Populate skb header fields from Rx descriptor
 523  * @rx_ring: rx descriptor ring packet is being transacted on
 524  * @rx_desc: pointer to the EOP Rx descriptor
 525  * @skb: pointer to current skb being populated
 526  *
 527  * This function checks the ring, descriptor, and packet information in
 528  * order to populate the checksum, VLAN, protocol, and other fields within
 529  * the skb.
 530  **/
 531 static void ixgbevf_process_skb_fields(struct ixgbevf_ring *rx_ring,
 532                                        union ixgbe_adv_rx_desc *rx_desc,
 533                                        struct sk_buff *skb)
 534 {
 535         ixgbevf_rx_hash(rx_ring, rx_desc, skb);
 536         ixgbevf_rx_checksum(rx_ring, rx_desc, skb);
 537
 538         if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
 539                 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
 540                 unsigned long *active_vlans = netdev_priv(rx_ring->netdev);
 541
 542                 if (test_bit(vid & VLAN_VID_MASK, active_vlans))
 543                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
 544         }
 545
 546         skb->protocol = eth_type_trans(skb, rx_ring->netdev);
 547 }
 548
 549 /**
 550  * ixgbevf_is_non_eop - process handling of non-EOP buffers
 551  * @rx_ring: Rx ring being processed
 552  * @rx_desc: Rx descriptor for current buffer
 553  * @skb: current socket buffer containing buffer in progress
 554  *
 555  * This function updates next to clean.  If the buffer is an EOP buffer
 556  * this function exits returning false, otherwise it will place the
 557  * sk_buff in the next buffer to be chained and return true indicating
 558  * that this is in fact a non-EOP buffer.
 559  **/
 560 static bool ixgbevf_is_non_eop(struct ixgbevf_ring *rx_ring,
 561                                union ixgbe_adv_rx_desc *rx_desc)
 562 {
 563         u32 ntc = rx_ring->next_to_clean + 1;
 564
 565         /* fetch, update, and store next to clean */
 566         ntc = (ntc < rx_ring->count) ? ntc : 0;
 567         rx_ring->next_to_clean = ntc;
 568
 569         prefetch(IXGBEVF_RX_DESC(rx_ring, ntc));
 570
 571         if (likely(ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
 572                 return false;
 573
 574         return true;
 575 }
 576
 577 static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring,
 578                                       struct ixgbevf_rx_buffer *bi)
 579 {
 580         struct page *page = bi->page;
 581         dma_addr_t dma = bi->dma;
 582
 583         /* since we are recycling buffers we should seldom need to alloc */
 584         if (likely(page))
 585                 return true;
 586
 587         /* alloc new page for storage */
 588         page = dev_alloc_page();
 589         if (unlikely(!page)) {
 590                 rx_ring->rx_stats.alloc_rx_page_failed++;
 591                 return false;
 592         }
 593
 594         /* map page for use */
 595         dma = dma_map_page(rx_ring->dev, page, 0,
 596                            PAGE_SIZE, DMA_FROM_DEVICE);
 597
 598         /* if mapping failed free memory back to system since
 599          * there isn't much point in holding memory we can't use
 600          */
 601         if (dma_mapping_error(rx_ring->dev, dma)) {
 602                 __free_page(page);
 603
 604                 rx_ring->rx_stats.alloc_rx_buff_failed++;
 605                 return false;
 606         }
 607
 608         bi->dma = dma;
 609         bi->page = page;
 610         bi->page_offset = 0;
 611
 612         return true;
 613 }
 614
 615 /**
 616  * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
 617  * @rx_ring: rx descriptor ring (for a specific queue) to setup buffers on
 618  * @cleaned_count: number of buffers to replace
 619  **/
 620 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring *rx_ring,
 621                                      u16 cleaned_count)
 622 {
 623         union ixgbe_adv_rx_desc *rx_desc;
 624         struct ixgbevf_rx_buffer *bi;
 625         unsigned int i = rx_ring->next_to_use;
 626
 627         /* nothing to do or no valid netdev defined */
 628         if (!cleaned_count || !rx_ring->netdev)
 629                 return;
 630
 631         rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
 632         bi = &rx_ring->rx_buffer_info[i];
 633         i -= rx_ring->count;
 634
 635         do {
 636                 if (!ixgbevf_alloc_mapped_page(rx_ring, bi))
 637                         break;
 638
 639                 /* Refresh the desc even if pkt_addr didn't change
 640                  * because each write-back erases this info.
 641                  */
 642                 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
 643
 644                 rx_desc++;
 645                 bi++;
 646                 i++;
 647                 if (unlikely(!i)) {
 648                         rx_desc = IXGBEVF_RX_DESC(rx_ring, 0);
 649                         bi = rx_ring->rx_buffer_info;
 650                         i -= rx_ring->count;
 651                 }
 652
 653                 /* clear the hdr_addr for the next_to_use descriptor */
 654                 rx_desc->read.hdr_addr = 0;
 655
 656                 cleaned_count--;
 657         } while (cleaned_count);
 658
 659         i += rx_ring->count;
 660
 661         if (rx_ring->next_to_use != i) {
 662                 /* record the next descriptor to use */
 663                 rx_ring->next_to_use = i;
 664
 665                 /* update next to alloc since we have filled the ring */
 666                 rx_ring->next_to_alloc = i;
 667
 668                 /* Force memory writes to complete before letting h/w
 669                  * know there are new descriptors to fetch.  (Only
 670                  * applicable for weak-ordered memory model archs,
 671                  * such as IA-64).
 672                  */
 673                 wmb();
 674                 ixgbevf_write_tail(rx_ring, i);
 675         }
 676 }
 677
 678 /**
 679  * ixgbevf_cleanup_headers - Correct corrupted or empty headers
 680  * @rx_ring: rx descriptor ring packet is being transacted on
 681  * @rx_desc: pointer to the EOP Rx descriptor
 682  * @skb: pointer to current skb being fixed
 683  *
 684  * Check for corrupted packet headers caused by senders on the local L2
 685  * embedded NIC switch not setting up their Tx Descriptors right.  These
 686  * should be very rare.
 687  *
 688  * Also address the case where we are pulling data in on pages only
 689  * and as such no data is present in the skb header.
 690  *
 691  * In addition if skb is not at least 60 bytes we need to pad it so that
 692  * it is large enough to qualify as a valid Ethernet frame.
 693  *
 694  * Returns true if an error was encountered and skb was freed.
 695  **/
 696 static bool ixgbevf_cleanup_headers(struct ixgbevf_ring *rx_ring,
 697                                     union ixgbe_adv_rx_desc *rx_desc,
 698                                     struct sk_buff *skb)
 699 {
 700         /* verify that the packet does not have any known errors */
 701         if (unlikely(ixgbevf_test_staterr(rx_desc,
 702                                           IXGBE_RXDADV_ERR_FRAME_ERR_MASK))) {
 703                 struct net_device *netdev = rx_ring->netdev;
 704
 705                 if (!(netdev->features & NETIF_F_RXALL)) {
 706                         dev_kfree_skb_any(skb);
 707                         return true;
 708                 }
 709         }
 710
 711         /* if eth_skb_pad returns an error the skb was freed */
 712         if (eth_skb_pad(skb))
 713                 return true;
 714
 715         return false;
 716 }
 717
 718 /**
 719  * ixgbevf_reuse_rx_page - page flip buffer and store it back on the ring
 720  * @rx_ring: rx descriptor ring to store buffers on
 721  * @old_buff: donor buffer to have page reused
 722  *
 723  * Synchronizes page for reuse by the adapter
 724  **/
 725 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
 726                                   struct ixgbevf_rx_buffer *old_buff)
 727 {
 728         struct ixgbevf_rx_buffer *new_buff;
 729         u16 nta = rx_ring->next_to_alloc;
 730
 731         new_buff = &rx_ring->rx_buffer_info[nta];
 732
 733         /* update, and store next to alloc */
 734         nta++;
 735         rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
 736
 737         /* transfer page from old buffer to new buffer */
 738         new_buff->page = old_buff->page;
 739         new_buff->dma = old_buff->dma;
 740         new_buff->page_offset = old_buff->page_offset;
 741
 742         /* sync the buffer for use by the device */
 743         dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
 744                                          new_buff->page_offset,
 745                                          IXGBEVF_RX_BUFSZ,
 746                                          DMA_FROM_DEVICE);
 747 }
 748
 749 static inline bool ixgbevf_page_is_reserved(struct page *page)
 750 {
 751         return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
 752 }
 753
 754 /**
 755  * ixgbevf_add_rx_frag - Add contents of Rx buffer to sk_buff
 756  * @rx_ring: rx descriptor ring to transact packets on
 757  * @rx_buffer: buffer containing page to add
 758  * @rx_desc: descriptor containing length of buffer written by hardware
 759  * @skb: sk_buff to place the data into
 760  *
 761  * This function will add the data contained in rx_buffer->page to the skb.
 762  * This is done either through a direct copy if the data in the buffer is
 763  * less than the skb header size, otherwise it will just attach the page as
 764  * a frag to the skb.
 765  *
 766  * The function will then update the page offset if necessary and return
 767  * true if the buffer can be reused by the adapter.
 768  **/
 769 static bool ixgbevf_add_rx_frag(struct ixgbevf_ring *rx_ring,
 770                                 struct ixgbevf_rx_buffer *rx_buffer,
 771                                 union ixgbe_adv_rx_desc *rx_desc,
 772                                 struct sk_buff *skb)
 773 {
 774         struct page *page = rx_buffer->page;
 775         unsigned char *va = page_address(page) + rx_buffer->page_offset;
 776         unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
 777 #if (PAGE_SIZE < 8192)
 778         unsigned int truesize = IXGBEVF_RX_BUFSZ;
 779 #else
 780         unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
 781 #endif
 782         unsigned int pull_len;
 783
 784         if (unlikely(skb_is_nonlinear(skb)))
 785                 goto add_tail_frag;
 786
 787         if (likely(size <= IXGBEVF_RX_HDR_SIZE)) {
 788                 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
 789
 790                 /* page is not reserved, we can reuse buffer as is */
 791                 if (likely(!ixgbevf_page_is_reserved(page)))
 792                         return true;
 793
 794                 /* this page cannot be reused so discard it */
 795                 put_page(page);
 796                 return false;
 797         }
 798
 799         /* we need the header to contain the greater of either ETH_HLEN or
 800          * 60 bytes if the skb->len is less than 60 for skb_pad.
 801          */
 802         pull_len = eth_get_headlen(va, IXGBEVF_RX_HDR_SIZE);
 803
 804         /* align pull length to size of long to optimize memcpy performance */
 805         memcpy(__skb_put(skb, pull_len), va, ALIGN(pull_len, sizeof(long)));
 806
 807         /* update all of the pointers */
 808         va += pull_len;
 809         size -= pull_len;
 810
 811 add_tail_frag:
 812         skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
 813                         (unsigned long)va & ~PAGE_MASK, size, truesize);
 814
 815         /* avoid re-using remote pages */
 816         if (unlikely(ixgbevf_page_is_reserved(page)))
 817                 return false;
 818
 819 #if (PAGE_SIZE < 8192)
 820         /* if we are only owner of page we can reuse it */
 821         if (unlikely(page_count(page) != 1))
 822                 return false;
 823
 824         /* flip page offset to other buffer */
 825         rx_buffer->page_offset ^= IXGBEVF_RX_BUFSZ;
 826
 827 #else
 828         /* move offset up to the next cache line */
 829         rx_buffer->page_offset += truesize;
 830
 831         if (rx_buffer->page_offset > (PAGE_SIZE - IXGBEVF_RX_BUFSZ))
 832                 return false;
 833
 834 #endif
 835         /* Even if we own the page, we are not allowed to use atomic_set()
 836          * This would break get_page_unless_zero() users.
 837          */
 838         atomic_inc(&page->_count);
 839
 840         return true;
 841 }
 842
 843 static struct sk_buff *ixgbevf_fetch_rx_buffer(struct ixgbevf_ring *rx_ring,
 844                                                union ixgbe_adv_rx_desc *rx_desc,
 845                                                struct sk_buff *skb)
 846 {
 847         struct ixgbevf_rx_buffer *rx_buffer;
 848         struct page *page;
 849
 850         rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
 851         page = rx_buffer->page;
 852         prefetchw(page);
 853
 854         if (likely(!skb)) {
 855                 void *page_addr = page_address(page) +
 856                                   rx_buffer->page_offset;
 857
 858                 /* prefetch first cache line of first page */
 859                 prefetch(page_addr);
 860 #if L1_CACHE_BYTES < 128
 861                 prefetch(page_addr + L1_CACHE_BYTES);
 862 #endif
 863
 864                 /* allocate a skb to store the frags */
 865                 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
 866                                                 IXGBEVF_RX_HDR_SIZE);
 867                 if (unlikely(!skb)) {
 868                         rx_ring->rx_stats.alloc_rx_buff_failed++;
 869                         return NULL;
 870                 }
 871
 872                 /* we will be copying header into skb->data in
 873                  * pskb_may_pull so it is in our interest to prefetch
 874                  * it now to avoid a possible cache miss
 875                  */
 876                 prefetchw(skb->data);
 877         }
 878
 879         /* we are reusing so sync this buffer for CPU use */
 880         dma_sync_single_range_for_cpu(rx_ring->dev,
 881                                       rx_buffer->dma,
 882                                       rx_buffer->page_offset,
 883                                       IXGBEVF_RX_BUFSZ,
 884                                       DMA_FROM_DEVICE);
 885
 886         /* pull page into skb */
 887         if (ixgbevf_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
 888                 /* hand second half of page back to the ring */
 889                 ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
 890         } else {
 891                 /* we are not reusing the buffer so unmap it */
 892                 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
 893                                PAGE_SIZE, DMA_FROM_DEVICE);
 894         }
 895
 896         /* clear contents of buffer_info */
 897         rx_buffer->dma = 0;
 898         rx_buffer->page = NULL;
 899
 900         return skb;
 901 }
 902
 903 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
 904                                              u32 qmask)
 905 {
 906         struct ixgbe_hw *hw = &adapter->hw;
 907
 908         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
 909 }
 910
 911 static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
 912                                 struct ixgbevf_ring *rx_ring,
 913                                 int budget)
 914 {
 915         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 916         u16 cleaned_count = ixgbevf_desc_unused(rx_ring);
 917         struct sk_buff *skb = rx_ring->skb;
 918
 919         while (likely(total_rx_packets < budget)) {
 920                 union ixgbe_adv_rx_desc *rx_desc;
 921
 922                 /* return some buffers to hardware, one at a time is too slow */
 923                 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
 924                         ixgbevf_alloc_rx_buffers(rx_ring, cleaned_count);
 925                         cleaned_count = 0;
 926                 }
 927
 928                 rx_desc = IXGBEVF_RX_DESC(rx_ring, rx_ring->next_to_clean);
 929
 930                 if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_DD))
 931                         break;
 932
 933                 /* This memory barrier is needed to keep us from reading
 934                  * any other fields out of the rx_desc until we know the
 935                  * RXD_STAT_DD bit is set
 936                  */
 937                 rmb();
 938
 939                 /* retrieve a buffer from the ring */
 940                 skb = ixgbevf_fetch_rx_buffer(rx_ring, rx_desc, skb);
 941
 942                 /* exit if we failed to retrieve a buffer */
 943                 if (!skb)
 944                         break;
 945
 946                 cleaned_count++;
 947
 948                 /* fetch next buffer in frame if non-eop */
 949                 if (ixgbevf_is_non_eop(rx_ring, rx_desc))
 950                         continue;
 951
 952                 /* verify the packet layout is correct */
 953                 if (ixgbevf_cleanup_headers(rx_ring, rx_desc, skb)) {
 954                         skb = NULL;
 955                         continue;
 956                 }
 957
 958                 /* probably a little skewed due to removing CRC */
 959                 total_rx_bytes += skb->len;
 960
 961                 /* Workaround hardware that can't do proper VEPA multicast
 962                  * source pruning.
