Pileus Git - ~andy/linux/blob - drivers/net/ixgbe/ixgbe_main.c

   1 /*******************************************************************************
   2
   3   Intel 10 Gigabit PCI Express Linux driver
   4   Copyright(c) 1999 - 2009 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, write to the Free Software Foundation, Inc.,
  17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18
  19   The full GNU General Public License is included in this distribution in
  20   the file called "COPYING".
  21
  22   Contact Information:
  23   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  24   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  25
  26 *******************************************************************************/
  27
  28 #include <linux/types.h>
  29 #include <linux/module.h>
  30 #include <linux/pci.h>
  31 #include <linux/netdevice.h>
  32 #include <linux/vmalloc.h>
  33 #include <linux/string.h>
  34 #include <linux/in.h>
  35 #include <linux/ip.h>
  36 #include <linux/tcp.h>
  37 #include <linux/pkt_sched.h>
  38 #include <linux/ipv6.h>
  39 #include <net/checksum.h>
  40 #include <net/ip6_checksum.h>
  41 #include <linux/ethtool.h>
  42 #include <linux/if_vlan.h>
  43 #include <scsi/fc/fc_fcoe.h>
  44
  45 #include "ixgbe.h"
  46 #include "ixgbe_common.h"
  47 #include "ixgbe_dcb_82599.h"
  48
  49 char ixgbe_driver_name[] = "ixgbe";
  50 static const char ixgbe_driver_string[] =
  51                               "Intel(R) 10 Gigabit PCI Express Network Driver";
  52
  53 #define DRV_VERSION "2.0.44-k2"
  54 const char ixgbe_driver_version[] = DRV_VERSION;
  55 static char ixgbe_copyright[] = "Copyright (c) 1999-2009 Intel Corporation.";
  56
  57 static const struct ixgbe_info *ixgbe_info_tbl[] = {
  58         [board_82598] = &ixgbe_82598_info,
  59         [board_82599] = &ixgbe_82599_info,
  60 };
  61
  62 /* ixgbe_pci_tbl - PCI Device ID Table
  63  *
  64  * Wildcard entries (PCI_ANY_ID) should come last
  65  * Last entry must be all 0s
  66  *
  67  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
  68  *   Class, Class Mask, private data (not used) }
  69  */
  70 static struct pci_device_id ixgbe_pci_tbl[] = {
  71         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598),
  72          board_82598 },
  73         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT),
  74          board_82598 },
  75         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT),
  76          board_82598 },
  77         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT),
  78          board_82598 },
  79         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2),
  80          board_82598 },
  81         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4),
  82          board_82598 },
  83         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT),
  84          board_82598 },
  85         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT),
  86          board_82598 },
  87         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM),
  88          board_82598 },
  89         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR),
  90          board_82598 },
  91         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM),
  92          board_82598 },
  93         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX),
  94          board_82598 },
  95         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4),
  96          board_82599 },
  97         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM),
  98          board_82599 },
  99         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP),
 100          board_82599 },
 101         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM),
 102          board_82599 },
 103         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ),
 104          board_82599 },
 105         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4),
 106          board_82599 },
 107         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE),
 108          board_82599 },
 109
 110         /* required last entry */
 111         {0, }
 112 };
 113 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
 114
 115 #ifdef CONFIG_IXGBE_DCA
 116 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
 117                             void *p);
 118 static struct notifier_block dca_notifier = {
 119         .notifier_call = ixgbe_notify_dca,
 120         .next          = NULL,
 121         .priority      = 0
 122 };
 123 #endif
 124
 125 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
 126 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
 127 MODULE_LICENSE("GPL");
 128 MODULE_VERSION(DRV_VERSION);
 129
 130 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
 131
 132 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
 133 {
 134         u32 ctrl_ext;
 135
 136         /* Let firmware take over control of h/w */
 137         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
 138         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
 139                         ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
 140 }
 141
 142 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
 143 {
 144         u32 ctrl_ext;
 145
 146         /* Let firmware know the driver has taken over */
 147         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
 148         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
 149                         ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
 150 }
 151
 152 /*
 153  * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
 154  * @adapter: pointer to adapter struct
 155  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
 156  * @queue: queue to map the corresponding interrupt to
 157  * @msix_vector: the vector to map to the corresponding queue
 158  *
 159  */
 160 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
 161                            u8 queue, u8 msix_vector)
 162 {
 163         u32 ivar, index;
 164         struct ixgbe_hw *hw = &adapter->hw;
 165         switch (hw->mac.type) {
 166         case ixgbe_mac_82598EB:
 167                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 168                 if (direction == -1)
 169                         direction = 0;
 170                 index = (((direction * 64) + queue) >> 2) & 0x1F;
 171                 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
 172                 ivar &= ~(0xFF << (8 * (queue & 0x3)));
 173                 ivar |= (msix_vector << (8 * (queue & 0x3)));
 174                 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
 175                 break;
 176         case ixgbe_mac_82599EB:
 177                 if (direction == -1) {
 178                         /* other causes */
 179                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 180                         index = ((queue & 1) * 8);
 181                         ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
 182                         ivar &= ~(0xFF << index);
 183                         ivar |= (msix_vector << index);
 184                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
 185                         break;
 186                 } else {
 187                         /* tx or rx causes */
 188                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 189                         index = ((16 * (queue & 1)) + (8 * direction));
 190                         ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
 191                         ivar &= ~(0xFF << index);
 192                         ivar |= (msix_vector << index);
 193                         IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
 194                         break;
 195                 }
 196         default:
 197                 break;
 198         }
 199 }
 200
 201 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
 202                                           u64 qmask)
 203 {
 204         u32 mask;
 205
 206         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
 207                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
 208                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
 209         } else {
 210                 mask = (qmask & 0xFFFFFFFF);
 211                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
 212                 mask = (qmask >> 32);
 213                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
 214         }
 215 }
 216
 217 static void ixgbe_unmap_and_free_tx_resource(struct ixgbe_adapter *adapter,
 218                                              struct ixgbe_tx_buffer
 219                                              *tx_buffer_info)
 220 {
 221         tx_buffer_info->dma = 0;
 222         if (tx_buffer_info->skb) {
 223                 skb_dma_unmap(&adapter->pdev->dev, tx_buffer_info->skb,
 224                               DMA_TO_DEVICE);
 225                 dev_kfree_skb_any(tx_buffer_info->skb);
 226                 tx_buffer_info->skb = NULL;
 227         }
 228         tx_buffer_info->time_stamp = 0;
 229         /* tx_buffer_info must be completely set up in the transmit path */
 230 }
 231
 232 /**
 233  * ixgbe_tx_is_paused - check if the tx ring is paused
 234  * @adapter: the ixgbe adapter
 235  * @tx_ring: the corresponding tx_ring
 236  *
 237  * If not in DCB mode, checks TFCS.TXOFF, otherwise, find out the
 238  * corresponding TC of this tx_ring when checking TFCS.