 963                  */
 964                 if ((skb->pkt_type == PACKET_BROADCAST ||
 965                      skb->pkt_type == PACKET_MULTICAST) &&
 966                     ether_addr_equal(rx_ring->netdev->dev_addr,
 967                                      eth_hdr(skb)->h_source)) {
 968                         dev_kfree_skb_irq(skb);
 969                         continue;
 970                 }
 971
 972                 /* populate checksum, VLAN, and protocol */
 973                 ixgbevf_process_skb_fields(rx_ring, rx_desc, skb);
 974
 975                 ixgbevf_rx_skb(q_vector, skb);
 976
 977                 /* reset skb pointer */
 978                 skb = NULL;
 979
 980                 /* update budget accounting */
 981                 total_rx_packets++;
 982         }
 983
 984         /* place incomplete frames back on ring for completion */
 985         rx_ring->skb = skb;
 986
 987         u64_stats_update_begin(&rx_ring->syncp);
 988         rx_ring->stats.packets += total_rx_packets;
 989         rx_ring->stats.bytes += total_rx_bytes;
 990         u64_stats_update_end(&rx_ring->syncp);
 991         q_vector->rx.total_packets += total_rx_packets;
 992         q_vector->rx.total_bytes += total_rx_bytes;
 993
 994         return total_rx_packets;
 995 }
 996
 997 /**
 998  * ixgbevf_poll - NAPI polling calback
 999  * @napi: napi struct with our devices info in it
1000  * @budget: amount of work driver is allowed to do this pass, in packets
1001  *
1002  * This function will clean more than one or more rings associated with a
1003  * q_vector.
1004  **/
1005 static int ixgbevf_poll(struct napi_struct *napi, int budget)
1006 {
1007         struct ixgbevf_q_vector *q_vector =
1008                 container_of(napi, struct ixgbevf_q_vector, napi);
1009         struct ixgbevf_adapter *adapter = q_vector->adapter;
1010         struct ixgbevf_ring *ring;
1011         int per_ring_budget, work_done = 0;
1012         bool clean_complete = true;
1013
1014         ixgbevf_for_each_ring(ring, q_vector->tx)
1015                 clean_complete &= ixgbevf_clean_tx_irq(q_vector, ring);
1016
1017         if (budget <= 0)
1018                 return budget;
1019 #ifdef CONFIG_NET_RX_BUSY_POLL
1020         if (!ixgbevf_qv_lock_napi(q_vector))
1021                 return budget;
1022 #endif
1023
1024         /* attempt to distribute budget to each queue fairly, but don't allow
1025          * the budget to go below 1 because we'll exit polling
1026          */
1027         if (q_vector->rx.count > 1)
1028                 per_ring_budget = max(budget/q_vector->rx.count, 1);
1029         else
1030                 per_ring_budget = budget;
1031
1032         ixgbevf_for_each_ring(ring, q_vector->rx) {
1033                 int cleaned = ixgbevf_clean_rx_irq(q_vector, ring,
1034                                                    per_ring_budget);
1035                 work_done += cleaned;
1036                 clean_complete &= (cleaned < per_ring_budget);
1037         }
1038
1039 #ifdef CONFIG_NET_RX_BUSY_POLL
1040         ixgbevf_qv_unlock_napi(q_vector);
1041 #endif
1042
1043         /* If all work not completed, return budget and keep polling */
1044         if (!clean_complete)
1045                 return budget;
1046         /* all work done, exit the polling mode */
1047         napi_complete_done(napi, work_done);
1048         if (adapter->rx_itr_setting & 1)
1049                 ixgbevf_set_itr(q_vector);
1050         if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
1051             !test_bit(__IXGBEVF_REMOVING, &adapter->state))
1052                 ixgbevf_irq_enable_queues(adapter,
1053                                           1 << q_vector->v_idx);
1054
1055         return 0;
1056 }
1057
1058 /**
1059  * ixgbevf_write_eitr - write VTEITR register in hardware specific way
1060  * @q_vector: structure containing interrupt and ring information
1061  **/
1062 void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
1063 {
1064         struct ixgbevf_adapter *adapter = q_vector->adapter;
1065         struct ixgbe_hw *hw = &adapter->hw;
1066         int v_idx = q_vector->v_idx;
1067         u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1068
1069         /* set the WDIS bit to not clear the timer bits and cause an
1070          * immediate assertion of the interrupt
1071          */
1072         itr_reg |= IXGBE_EITR_CNT_WDIS;
1073
1074         IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
1075 }
1076
1077 #ifdef CONFIG_NET_RX_BUSY_POLL
1078 /* must be called with local_bh_disable()d */
1079 static int ixgbevf_busy_poll_recv(struct napi_struct *napi)
1080 {
1081         struct ixgbevf_q_vector *q_vector =
1082                         container_of(napi, struct ixgbevf_q_vector, napi);
1083         struct ixgbevf_adapter *adapter = q_vector->adapter;
1084         struct ixgbevf_ring  *ring;
1085         int found = 0;
1086
1087         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
1088                 return LL_FLUSH_FAILED;
1089
1090         if (!ixgbevf_qv_lock_poll(q_vector))
1091                 return LL_FLUSH_BUSY;
1092
1093         ixgbevf_for_each_ring(ring, q_vector->rx) {
1094                 found = ixgbevf_clean_rx_irq(q_vector, ring, 4);
1095 #ifdef BP_EXTENDED_STATS
1096                 if (found)
1097                         ring->stats.cleaned += found;
1098                 else
1099                         ring->stats.misses++;
1100 #endif
1101                 if (found)
1102                         break;
1103         }
1104
1105         ixgbevf_qv_unlock_poll(q_vector);
1106
1107         return found;
1108 }
1109 #endif /* CONFIG_NET_RX_BUSY_POLL */
1110
1111 /**
1112  * ixgbevf_configure_msix - Configure MSI-X hardware
1113  * @adapter: board private structure
1114  *
1115  * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
1116  * interrupts.
1117  **/
1118 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
1119 {
1120         struct ixgbevf_q_vector *q_vector;
1121         int q_vectors, v_idx;
1122
1123         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1124         adapter->eims_enable_mask = 0;
1125
1126         /* Populate the IVAR table and set the ITR values to the
1127          * corresponding register.
1128          */
1129         for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1130                 struct ixgbevf_ring *ring;
1131
1132                 q_vector = adapter->q_vector[v_idx];
1133
1134                 ixgbevf_for_each_ring(ring, q_vector->rx)
1135                         ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1136
1137                 ixgbevf_for_each_ring(ring, q_vector->tx)
1138                         ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1139
1140                 if (q_vector->tx.ring && !q_vector->rx.ring) {
1141                         /* Tx only vector */
1142                         if (adapter->tx_itr_setting == 1)
1143                                 q_vector->itr = IXGBE_10K_ITR;
1144                         else
1145                                 q_vector->itr = adapter->tx_itr_setting;
1146                 } else {
1147                         /* Rx or Rx/Tx vector */
1148                         if (adapter->rx_itr_setting == 1)
1149                                 q_vector->itr = IXGBE_20K_ITR;
1150                         else
1151                                 q_vector->itr = adapter->rx_itr_setting;
1152                 }
1153
1154                 /* add q_vector eims value to global eims_enable_mask */
1155                 adapter->eims_enable_mask |= 1 << v_idx;
1156
1157                 ixgbevf_write_eitr(q_vector);
1158         }
1159
1160         ixgbevf_set_ivar(adapter, -1, 1, v_idx);
1161         /* setup eims_other and add value to global eims_enable_mask */
1162         adapter->eims_other = 1 << v_idx;
1163         adapter->eims_enable_mask |= adapter->eims_other;
1164 }
1165
1166 enum latency_range {
1167         lowest_latency = 0,
1168         low_latency = 1,
1169         bulk_latency = 2,
1170         latency_invalid = 255
1171 };
1172
1173 /**
1174  * ixgbevf_update_itr - update the dynamic ITR value based on statistics
1175  * @q_vector: structure containing interrupt and ring information
1176  * @ring_container: structure containing ring performance data
1177  *
1178  * Stores a new ITR value based on packets and byte
1179  * counts during the last interrupt.  The advantage of per interrupt
1180  * computation is faster updates and more accurate ITR for the current
1181  * traffic pattern.  Constants in this function were computed
1182  * based on theoretical maximum wire speed and thresholds were set based
1183  * on testing data as well as attempting to minimize response time
1184  * while increasing bulk throughput.
1185  **/
1186 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
1187                                struct ixgbevf_ring_container *ring_container)
1188 {
1189         int bytes = ring_container->total_bytes;
1190         int packets = ring_container->total_packets;
1191         u32 timepassed_us;
1192         u64 bytes_perint;
1193         u8 itr_setting = ring_container->itr;
1194
1195         if (packets == 0)
1196                 return;
1197
1198         /* simple throttle rate management
1199          *    0-20MB/s lowest (100000 ints/s)
1200          *   20-100MB/s low   (20000 ints/s)
1201          *  100-1249MB/s bulk (8000 ints/s)
1202          */
1203         /* what was last interrupt timeslice? */
1204         timepassed_us = q_vector->itr >> 2;
1205         bytes_perint = bytes / timepassed_us; /* bytes/usec */
1206
1207         switch (itr_setting) {
1208         case lowest_latency:
1209                 if (bytes_perint > 10)
1210                         itr_setting = low_latency;
1211                 break;
1212         case low_latency:
1213                 if (bytes_perint > 20)
1214                         itr_setting = bulk_latency;
1215                 else if (bytes_perint <= 10)
1216                         itr_setting = lowest_latency;
1217                 break;
1218         case bulk_latency:
1219                 if (bytes_perint <= 20)
1220                         itr_setting = low_latency;
1221                 break;
1222         }
1223
1224         /* clear work counters since we have the values we need */
1225         ring_container->total_bytes = 0;
1226         ring_container->total_packets = 0;
1227
1228         /* write updated itr to ring container */
1229         ring_container->itr = itr_setting;
1230 }
1231
1232 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
1233 {
1234         u32 new_itr = q_vector->itr;
1235         u8 current_itr;
1236
1237         ixgbevf_update_itr(q_vector, &q_vector->tx);
1238         ixgbevf_update_itr(q_vector, &q_vector->rx);
1239
1240         current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1241
1242         switch (current_itr) {
1243         /* counts and packets in update_itr are dependent on these numbers */
1244         case lowest_latency:
1245                 new_itr = IXGBE_100K_ITR;
1246                 break;
1247         case low_latency:
1248                 new_itr = IXGBE_20K_ITR;
1249                 break;
1250         case bulk_latency:
1251         default:
1252                 new_itr = IXGBE_8K_ITR;
1253                 break;
1254         }
1255
1256         if (new_itr != q_vector->itr) {
1257                 /* do an exponential smoothing */
1258                 new_itr = (10 * new_itr * q_vector->itr) /
1259                           ((9 * new_itr) + q_vector->itr);
1260
1261                 /* save the algorithm value here */
1262                 q_vector->itr = new_itr;
1263
1264                 ixgbevf_write_eitr(q_vector);
1265         }
1266 }
1267
1268 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
1269 {
1270         struct ixgbevf_adapter *adapter = data;
1271         struct ixgbe_hw *hw = &adapter->hw;
1272
1273         hw->mac.get_link_status = 1;
1274
1275         ixgbevf_service_event_schedule(adapter);
1276
1277         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
1278
1279         return IRQ_HANDLED;
1280 }
1281
1282 /**
1283  * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
1284  * @irq: unused
1285  * @data: pointer to our q_vector struct for this interrupt vector
1286  **/
1287 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
1288 {
1289         struct ixgbevf_q_vector *q_vector = data;
1290
1291         /* EIAM disabled interrupts (on this vector) for us */
1292         if (q_vector->rx.ring || q_vector->tx.ring)
1293                 napi_schedule(&q_vector->napi);
1294
1295         return IRQ_HANDLED;
1296 }
1297
1298 static inline void map_vector_to_rxq(struct ixgbevf_adapter *a, int v_idx,
1299                                      int r_idx)
1300 {
1301         struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1302
1303         a->rx_ring[r_idx]->next = q_vector->rx.ring;
1304         q_vector->rx.ring = a->rx_ring[r_idx];
1305         q_vector->rx.count++;
1306 }
1307
1308 static inline void map_vector_to_txq(struct ixgbevf_adapter *a, int v_idx,
1309                                      int t_idx)
1310 {
1311         struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
1312
1313         a->tx_ring[t_idx]->next = q_vector->tx.ring;
1314         q_vector->tx.ring = a->tx_ring[t_idx];
1315         q_vector->tx.count++;
1316 }
1317
1318 /**
1319  * ixgbevf_map_rings_to_vectors - Maps descriptor rings to vectors
1320  * @adapter: board private structure to initialize
1321  *
1322  * This function maps descriptor rings to the queue-specific vectors
1323  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
1324  * one vector per ring/queue, but on a constrained vector budget, we
1325  * group the rings as "efficiently" as possible.  You would add new
1326  * mapping configurations in here.
1327  **/
1328 static int ixgbevf_map_rings_to_vectors(struct ixgbevf_adapter *adapter)
1329 {
1330         int q_vectors;
1331         int v_start = 0;
1332         int rxr_idx = 0, txr_idx = 0;
1333         int rxr_remaining = adapter->num_rx_queues;
1334         int txr_remaining = adapter->num_tx_queues;
1335         int i, j;
1336         int rqpv, tqpv;
1337         int err = 0;
1338
1339         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1340
1341         /* The ideal configuration...
1342          * We have enough vectors to map one per queue.
1343          */
1344         if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1345                 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1346                         map_vector_to_rxq(adapter, v_start, rxr_idx);
1347
1348                 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1349                         map_vector_to_txq(adapter, v_start, txr_idx);
1350                 goto out;
1351         }
1352
1353         /* If we don't have enough vectors for a 1-to-1
1354          * mapping, we'll have to group them so there are
1355          * multiple queues per vector.
1356          */
1357         /* Re-adjusting *qpv takes care of the remainder. */
1358         for (i = v_start; i < q_vectors; i++) {
1359                 rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
1360                 for (j = 0; j < rqpv; j++) {
1361                         map_vector_to_rxq(adapter, i, rxr_idx);
1362                         rxr_idx++;
1363                         rxr_remaining--;
1364                 }
1365         }
1366         for (i = v_start; i < q_vectors; i++) {
1367                 tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
1368                 for (j = 0; j < tqpv; j++) {
1369                         map_vector_to_txq(adapter, i, txr_idx);
1370                         txr_idx++;
1371                         txr_remaining--;
1372                 }
1373         }
1374
1375 out:
1376         return err;
1377 }
1378
1379 /**
1380  * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1381  * @adapter: board private structure
1382  *
1383  * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1384  * interrupts from the kernel.
1385  **/
1386 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
1387 {
1388         struct net_device *netdev = adapter->netdev;
1389         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1390         int vector, err;
1391         int ri = 0, ti = 0;
1392
1393         for (vector = 0; vector < q_vectors; vector++) {
1394                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
1395                 struct msix_entry *entry = &adapter->msix_entries[vector];
1396
1397                 if (q_vector->tx.ring && q_vector->rx.ring) {
1398                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1399                                  "%s-%s-%d", netdev->name, "TxRx", ri++);
1400                         ti++;
1401                 } else if (q_vector->rx.ring) {
1402                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1403                                  "%s-%s-%d", netdev->name, "rx", ri++);
1404                 } else if (q_vector->tx.ring) {
1405                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
1406                                  "%s-%s-%d", netdev->name, "tx", ti++);
1407                 } else {
1408                         /* skip this unused q_vector */
1409                         continue;
1410                 }
1411                 err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
1412                                   q_vector->name, q_vector);
1413                 if (err) {
1414                         hw_dbg(&adapter->hw,
1415                                "request_irq failed for MSIX interrupt Error: %d\n",
1416                                err);
1417                         goto free_queue_irqs;
1418                 }
1419         }
1420
1421         err = request_irq(adapter->msix_entries[vector].vector,
1422                           &ixgbevf_msix_other, 0, netdev->name, adapter);
1423         if (err) {
1424                 hw_dbg(&adapter->hw, "request_irq for msix_other failed: %d\n",
1425                        err);
1426                 goto free_queue_irqs;
1427         }
1428
1429         return 0;
1430
1431 free_queue_irqs:
1432         while (vector) {
1433                 vector--;
1434                 free_irq(adapter->msix_entries[vector].vector,
1435                          adapter->q_vector[vector]);
1436         }
1437         /* This failure is non-recoverable - it indicates the system is
1438          * out of MSIX vector resources and the VF driver cannot run
1439          * without them.  Set the number of msix vectors to zero
1440          * indicating that not enough can be allocated.  The error
1441          * will be returned to the user indicating device open failed.
1442          * Any further attempts to force the driver to open will also
1443          * fail.  The only way to recover is to unload the driver and
1444          * reload it again.  If the system has recovered some MSIX
1445          * vectors then it may succeed.
1446          */
1447         adapter->num_msix_vectors = 0;
1448         return err;
1449 }
1450
1451 static inline void ixgbevf_reset_q_vectors(struct ixgbevf_adapter *adapter)
1452 {
1453         int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1454
1455         for (i = 0; i < q_vectors; i++) {
1456                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[i];
1457
1458                 q_vector->rx.ring = NULL;
1459                 q_vector->tx.ring = NULL;
1460                 q_vector->rx.count = 0;
1461                 q_vector->tx.count = 0;
1462         }
1463 }
1464
1465 /**
1466  * ixgbevf_request_irq - initialize interrupts
1467  * @adapter: board private structure
1468  *
1469  * Attempts to configure interrupts using the best available
1470  * capabilities of the hardware and kernel.
1471  **/
1472 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1473 {
1474         int err = 0;
1475
1476         err = ixgbevf_request_msix_irqs(adapter);
1477
1478         if (err)
1479                 hw_dbg(&adapter->hw, "request_irq failed, Error %d\n", err);
1480
1481         return err;
1482 }
1483
1484 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1485 {
1486         int i, q_vectors;
1487
1488         q_vectors = adapter->num_msix_vectors;
1489         i = q_vectors - 1;
1490
1491         free_irq(adapter->msix_entries[i].vector, adapter);
1492         i--;
1493
1494         for (; i >= 0; i--) {
1495                 /* free only the irqs that were actually requested */
1496                 if (!adapter->q_vector[i]->rx.ring &&
1497                     !adapter->q_vector[i]->tx.ring)
1498                         continue;
1499
1500                 free_irq(adapter->msix_entries[i].vector,
1501                          adapter->q_vector[i]);
1502         }
1503
1504         ixgbevf_reset_q_vectors(adapter);
1505 }
1506
1507 /**
1508  * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1509  * @adapter: board private structure
1510  **/
1511 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1512 {
1513         struct ixgbe_hw *hw = &adapter->hw;
1514         int i;
1515
1516         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1517         IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1518         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1519
1520         IXGBE_WRITE_FLUSH(hw);
1521
1522         for (i = 0; i < adapter->num_msix_vectors; i++)
1523                 synchronize_irq(adapter->msix_entries[i].vector);
1524 }
1525
1526 /**
1527  * ixgbevf_irq_enable - Enable default interrupt generation settings
1528  * @adapter: board private structure
1529  **/
1530 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1531 {
1532         struct ixgbe_hw *hw = &adapter->hw;
1533
1534         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1535         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1536         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1537 }
1538
1539 /**
1540  * ixgbevf_configure_tx_ring - Configure 82599 VF Tx ring after Reset
1541  * @adapter: board private structure
1542  * @ring: structure containing ring specific data
1543  *
1544  * Configure the Tx descriptor ring after a reset.
1545  **/
1546 static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter *adapter,
1547                                       struct ixgbevf_ring *ring)
1548 {
1549         struct ixgbe_hw *hw = &adapter->hw;
1550         u64 tdba = ring->dma;
1551         int wait_loop = 10;
1552         u32 txdctl = IXGBE_TXDCTL_ENABLE;
1553         u8 reg_idx = ring->reg_idx;
1554
1555         /* disable queue to avoid issues while updating state */
1556         IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
1557         IXGBE_WRITE_FLUSH(hw);
1558
1559         IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(reg_idx), tdba & DMA_BIT_MASK(32));
1560         IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(reg_idx), tdba >> 32);
1561         IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(reg_idx),
1562                         ring->count * sizeof(union ixgbe_adv_tx_desc));
1563
1564         /* disable head writeback */
1565         IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAH(reg_idx), 0);
1566         IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAL(reg_idx), 0);
1567
1568         /* enable relaxed ordering */
1569         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(reg_idx),
1570                         (IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1571                          IXGBE_DCA_TXCTRL_DATA_RRO_EN));
1572
1573         /* reset head and tail pointers */
1574         IXGBE_WRITE_REG(hw, IXGBE_VFTDH(reg_idx), 0);
1575         IXGBE_WRITE_REG(hw, IXGBE_VFTDT(reg_idx), 0);
1576         ring->tail = adapter->io_addr + IXGBE_VFTDT(reg_idx);
1577
1578         /* reset ntu and ntc to place SW in sync with hardwdare */
1579         ring->next_to_clean = 0;
1580         ring->next_to_use = 0;
1581
1582         /* In order to avoid issues WTHRESH + PTHRESH should always be equal
1583          * to or less than the number of on chip descriptors, which is
1584          * currently 40.
1585          */
1586         txdctl |= (8 << 16);    /* WTHRESH = 8 */
1587
1588         /* Setting PTHRESH to 32 both improves performance */
1589         txdctl |= (1 << 8) |    /* HTHRESH = 1 */
1590                   32;          /* PTHRESH = 32 */
1591
1592         clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &ring->state);
1593
1594         IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
1595
1596         /* poll to verify queue is enabled */
1597         do {
1598                 usleep_range(1000, 2000);
1599                 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(reg_idx));
1600         }  while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
1601         if (!wait_loop)
1602                 pr_err("Could not enable Tx Queue %d\n", reg_idx);
1603 }
1604
1605 /**
1606  * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1607  * @adapter: board private structure
1608  *
1609  * Configure the Tx unit of the MAC after a reset.
1610  **/
1611 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1612 {
1613         u32 i;
1614
1615         /* Setup the HW Tx Head and Tail descriptor pointers */
1616         for (i = 0; i < adapter->num_tx_queues; i++)
1617                 ixgbevf_configure_tx_ring(adapter, adapter->tx_ring[i]);
1618 }
1619
1620 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1621
1622 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter, int index)
1623 {
1624         struct ixgbe_hw *hw = &adapter->hw;
1625         u32 srrctl;
1626
1627         srrctl = IXGBE_SRRCTL_DROP_EN;
1628
1629         srrctl |= IXGBEVF_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
1630         srrctl |= IXGBEVF_RX_BUFSZ >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1631         srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1632
1633         IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1634 }
1635
1636 static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter)
1637 {
1638         struct ixgbe_hw *hw = &adapter->hw;
1639
1640         /* PSRTYPE must be initialized in 82599 */
1641         u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
1642                       IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR |
1643                       IXGBE_PSRTYPE_L2HDR;
1644
1645         if (adapter->num_rx_queues > 1)
1646                 psrtype |= 1 << 29;
1647
1648         IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1649 }
1650
1651 #define IXGBEVF_MAX_RX_DESC_POLL 10
1652 static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter,
1653                                      struct ixgbevf_ring *ring)
1654 {
1655         struct ixgbe_hw *hw = &adapter->hw;
1656         int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1657         u32 rxdctl;
1658         u8 reg_idx = ring->reg_idx;
1659
1660         if (IXGBE_REMOVED(hw->hw_addr))
1661                 return;
1662         rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1663         rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1664
1665         /* write value back with RXDCTL.ENABLE bit cleared */
1666         IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1667
1668         /* the hardware may take up to 100us to really disable the Rx queue */
1669         do {
1670                 udelay(10);
1671                 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1672         } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
1673
1674         if (!wait_loop)
1675                 pr_err("RXDCTL.ENABLE queue %d not cleared while polling\n",
1676                        reg_idx);
1677 }
1678
1679 static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1680                                          struct ixgbevf_ring *ring)
1681 {
1682         struct ixgbe_hw *hw = &adapter->hw;
1683         int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1684         u32 rxdctl;
1685         u8 reg_idx = ring->reg_idx;
1686
1687         if (IXGBE_REMOVED(hw->hw_addr))
1688                 return;
1689         do {
1690                 usleep_range(1000, 2000);
1691                 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1692         } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
1693
1694         if (!wait_loop)
1695                 pr_err("RXDCTL.ENABLE queue %d not set while polling\n",
1696                        reg_idx);
1697 }
1698
1699 static void ixgbevf_setup_vfmrqc(struct ixgbevf_adapter *adapter)
1700 {
1701         struct ixgbe_hw *hw = &adapter->hw;
1702         u32 vfmrqc = 0, vfreta = 0;
1703         u16 rss_i = adapter->num_rx_queues;
1704         u8 i, j;
1705
1706         /* Fill out hash function seeds */
1707         netdev_rss_key_fill(adapter->rss_key, sizeof(adapter->rss_key));
1708         for (i = 0; i < IXGBEVF_VFRSSRK_REGS; i++)
1709                 IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), adapter->rss_key[i]);
1710
1711         for (i = 0, j = 0; i < IXGBEVF_X550_VFRETA_SIZE; i++, j++) {
1712                 if (j == rss_i)
1713                         j = 0;
1714
1715                 adapter->rss_indir_tbl[i] = j;
1716
1717                 vfreta |= j << (i & 0x3) * 8;
1718                 if ((i & 3) == 3) {
1719                         IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), vfreta);
1720                         vfreta = 0;
1721                 }
1722         }
1723
1724         /* Perform hash on these packet types */
1725         vfmrqc |= IXGBE_VFMRQC_RSS_FIELD_IPV4 |
1726                 IXGBE_VFMRQC_RSS_FIELD_IPV4_TCP |
1727                 IXGBE_VFMRQC_RSS_FIELD_IPV6 |
1728                 IXGBE_VFMRQC_RSS_FIELD_IPV6_TCP;
1729
1730         vfmrqc |= IXGBE_VFMRQC_RSSEN;
1731
1732         IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, vfmrqc);
1733 }
1734
1735 static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
1736                                       struct ixgbevf_ring *ring)
1737 {
1738         struct ixgbe_hw *hw = &adapter->hw;
1739         u64 rdba = ring->dma;
1740         u32 rxdctl;
1741         u8 reg_idx = ring->reg_idx;
1742
1743         /* disable queue to avoid issues while updating state */
1744         rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1745         ixgbevf_disable_rx_queue(adapter, ring);
1746
1747         IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(reg_idx), rdba & DMA_BIT_MASK(32));
1748         IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(reg_idx), rdba >> 32);
1749         IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(reg_idx),
1750                         ring->count * sizeof(union ixgbe_adv_rx_desc));
1751
1752         /* enable relaxed ordering */
1753         IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1754                         IXGBE_DCA_RXCTRL_DESC_RRO_EN);
1755
1756         /* reset head and tail pointers */
1757         IXGBE_WRITE_REG(hw, IXGBE_VFRDH(reg_idx), 0);
1758         IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
1759         ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
1760
1761         /* reset ntu and ntc to place SW in sync with hardwdare */
1762         ring->next_to_clean = 0;
1763         ring->next_to_use = 0;
1764         ring->next_to_alloc = 0;
1765
1766         ixgbevf_configure_srrctl(adapter, reg_idx);
1767
1768         /* allow any size packet since we can handle overflow */
1769         rxdctl &= ~IXGBE_RXDCTL_RLPML_EN;
1770
1771         rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1772         IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1773
1774         ixgbevf_rx_desc_queue_enable(adapter, ring);
1775         ixgbevf_alloc_rx_buffers(ring, ixgbevf_desc_unused(ring));
1776 }
1777
1778 /**
1779  * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1780  * @adapter: board private structure
1781  *
1782  * Configure the Rx unit of the MAC after a reset.
1783  **/
1784 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1785 {
1786         int i;
1787         struct ixgbe_hw *hw = &adapter->hw;
1788         struct net_device *netdev = adapter->netdev;
1789
1790         ixgbevf_setup_psrtype(adapter);
1791         if (hw->mac.type >= ixgbe_mac_X550_vf)
1792                 ixgbevf_setup_vfmrqc(adapter);
1793
1794         /* notify the PF of our intent to use this size of frame */
1795         ixgbevf_rlpml_set_vf(hw, netdev->mtu + ETH_HLEN + ETH_FCS_LEN);
1796
1797         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1798          * the Base and Length of the Rx Descriptor Ring
1799          */
1800         for (i = 0; i < adapter->num_rx_queues; i++)
1801                 ixgbevf_configure_rx_ring(adapter, adapter->rx_ring[i]);
1802 }
1803
1804 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev,
1805                                    __be16 proto, u16 vid)
1806 {
1807         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1808         struct ixgbe_hw *hw = &adapter->hw;
1809         int err;
1810
1811         spin_lock_bh(&adapter->mbx_lock);
1812
1813         /* add VID to filter table */
1814         err = hw->mac.ops.set_vfta(hw, vid, 0, true);
1815
1816         spin_unlock_bh(&adapter->mbx_lock);
1817
1818         /* translate error return types so error makes sense */
1819         if (err == IXGBE_ERR_MBX)
1820                 return -EIO;
1821
1822         if (err == IXGBE_ERR_INVALID_ARGUMENT)
1823                 return -EACCES;
1824
1825         set_bit(vid, adapter->active_vlans);
1826
1827         return err;
1828 }
1829
1830 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev,
1831                                     __be16 proto, u16 vid)
1832 {
1833         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1834         struct ixgbe_hw *hw = &adapter->hw;
1835         int err = -EOPNOTSUPP;
1836
1837         spin_lock_bh(&adapter->mbx_lock);
1838
1839         /* remove VID from filter table */
1840         err = hw->mac.ops.set_vfta(hw, vid, 0, false);
1841
1842         spin_unlock_bh(&adapter->mbx_lock);
1843
1844         clear_bit(vid, adapter->active_vlans);
1845
1846         return err;
1847 }
1848
1849 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
1850 {
1851         u16 vid;
1852
1853         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1854                 ixgbevf_vlan_rx_add_vid(adapter->netdev,
1855                                         htons(ETH_P_8021Q), vid);
1856 }
1857
1858 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
1859 {
1860         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1861         struct ixgbe_hw *hw = &adapter->hw;
1862         int count = 0;
1863
1864         if ((netdev_uc_count(netdev)) > 10) {
1865                 pr_err("Too many unicast filters - No Space\n");
1866                 return -ENOSPC;
1867         }
1868
1869         if (!netdev_uc_empty(netdev)) {
1870                 struct netdev_hw_addr *ha;
1871
1872                 netdev_for_each_uc_addr(ha, netdev) {
1873                         hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
1874                         udelay(200);
1875                 }
1876         } else {
1877                 /* If the list is empty then send message to PF driver to
1878                  * clear all MAC VLANs on this VF.
1879                  */
1880                 hw->mac.ops.set_uc_addr(hw, 0, NULL);
1881         }
1882
1883         return count;
1884 }
1885
1886 /**
1887  * ixgbevf_set_rx_mode - Multicast and unicast set
1888  * @netdev: network interface device structure
1889  *
1890  * The set_rx_method entry point is called whenever the multicast address
1891  * list, unicast address list or the network interface flags are updated.