 239  *
 240  * Returns : true if paused
 241  */
 242 static inline bool ixgbe_tx_is_paused(struct ixgbe_adapter *adapter,
 243                                       struct ixgbe_ring *tx_ring)
 244 {
 245         int tc;
 246         u32 txoff = IXGBE_TFCS_TXOFF;
 247
 248 #ifdef CONFIG_IXGBE_DCB
 249         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
 250                 int reg_idx = tx_ring->reg_idx;
 251                 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
 252
 253                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
 254                         tc = reg_idx >> 2;
 255                         txoff = IXGBE_TFCS_TXOFF0;
 256                 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
 257                         tc = 0;
 258                         txoff = IXGBE_TFCS_TXOFF;
 259                         if (dcb_i == 8) {
 260                                 /* TC0, TC1 */
 261                                 tc = reg_idx >> 5;
 262                                 if (tc == 2) /* TC2, TC3 */
 263                                         tc += (reg_idx - 64) >> 4;
 264                                 else if (tc == 3) /* TC4, TC5, TC6, TC7 */
 265                                         tc += 1 + ((reg_idx - 96) >> 3);
 266                         } else if (dcb_i == 4) {
 267                                 /* TC0, TC1 */
 268                                 tc = reg_idx >> 6;
 269                                 if (tc == 1) {
 270                                         tc += (reg_idx - 64) >> 5;
 271                                         if (tc == 2) /* TC2, TC3 */
 272                                                 tc += (reg_idx - 96) >> 4;
 273                                 }
 274                         }
 275                 }
 276                 txoff <<= tc;
 277         }
 278 #endif
 279         return IXGBE_READ_REG(&adapter->hw, IXGBE_TFCS) & txoff;
 280 }
 281
 282 static inline bool ixgbe_check_tx_hang(struct ixgbe_adapter *adapter,
 283                                        struct ixgbe_ring *tx_ring,
 284                                        unsigned int eop)
 285 {
 286         struct ixgbe_hw *hw = &adapter->hw;
 287
 288         /* Detect a transmit hang in hardware, this serializes the
 289          * check with the clearing of time_stamp and movement of eop */
 290         adapter->detect_tx_hung = false;
 291         if (tx_ring->tx_buffer_info[eop].time_stamp &&
 292             time_after(jiffies, tx_ring->tx_buffer_info[eop].time_stamp + HZ) &&
 293             !ixgbe_tx_is_paused(adapter, tx_ring)) {
 294                 /* detected Tx unit hang */
 295                 union ixgbe_adv_tx_desc *tx_desc;
 296                 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
 297                 DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
 298                         "  Tx Queue             <%d>\n"
 299                         "  TDH, TDT             <%x>, <%x>\n"
 300                         "  next_to_use          <%x>\n"
 301                         "  next_to_clean        <%x>\n"
 302                         "tx_buffer_info[next_to_clean]\n"
 303                         "  time_stamp           <%lx>\n"
 304                         "  jiffies              <%lx>\n",
 305                         tx_ring->queue_index,
 306                         IXGBE_READ_REG(hw, tx_ring->head),
 307                         IXGBE_READ_REG(hw, tx_ring->tail),
 308                         tx_ring->next_to_use, eop,
 309                         tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
 310                 return true;
 311         }
 312
 313         return false;
 314 }
 315
 316 #define IXGBE_MAX_TXD_PWR       14
 317 #define IXGBE_MAX_DATA_PER_TXD  (1 << IXGBE_MAX_TXD_PWR)
 318
 319 /* Tx Descriptors needed, worst case */
 320 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
 321                          (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
 322 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
 323         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
 324
 325 static void ixgbe_tx_timeout(struct net_device *netdev);
 326
 327 /**
 328  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
 329  * @q_vector: structure containing interrupt and ring information
 330  * @tx_ring: tx ring to clean
 331  **/
 332 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
 333                                struct ixgbe_ring *tx_ring)
 334 {
 335         struct ixgbe_adapter *adapter = q_vector->adapter;
 336         struct net_device *netdev = adapter->netdev;
 337         union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
 338         struct ixgbe_tx_buffer *tx_buffer_info;
 339         unsigned int i, eop, count = 0;
 340         unsigned int total_bytes = 0, total_packets = 0;
 341
 342         i = tx_ring->next_to_clean;
 343         eop = tx_ring->tx_buffer_info[i].next_to_watch;
 344         eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
 345
 346         while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
 347                (count < tx_ring->work_limit)) {
 348                 bool cleaned = false;
 349                 for ( ; !cleaned; count++) {
 350                         struct sk_buff *skb;
 351                         tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
 352                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
 353                         cleaned = (i == eop);
 354                         skb = tx_buffer_info->skb;
 355
 356                         if (cleaned && skb) {
 357                                 unsigned int segs, bytecount;
 358                                 unsigned int hlen = skb_headlen(skb);
 359
 360                                 /* gso_segs is currently only valid for tcp */
 361                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 362 #ifdef IXGBE_FCOE
 363                                 /* adjust for FCoE Sequence Offload */
 364                                 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
 365                                     && (skb->protocol == htons(ETH_P_FCOE)) &&
 366                                     skb_is_gso(skb)) {
 367                                         hlen = skb_transport_offset(skb) +
 368                                                 sizeof(struct fc_frame_header) +
 369                                                 sizeof(struct fcoe_crc_eof);
 370                                         segs = DIV_ROUND_UP(skb->len - hlen,
 371                                                 skb_shinfo(skb)->gso_size);
 372                                 }
 373 #endif /* IXGBE_FCOE */
 374                                 /* multiply data chunks by size of headers */
 375                                 bytecount = ((segs - 1) * hlen) + skb->len;
 376                                 total_packets += segs;
 377                                 total_bytes += bytecount;
 378                         }
 379
 380                         ixgbe_unmap_and_free_tx_resource(adapter,
 381                                                          tx_buffer_info);
 382
 383                         tx_desc->wb.status = 0;
 384
 385                         i++;
 386                         if (i == tx_ring->count)
 387                                 i = 0;
 388                 }
 389
 390                 eop = tx_ring->tx_buffer_info[i].next_to_watch;
 391                 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
 392         }
 393
 394         tx_ring->next_to_clean = i;
 395
 396 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
 397         if (unlikely(count && netif_carrier_ok(netdev) &&
 398                      (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
 399                 /* Make sure that anybody stopping the queue after this
 400                  * sees the new next_to_clean.
 401                  */
 402                 smp_mb();
 403                 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
 404                     !test_bit(__IXGBE_DOWN, &adapter->state)) {
 405                         netif_wake_subqueue(netdev, tx_ring->queue_index);
 406                         ++adapter->restart_queue;
 407                 }
 408         }
 409
 410         if (adapter->detect_tx_hung) {
 411                 if (ixgbe_check_tx_hang(adapter, tx_ring, i)) {
 412                         /* schedule immediate reset if we believe we hung */
 413                         DPRINTK(PROBE, INFO,
 414                                 "tx hang %d detected, resetting adapter\n",
 415                                 adapter->tx_timeout_count + 1);
 416                         ixgbe_tx_timeout(adapter->netdev);
 417                 }
 418         }
 419
 420         /* re-arm the interrupt */
 421         if (count >= tx_ring->work_limit)
 422                 ixgbe_irq_rearm_queues(adapter, ((u64)1 << q_vector->v_idx));
 423
 424         tx_ring->total_bytes += total_bytes;
 425         tx_ring->total_packets += total_packets;
 426         tx_ring->stats.packets += total_packets;
 427         tx_ring->stats.bytes += total_bytes;
 428         netdev->stats.tx_bytes += total_bytes;
 429         netdev->stats.tx_packets += total_packets;
 430         return (count < tx_ring->work_limit);
 431 }
 432
 433 #ifdef CONFIG_IXGBE_DCA
 434 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
 435                                 struct ixgbe_ring *rx_ring)
 436 {
 437         u32 rxctrl;
 438         int cpu = get_cpu();
 439         int q = rx_ring - adapter->rx_ring;
 440
 441         if (rx_ring->cpu != cpu) {
 442                 rxctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q));
 443                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
 444                         rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
 445                         rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
 446                 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
 447                         rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
 448                         rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
 449                                    IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
 450                 }
 451                 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
 452                 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
 453                 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
 454                 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN |
 455                             IXGBE_DCA_RXCTRL_DESC_HSRO_EN);
 456                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q), rxctrl);
 457                 rx_ring->cpu = cpu;
 458         }
 459         put_cpu();
 460 }
 461
 462 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
 463                                 struct ixgbe_ring *tx_ring)
 464 {
 465         u32 txctrl;
 466         int cpu = get_cpu();
 467         int q = tx_ring - adapter->tx_ring;
 468         struct ixgbe_hw *hw = &adapter->hw;
 469
 470         if (tx_ring->cpu != cpu) {
 471                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
 472                         txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(q));
 473                         txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
 474                         txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
 475                         txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
 476                         IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(q), txctrl);
 477                 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
 478                         txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(q));
 479                         txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
 480                         txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
 481                                   IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
 482                         txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
 483                         IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(q), txctrl);
 484                 }
 485                 tx_ring->cpu = cpu;
 486         }
 487         put_cpu();
 488 }
 489
 490 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
 491 {
 492         int i;
 493
 494         if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
 495                 return;
 496
 497         /* always use CB2 mode, difference is masked in the CB driver */
 498         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
 499
 500         for (i = 0; i < adapter->num_tx_queues; i++) {
 501                 adapter->tx_ring[i].cpu = -1;
 502                 ixgbe_update_tx_dca(adapter, &adapter->tx_ring[i]);
 503         }
 504         for (i = 0; i < adapter->num_rx_queues; i++) {
 505                 adapter->rx_ring[i].cpu = -1;
 506                 ixgbe_update_rx_dca(adapter, &adapter->rx_ring[i]);
 507         }
 508 }
 509
 510 static int __ixgbe_notify_dca(struct device *dev, void *data)
 511 {
 512         struct net_device *netdev = dev_get_drvdata(dev);
 513         struct ixgbe_adapter *adapter = netdev_priv(netdev);
 514         unsigned long event = *(unsigned long *)data;
 515
 516         switch (event) {
 517         case DCA_PROVIDER_ADD:
 518                 /* if we're already enabled, don't do it again */
 519                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
 520                         break;
 521                 if (dca_add_requester(dev) == 0) {
 522                         adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
 523                         ixgbe_setup_dca(adapter);
 524                         break;
 525                 }
 526                 /* Fall Through since DCA is disabled. */
 527         case DCA_PROVIDER_REMOVE:
 528                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
 529                         dca_remove_requester(dev);
 530                         adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
 531                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
 532                 }
 533                 break;
 534         }
 535
 536         return 0;
 537 }
 538
 539 #endif /* CONFIG_IXGBE_DCA */
 540 /**
 541  * ixgbe_receive_skb - Send a completed packet up the stack
 542  * @adapter: board private structure
 543  * @skb: packet to send up
 544  * @status: hardware indication of status of receive
 545  * @rx_ring: rx descriptor ring (for a specific queue) to setup
 546  * @rx_desc: rx descriptor
 547  **/
 548 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
 549                               struct sk_buff *skb, u8 status,
 550                               struct ixgbe_ring *ring,
 551                               union ixgbe_adv_rx_desc *rx_desc)
 552 {
 553         struct ixgbe_adapter *adapter = q_vector->adapter;
 554         struct napi_struct *napi = &q_vector->napi;
 555         bool is_vlan = (status & IXGBE_RXD_STAT_VP);
 556         u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
 557
 558         skb_record_rx_queue(skb, ring->queue_index);
 559         if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL)) {
 560                 if (adapter->vlgrp && is_vlan && (tag & VLAN_VID_MASK))
 561                         vlan_gro_receive(napi, adapter->vlgrp, tag, skb);
 562                 else
 563                         napi_gro_receive(napi, skb);
 564         } else {
 565                 if (adapter->vlgrp && is_vlan && (tag & VLAN_VID_MASK))
 566                         vlan_hwaccel_rx(skb, adapter->vlgrp, tag);
 567                 else
 568                         netif_rx(skb);
 569         }
 570 }
 571
 572 /**
 573  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
 574  * @adapter: address of board private structure
 575  * @status_err: hardware indication of status of receive
 576  * @skb: skb currently being received and modified
 577  **/
 578 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
 579                                      union ixgbe_adv_rx_desc *rx_desc,
 580                                      struct sk_buff *skb)
 581 {
 582         u32 status_err = le32_to_cpu(rx_desc->wb.upper.status_error);
 583
 584         skb->ip_summed = CHECKSUM_NONE;
 585
 586         /* Rx csum disabled */
 587         if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
 588                 return;
 589
 590         /* if IP and error */
 591         if ((status_err & IXGBE_RXD_STAT_IPCS) &&
 592             (status_err & IXGBE_RXDADV_ERR_IPE)) {
 593                 adapter->hw_csum_rx_error++;
 594                 return;
 595         }
 596
 597         if (!(status_err & IXGBE_RXD_STAT_L4CS))
 598                 return;
 599
 600         if (status_err & IXGBE_RXDADV_ERR_TCPE) {
 601                 u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
 602
 603                 /*
 604                  * 82599 errata, UDP frames with a 0 checksum can be marked as
 605                  * checksum errors.