1892  * This routine is responsible for configuring the hardware for proper
1893  * multicast mode and configuring requested unicast filters.
1894  **/
1895 static void ixgbevf_set_rx_mode(struct net_device *netdev)
1896 {
1897         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1898         struct ixgbe_hw *hw = &adapter->hw;
1899         unsigned int flags = netdev->flags;
1900         int xcast_mode;
1901
1902         xcast_mode = (flags & IFF_ALLMULTI) ? IXGBEVF_XCAST_MODE_ALLMULTI :
1903                      (flags & (IFF_BROADCAST | IFF_MULTICAST)) ?
1904                      IXGBEVF_XCAST_MODE_MULTI : IXGBEVF_XCAST_MODE_NONE;
1905
1906         spin_lock_bh(&adapter->mbx_lock);
1907
1908         hw->mac.ops.update_xcast_mode(hw, netdev, xcast_mode);
1909
1910         /* reprogram multicast list */
1911         hw->mac.ops.update_mc_addr_list(hw, netdev);
1912
1913         ixgbevf_write_uc_addr_list(netdev);
1914
1915         spin_unlock_bh(&adapter->mbx_lock);
1916 }
1917
1918 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
1919 {
1920         int q_idx;
1921         struct ixgbevf_q_vector *q_vector;
1922         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1923
1924         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1925                 q_vector = adapter->q_vector[q_idx];
1926 #ifdef CONFIG_NET_RX_BUSY_POLL
1927                 ixgbevf_qv_init_lock(adapter->q_vector[q_idx]);
1928 #endif
1929                 napi_enable(&q_vector->napi);
1930         }
1931 }
1932
1933 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
1934 {
1935         int q_idx;
1936         struct ixgbevf_q_vector *q_vector;
1937         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1938
1939         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1940                 q_vector = adapter->q_vector[q_idx];
1941                 napi_disable(&q_vector->napi);
1942 #ifdef CONFIG_NET_RX_BUSY_POLL
1943                 while (!ixgbevf_qv_disable(adapter->q_vector[q_idx])) {
1944                         pr_info("QV %d locked\n", q_idx);
1945                         usleep_range(1000, 20000);
1946                 }
1947 #endif /* CONFIG_NET_RX_BUSY_POLL */
1948         }
1949 }
1950
1951 static int ixgbevf_configure_dcb(struct ixgbevf_adapter *adapter)
1952 {
1953         struct ixgbe_hw *hw = &adapter->hw;
1954         unsigned int def_q = 0;
1955         unsigned int num_tcs = 0;
1956         unsigned int num_rx_queues = adapter->num_rx_queues;
1957         unsigned int num_tx_queues = adapter->num_tx_queues;
1958         int err;
1959
1960         spin_lock_bh(&adapter->mbx_lock);
1961
1962         /* fetch queue configuration from the PF */
1963         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
1964
1965         spin_unlock_bh(&adapter->mbx_lock);
1966
1967         if (err)
1968                 return err;
1969
1970         if (num_tcs > 1) {
1971                 /* we need only one Tx queue */
1972                 num_tx_queues = 1;
1973
1974                 /* update default Tx ring register index */
1975                 adapter->tx_ring[0]->reg_idx = def_q;
1976
1977                 /* we need as many queues as traffic classes */
1978                 num_rx_queues = num_tcs;
1979         }
1980
1981         /* if we have a bad config abort request queue reset */
1982         if ((adapter->num_rx_queues != num_rx_queues) ||
1983             (adapter->num_tx_queues != num_tx_queues)) {
1984                 /* force mailbox timeout to prevent further messages */
1985                 hw->mbx.timeout = 0;
1986
1987                 /* wait for watchdog to come around and bail us out */
1988                 adapter->flags |= IXGBEVF_FLAG_QUEUE_RESET_REQUESTED;
1989         }
1990
1991         return 0;
1992 }
1993
1994 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
1995 {
1996         ixgbevf_configure_dcb(adapter);
1997
1998         ixgbevf_set_rx_mode(adapter->netdev);
1999
2000         ixgbevf_restore_vlan(adapter);
2001
2002         ixgbevf_configure_tx(adapter);
2003         ixgbevf_configure_rx(adapter);
2004 }
2005
2006 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
2007 {
2008         /* Only save pre-reset stats if there are some */
2009         if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
2010                 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
2011                         adapter->stats.base_vfgprc;
2012                 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
2013                         adapter->stats.base_vfgptc;
2014                 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
2015                         adapter->stats.base_vfgorc;
2016                 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
2017                         adapter->stats.base_vfgotc;
2018                 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
2019                         adapter->stats.base_vfmprc;
2020         }
2021 }
2022
2023 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
2024 {
2025         struct ixgbe_hw *hw = &adapter->hw;
2026
2027         adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
2028         adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
2029         adapter->stats.last_vfgorc |=
2030                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
2031         adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
2032         adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
2033         adapter->stats.last_vfgotc |=
2034                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
2035         adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
2036
2037         adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
2038         adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
2039         adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
2040         adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
2041         adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
2042 }
2043
2044 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
2045 {
2046         struct ixgbe_hw *hw = &adapter->hw;
2047         int api[] = { ixgbe_mbox_api_12,
2048                       ixgbe_mbox_api_11,
2049                       ixgbe_mbox_api_10,
2050                       ixgbe_mbox_api_unknown };
2051         int err = 0, idx = 0;
2052
2053         spin_lock_bh(&adapter->mbx_lock);
2054
2055         while (api[idx] != ixgbe_mbox_api_unknown) {
2056                 err = ixgbevf_negotiate_api_version(hw, api[idx]);
2057                 if (!err)
2058                         break;
2059                 idx++;
2060         }
2061
2062         spin_unlock_bh(&adapter->mbx_lock);
2063 }
2064
2065 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
2066 {
2067         struct net_device *netdev = adapter->netdev;
2068         struct ixgbe_hw *hw = &adapter->hw;
2069
2070         ixgbevf_configure_msix(adapter);
2071
2072         spin_lock_bh(&adapter->mbx_lock);
2073
2074         if (is_valid_ether_addr(hw->mac.addr))
2075                 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
2076         else
2077                 hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
2078
2079         spin_unlock_bh(&adapter->mbx_lock);
2080
2081         smp_mb__before_atomic();
2082         clear_bit(__IXGBEVF_DOWN, &adapter->state);
2083         ixgbevf_napi_enable_all(adapter);
2084
2085         /* clear any pending interrupts, may auto mask */
2086         IXGBE_READ_REG(hw, IXGBE_VTEICR);
2087         ixgbevf_irq_enable(adapter);
2088
2089         /* enable transmits */
2090         netif_tx_start_all_queues(netdev);
2091
2092         ixgbevf_save_reset_stats(adapter);
2093         ixgbevf_init_last_counter_stats(adapter);
2094
2095         hw->mac.get_link_status = 1;
2096         mod_timer(&adapter->service_timer, jiffies);
2097 }
2098
2099 void ixgbevf_up(struct ixgbevf_adapter *adapter)
2100 {
2101         ixgbevf_configure(adapter);
2102
2103         ixgbevf_up_complete(adapter);
2104 }
2105
2106 /**
2107  * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
2108  * @rx_ring: ring to free buffers from
2109  **/
2110 static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
2111 {
2112         struct device *dev = rx_ring->dev;
2113         unsigned long size;
2114         unsigned int i;
2115
2116         /* Free Rx ring sk_buff */
2117         if (rx_ring->skb) {
2118                 dev_kfree_skb(rx_ring->skb);
2119                 rx_ring->skb = NULL;
2120         }
2121
2122         /* ring already cleared, nothing to do */
2123         if (!rx_ring->rx_buffer_info)
2124                 return;
2125
2126         /* Free all the Rx ring pages */
2127         for (i = 0; i < rx_ring->count; i++) {
2128                 struct ixgbevf_rx_buffer *rx_buffer;
2129
2130                 rx_buffer = &rx_ring->rx_buffer_info[i];
2131                 if (rx_buffer->dma)
2132                         dma_unmap_page(dev, rx_buffer->dma,
2133                                        PAGE_SIZE, DMA_FROM_DEVICE);
2134                 rx_buffer->dma = 0;
2135                 if (rx_buffer->page)
2136                         __free_page(rx_buffer->page);
2137                 rx_buffer->page = NULL;
2138         }
2139
2140         size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
2141         memset(rx_ring->rx_buffer_info, 0, size);
2142
2143         /* Zero out the descriptor ring */
2144         memset(rx_ring->desc, 0, rx_ring->size);
2145 }
2146
2147 /**
2148  * ixgbevf_clean_tx_ring - Free Tx Buffers
2149  * @tx_ring: ring to be cleaned
2150  **/
2151 static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
2152 {
2153         struct ixgbevf_tx_buffer *tx_buffer_info;
2154         unsigned long size;
2155         unsigned int i;
2156
2157         if (!tx_ring->tx_buffer_info)
2158                 return;
2159
2160         /* Free all the Tx ring sk_buffs */
2161         for (i = 0; i < tx_ring->count; i++) {
2162                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2163                 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
2164         }
2165
2166         size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2167         memset(tx_ring->tx_buffer_info, 0, size);
2168
2169         memset(tx_ring->desc, 0, tx_ring->size);
2170 }
2171
2172 /**
2173  * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
2174  * @adapter: board private structure
2175  **/
2176 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
2177 {
2178         int i;
2179
2180         for (i = 0; i < adapter->num_rx_queues; i++)
2181                 ixgbevf_clean_rx_ring(adapter->rx_ring[i]);
2182 }
2183
2184 /**
2185  * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
2186  * @adapter: board private structure
2187  **/
2188 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
2189 {
2190         int i;
2191
2192         for (i = 0; i < adapter->num_tx_queues; i++)
2193                 ixgbevf_clean_tx_ring(adapter->tx_ring[i]);
2194 }
2195
2196 void ixgbevf_down(struct ixgbevf_adapter *adapter)
2197 {
2198         struct net_device *netdev = adapter->netdev;
2199         struct ixgbe_hw *hw = &adapter->hw;
2200         int i;
2201
2202         /* signal that we are down to the interrupt handler */
2203         if (test_and_set_bit(__IXGBEVF_DOWN, &adapter->state))
2204                 return; /* do nothing if already down */
2205
2206         /* disable all enabled Rx queues */
2207         for (i = 0; i < adapter->num_rx_queues; i++)
2208                 ixgbevf_disable_rx_queue(adapter, adapter->rx_ring[i]);
2209
2210         usleep_range(10000, 20000);
2211
2212         netif_tx_stop_all_queues(netdev);
2213
2214         /* call carrier off first to avoid false dev_watchdog timeouts */
2215         netif_carrier_off(netdev);
2216         netif_tx_disable(netdev);
2217
2218         ixgbevf_irq_disable(adapter);
2219
2220         ixgbevf_napi_disable_all(adapter);
2221
2222         del_timer_sync(&adapter->service_timer);
2223
2224         /* disable transmits in the hardware now that interrupts are off */
2225         for (i = 0; i < adapter->num_tx_queues; i++) {
2226                 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
2227
2228                 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2229                                 IXGBE_TXDCTL_SWFLSH);
2230         }
2231
2232         if (!pci_channel_offline(adapter->pdev))
2233                 ixgbevf_reset(adapter);
2234
2235         ixgbevf_clean_all_tx_rings(adapter);
2236         ixgbevf_clean_all_rx_rings(adapter);
2237 }
2238
2239 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
2240 {
2241         WARN_ON(in_interrupt());
2242
2243         while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
2244                 msleep(1);
2245
2246         ixgbevf_down(adapter);
2247         ixgbevf_up(adapter);
2248
2249         clear_bit(__IXGBEVF_RESETTING, &adapter->state);
2250 }
2251
2252 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
2253 {
2254         struct ixgbe_hw *hw = &adapter->hw;
2255         struct net_device *netdev = adapter->netdev;
2256
2257         if (hw->mac.ops.reset_hw(hw)) {
2258                 hw_dbg(hw, "PF still resetting\n");
2259         } else {
2260                 hw->mac.ops.init_hw(hw);
2261                 ixgbevf_negotiate_api(adapter);
2262         }
2263
2264         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
2265                 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
2266                        netdev->addr_len);
2267                 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
2268                        netdev->addr_len);
2269         }
2270
2271         adapter->last_reset = jiffies;
2272 }
2273
2274 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
2275                                         int vectors)
2276 {
2277         int vector_threshold;
2278
2279         /* We'll want at least 2 (vector_threshold):
2280          * 1) TxQ[0] + RxQ[0] handler
2281          * 2) Other (Link Status Change, etc.)
2282          */
2283         vector_threshold = MIN_MSIX_COUNT;
2284
2285         /* The more we get, the more we will assign to Tx/Rx Cleanup
2286          * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2287          * Right now, we simply care about how many we'll get; we'll
2288          * set them up later while requesting irq's.
2289          */
2290         vectors = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
2291                                         vector_threshold, vectors);
2292
2293         if (vectors < 0) {
2294                 dev_err(&adapter->pdev->dev,
2295                         "Unable to allocate MSI-X interrupts\n");
2296                 kfree(adapter->msix_entries);
2297                 adapter->msix_entries = NULL;
2298                 return vectors;
2299         }
2300
2301         /* Adjust for only the vectors we'll use, which is minimum
2302          * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2303          * vectors we were allocated.
2304          */
2305         adapter->num_msix_vectors = vectors;
2306
2307         return 0;
2308 }
2309
2310 /**
2311  * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
2312  * @adapter: board private structure to initialize
2313  *
2314  * This is the top level queue allocation routine.  The order here is very
2315  * important, starting with the "most" number of features turned on at once,
2316  * and ending with the smallest set of features.  This way large combinations
2317  * can be allocated if they're turned on, and smaller combinations are the
2318  * fallthrough conditions.
2319  *
2320  **/
2321 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
2322 {
2323         struct ixgbe_hw *hw = &adapter->hw;
2324         unsigned int def_q = 0;
2325         unsigned int num_tcs = 0;
2326         int err;
2327
2328         /* Start with base case */
2329         adapter->num_rx_queues = 1;
2330         adapter->num_tx_queues = 1;
2331
2332         spin_lock_bh(&adapter->mbx_lock);
2333
2334         /* fetch queue configuration from the PF */
2335         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2336
2337         spin_unlock_bh(&adapter->mbx_lock);
2338
2339         if (err)
2340                 return;
2341
2342         /* we need as many queues as traffic classes */
2343         if (num_tcs > 1) {
2344                 adapter->num_rx_queues = num_tcs;
2345         } else {
2346                 u16 rss = min_t(u16, num_online_cpus(), IXGBEVF_MAX_RSS_QUEUES);
2347
2348                 switch (hw->api_version) {
2349                 case ixgbe_mbox_api_11:
2350                 case ixgbe_mbox_api_12:
2351                         adapter->num_rx_queues = rss;
2352                         adapter->num_tx_queues = rss;
2353                 default:
2354                         break;
2355                 }
2356         }
2357 }
2358
2359 /**
2360  * ixgbevf_alloc_queues - Allocate memory for all rings
2361  * @adapter: board private structure to initialize
2362  *
2363  * We allocate one ring per queue at run-time since we don't know the
2364  * number of queues at compile-time.  The polling_netdev array is
2365  * intended for Multiqueue, but should work fine with a single queue.