 606                  */
 607                 if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
 608                     (adapter->hw.mac.type == ixgbe_mac_82599EB))
 609                         return;
 610
 611                 adapter->hw_csum_rx_error++;
 612                 return;
 613         }
 614
 615         /* It must be a TCP or UDP packet with a valid checksum */
 616         skb->ip_summed = CHECKSUM_UNNECESSARY;
 617         adapter->hw_csum_rx_good++;
 618 }
 619
 620 static inline void ixgbe_release_rx_desc(struct ixgbe_hw *hw,
 621                                          struct ixgbe_ring *rx_ring, u32 val)
 622 {
 623         /*
 624          * Force memory writes to complete before letting h/w
 625          * know there are new descriptors to fetch.  (Only
 626          * applicable for weak-ordered memory model archs,
 627          * such as IA-64).
 628          */
 629         wmb();
 630         IXGBE_WRITE_REG(hw, IXGBE_RDT(rx_ring->reg_idx), val);
 631 }
 632
 633 /**
 634  * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
 635  * @adapter: address of board private structure
 636  **/
 637 static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
 638                                    struct ixgbe_ring *rx_ring,
 639                                    int cleaned_count)
 640 {
 641         struct pci_dev *pdev = adapter->pdev;
 642         union ixgbe_adv_rx_desc *rx_desc;
 643         struct ixgbe_rx_buffer *bi;
 644         unsigned int i;
 645
 646         i = rx_ring->next_to_use;
 647         bi = &rx_ring->rx_buffer_info[i];
 648
 649         while (cleaned_count--) {
 650                 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
 651
 652                 if (!bi->page_dma &&
 653                     (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED)) {
 654                         if (!bi->page) {
 655                                 bi->page = alloc_page(GFP_ATOMIC);
 656                                 if (!bi->page) {
 657                                         adapter->alloc_rx_page_failed++;
 658                                         goto no_buffers;
 659                                 }
 660                                 bi->page_offset = 0;
 661                         } else {
 662                                 /* use a half page if we're re-using */
 663                                 bi->page_offset ^= (PAGE_SIZE / 2);
 664                         }
 665
 666                         bi->page_dma = pci_map_page(pdev, bi->page,
 667                                                     bi->page_offset,
 668                                                     (PAGE_SIZE / 2),
 669                                                     PCI_DMA_FROMDEVICE);
 670                 }
 671
 672                 if (!bi->skb) {
 673                         struct sk_buff *skb;
 674                         skb = netdev_alloc_skb_ip_align(adapter->netdev,
 675                                                         rx_ring->rx_buf_len);
 676
 677                         if (!skb) {
 678                                 adapter->alloc_rx_buff_failed++;
 679                                 goto no_buffers;
 680                         }
 681
 682                         bi->skb = skb;
 683                         bi->dma = pci_map_single(pdev, skb->data,
 684                                                  rx_ring->rx_buf_len,
 685                                                  PCI_DMA_FROMDEVICE);
 686                 }
 687                 /* Refresh the desc even if buffer_addrs didn't change because
 688                  * each write-back erases this info. */
 689                 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
 690                         rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
 691                         rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
 692                 } else {
 693                         rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
 694                 }
 695
 696                 i++;
 697                 if (i == rx_ring->count)
 698                         i = 0;
 699                 bi = &rx_ring->rx_buffer_info[i];
 700         }
 701
 702 no_buffers:
 703         if (rx_ring->next_to_use != i) {
 704                 rx_ring->next_to_use = i;
 705                 if (i-- == 0)
 706                         i = (rx_ring->count - 1);
 707
 708                 ixgbe_release_rx_desc(&adapter->hw, rx_ring, i);
 709         }
 710 }
 711
 712 static inline u16 ixgbe_get_hdr_info(union ixgbe_adv_rx_desc *rx_desc)
 713 {
 714         return rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
 715 }
 716
 717 static inline u16 ixgbe_get_pkt_info(union ixgbe_adv_rx_desc *rx_desc)
 718 {
 719         return rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
 720 }
 721
 722 static inline u32 ixgbe_get_rsc_count(union ixgbe_adv_rx_desc *rx_desc)
 723 {
 724         return (le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
 725                 IXGBE_RXDADV_RSCCNT_MASK) >>
 726                 IXGBE_RXDADV_RSCCNT_SHIFT;
 727 }
 728
 729 /**
 730  * ixgbe_transform_rsc_queue - change rsc queue into a full packet
 731  * @skb: pointer to the last skb in the rsc queue
 732  *
 733  * This function changes a queue full of hw rsc buffers into a completed
 734  * packet.  It uses the ->prev pointers to find the first packet and then
 735  * turns it into the frag list owner.
 736  **/
 737 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb)
 738 {
 739         unsigned int frag_list_size = 0;
 740
 741         while (skb->prev) {
 742                 struct sk_buff *prev = skb->prev;
 743                 frag_list_size += skb->len;
 744                 skb->prev = NULL;
 745                 skb = prev;
 746         }
 747
 748         skb_shinfo(skb)->frag_list = skb->next;
 749         skb->next = NULL;
 750         skb->len += frag_list_size;
 751         skb->data_len += frag_list_size;
 752         skb->truesize += frag_list_size;
 753         return skb;
 754 }
 755
 756 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
 757                                struct ixgbe_ring *rx_ring,
 758                                int *work_done, int work_to_do)
 759 {
 760         struct ixgbe_adapter *adapter = q_vector->adapter;
 761         struct net_device *netdev = adapter->netdev;
 762         struct pci_dev *pdev = adapter->pdev;
 763         union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
 764         struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
 765         struct sk_buff *skb;
 766         unsigned int i, rsc_count = 0;
 767         u32 len, staterr;
 768         u16 hdr_info;
 769         bool cleaned = false;
 770         int cleaned_count = 0;
 771         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 772 #ifdef IXGBE_FCOE
 773         int ddp_bytes = 0;
 774 #endif /* IXGBE_FCOE */
 775
 776         i = rx_ring->next_to_clean;
 777         rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
 778         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 779         rx_buffer_info = &rx_ring->rx_buffer_info[i];
 780
 781         while (staterr & IXGBE_RXD_STAT_DD) {
 782                 u32 upper_len = 0;
 783                 if (*work_done >= work_to_do)
 784                         break;
 785                 (*work_done)++;
 786
 787                 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
 788                         hdr_info = le16_to_cpu(ixgbe_get_hdr_info(rx_desc));
 789                         len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
 790                                IXGBE_RXDADV_HDRBUFLEN_SHIFT;
 791                         if (hdr_info & IXGBE_RXDADV_SPH)
 792                                 adapter->rx_hdr_split++;
 793                         if (len > IXGBE_RX_HDR_SIZE)
 794                                 len = IXGBE_RX_HDR_SIZE;
 795                         upper_len = le16_to_cpu(rx_desc->wb.upper.length);
 796                 } else {
 797                         len = le16_to_cpu(rx_desc->wb.upper.length);
 798                 }
 799
 800                 cleaned = true;
 801                 skb = rx_buffer_info->skb;
 802                 prefetch(skb->data - NET_IP_ALIGN);
 803                 rx_buffer_info->skb = NULL;
 804
 805                 if (rx_buffer_info->dma) {
 806                         pci_unmap_single(pdev, rx_buffer_info->dma,
 807                                          rx_ring->rx_buf_len,
 808                                          PCI_DMA_FROMDEVICE);
 809                         rx_buffer_info->dma = 0;
 810                         skb_put(skb, len);
 811                 }
 812
 813                 if (upper_len) {
 814                         pci_unmap_page(pdev, rx_buffer_info->page_dma,
 815                                        PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
 816                         rx_buffer_info->page_dma = 0;
 817                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
 818                                            rx_buffer_info->page,
 819                                            rx_buffer_info->page_offset,