2366  **/
2367 static int ixgbevf_alloc_queues(struct ixgbevf_adapter *adapter)
2368 {
2369         struct ixgbevf_ring *ring;
2370         int rx = 0, tx = 0;
2371
2372         for (; tx < adapter->num_tx_queues; tx++) {
2373                 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
2374                 if (!ring)
2375                         goto err_allocation;
2376
2377                 ring->dev = &adapter->pdev->dev;
2378                 ring->netdev = adapter->netdev;
2379                 ring->count = adapter->tx_ring_count;
2380                 ring->queue_index = tx;
2381                 ring->reg_idx = tx;
2382
2383                 adapter->tx_ring[tx] = ring;
2384         }
2385
2386         for (; rx < adapter->num_rx_queues; rx++) {
2387                 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
2388                 if (!ring)
2389                         goto err_allocation;
2390
2391                 ring->dev = &adapter->pdev->dev;
2392                 ring->netdev = adapter->netdev;
2393
2394                 ring->count = adapter->rx_ring_count;
2395                 ring->queue_index = rx;
2396                 ring->reg_idx = rx;
2397
2398                 adapter->rx_ring[rx] = ring;
2399         }
2400
2401         return 0;
2402
2403 err_allocation:
2404         while (tx) {
2405                 kfree(adapter->tx_ring[--tx]);
2406                 adapter->tx_ring[tx] = NULL;
2407         }
2408
2409         while (rx) {
2410                 kfree(adapter->rx_ring[--rx]);
2411                 adapter->rx_ring[rx] = NULL;
2412         }
2413         return -ENOMEM;
2414 }
2415
2416 /**
2417  * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2418  * @adapter: board private structure to initialize
2419  *
2420  * Attempt to configure the interrupts using the best available
2421  * capabilities of the hardware and the kernel.
2422  **/
2423 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
2424 {
2425         struct net_device *netdev = adapter->netdev;
2426         int err = 0;
2427         int vector, v_budget;
2428
2429         /* It's easy to be greedy for MSI-X vectors, but it really
2430          * doesn't do us much good if we have a lot more vectors
2431          * than CPU's.  So let's be conservative and only ask for
2432          * (roughly) the same number of vectors as there are CPU's.
2433          * The default is to use pairs of vectors.
2434          */
2435         v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
2436         v_budget = min_t(int, v_budget, num_online_cpus());
2437         v_budget += NON_Q_VECTORS;
2438
2439         /* A failure in MSI-X entry allocation isn't fatal, but it does
2440          * mean we disable MSI-X capabilities of the adapter.
2441          */
2442         adapter->msix_entries = kcalloc(v_budget,
2443                                         sizeof(struct msix_entry), GFP_KERNEL);
2444         if (!adapter->msix_entries) {
2445                 err = -ENOMEM;
2446                 goto out;
2447         }
2448
2449         for (vector = 0; vector < v_budget; vector++)
2450                 adapter->msix_entries[vector].entry = vector;
2451
2452         err = ixgbevf_acquire_msix_vectors(adapter, v_budget);
2453         if (err)
2454                 goto out;
2455
2456         err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
2457         if (err)
2458                 goto out;
2459
2460         err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
2461
2462 out:
2463         return err;
2464 }
2465
2466 /**
2467  * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2468  * @adapter: board private structure to initialize
2469  *
2470  * We allocate one q_vector per queue interrupt.  If allocation fails we
2471  * return -ENOMEM.
2472  **/
2473 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2474 {
2475         int q_idx, num_q_vectors;
2476         struct ixgbevf_q_vector *q_vector;
2477
2478         num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2479
2480         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2481                 q_vector = kzalloc(sizeof(struct ixgbevf_q_vector), GFP_KERNEL);
2482                 if (!q_vector)
2483                         goto err_out;
2484                 q_vector->adapter = adapter;
2485                 q_vector->v_idx = q_idx;
2486                 netif_napi_add(adapter->netdev, &q_vector->napi,
2487                                ixgbevf_poll, 64);
2488 #ifdef CONFIG_NET_RX_BUSY_POLL
2489                 napi_hash_add(&q_vector->napi);
2490 #endif
2491                 adapter->q_vector[q_idx] = q_vector;
2492         }
2493
2494         return 0;
2495
2496 err_out:
2497         while (q_idx) {
2498                 q_idx--;
2499                 q_vector = adapter->q_vector[q_idx];
2500 #ifdef CONFIG_NET_RX_BUSY_POLL
2501                 napi_hash_del(&q_vector->napi);
2502 #endif
2503                 netif_napi_del(&q_vector->napi);
2504                 kfree(q_vector);
2505                 adapter->q_vector[q_idx] = NULL;
2506         }
2507         return -ENOMEM;
2508 }
2509
2510 /**
2511  * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2512  * @adapter: board private structure to initialize
2513  *
2514  * This function frees the memory allocated to the q_vectors.  In addition if
2515  * NAPI is enabled it will delete any references to the NAPI struct prior
2516  * to freeing the q_vector.
2517  **/
2518 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2519 {
2520         int q_idx, num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2521
2522         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2523                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[q_idx];
2524
2525                 adapter->q_vector[q_idx] = NULL;
2526 #ifdef CONFIG_NET_RX_BUSY_POLL
2527                 napi_hash_del(&q_vector->napi);
2528 #endif
2529                 netif_napi_del(&q_vector->napi);
2530                 kfree(q_vector);
2531         }
2532 }
2533
2534 /**
2535  * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2536  * @adapter: board private structure
2537  *
2538  **/
2539 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2540 {
2541         pci_disable_msix(adapter->pdev);
2542         kfree(adapter->msix_entries);
2543         adapter->msix_entries = NULL;
2544 }
2545
2546 /**
2547  * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2548  * @adapter: board private structure to initialize
2549  *
2550  **/
2551 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2552 {
2553         int err;
2554
2555         /* Number of supported queues */
2556         ixgbevf_set_num_queues(adapter);
2557
2558         err = ixgbevf_set_interrupt_capability(adapter);
2559         if (err) {
2560                 hw_dbg(&adapter->hw,
2561                        "Unable to setup interrupt capabilities\n");
2562                 goto err_set_interrupt;
2563         }
2564
2565         err = ixgbevf_alloc_q_vectors(adapter);
2566         if (err) {
2567                 hw_dbg(&adapter->hw, "Unable to allocate memory for queue vectors\n");
2568                 goto err_alloc_q_vectors;
2569         }
2570
2571         err = ixgbevf_alloc_queues(adapter);
2572         if (err) {
2573                 pr_err("Unable to allocate memory for queues\n");
2574                 goto err_alloc_queues;
2575         }
2576
2577         hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
2578                (adapter->num_rx_queues > 1) ? "Enabled" :
2579                "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2580
2581         set_bit(__IXGBEVF_DOWN, &adapter->state);
2582
2583         return 0;
2584 err_alloc_queues:
2585         ixgbevf_free_q_vectors(adapter);
2586 err_alloc_q_vectors:
2587         ixgbevf_reset_interrupt_capability(adapter);
2588 err_set_interrupt:
2589         return err;
2590 }
2591
2592 /**
2593  * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2594  * @adapter: board private structure to clear interrupt scheme on
2595  *
2596  * We go through and clear interrupt specific resources and reset the structure
2597  * to pre-load conditions
2598  **/
2599 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2600 {
2601         int i;
2602
2603         for (i = 0; i < adapter->num_tx_queues; i++) {
2604                 kfree(adapter->tx_ring[i]);
2605                 adapter->tx_ring[i] = NULL;
2606         }
2607         for (i = 0; i < adapter->num_rx_queues; i++) {
2608                 kfree(adapter->rx_ring[i]);
2609                 adapter->rx_ring[i] = NULL;
2610         }
2611
2612         adapter->num_tx_queues = 0;
2613         adapter->num_rx_queues = 0;
2614
2615         ixgbevf_free_q_vectors(adapter);
2616         ixgbevf_reset_interrupt_capability(adapter);
2617 }
2618
2619 /**
2620  * ixgbevf_sw_init - Initialize general software structures
2621  * @adapter: board private structure to initialize
2622  *
2623  * ixgbevf_sw_init initializes the Adapter private data structure.
2624  * Fields are initialized based on PCI device information and
2625  * OS network device settings (MTU size).
2626  **/
2627 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
2628 {
2629         struct ixgbe_hw *hw = &adapter->hw;
2630         struct pci_dev *pdev = adapter->pdev;
2631         struct net_device *netdev = adapter->netdev;
2632         int err;
2633
2634         /* PCI config space info */
2635         hw->vendor_id = pdev->vendor;
2636         hw->device_id = pdev->device;
2637         hw->revision_id = pdev->revision;
2638         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2639         hw->subsystem_device_id = pdev->subsystem_device;
2640
2641         hw->mbx.ops.init_params(hw);
2642
2643         /* assume legacy case in which PF would only give VF 2 queues */
2644         hw->mac.max_tx_queues = 2;
2645         hw->mac.max_rx_queues = 2;
2646
2647         /* lock to protect mailbox accesses */
2648         spin_lock_init(&adapter->mbx_lock);
2649
2650         err = hw->mac.ops.reset_hw(hw);
2651         if (err) {
2652                 dev_info(&pdev->dev,
2653                          "PF still in reset state.  Is the PF interface up?\n");
2654         } else {
2655                 err = hw->mac.ops.init_hw(hw);
2656                 if (err) {
2657                         pr_err("init_shared_code failed: %d\n", err);
2658                         goto out;
2659                 }
2660                 ixgbevf_negotiate_api(adapter);
2661                 err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
2662                 if (err)
2663                         dev_info(&pdev->dev, "Error reading MAC address\n");
2664                 else if (is_zero_ether_addr(adapter->hw.mac.addr))
2665                         dev_info(&pdev->dev,
2666                                  "MAC address not assigned by administrator.\n");
2667                 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
2668         }
2669
2670         if (!is_valid_ether_addr(netdev->dev_addr)) {
2671                 dev_info(&pdev->dev, "Assigning random MAC address\n");
2672                 eth_hw_addr_random(netdev);
2673                 memcpy(hw->mac.addr, netdev->dev_addr, netdev->addr_len);
2674         }
2675
2676         /* Enable dynamic interrupt throttling rates */
2677         adapter->rx_itr_setting = 1;
2678         adapter->tx_itr_setting = 1;
2679
2680         /* set default ring sizes */
2681         adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
2682         adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
2683
2684         set_bit(__IXGBEVF_DOWN, &adapter->state);
2685         return 0;
2686
2687 out:
2688         return err;
2689 }
2690
2691 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter)     \
2692         {                                                       \
2693                 u32 current_counter = IXGBE_READ_REG(hw, reg);  \
2694                 if (current_counter < last_counter)             \
2695                         counter += 0x100000000LL;               \
2696                 last_counter = current_counter;                 \
2697                 counter &= 0xFFFFFFFF00000000LL;                \
2698                 counter |= current_counter;                     \
2699         }
2700
2701 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
2702         {                                                                \
2703                 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb);   \
2704                 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb);   \
2705                 u64 current_counter = (current_counter_msb << 32) |      \
2706                         current_counter_lsb;                             \
2707                 if (current_counter < last_counter)                      \
2708                         counter += 0x1000000000LL;                       \
2709                 last_counter = current_counter;                          \
2710                 counter &= 0xFFFFFFF000000000LL;                         \
2711                 counter |= current_counter;                              \
2712         }
2713 /**
2714  * ixgbevf_update_stats - Update the board statistics counters.
2715  * @adapter: board private structure
2716  **/
2717 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
2718 {
2719         struct ixgbe_hw *hw = &adapter->hw;
2720         int i;
2721
2722         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2723             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2724                 return;
2725
2726         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
2727                                 adapter->stats.vfgprc);
2728         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
2729                                 adapter->stats.vfgptc);
2730         UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
2731                                 adapter->stats.last_vfgorc,
2732                                 adapter->stats.vfgorc);
2733         UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
2734                                 adapter->stats.last_vfgotc,
2735                                 adapter->stats.vfgotc);
2736         UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
2737                                 adapter->stats.vfmprc);
2738
2739         for (i = 0;  i  < adapter->num_rx_queues;  i++) {
2740                 adapter->hw_csum_rx_error +=
2741                         adapter->rx_ring[i]->hw_csum_rx_error;
2742                 adapter->rx_ring[i]->hw_csum_rx_error = 0;
2743         }
2744 }
2745
2746 /**
2747  * ixgbevf_service_timer - Timer Call-back
2748  * @data: pointer to adapter cast into an unsigned long
2749  **/
2750 static void ixgbevf_service_timer(unsigned long data)
2751 {
2752         struct ixgbevf_adapter *adapter = (struct ixgbevf_adapter *)data;
2753
2754         /* Reset the timer */
2755         mod_timer(&adapter->service_timer, (HZ * 2) + jiffies);
2756
2757         ixgbevf_service_event_schedule(adapter);
2758 }
2759
2760 static void ixgbevf_reset_subtask(struct ixgbevf_adapter *adapter)
2761 {
2762         if (!(adapter->flags & IXGBEVF_FLAG_RESET_REQUESTED))
2763                 return;
2764
2765         adapter->flags &= ~IXGBEVF_FLAG_RESET_REQUESTED;
2766
2767         /* If we're already down or resetting, just bail */
2768         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2769             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2770                 return;
2771
2772         adapter->tx_timeout_count++;
2773
2774         ixgbevf_reinit_locked(adapter);
2775 }
2776
2777 /**
2778  * ixgbevf_check_hang_subtask - check for hung queues and dropped interrupts
2779  * @adapter: pointer to the device adapter structure
2780  *
2781  * This function serves two purposes.  First it strobes the interrupt lines
2782  * in order to make certain interrupts are occurring.  Secondly it sets the
2783  * bits needed to check for TX hangs.  As a result we should immediately
2784  * determine if a hang has occurred.
2785  **/
2786 static void ixgbevf_check_hang_subtask(struct ixgbevf_adapter *adapter)
2787 {
2788         struct ixgbe_hw *hw = &adapter->hw;
2789         u32 eics = 0;
2790         int i;
2791
2792         /* If we're down or resetting, just bail */
2793         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2794             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2795                 return;
2796
2797         /* Force detection of hung controller */
2798         if (netif_carrier_ok(adapter->netdev)) {
2799                 for (i = 0; i < adapter->num_tx_queues; i++)
2800                         set_check_for_tx_hang(adapter->tx_ring[i]);
2801         }
2802
2803         /* get one bit for every active Tx/Rx interrupt vector */
2804         for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
2805                 struct ixgbevf_q_vector *qv = adapter->q_vector[i];
2806
2807                 if (qv->rx.ring || qv->tx.ring)
2808                         eics |= 1 << i;
2809         }
2810
2811         /* Cause software interrupt to ensure rings are cleaned */
2812         IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
2813 }
2814
2815 /**
2816  * ixgbevf_watchdog_update_link - update the link status
2817  * @adapter: pointer to the device adapter structure
2818  **/
2819 static void ixgbevf_watchdog_update_link(struct ixgbevf_adapter *adapter)
2820 {
2821         struct ixgbe_hw *hw = &adapter->hw;
2822         u32 link_speed = adapter->link_speed;
2823         bool link_up = adapter->link_up;
2824         s32 err;
2825
2826         spin_lock_bh(&adapter->mbx_lock);
2827
2828         err = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
2829
2830         spin_unlock_bh(&adapter->mbx_lock);
2831
2832         /* if check for link returns error we will need to reset */
2833         if (err && time_after(jiffies, adapter->last_reset + (10 * HZ))) {
2834                 adapter->flags |= IXGBEVF_FLAG_RESET_REQUESTED;
2835                 link_up = false;
2836         }
2837
2838         adapter->link_up = link_up;
2839         adapter->link_speed = link_speed;
2840 }
2841
2842 /**
2843  * ixgbevf_watchdog_link_is_up - update netif_carrier status and
2844  *                               print link up message
2845  * @adapter: pointer to the device adapter structure
2846  **/
2847 static void ixgbevf_watchdog_link_is_up(struct ixgbevf_adapter *adapter)
2848 {
2849         struct net_device *netdev = adapter->netdev;
2850
2851         /* only continue if link was previously down */
2852         if (netif_carrier_ok(netdev))
2853                 return;
2854
2855         dev_info(&adapter->pdev->dev, "NIC Link is Up %s\n",
2856                  (adapter->link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
2857                  "10 Gbps" :
2858                  (adapter->link_speed == IXGBE_LINK_SPEED_1GB_FULL) ?
2859                  "1 Gbps" :
2860                  (adapter->link_speed == IXGBE_LINK_SPEED_100_FULL) ?
2861                  "100 Mbps" :
2862                  "unknown speed");
2863
2864         netif_carrier_on(netdev);
2865 }
2866
2867 /**
2868  * ixgbevf_watchdog_link_is_down - update netif_carrier status and
2869  *                                 print link down message
2870  * @adapter: pointer to the adapter structure
2871  **/
2872 static void ixgbevf_watchdog_link_is_down(struct ixgbevf_adapter *adapter)
2873 {
2874         struct net_device *netdev = adapter->netdev;
2875
2876         adapter->link_speed = 0;
2877
2878         /* only continue if link was up previously */
2879         if (!netif_carrier_ok(netdev))
2880                 return;
2881
2882         dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
2883
2884         netif_carrier_off(netdev);
2885 }
2886
2887 /**
2888  * ixgbevf_watchdog_subtask - worker thread to bring link up
2889  * @work: pointer to work_struct containing our data
2890  **/
2891 static void ixgbevf_watchdog_subtask(struct ixgbevf_adapter *adapter)
2892 {
2893         /* if interface is down do nothing */
2894         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2895             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2896                 return;
2897
2898         ixgbevf_watchdog_update_link(adapter);
2899
2900         if (adapter->link_up)
2901                 ixgbevf_watchdog_link_is_up(adapter);
2902         else
2903                 ixgbevf_watchdog_link_is_down(adapter);
2904
2905         ixgbevf_update_stats(adapter);
2906 }
2907
2908 /**
2909  * ixgbevf_service_task - manages and runs subtasks
2910  * @work: pointer to work_struct containing our data
2911  **/
2912 static void ixgbevf_service_task(struct work_struct *work)
2913 {
2914         struct ixgbevf_adapter *adapter = container_of(work,
2915                                                        struct ixgbevf_adapter,
2916                                                        service_task);
2917         struct ixgbe_hw *hw = &adapter->hw;
2918
2919         if (IXGBE_REMOVED(hw->hw_addr)) {
2920                 if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
2921                         rtnl_lock();
2922                         ixgbevf_down(adapter);
2923                         rtnl_unlock();
2924                 }
2925                 return;
2926         }
2927
2928         ixgbevf_queue_reset_subtask(adapter);
2929         ixgbevf_reset_subtask(adapter);
2930         ixgbevf_watchdog_subtask(adapter);
2931         ixgbevf_check_hang_subtask(adapter);
2932
2933         ixgbevf_service_event_complete(adapter);
2934 }
2935
2936 /**
2937  * ixgbevf_free_tx_resources - Free Tx Resources per Queue
2938  * @tx_ring: Tx descriptor ring for a specific queue
2939  *
2940  * Free all transmit software resources
2941  **/
2942 void ixgbevf_free_tx_resources(struct ixgbevf_ring *tx_ring)
2943 {
2944         ixgbevf_clean_tx_ring(tx_ring);
2945
2946         vfree(tx_ring->tx_buffer_info);
2947         tx_ring->tx_buffer_info = NULL;
2948
2949         /* if not set, then don't free */
2950         if (!tx_ring->desc)
2951                 return;
2952
2953         dma_free_coherent(tx_ring->dev, tx_ring->size, tx_ring->desc,
2954                           tx_ring->dma);
2955
2956         tx_ring->desc = NULL;
2957 }
2958
2959 /**
2960  * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
2961  * @adapter: board private structure
2962  *
2963  * Free all transmit software resources
2964  **/
2965 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
2966 {
2967         int i;
2968
2969         for (i = 0; i < adapter->num_tx_queues; i++)
2970                 if (adapter->tx_ring[i]->desc)
2971                         ixgbevf_free_tx_resources(adapter->tx_ring[i]);
2972 }
2973
2974 /**
2975  * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
2976  * @tx_ring: Tx descriptor ring (for a specific queue) to setup
2977  *
2978  * Return 0 on success, negative on failure
2979  **/
2980 int ixgbevf_setup_tx_resources(struct ixgbevf_ring *tx_ring)
2981 {
2982         int size;
2983
2984         size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2985         tx_ring->tx_buffer_info = vzalloc(size);
2986         if (!tx_ring->tx_buffer_info)
2987                 goto err;
2988
2989         /* round up to nearest 4K */
2990         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
2991         tx_ring->size = ALIGN(tx_ring->size, 4096);
2992
2993         tx_ring->desc = dma_alloc_coherent(tx_ring->dev, tx_ring->size,
2994                                            &tx_ring->dma, GFP_KERNEL);
2995         if (!tx_ring->desc)
2996                 goto err;
2997
2998         return 0;
2999
3000 err:
3001         vfree(tx_ring->tx_buffer_info);
3002         tx_ring->tx_buffer_info = NULL;
3003         hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit descriptor ring\n");
3004         return -ENOMEM;
3005 }
3006
3007 /**
3008  * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
3009  * @adapter: board private structure
3010  *
3011  * If this function returns with an error, then it's possible one or
3012  * more of the rings is populated (while the rest are not).  It is the
3013  * callers duty to clean those orphaned rings.
3014  *
3015  * Return 0 on success, negative on failure
3016  **/
3017 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
3018 {
3019         int i, err = 0;
3020
3021         for (i = 0; i < adapter->num_tx_queues; i++) {
3022                 err = ixgbevf_setup_tx_resources(adapter->tx_ring[i]);
3023                 if (!err)
3024                         continue;
3025                 hw_dbg(&adapter->hw, "Allocation for Tx Queue %u failed\n", i);
3026                 break;
3027         }
3028
3029         return err;
3030 }
3031
3032 /**
3033  * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
3034  * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3035  *
3036  * Returns 0 on success, negative on failure
3037  **/
3038 int ixgbevf_setup_rx_resources(struct ixgbevf_ring *rx_ring)
3039 {
3040         int size;
3041
3042         size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
3043         rx_ring->rx_buffer_info = vzalloc(size);
3044         if (!rx_ring->rx_buffer_info)
3045                 goto err;
3046
3047         /* Round up to nearest 4K */
3048         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3049         rx_ring->size = ALIGN(rx_ring->size, 4096);
3050
3051         rx_ring->desc = dma_alloc_coherent(rx_ring->dev, rx_ring->size,
3052                                            &rx_ring->dma, GFP_KERNEL);
3053
3054         if (!rx_ring->desc)
3055                 goto err;
3056
3057         return 0;
3058 err:
3059         vfree(rx_ring->rx_buffer_info);
3060         rx_ring->rx_buffer_info = NULL;
3061         dev_err(rx_ring->dev, "Unable to allocate memory for the Rx descriptor ring\n");
3062         return -ENOMEM;
3063 }
3064
3065 /**
3066  * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
3067  * @adapter: board private structure
3068  *
3069  * If this function returns with an error, then it's possible one or
3070  * more of the rings is populated (while the rest are not).  It is the
3071  * callers duty to clean those orphaned rings.
3072  *
3073  * Return 0 on success, negative on failure
3074  **/
3075 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
3076 {
3077         int i, err = 0;
3078
3079         for (i = 0; i < adapter->num_rx_queues; i++) {
3080                 err = ixgbevf_setup_rx_resources(adapter->rx_ring[i]);
3081                 if (!err)
3082                         continue;
3083                 hw_dbg(&adapter->hw, "Allocation for Rx Queue %u failed\n", i);
3084                 break;
3085         }
3086         return err;
3087 }
3088
3089 /**
3090  * ixgbevf_free_rx_resources - Free Rx Resources
3091  * @rx_ring: ring to clean the resources from
3092  *
3093  * Free all receive software resources
3094  **/
3095 void ixgbevf_free_rx_resources(struct ixgbevf_ring *rx_ring)
3096 {
3097         ixgbevf_clean_rx_ring(rx_ring);
3098
3099         vfree(rx_ring->rx_buffer_info);
3100         rx_ring->rx_buffer_info = NULL;
3101
3102         dma_free_coherent(rx_ring->dev, rx_ring->size, rx_ring->desc,
3103                           rx_ring->dma);
3104
3105         rx_ring->desc = NULL;
3106 }
3107
3108 /**
3109  * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
3110  * @adapter: board private structure
3111  *
3112  * Free all receive software resources
3113  **/
3114 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
3115 {
3116         int i;
3117
3118         for (i = 0; i < adapter->num_rx_queues; i++)
3119                 if (adapter->rx_ring[i]->desc)
3120                         ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3121 }
3122
3123 /**
3124  * ixgbevf_open - Called when a network interface is made active
3125  * @netdev: network interface device structure
3126  *
3127  * Returns 0 on success, negative value on failure
3128  *
3129  * The open entry point is called when a network interface is made
3130  * active by the system (IFF_UP).  At this point all resources needed
3131  * for transmit and receive operations are allocated, the interrupt
3132  * handler is registered with the OS, the watchdog timer is started,
3133  * and the stack is notified that the interface is ready.
3134  **/
3135 static int ixgbevf_open(struct net_device *netdev)
3136 {
3137         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3138         struct ixgbe_hw *hw = &adapter->hw;
3139         int err;
3140
3141         /* A previous failure to open the device because of a lack of
3142          * available MSIX vector resources may have reset the number
3143          * of msix vectors variable to zero.  The only way to recover
3144          * is to unload/reload the driver and hope that the system has
3145          * been able to recover some MSIX vector resources.
3146          */
3147         if (!adapter->num_msix_vectors)
3148                 return -ENOMEM;
3149
3150         if (hw->adapter_stopped) {
3151                 ixgbevf_reset(adapter);
3152                 /* if adapter is still stopped then PF isn't up and
3153                  * the VF can't start.
3154                  */
3155                 if (hw->adapter_stopped) {
3156                         err = IXGBE_ERR_MBX;
3157                         pr_err("Unable to start - perhaps the PF Driver isn't up yet\n");
3158                         goto err_setup_reset;
3159                 }
3160         }
3161
3162         /* disallow open during test */
3163         if (test_bit(__IXGBEVF_TESTING, &adapter->state))
3164                 return -EBUSY;
3165
3166         netif_carrier_off(netdev);
3167
3168         /* allocate transmit descriptors */
3169         err = ixgbevf_setup_all_tx_resources(adapter);
3170         if (err)
3171                 goto err_setup_tx;
3172
3173         /* allocate receive descriptors */
3174         err = ixgbevf_setup_all_rx_resources(adapter);
3175         if (err)
3176                 goto err_setup_rx;
3177
3178         ixgbevf_configure(adapter);
3179
3180         /* Map the Tx/Rx rings to the vectors we were allotted.
3181          * if request_irq will be called in this function map_rings
3182          * must be called *before* up_complete
3183          */
3184         ixgbevf_map_rings_to_vectors(adapter);
3185
3186         err = ixgbevf_request_irq(adapter);
3187         if (err)
3188                 goto err_req_irq;
3189
3190         ixgbevf_up_complete(adapter);
3191
3192         return 0;
3193
3194 err_req_irq:
3195         ixgbevf_down(adapter);
3196 err_setup_rx:
3197         ixgbevf_free_all_rx_resources(adapter);
3198 err_setup_tx:
3199         ixgbevf_free_all_tx_resources(adapter);
3200         ixgbevf_reset(adapter);
3201
3202 err_setup_reset:
3203
3204         return err;
3205 }
3206
3207 /**
3208  * ixgbevf_close - Disables a network interface
3209  * @netdev: network interface device structure
3210  *
3211  * Returns 0, this is not allowed to fail
3212  *
3213  * The close entry point is called when an interface is de-activated
3214  * by the OS.  The hardware is still under the drivers control, but
3215  * needs to be disabled.  A global MAC reset is issued to stop the
3216  * hardware, and all transmit and receive resources are freed.
3217  **/
3218 static int ixgbevf_close(struct net_device *netdev)
3219 {
3220         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3221
3222         ixgbevf_down(adapter);
3223         ixgbevf_free_irq(adapter);
3224
3225         ixgbevf_free_all_tx_resources(adapter);
3226         ixgbevf_free_all_rx_resources(adapter);
3227
3228         return 0;
3229 }
3230
3231 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter)
3232 {
3233         struct net_device *dev = adapter->netdev;
3234
3235         if (!(adapter->flags & IXGBEVF_FLAG_QUEUE_RESET_REQUESTED))
3236                 return;
3237
3238         adapter->flags &= ~IXGBEVF_FLAG_QUEUE_RESET_REQUESTED;
3239
3240         /* if interface is down do nothing */
3241         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3242             test_bit(__IXGBEVF_RESETTING, &adapter->state))
3243                 return;
3244
3245         /* Hardware has to reinitialize queues and interrupts to
3246          * match packet buffer alignment. Unfortunately, the
3247          * hardware is not flexible enough to do this dynamically.