 820                                            upper_len);
 821
 822                         if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
 823                             (page_count(rx_buffer_info->page) != 1))
 824                                 rx_buffer_info->page = NULL;
 825                         else
 826                                 get_page(rx_buffer_info->page);
 827
 828                         skb->len += upper_len;
 829                         skb->data_len += upper_len;
 830                         skb->truesize += upper_len;
 831                 }
 832
 833                 i++;
 834                 if (i == rx_ring->count)
 835                         i = 0;
 836
 837                 next_rxd = IXGBE_RX_DESC_ADV(*rx_ring, i);
 838                 prefetch(next_rxd);
 839                 cleaned_count++;
 840
 841                 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
 842                         rsc_count = ixgbe_get_rsc_count(rx_desc);
 843
 844                 if (rsc_count) {
 845                         u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
 846                                      IXGBE_RXDADV_NEXTP_SHIFT;
 847                         next_buffer = &rx_ring->rx_buffer_info[nextp];
 848                         rx_ring->rsc_count += (rsc_count - 1);
 849                 } else {
 850                         next_buffer = &rx_ring->rx_buffer_info[i];
 851                 }
 852
 853                 if (staterr & IXGBE_RXD_STAT_EOP) {
 854                         if (skb->prev)
 855                                 skb = ixgbe_transform_rsc_queue(skb);
 856                         rx_ring->stats.packets++;
 857                         rx_ring->stats.bytes += skb->len;
 858                 } else {
 859                         if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
 860                                 rx_buffer_info->skb = next_buffer->skb;
 861                                 rx_buffer_info->dma = next_buffer->dma;
 862                                 next_buffer->skb = skb;
 863                                 next_buffer->dma = 0;
 864                         } else {
 865                                 skb->next = next_buffer->skb;
 866                                 skb->next->prev = skb;
 867                         }
 868                         adapter->non_eop_descs++;
 869                         goto next_desc;
 870                 }
 871
 872                 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
 873                         dev_kfree_skb_irq(skb);
 874                         goto next_desc;
 875                 }
 876
 877                 ixgbe_rx_checksum(adapter, rx_desc, skb);
 878
 879                 /* probably a little skewed due to removing CRC */
 880                 total_rx_bytes += skb->len;
 881                 total_rx_packets++;
 882
 883                 skb->protocol = eth_type_trans(skb, adapter->netdev);
 884 #ifdef IXGBE_FCOE
 885                 /* if ddp, not passing to ULD unless for FCP_RSP or error */
 886                 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
 887                         ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
 888                         if (!ddp_bytes)
 889                                 goto next_desc;
 890                 }
 891 #endif /* IXGBE_FCOE */
 892                 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
 893
 894 next_desc:
 895                 rx_desc->wb.upper.status_error = 0;
 896
 897                 /* return some buffers to hardware, one at a time is too slow */
 898                 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
 899                         ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
 900                         cleaned_count = 0;
 901                 }
 902
 903                 /* use prefetched values */
 904                 rx_desc = next_rxd;
 905                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
 906
 907                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 908         }
 909
 910         rx_ring->next_to_clean = i;
 911         cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
 912
 913         if (cleaned_count)
 914                 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
 915
 916 #ifdef IXGBE_FCOE
 917         /* include DDPed FCoE data */
 918         if (ddp_bytes > 0) {
 919                 unsigned int mss;
 920
 921                 mss = adapter->netdev->mtu - sizeof(struct fcoe_hdr) -
 922                         sizeof(struct fc_frame_header) -
 923                         sizeof(struct fcoe_crc_eof);
 924                 if (mss > 512)
 925                         mss &= ~511;
 926                 total_rx_bytes += ddp_bytes;
 927                 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
 928         }
 929 #endif /* IXGBE_FCOE */
 930
 931         rx_ring->total_packets += total_rx_packets;
 932         rx_ring->total_bytes += total_rx_bytes;
 933         netdev->stats.rx_bytes += total_rx_bytes;
 934         netdev->stats.rx_packets += total_rx_packets;
 935
 936         return cleaned;
 937 }
 938
 939 static int ixgbe_clean_rxonly(struct napi_struct *, int);
 940 /**
 941  * ixgbe_configure_msix - Configure MSI-X hardware
 942  * @adapter: board private structure
 943  *
 944  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
 945  * interrupts.
 946  **/
 947 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
 948 {
 949         struct ixgbe_q_vector *q_vector;
 950         int i, j, q_vectors, v_idx, r_idx;
 951         u32 mask;
 952
 953         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
 954
 955         /*
 956          * Populate the IVAR table and set the ITR values to the
 957          * corresponding register.
 958          */
 959         for (v_idx = 0; v_idx < q_vectors; v_idx++) {
 960                 q_vector = adapter->q_vector[v_idx];
 961                 /* XXX for_each_bit(...) */
 962                 r_idx = find_first_bit(q_vector->rxr_idx,
 963                                        adapter->num_rx_queues);
 964
 965                 for (i = 0; i < q_vector->rxr_count; i++) {
 966                         j = adapter->rx_ring[r_idx].reg_idx;
 967                         ixgbe_set_ivar(adapter, 0, j, v_idx);
 968                         r_idx = find_next_bit(q_vector->rxr_idx,
 969                                               adapter->num_rx_queues,
 970                                               r_idx + 1);
 971                 }
 972                 r_idx = find_first_bit(q_vector->txr_idx,
 973                                        adapter->num_tx_queues);
 974
 975                 for (i = 0; i < q_vector->txr_count; i++) {
 976                         j = adapter->tx_ring[r_idx].reg_idx;
 977                         ixgbe_set_ivar(adapter, 1, j, v_idx);
 978                         r_idx = find_next_bit(q_vector->txr_idx,
 979                                               adapter->num_tx_queues,
 980                                               r_idx + 1);
 981                 }
 982
 983                 if (q_vector->txr_count && !q_vector->rxr_count)
 984                         /* tx only */
 985                         q_vector->eitr = adapter->tx_eitr_param;
 986                 else if (q_vector->rxr_count)
 987                         /* rx or mixed */
 988                         q_vector->eitr = adapter->rx_eitr_param;
 989
 990                 ixgbe_write_eitr(q_vector);
 991         }
 992
 993         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
 994                 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
 995                                v_idx);
 996         else if (adapter->hw.mac.type == ixgbe_mac_82599EB)
 997                 ixgbe_set_ivar(adapter, -1, 1, v_idx);
 998         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
 999
1000         /* set up to autoclear timer, and the vectors */
1001         mask = IXGBE_EIMS_ENABLE_MASK;
1002         mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
1003         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1004 }
1005
1006 enum latency_range {
1007         lowest_latency = 0,
1008         low_latency = 1,
1009         bulk_latency = 2,
1010         latency_invalid = 255
1011 };
1012
1013 /**
1014  * ixgbe_update_itr - update the dynamic ITR value based on statistics
1015  * @adapter: pointer to adapter
1016  * @eitr: eitr setting (ints per sec) to give last timeslice
1017  * @itr_setting: current throttle rate in ints/second
1018  * @packets: the number of packets during this measurement interval
1019  * @bytes: the number of bytes during this measurement interval
1020  *
1021  *      Stores a new ITR value based on packets and byte
1022  *      counts during the last interrupt.  The advantage of per interrupt
1023  *      computation is faster updates and more accurate ITR for the current
1024  *      traffic pattern.  Constants in this function were computed
1025  *      based on theoretical maximum wire speed and thresholds were set based
1026  *      on testing data as well as attempting to minimize response time
1027  *      while increasing bulk throughput.
1028  *      this functionality is controlled by the InterruptThrottleRate module
1029  *      parameter (see ixgbe_param.c)
1030  **/
1031 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
1032                            u32 eitr, u8 itr_setting,
1033                            int packets, int bytes)
1034 {
1035         unsigned int retval = itr_setting;
1036         u32 timepassed_us;
1037         u64 bytes_perint;
1038
1039         if (packets == 0)
1040                 goto update_itr_done;
1041
1042
1043         /* simple throttlerate management
1044          *    0-20MB/s lowest (100000 ints/s)
1045          *   20-100MB/s low   (20000 ints/s)
1046          *  100-1249MB/s bulk (8000 ints/s)
1047          */
1048         /* what was last interrupt timeslice? */
1049         timepassed_us = 1000000/eitr;
1050         bytes_perint = bytes / timepassed_us; /* bytes/usec */
1051
1052         switch (itr_setting) {
1053         case lowest_latency:
1054                 if (bytes_perint > adapter->eitr_low)
1055                         retval = low_latency;
1056                 break;
1057         case low_latency:
1058                 if (bytes_perint > adapter->eitr_high)
1059                         retval = bulk_latency;
1060                 else if (bytes_perint <= adapter->eitr_low)
1061                         retval = lowest_latency;
1062                 break;
1063         case bulk_latency:
1064                 if (bytes_perint <= adapter->eitr_high)
1065                         retval = low_latency;
1066                 break;
1067         }
1068
1069 update_itr_done:
1070         return retval;
1071 }
1072
1073 /**
1074  * ixgbe_write_eitr - write EITR register in hardware specific way
1075  * @q_vector: structure containing interrupt and ring information
1076  *
1077  * This function is made to be called by ethtool and by the driver
1078  * when it needs to update EITR registers at runtime.  Hardware
1079  * specific quirks/differences are taken care of here.
1080  */
1081 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1082 {
1083         struct ixgbe_adapter *adapter = q_vector->adapter;
1084         struct ixgbe_hw *hw = &adapter->hw;
1085         int v_idx = q_vector->v_idx;
1086         u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1087
1088         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1089                 /* must write high and low 16 bits to reset counter */
1090                 itr_reg |= (itr_reg << 16);
1091         } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1092                 /*
1093                  * set the WDIS bit to not clear the timer bits and cause an
1094                  * immediate assertion of the interrupt
1095                  */
1096                 itr_reg |= IXGBE_EITR_CNT_WDIS;
1097         }
1098         IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1099 }
1100
1101 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1102 {
1103         struct ixgbe_adapter *adapter = q_vector->adapter;
1104         u32 new_itr;
1105         u8 current_itr, ret_itr;
1106         int i, r_idx;
1107         struct ixgbe_ring *rx_ring, *tx_ring;
1108
1109         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1110         for (i = 0; i < q_vector->txr_count; i++) {
1111                 tx_ring = &(adapter->tx_ring[r_idx]);
1112                 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1113                                            q_vector->tx_itr,
1114                                            tx_ring->total_packets,
1115                                            tx_ring->total_bytes);
1116                 /* if the result for this queue would decrease interrupt
1117                  * rate for this vector then use that result */
1118                 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1119                                     q_vector->tx_itr - 1 : ret_itr);
1120                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1121                                       r_idx + 1);
1122         }
1123
1124         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1125         for (i = 0; i < q_vector->rxr_count; i++) {
1126                 rx_ring = &(adapter->rx_ring[r_idx]);
1127                 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1128                                            q_vector->rx_itr,
1129                                            rx_ring->total_packets,
1130                                            rx_ring->total_bytes);
1131                 /* if the result for this queue would decrease interrupt
1132                  * rate for this vector then use that result */
1133                 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1134                                     q_vector->rx_itr - 1 : ret_itr);
1135                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1136                                       r_idx + 1);
1137         }
1138
1139         current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1140
1141         switch (current_itr) {
1142         /* counts and packets in update_itr are dependent on these numbers */
1143         case lowest_latency:
1144                 new_itr = 100000;
1145                 break;
1146         case low_latency:
1147                 new_itr = 20000; /* aka hwitr = ~200 */
1148                 break;
1149         case bulk_latency:
1150         default:
1151                 new_itr = 8000;
1152                 break;
1153         }
1154
1155         if (new_itr != q_vector->eitr) {
1156                 /* do an exponential smoothing */
1157                 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1158
1159                 /* save the algorithm value here, not the smoothed one */
1160                 q_vector->eitr = new_itr;
1161
1162                 ixgbe_write_eitr(q_vector);
1163         }
1164
1165         return;
1166 }
1167
1168 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1169 {
1170         struct ixgbe_hw *hw = &adapter->hw;
1171
1172         if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1173             (eicr & IXGBE_EICR_GPI_SDP1)) {
1174                 DPRINTK(PROBE, CRIT, "Fan has stopped, replace the adapter\n");
1175                 /* write to clear the interrupt */
1176                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1177         }
1178 }
1179
1180 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1181 {
1182         struct ixgbe_hw *hw = &adapter->hw;
1183
1184         if (eicr & IXGBE_EICR_GPI_SDP1) {
1185                 /* Clear the interrupt */
1186                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1187                 schedule_work(&adapter->multispeed_fiber_task);
1188         } else if (eicr & IXGBE_EICR_GPI_SDP2) {
1189                 /* Clear the interrupt */
1190                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1191                 schedule_work(&adapter->sfp_config_module_task);
1192         } else {
1193                 /* Interrupt isn't for us... */
1194                 return;
1195         }
1196 }
1197
1198 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1199 {
1200         struct ixgbe_hw *hw = &adapter->hw;
1201
1202         adapter->lsc_int++;
1203         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1204         adapter->link_check_timeout = jiffies;
1205         if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1206                 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1207                 IXGBE_WRITE_FLUSH(hw);
1208                 schedule_work(&adapter->watchdog_task);
1209         }
1210 }
1211
1212 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1213 {
1214         struct net_device *netdev = data;
1215         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1216         struct ixgbe_hw *hw = &adapter->hw;
1217         u32 eicr;
1218
1219         /*
1220          * Workaround for Silicon errata.  Use clear-by-write instead
1221          * of clear-by-read.  Reading with EICS will return the
1222          * interrupt causes without clearing, which later be done
1223          * with the write to EICR.