3248          */
3249         if (netif_running(dev))
3250                 ixgbevf_close(dev);
3251
3252         ixgbevf_clear_interrupt_scheme(adapter);
3253         ixgbevf_init_interrupt_scheme(adapter);
3254
3255         if (netif_running(dev))
3256                 ixgbevf_open(dev);
3257 }
3258
3259 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
3260                                 u32 vlan_macip_lens, u32 type_tucmd,
3261                                 u32 mss_l4len_idx)
3262 {
3263         struct ixgbe_adv_tx_context_desc *context_desc;
3264         u16 i = tx_ring->next_to_use;
3265
3266         context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
3267
3268         i++;
3269         tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3270
3271         /* set bits to identify this as an advanced context descriptor */
3272         type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
3273
3274         context_desc->vlan_macip_lens   = cpu_to_le32(vlan_macip_lens);
3275         context_desc->seqnum_seed       = 0;
3276         context_desc->type_tucmd_mlhl   = cpu_to_le32(type_tucmd);
3277         context_desc->mss_l4len_idx     = cpu_to_le32(mss_l4len_idx);
3278 }
3279
3280 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
3281                        struct ixgbevf_tx_buffer *first,
3282                        u8 *hdr_len)
3283 {
3284         struct sk_buff *skb = first->skb;
3285         u32 vlan_macip_lens, type_tucmd;
3286         u32 mss_l4len_idx, l4len;
3287         int err;
3288
3289         if (skb->ip_summed != CHECKSUM_PARTIAL)
3290                 return 0;
3291
3292         if (!skb_is_gso(skb))
3293                 return 0;
3294
3295         err = skb_cow_head(skb, 0);
3296         if (err < 0)
3297                 return err;
3298
3299         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3300         type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3301
3302         if (first->protocol == htons(ETH_P_IP)) {
3303                 struct iphdr *iph = ip_hdr(skb);
3304
3305                 iph->tot_len = 0;
3306                 iph->check = 0;
3307                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3308                                                          iph->daddr, 0,
3309                                                          IPPROTO_TCP,
3310                                                          0);
3311                 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3312                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3313                                    IXGBE_TX_FLAGS_CSUM |
3314                                    IXGBE_TX_FLAGS_IPV4;
3315         } else if (skb_is_gso_v6(skb)) {
3316                 ipv6_hdr(skb)->payload_len = 0;
3317                 tcp_hdr(skb)->check =
3318                     ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3319                                      &ipv6_hdr(skb)->daddr,
3320                                      0, IPPROTO_TCP, 0);
3321                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3322                                    IXGBE_TX_FLAGS_CSUM;
3323         }
3324
3325         /* compute header lengths */
3326         l4len = tcp_hdrlen(skb);
3327         *hdr_len += l4len;
3328         *hdr_len = skb_transport_offset(skb) + l4len;
3329
3330         /* update GSO size and bytecount with header size */
3331         first->gso_segs = skb_shinfo(skb)->gso_segs;
3332         first->bytecount += (first->gso_segs - 1) * *hdr_len;
3333
3334         /* mss_l4len_id: use 1 as index for TSO */
3335         mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
3336         mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
3337         mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
3338
3339         /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
3340         vlan_macip_lens = skb_network_header_len(skb);
3341         vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3342         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3343
3344         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3345                             type_tucmd, mss_l4len_idx);
3346
3347         return 1;
3348 }
3349
3350 static void ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
3351                             struct ixgbevf_tx_buffer *first)
3352 {
3353         struct sk_buff *skb = first->skb;
3354         u32 vlan_macip_lens = 0;
3355         u32 mss_l4len_idx = 0;
3356         u32 type_tucmd = 0;
3357
3358         if (skb->ip_summed == CHECKSUM_PARTIAL) {
3359                 u8 l4_hdr = 0;
3360
3361                 switch (first->protocol) {
3362                 case htons(ETH_P_IP):
3363                         vlan_macip_lens |= skb_network_header_len(skb);
3364                         type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3365                         l4_hdr = ip_hdr(skb)->protocol;
3366                         break;
3367                 case htons(ETH_P_IPV6):
3368                         vlan_macip_lens |= skb_network_header_len(skb);
3369                         l4_hdr = ipv6_hdr(skb)->nexthdr;
3370                         break;
3371                 default:
3372                         if (unlikely(net_ratelimit())) {
3373                                 dev_warn(tx_ring->dev,
3374                                          "partial checksum but proto=%x!\n",
3375                                          first->protocol);
3376                         }
3377                         break;
3378                 }
3379
3380                 switch (l4_hdr) {
3381                 case IPPROTO_TCP:
3382                         type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
3383                         mss_l4len_idx = tcp_hdrlen(skb) <<
3384                                         IXGBE_ADVTXD_L4LEN_SHIFT;
3385                         break;
3386                 case IPPROTO_SCTP:
3387                         type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
3388                         mss_l4len_idx = sizeof(struct sctphdr) <<
3389                                         IXGBE_ADVTXD_L4LEN_SHIFT;
3390                         break;
3391                 case IPPROTO_UDP:
3392                         mss_l4len_idx = sizeof(struct udphdr) <<
3393                                         IXGBE_ADVTXD_L4LEN_SHIFT;
3394                         break;
3395                 default:
3396                         if (unlikely(net_ratelimit())) {
3397                                 dev_warn(tx_ring->dev,
3398                                          "partial checksum but l4 proto=%x!\n",
3399                                          l4_hdr);
3400                         }
3401                         break;
3402                 }
3403
3404                 /* update TX checksum flag */
3405                 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
3406         }
3407
3408         /* vlan_macip_lens: MACLEN, VLAN tag */
3409         vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3410         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3411
3412         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3413                             type_tucmd, mss_l4len_idx);
3414 }
3415
3416 static __le32 ixgbevf_tx_cmd_type(u32 tx_flags)
3417 {
3418         /* set type for advanced descriptor with frame checksum insertion */
3419         __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
3420                                       IXGBE_ADVTXD_DCMD_IFCS |
3421                                       IXGBE_ADVTXD_DCMD_DEXT);
3422
3423         /* set HW VLAN bit if VLAN is present */
3424         if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3425                 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
3426
3427         /* set segmentation enable bits for TSO/FSO */
3428         if (tx_flags & IXGBE_TX_FLAGS_TSO)
3429                 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
3430
3431         return cmd_type;
3432 }
3433
3434 static void ixgbevf_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
3435                                      u32 tx_flags, unsigned int paylen)
3436 {
3437         __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
3438
3439         /* enable L4 checksum for TSO and TX checksum offload */
3440         if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3441                 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
3442
3443         /* enble IPv4 checksum for TSO */
3444         if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3445                 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
3446
3447         /* use index 1 context for TSO/FSO/FCOE */
3448         if (tx_flags & IXGBE_TX_FLAGS_TSO)
3449                 olinfo_status |= cpu_to_le32(1 << IXGBE_ADVTXD_IDX_SHIFT);
3450
3451         /* Check Context must be set if Tx switch is enabled, which it
3452          * always is for case where virtual functions are running
3453          */
3454         olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
3455
3456         tx_desc->read.olinfo_status = olinfo_status;
3457 }
3458
3459 static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
3460                            struct ixgbevf_tx_buffer *first,
3461                            const u8 hdr_len)
3462 {
3463         dma_addr_t dma;
3464         struct sk_buff *skb = first->skb;
3465         struct ixgbevf_tx_buffer *tx_buffer;
3466         union ixgbe_adv_tx_desc *tx_desc;
3467         struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
3468         unsigned int data_len = skb->data_len;
3469         unsigned int size = skb_headlen(skb);
3470         unsigned int paylen = skb->len - hdr_len;
3471         u32 tx_flags = first->tx_flags;
3472         __le32 cmd_type;
3473         u16 i = tx_ring->next_to_use;
3474
3475         tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
3476
3477         ixgbevf_tx_olinfo_status(tx_desc, tx_flags, paylen);
3478         cmd_type = ixgbevf_tx_cmd_type(tx_flags);
3479
3480         dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3481         if (dma_mapping_error(tx_ring->dev, dma))
3482                 goto dma_error;
3483
3484         /* record length, and DMA address */
3485         dma_unmap_len_set(first, len, size);
3486         dma_unmap_addr_set(first, dma, dma);
3487
3488         tx_desc->read.buffer_addr = cpu_to_le64(dma);
3489
3490         for (;;) {
3491                 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
3492                         tx_desc->read.cmd_type_len =
3493                                 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
3494
3495                         i++;
3496                         tx_desc++;
3497                         if (i == tx_ring->count) {
3498                                 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3499                                 i = 0;
3500                         }
3501
3502                         dma += IXGBE_MAX_DATA_PER_TXD;
3503                         size -= IXGBE_MAX_DATA_PER_TXD;
3504
3505                         tx_desc->read.buffer_addr = cpu_to_le64(dma);
3506                         tx_desc->read.olinfo_status = 0;
3507                 }
3508
3509                 if (likely(!data_len))
3510                         break;
3511
3512                 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
3513
3514                 i++;
3515                 tx_desc++;
3516                 if (i == tx_ring->count) {
3517                         tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3518                         i = 0;
3519                 }
3520
3521                 size = skb_frag_size(frag);
3522                 data_len -= size;
3523
3524                 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
3525                                        DMA_TO_DEVICE);
3526                 if (dma_mapping_error(tx_ring->dev, dma))
3527                         goto dma_error;
3528
3529                 tx_buffer = &tx_ring->tx_buffer_info[i];
3530                 dma_unmap_len_set(tx_buffer, len, size);
3531                 dma_unmap_addr_set(tx_buffer, dma, dma);
3532
3533                 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3534                 tx_desc->read.olinfo_status = 0;
3535
3536                 frag++;
3537         }
3538
3539         /* write last descriptor with RS and EOP bits */
3540         cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
3541         tx_desc->read.cmd_type_len = cmd_type;
3542
3543         /* set the timestamp */
3544         first->time_stamp = jiffies;
3545
3546         /* Force memory writes to complete before letting h/w know there
3547          * are new descriptors to fetch.  (Only applicable for weak-ordered
3548          * memory model archs, such as IA-64).
3549          *
3550          * We also need this memory barrier (wmb) to make certain all of the
3551          * status bits have been updated before next_to_watch is written.
3552          */
3553         wmb();
3554
3555         /* set next_to_watch value indicating a packet is present */
3556         first->next_to_watch = tx_desc;
3557
3558         i++;
3559         if (i == tx_ring->count)
3560                 i = 0;
3561
3562         tx_ring->next_to_use = i;
3563
3564         /* notify HW of packet */
3565         ixgbevf_write_tail(tx_ring, i);
3566
3567         return;
3568 dma_error:
3569         dev_err(tx_ring->dev, "TX DMA map failed\n");
3570
3571         /* clear dma mappings for failed tx_buffer_info map */
3572         for (;;) {
3573                 tx_buffer = &tx_ring->tx_buffer_info[i];
3574                 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer);
3575                 if (tx_buffer == first)
3576                         break;
3577                 if (i == 0)
3578                         i = tx_ring->count;
3579                 i--;
3580         }
3581
3582         tx_ring->next_to_use = i;
3583 }
3584
3585 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3586 {
3587         netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
3588         /* Herbert's original patch had:
3589          *  smp_mb__after_netif_stop_queue();
3590          * but since that doesn't exist yet, just open code it.
3591          */
3592         smp_mb();
3593
3594         /* We need to check again in a case another CPU has just
3595          * made room available.
3596          */
3597         if (likely(ixgbevf_desc_unused(tx_ring) < size))
3598                 return -EBUSY;
3599
3600         /* A reprieve! - use start_queue because it doesn't call schedule */
3601         netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
3602         ++tx_ring->tx_stats.restart_queue;
3603
3604         return 0;
3605 }
3606
3607 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3608 {
3609         if (likely(ixgbevf_desc_unused(tx_ring) >= size))
3610                 return 0;
3611         return __ixgbevf_maybe_stop_tx(tx_ring, size);
3612 }
3613
3614 static int ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3615 {
3616         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3617         struct ixgbevf_tx_buffer *first;
3618         struct ixgbevf_ring *tx_ring;
3619         int tso;
3620         u32 tx_flags = 0;
3621         u16 count = TXD_USE_COUNT(skb_headlen(skb));
3622 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3623         unsigned short f;
3624 #endif
3625         u8 hdr_len = 0;
3626         u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
3627
3628         if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
3629                 dev_kfree_skb_any(skb);
3630                 return NETDEV_TX_OK;
3631         }
3632
3633         tx_ring = adapter->tx_ring[skb->queue_mapping];
3634
3635         /* need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
3636          *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
3637          *       + 2 desc gap to keep tail from touching head,
3638          *       + 1 desc for context descriptor,
3639          * otherwise try next time
3640          */
3641 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3642         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
3643                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
3644 #else
3645         count += skb_shinfo(skb)->nr_frags;
3646 #endif
3647         if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
3648                 tx_ring->tx_stats.tx_busy++;
3649                 return NETDEV_TX_BUSY;
3650         }
3651
3652         /* record the location of the first descriptor for this packet */
3653         first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
3654         first->skb = skb;
3655         first->bytecount = skb->len;
3656         first->gso_segs = 1;
3657
3658         if (skb_vlan_tag_present(skb)) {
3659                 tx_flags |= skb_vlan_tag_get(skb);
3660                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
3661                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
3662         }
3663
3664         /* record initial flags and protocol */
3665         first->tx_flags = tx_flags;
3666         first->protocol = vlan_get_protocol(skb);
3667
3668         tso = ixgbevf_tso(tx_ring, first, &hdr_len);
3669         if (tso < 0)
3670                 goto out_drop;
3671         else if (!tso)
3672                 ixgbevf_tx_csum(tx_ring, first);
3673
3674         ixgbevf_tx_map(tx_ring, first, hdr_len);
3675
3676         ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
3677
3678         return NETDEV_TX_OK;
3679
3680 out_drop:
3681         dev_kfree_skb_any(first->skb);
3682         first->skb = NULL;
3683
3684         return NETDEV_TX_OK;
3685 }
3686
3687 /**
3688  * ixgbevf_set_mac - Change the Ethernet Address of the NIC
3689  * @netdev: network interface device structure
3690  * @p: pointer to an address structure
3691  *
3692  * Returns 0 on success, negative on failure
3693  **/
3694 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
3695 {
3696         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3697         struct ixgbe_hw *hw = &adapter->hw;
3698         struct sockaddr *addr = p;
3699
3700         if (!is_valid_ether_addr(addr->sa_data))
3701                 return -EADDRNOTAVAIL;
3702
3703         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3704         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3705
3706         spin_lock_bh(&adapter->mbx_lock);
3707
3708         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
3709
3710         spin_unlock_bh(&adapter->mbx_lock);
3711
3712         return 0;
3713 }
3714
3715 /**
3716  * ixgbevf_change_mtu - Change the Maximum Transfer Unit
3717  * @netdev: network interface device structure
3718  * @new_mtu: new value for maximum frame size
3719  *
3720  * Returns 0 on success, negative on failure
3721  **/
3722 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
3723 {
3724         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3725         struct ixgbe_hw *hw = &adapter->hw;
3726         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3727         int max_possible_frame = MAXIMUM_ETHERNET_VLAN_SIZE;
3728
3729         switch (adapter->hw.api_version) {
3730         case ixgbe_mbox_api_11:
3731         case ixgbe_mbox_api_12:
3732                 max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3733                 break;
3734         default:
3735                 if (adapter->hw.mac.type != ixgbe_mac_82599_vf)
3736                         max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3737                 break;
3738         }
3739
3740         /* MTU < 68 is an error and causes problems on some kernels */
3741         if ((new_mtu < 68) || (max_frame > max_possible_frame))
3742                 return -EINVAL;
3743
3744         hw_dbg(hw, "changing MTU from %d to %d\n",
3745                netdev->mtu, new_mtu);
3746         /* must set new MTU before calling down or up */
3747         netdev->mtu = new_mtu;
3748
3749         /* notify the PF of our intent to use this size of frame */
3750         ixgbevf_rlpml_set_vf(hw, max_frame);
3751
3752         return 0;
3753 }
3754
3755 #ifdef CONFIG_NET_POLL_CONTROLLER
3756 /* Polling 'interrupt' - used by things like netconsole to send skbs
3757  * without having to re-enable interrupts. It's not called while
3758  * the interrupt routine is executing.
3759  */
3760 static void ixgbevf_netpoll(struct net_device *netdev)
3761 {
3762         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3763         int i;
3764
3765         /* if interface is down do nothing */
3766         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
3767                 return;
3768         for (i = 0; i < adapter->num_rx_queues; i++)
3769                 ixgbevf_msix_clean_rings(0, adapter->q_vector[i]);
3770 }
3771 #endif /* CONFIG_NET_POLL_CONTROLLER */
3772
3773 static int ixgbevf_suspend(struct pci_dev *pdev, pm_message_t state)
3774 {
3775         struct net_device *netdev = pci_get_drvdata(pdev);
3776         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3777 #ifdef CONFIG_PM
3778         int retval = 0;
3779 #endif
3780
3781         netif_device_detach(netdev);
3782
3783         if (netif_running(netdev)) {
3784                 rtnl_lock();
3785                 ixgbevf_down(adapter);
3786                 ixgbevf_free_irq(adapter);
3787                 ixgbevf_free_all_tx_resources(adapter);
3788                 ixgbevf_free_all_rx_resources(adapter);
3789                 rtnl_unlock();
3790         }
3791
3792         ixgbevf_clear_interrupt_scheme(adapter);
3793
3794 #ifdef CONFIG_PM
3795         retval = pci_save_state(pdev);
3796         if (retval)
3797                 return retval;
3798
3799 #endif
3800         if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
3801                 pci_disable_device(pdev);
3802
3803         return 0;
3804 }
3805
3806 #ifdef CONFIG_PM
3807 static int ixgbevf_resume(struct pci_dev *pdev)
3808 {
3809         struct net_device *netdev = pci_get_drvdata(pdev);
3810         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3811         u32 err;
3812
3813         pci_restore_state(pdev);
3814         /* pci_restore_state clears dev->state_saved so call
3815          * pci_save_state to restore it.