1224          */
1225         eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1226         IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1227
1228         if (eicr & IXGBE_EICR_LSC)
1229                 ixgbe_check_lsc(adapter);
1230
1231         if (hw->mac.type == ixgbe_mac_82598EB)
1232                 ixgbe_check_fan_failure(adapter, eicr);
1233
1234         if (hw->mac.type == ixgbe_mac_82599EB) {
1235                 ixgbe_check_sfp_event(adapter, eicr);
1236
1237                 /* Handle Flow Director Full threshold interrupt */
1238                 if (eicr & IXGBE_EICR_FLOW_DIR) {
1239                         int i;
1240                         IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1241                         /* Disable transmits before FDIR Re-initialization */
1242                         netif_tx_stop_all_queues(netdev);
1243                         for (i = 0; i < adapter->num_tx_queues; i++) {
1244                                 struct ixgbe_ring *tx_ring =
1245                                                            &adapter->tx_ring[i];
1246                                 if (test_and_clear_bit(__IXGBE_FDIR_INIT_DONE,
1247                                                        &tx_ring->reinit_state))
1248                                         schedule_work(&adapter->fdir_reinit_task);
1249                         }
1250                 }
1251         }
1252         if (!test_bit(__IXGBE_DOWN, &adapter->state))
1253                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1254
1255         return IRQ_HANDLED;
1256 }
1257
1258 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1259                                            u64 qmask)
1260 {
1261         u32 mask;
1262
1263         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1264                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1265                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1266         } else {
1267                 mask = (qmask & 0xFFFFFFFF);
1268                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(0), mask);
1269                 mask = (qmask >> 32);
1270                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(1), mask);
1271         }
1272         /* skip the flush */
1273 }
1274
1275 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1276                                             u64 qmask)
1277 {
1278         u32 mask;
1279
1280         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1281                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1282                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, mask);
1283         } else {
1284                 mask = (qmask & 0xFFFFFFFF);
1285                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), mask);
1286                 mask = (qmask >> 32);
1287                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), mask);
1288         }
1289         /* skip the flush */
1290 }
1291
1292 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1293 {
1294         struct ixgbe_q_vector *q_vector = data;
1295         struct ixgbe_adapter  *adapter = q_vector->adapter;
1296         struct ixgbe_ring     *tx_ring;
1297         int i, r_idx;
1298
1299         if (!q_vector->txr_count)
1300                 return IRQ_HANDLED;
1301
1302         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1303         for (i = 0; i < q_vector->txr_count; i++) {
1304                 tx_ring = &(adapter->tx_ring[r_idx]);
1305                 tx_ring->total_bytes = 0;
1306                 tx_ring->total_packets = 0;
1307                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1308                                       r_idx + 1);
1309         }
1310
1311         /* disable interrupts on this vector only */
1312         ixgbe_irq_disable_queues(adapter, ((u64)1 << q_vector->v_idx));
1313         napi_schedule(&q_vector->napi);
1314
1315         return IRQ_HANDLED;
1316 }
1317
1318 /**
1319  * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
1320  * @irq: unused
1321  * @data: pointer to our q_vector struct for this interrupt vector
1322  **/
1323 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
1324 {
1325         struct ixgbe_q_vector *q_vector = data;
1326         struct ixgbe_adapter  *adapter = q_vector->adapter;
1327         struct ixgbe_ring  *rx_ring;
1328         int r_idx;
1329         int i;
1330
1331         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1332         for (i = 0;  i < q_vector->rxr_count; i++) {
1333                 rx_ring = &(adapter->rx_ring[r_idx]);
1334                 rx_ring->total_bytes = 0;
1335                 rx_ring->total_packets = 0;
1336                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1337                                       r_idx + 1);
1338         }
1339
1340         if (!q_vector->rxr_count)
1341                 return IRQ_HANDLED;
1342
1343         /* disable interrupts on this vector only */
1344         ixgbe_irq_disable_queues(adapter, ((u64)1 << q_vector->v_idx));
1345         napi_schedule(&q_vector->napi);
1346
1347         return IRQ_HANDLED;
1348 }
1349
1350 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1351 {
1352         struct ixgbe_q_vector *q_vector = data;
1353         struct ixgbe_adapter  *adapter = q_vector->adapter;
1354         struct ixgbe_ring  *ring;
1355         int r_idx;
1356         int i;
1357
1358         if (!q_vector->txr_count && !q_vector->rxr_count)
1359                 return IRQ_HANDLED;
1360
1361         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1362         for (i = 0; i < q_vector->txr_count; i++) {
1363                 ring = &(adapter->tx_ring[r_idx]);
1364                 ring->total_bytes = 0;
1365                 ring->total_packets = 0;
1366                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1367                                       r_idx + 1);
1368         }
1369
1370         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1371         for (i = 0; i < q_vector->rxr_count; i++) {
1372                 ring = &(adapter->rx_ring[r_idx]);
1373                 ring->total_bytes = 0;
1374                 ring->total_packets = 0;
1375                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1376                                       r_idx + 1);
1377         }
1378
1379         /* disable interrupts on this vector only */
1380         ixgbe_irq_disable_queues(adapter, ((u64)1 << q_vector->v_idx));
1381         napi_schedule(&q_vector->napi);
1382
1383         return IRQ_HANDLED;
1384 }
1385
1386 /**
1387  * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1388  * @napi: napi struct with our devices info in it
1389  * @budget: amount of work driver is allowed to do this pass, in packets
1390  *
1391  * This function is optimized for cleaning one queue only on a single
1392  * q_vector!!!
1393  **/
1394 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1395 {
1396         struct ixgbe_q_vector *q_vector =
1397                                container_of(napi, struct ixgbe_q_vector, napi);
1398         struct ixgbe_adapter *adapter = q_vector->adapter;
1399         struct ixgbe_ring *rx_ring = NULL;
1400         int work_done = 0;
1401         long r_idx;
1402
1403         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1404         rx_ring = &(adapter->rx_ring[r_idx]);
1405 #ifdef CONFIG_IXGBE_DCA
1406         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1407                 ixgbe_update_rx_dca(adapter, rx_ring);
1408 #endif
1409
1410         ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1411
1412         /* If all Rx work done, exit the polling mode */
1413         if (work_done < budget) {
1414                 napi_complete(napi);
1415                 if (adapter->rx_itr_setting & 1)
1416                         ixgbe_set_itr_msix(q_vector);
1417                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1418                         ixgbe_irq_enable_queues(adapter,
1419                                                 ((u64)1 << q_vector->v_idx));
1420         }
1421
1422         return work_done;
1423 }
1424
1425 /**
1426  * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
1427  * @napi: napi struct with our devices info in it
1428  * @budget: amount of work driver is allowed to do this pass, in packets
1429  *
1430  * This function will clean more than one rx queue associated with a
1431  * q_vector.