3816          */
3817         pci_save_state(pdev);
3818
3819         err = pci_enable_device_mem(pdev);
3820         if (err) {
3821                 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
3822                 return err;
3823         }
3824         smp_mb__before_atomic();
3825         clear_bit(__IXGBEVF_DISABLED, &adapter->state);
3826         pci_set_master(pdev);
3827
3828         ixgbevf_reset(adapter);
3829
3830         rtnl_lock();
3831         err = ixgbevf_init_interrupt_scheme(adapter);
3832         rtnl_unlock();
3833         if (err) {
3834                 dev_err(&pdev->dev, "Cannot initialize interrupts\n");
3835                 return err;
3836         }
3837
3838         if (netif_running(netdev)) {
3839                 err = ixgbevf_open(netdev);
3840                 if (err)
3841                         return err;
3842         }
3843
3844         netif_device_attach(netdev);
3845
3846         return err;
3847 }
3848
3849 #endif /* CONFIG_PM */
3850 static void ixgbevf_shutdown(struct pci_dev *pdev)
3851 {
3852         ixgbevf_suspend(pdev, PMSG_SUSPEND);
3853 }
3854
3855 static struct rtnl_link_stats64 *ixgbevf_get_stats(struct net_device *netdev,
3856                                                 struct rtnl_link_stats64 *stats)
3857 {
3858         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3859         unsigned int start;
3860         u64 bytes, packets;
3861         const struct ixgbevf_ring *ring;
3862         int i;
3863
3864         ixgbevf_update_stats(adapter);
3865
3866         stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
3867
3868         for (i = 0; i < adapter->num_rx_queues; i++) {
3869                 ring = adapter->rx_ring[i];
3870                 do {
3871                         start = u64_stats_fetch_begin_irq(&ring->syncp);
3872                         bytes = ring->stats.bytes;
3873                         packets = ring->stats.packets;
3874                 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3875                 stats->rx_bytes += bytes;
3876                 stats->rx_packets += packets;
3877         }
3878
3879         for (i = 0; i < adapter->num_tx_queues; i++) {
3880                 ring = adapter->tx_ring[i];
3881                 do {
3882                         start = u64_stats_fetch_begin_irq(&ring->syncp);
3883                         bytes = ring->stats.bytes;
3884                         packets = ring->stats.packets;
3885                 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
3886                 stats->tx_bytes += bytes;
3887                 stats->tx_packets += packets;
3888         }
3889
3890         return stats;
3891 }
3892
3893 static const struct net_device_ops ixgbevf_netdev_ops = {
3894         .ndo_open               = ixgbevf_open,
3895         .ndo_stop               = ixgbevf_close,
3896         .ndo_start_xmit         = ixgbevf_xmit_frame,
3897         .ndo_set_rx_mode        = ixgbevf_set_rx_mode,
3898         .ndo_get_stats64        = ixgbevf_get_stats,
3899         .ndo_validate_addr      = eth_validate_addr,
3900         .ndo_set_mac_address    = ixgbevf_set_mac,
3901         .ndo_change_mtu         = ixgbevf_change_mtu,
3902         .ndo_tx_timeout         = ixgbevf_tx_timeout,
3903         .ndo_vlan_rx_add_vid    = ixgbevf_vlan_rx_add_vid,
3904         .ndo_vlan_rx_kill_vid   = ixgbevf_vlan_rx_kill_vid,
3905 #ifdef CONFIG_NET_RX_BUSY_POLL
3906         .ndo_busy_poll          = ixgbevf_busy_poll_recv,
3907 #endif
3908 #ifdef CONFIG_NET_POLL_CONTROLLER
3909         .ndo_poll_controller    = ixgbevf_netpoll,
3910 #endif
3911         .ndo_features_check     = passthru_features_check,
3912 };
3913
3914 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
3915 {
3916         dev->netdev_ops = &ixgbevf_netdev_ops;
3917         ixgbevf_set_ethtool_ops(dev);
3918         dev->watchdog_timeo = 5 * HZ;
3919 }
3920
3921 /**
3922  * ixgbevf_probe - Device Initialization Routine
3923  * @pdev: PCI device information struct
3924  * @ent: entry in ixgbevf_pci_tbl
3925  *
3926  * Returns 0 on success, negative on failure
3927  *
3928  * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
3929  * The OS initialization, configuring of the adapter private structure,
3930  * and a hardware reset occur.
3931  **/
3932 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3933 {
3934         struct net_device *netdev;
3935         struct ixgbevf_adapter *adapter = NULL;
3936         struct ixgbe_hw *hw = NULL;
3937         const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
3938         int err, pci_using_dac;
3939         bool disable_dev = false;
3940
3941         err = pci_enable_device(pdev);
3942         if (err)
3943                 return err;
3944
3945         if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
3946                 pci_using_dac = 1;
3947         } else {
3948                 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
3949                 if (err) {
3950                         dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
3951                         goto err_dma;
3952                 }
3953                 pci_using_dac = 0;
3954         }
3955
3956         err = pci_request_regions(pdev, ixgbevf_driver_name);
3957         if (err) {
3958                 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
3959                 goto err_pci_reg;
3960         }
3961
3962         pci_set_master(pdev);
3963
3964         netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
3965                                    MAX_TX_QUEUES);
3966         if (!netdev) {
3967                 err = -ENOMEM;
3968                 goto err_alloc_etherdev;
3969         }
3970
3971         SET_NETDEV_DEV(netdev, &pdev->dev);
3972
3973         adapter = netdev_priv(netdev);
3974
3975         adapter->netdev = netdev;
3976         adapter->pdev = pdev;
3977         hw = &adapter->hw;
3978         hw->back = adapter;
3979         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3980
3981         /* call save state here in standalone driver because it relies on
3982          * adapter struct to exist, and needs to call netdev_priv
3983          */
3984         pci_save_state(pdev);
3985
3986         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
3987                               pci_resource_len(pdev, 0));
3988         adapter->io_addr = hw->hw_addr;
3989         if (!hw->hw_addr) {
3990                 err = -EIO;
3991                 goto err_ioremap;
3992         }
3993
3994         ixgbevf_assign_netdev_ops(netdev);
3995
3996         /* Setup HW API */
3997         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
3998         hw->mac.type  = ii->mac;
3999
4000         memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
4001                sizeof(struct ixgbe_mbx_operations));
4002
4003         /* setup the private structure */
4004         err = ixgbevf_sw_init(adapter);
4005         if (err)
4006                 goto err_sw_init;
4007
4008         /* The HW MAC address was set and/or determined in sw_init */
4009         if (!is_valid_ether_addr(netdev->dev_addr)) {
4010                 pr_err("invalid MAC address\n");
4011                 err = -EIO;
4012                 goto err_sw_init;
4013         }
4014
4015         netdev->hw_features = NETIF_F_SG |
4016                               NETIF_F_IP_CSUM |
4017                               NETIF_F_IPV6_CSUM |
4018                               NETIF_F_TSO |
4019                               NETIF_F_TSO6 |
4020                               NETIF_F_RXCSUM;
4021
4022         netdev->features = netdev->hw_features |
4023                            NETIF_F_HW_VLAN_CTAG_TX |
4024                            NETIF_F_HW_VLAN_CTAG_RX |
4025                            NETIF_F_HW_VLAN_CTAG_FILTER;
4026
4027         netdev->vlan_features |= NETIF_F_TSO |
4028                                  NETIF_F_TSO6 |
4029                                  NETIF_F_IP_CSUM |
4030                                  NETIF_F_IPV6_CSUM |
4031                                  NETIF_F_SG;
4032
4033         if (pci_using_dac)
4034                 netdev->features |= NETIF_F_HIGHDMA;
4035
4036         netdev->priv_flags |= IFF_UNICAST_FLT;
4037
4038         if (IXGBE_REMOVED(hw->hw_addr)) {
4039                 err = -EIO;
4040                 goto err_sw_init;
4041         }
4042
4043         setup_timer(&adapter->service_timer, &ixgbevf_service_timer,
4044                     (unsigned long)adapter);
4045
4046         INIT_WORK(&adapter->service_task, ixgbevf_service_task);
4047         set_bit(__IXGBEVF_SERVICE_INITED, &adapter->state);
4048         clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
4049
4050         err = ixgbevf_init_interrupt_scheme(adapter);
4051         if (err)
4052                 goto err_sw_init;
4053
4054         strcpy(netdev->name, "eth%d");
4055
4056         err = register_netdev(netdev);
4057         if (err)
4058                 goto err_register;
4059
4060         pci_set_drvdata(pdev, netdev);
4061         netif_carrier_off(netdev);
4062
4063         ixgbevf_init_last_counter_stats(adapter);
4064
4065         /* print the VF info */
4066         dev_info(&pdev->dev, "%pM\n", netdev->dev_addr);
4067         dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
4068
4069         switch (hw->mac.type) {
4070         case ixgbe_mac_X550_vf:
4071                 dev_info(&pdev->dev, "Intel(R) X550 Virtual Function\n");
4072                 break;
4073         case ixgbe_mac_X540_vf:
4074                 dev_info(&pdev->dev, "Intel(R) X540 Virtual Function\n");
4075                 break;
4076         case ixgbe_mac_82599_vf:
4077         default:
4078                 dev_info(&pdev->dev, "Intel(R) 82599 Virtual Function\n");
4079                 break;
4080         }
4081
4082         return 0;
4083
4084 err_register:
4085         ixgbevf_clear_interrupt_scheme(adapter);
4086 err_sw_init:
4087         ixgbevf_reset_interrupt_capability(adapter);
4088         iounmap(adapter->io_addr);
4089 err_ioremap:
4090         disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4091         free_netdev(netdev);
4092 err_alloc_etherdev:
4093         pci_release_regions(pdev);
4094 err_pci_reg:
4095 err_dma:
4096         if (!adapter || disable_dev)
4097                 pci_disable_device(pdev);
4098         return err;
4099 }
4100
4101 /**
4102  * ixgbevf_remove - Device Removal Routine
4103  * @pdev: PCI device information struct
4104  *
4105  * ixgbevf_remove is called by the PCI subsystem to alert the driver
4106  * that it should release a PCI device.  The could be caused by a
4107  * Hot-Plug event, or because the driver is going to be removed from
4108  * memory.
4109  **/
4110 static void ixgbevf_remove(struct pci_dev *pdev)
4111 {
4112         struct net_device *netdev = pci_get_drvdata(pdev);
4113         struct ixgbevf_adapter *adapter;
4114         bool disable_dev;
4115
4116         if (!netdev)
4117                 return;
4118
4119         adapter = netdev_priv(netdev);
4120
4121         set_bit(__IXGBEVF_REMOVING, &adapter->state);
4122         cancel_work_sync(&adapter->service_task);
4123
4124         if (netdev->reg_state == NETREG_REGISTERED)
4125                 unregister_netdev(netdev);
4126
4127         ixgbevf_clear_interrupt_scheme(adapter);
4128         ixgbevf_reset_interrupt_capability(adapter);
4129
4130         iounmap(adapter->io_addr);
4131         pci_release_regions(pdev);
4132
4133         hw_dbg(&adapter->hw, "Remove complete\n");
4134
4135         disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4136         free_netdev(netdev);
4137
4138         if (disable_dev)
4139                 pci_disable_device(pdev);
4140 }
4141
4142 /**
4143  * ixgbevf_io_error_detected - called when PCI error is detected
4144  * @pdev: Pointer to PCI device
4145  * @state: The current pci connection state
4146  *
4147  * This function is called after a PCI bus error affecting
4148  * this device has been detected.
4149  **/
4150 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
4151                                                   pci_channel_state_t state)
4152 {
4153         struct net_device *netdev = pci_get_drvdata(pdev);
4154         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4155
4156         if (!test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
4157                 return PCI_ERS_RESULT_DISCONNECT;
4158
4159         rtnl_lock();
4160         netif_device_detach(netdev);
4161
4162         if (state == pci_channel_io_perm_failure) {
4163                 rtnl_unlock();
4164                 return PCI_ERS_RESULT_DISCONNECT;
4165         }
4166
4167         if (netif_running(netdev))
4168                 ixgbevf_down(adapter);
4169
4170         if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4171                 pci_disable_device(pdev);
4172         rtnl_unlock();
4173
4174         /* Request a slot slot reset. */
4175         return PCI_ERS_RESULT_NEED_RESET;
4176 }
4177
4178 /**
4179  * ixgbevf_io_slot_reset - called after the pci bus has been reset.
4180  * @pdev: Pointer to PCI device
4181  *
4182  * Restart the card from scratch, as if from a cold-boot. Implementation
4183  * resembles the first-half of the ixgbevf_resume routine.
4184  **/
4185 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
4186 {
4187         struct net_device *netdev = pci_get_drvdata(pdev);
4188         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4189
4190         if (pci_enable_device_mem(pdev)) {
4191                 dev_err(&pdev->dev,
4192                         "Cannot re-enable PCI device after reset.\n");
4193                 return PCI_ERS_RESULT_DISCONNECT;
4194         }
4195
4196         smp_mb__before_atomic();
4197         clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4198         pci_set_master(pdev);
4199
4200         ixgbevf_reset(adapter);
4201
4202         return PCI_ERS_RESULT_RECOVERED;
4203 }
4204
4205 /**
4206  * ixgbevf_io_resume - called when traffic can start flowing again.
4207  * @pdev: Pointer to PCI device
4208  *
4209  * This callback is called when the error recovery driver tells us that
4210  * its OK to resume normal operation. Implementation resembles the
4211  * second-half of the ixgbevf_resume routine.
4212  **/
4213 static void ixgbevf_io_resume(struct pci_dev *pdev)
4214 {
4215         struct net_device *netdev = pci_get_drvdata(pdev);
4216         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4217
4218         if (netif_running(netdev))
4219                 ixgbevf_up(adapter);
4220
4221         netif_device_attach(netdev);
4222 }
4223
4224 /* PCI Error Recovery (ERS) */
4225 static const struct pci_error_handlers ixgbevf_err_handler = {
4226         .error_detected = ixgbevf_io_error_detected,
4227         .slot_reset = ixgbevf_io_slot_reset,
4228         .resume = ixgbevf_io_resume,
4229 };
4230
4231 static struct pci_driver ixgbevf_driver = {
4232         .name           = ixgbevf_driver_name,
4233         .id_table       = ixgbevf_pci_tbl,
4234         .probe          = ixgbevf_probe,
4235         .remove         = ixgbevf_remove,
4236 #ifdef CONFIG_PM
4237         /* Power Management Hooks */
4238         .suspend        = ixgbevf_suspend,
4239         .resume         = ixgbevf_resume,
4240 #endif
4241         .shutdown       = ixgbevf_shutdown,
4242         .err_handler    = &ixgbevf_err_handler
4243 };
4244
4245 /**
4246  * ixgbevf_init_module - Driver Registration Routine
4247  *
4248  * ixgbevf_init_module is the first routine called when the driver is
4249  * loaded. All it does is register with the PCI subsystem.
4250  **/
4251 static int __init ixgbevf_init_module(void)
4252 {
4253         int ret;
4254
4255         pr_info("%s - version %s\n", ixgbevf_driver_string,
4256                 ixgbevf_driver_version);
4257
4258         pr_info("%s\n", ixgbevf_copyright);
4259
4260         ret = pci_register_driver(&ixgbevf_driver);
4261         return ret;
4262 }
4263
4264 module_init(ixgbevf_init_module);
4265
4266 /**
4267  * ixgbevf_exit_module - Driver Exit Cleanup Routine
4268  *
4269  * ixgbevf_exit_module is called just before the driver is removed
4270  * from memory.
4271  **/
4272 static void __exit ixgbevf_exit_module(void)
4273 {
4274         pci_unregister_driver(&ixgbevf_driver);
4275 }
4276
4277 #ifdef DEBUG
4278 /**
4279  * ixgbevf_get_hw_dev_name - return device name string
4280  * used by hardware layer to print debugging information
4281  **/
4282 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
4283 {
4284         struct ixgbevf_adapter *adapter = hw->back;
4285
4286         return adapter->netdev->name;
4287 }
4288
4289 #endif
4290 module_exit(ixgbevf_exit_module);
4291
4292 /* ixgbevf_main.c */