1432  **/
1433 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
1434 {
1435         struct ixgbe_q_vector *q_vector =
1436                                container_of(napi, struct ixgbe_q_vector, napi);
1437         struct ixgbe_adapter *adapter = q_vector->adapter;
1438         struct ixgbe_ring *ring = NULL;
1439         int work_done = 0, i;
1440         long r_idx;
1441         bool tx_clean_complete = true;
1442
1443         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1444         for (i = 0; i < q_vector->txr_count; i++) {
1445                 ring = &(adapter->tx_ring[r_idx]);
1446 #ifdef CONFIG_IXGBE_DCA
1447                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1448                         ixgbe_update_tx_dca(adapter, ring);
1449 #endif
1450                 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
1451                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1452                                       r_idx + 1);
1453         }
1454
1455         /* attempt to distribute budget to each queue fairly, but don't allow
1456          * the budget to go below 1 because we'll exit polling */
1457         budget /= (q_vector->rxr_count ?: 1);
1458         budget = max(budget, 1);
1459         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1460         for (i = 0; i < q_vector->rxr_count; i++) {
1461                 ring = &(adapter->rx_ring[r_idx]);
1462 #ifdef CONFIG_IXGBE_DCA
1463                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1464                         ixgbe_update_rx_dca(adapter, ring);
1465 #endif
1466                 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
1467                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1468                                       r_idx + 1);
1469         }
1470
1471         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1472         ring = &(adapter->rx_ring[r_idx]);
1473         /* If all Rx work done, exit the polling mode */
1474         if (work_done < budget) {
1475                 napi_complete(napi);
1476                 if (adapter->rx_itr_setting & 1)
1477                         ixgbe_set_itr_msix(q_vector);
1478                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1479                         ixgbe_irq_enable_queues(adapter,
1480                                                 ((u64)1 << q_vector->v_idx));
1481                 return 0;
1482         }
1483
1484         return work_done;
1485 }
1486
1487 /**
1488  * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
1489  * @napi: napi struct with our devices info in it
1490  * @budget: amount of work driver is allowed to do this pass, in packets
1491  *
1492  * This function is optimized for cleaning one queue only on a single
1493  * q_vector!!!
1494  **/
1495 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
1496 {
1497         struct ixgbe_q_vector *q_vector =
1498                                container_of(napi, struct ixgbe_q_vector, napi);
1499         struct ixgbe_adapter *adapter = q_vector->adapter;
1500         struct ixgbe_ring *tx_ring = NULL;
1501         int work_done = 0;
1502         long r_idx;
1503
1504         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1505         tx_ring = &(adapter->tx_ring[r_idx]);
1506 #ifdef CONFIG_IXGBE_DCA
1507         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1508                 ixgbe_update_tx_dca(adapter, tx_ring);
1509 #endif
1510
1511         if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
1512                 work_done = budget;
1513
1514         /* If all Tx work done, exit the polling mode */
1515         if (work_done < budget) {
1516                 napi_complete(napi);
1517                 if (adapter->tx_itr_setting & 1)
1518                         ixgbe_set_itr_msix(q_vector);
1519                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1520                         ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
1521         }
1522
1523         return work_done;
1524 }
1525
1526 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
1527                                      int r_idx)
1528 {
1529         struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
1530
1531         set_bit(r_idx, q_vector->rxr_idx);
1532         q_vector->rxr_count++;
1533 }
1534
1535 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
1536                                      int t_idx)
1537 {
1538         struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
1539
1540         set_bit(t_idx, q_vector->txr_idx);
1541         q_vector->txr_count++;
1542 }
1543
1544 /**
1545  * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
1546  * @adapter: board private structure to initialize
1547  * @vectors: allotted vector count for descriptor rings
1548  *
1549  * This function maps descriptor rings to the queue-specific vectors
1550  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
1551  * one vector per ring/queue, but on a constrained vector budget, we
1552  * group the rings as "efficiently" as possible.  You would add new
1553  * mapping configurations in here.
1554  **/
1555 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
1556                                       int vectors)
1557 {
1558         int v_start = 0;
1559         int rxr_idx = 0, txr_idx = 0;
1560         int rxr_remaining = adapter->num_rx_queues;
1561         int txr_remaining = adapter->num_tx_queues;
1562         int i, j;
1563         int rqpv, tqpv;
1564         int err = 0;
1565
1566         /* No mapping required if MSI-X is disabled. */
1567         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
1568                 goto out;
1569
1570         /*
1571          * The ideal configuration...
1572          * We have enough vectors to map one per queue.
1573          */
1574         if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1575                 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1576                         map_vector_to_rxq(adapter, v_start, rxr_idx);
1577
1578                 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1579                         map_vector_to_txq(adapter, v_start, txr_idx);
1580
1581                 goto out;
1582         }
1583
1584         /*
1585          * If we don't have enough vectors for a 1-to-1
1586          * mapping, we'll have to group them so there are
1587          * multiple queues per vector.
1588          */
1589         /* Re-adjusting *qpv takes care of the remainder. */
1590         for (i = v_start; i < vectors; i++) {
1591                 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
1592                 for (j = 0; j < rqpv; j++) {
1593                         map_vector_to_rxq(adapter, i, rxr_idx);
1594                         rxr_idx++;
1595                         rxr_remaining--;
1596                 }
1597         }
1598         for (i = v_start; i < vectors; i++) {
1599                 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
1600                 for (j = 0; j < tqpv; j++) {
1601                         map_vector_to_txq(adapter, i, txr_idx);
1602                         txr_idx++;
1603                         txr_remaining--;
1604                 }
1605         }
1606
1607 out:
1608         return err;
1609 }
1610
1611 /**
1612  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
1613  * @adapter: board private structure
1614  *
1615  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
1616  * interrupts from the kernel.
1617  **/
1618 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
1619 {
1620         struct net_device *netdev = adapter->netdev;
1621         irqreturn_t (*handler)(int, void *);
1622         int i, vector, q_vectors, err;
1623         int ri=0, ti=0;
1624
1625         /* Decrement for Other and TCP Timer vectors */
1626         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1627
1628         /* Map the Tx/Rx rings to the vectors we were allotted. */
1629         err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
1630         if (err)
1631                 goto out;
1632
1633 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
1634                          (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
1635                          &ixgbe_msix_clean_many)
1636         for (vector = 0; vector < q_vectors; vector++) {
1637                 handler = SET_HANDLER(adapter->q_vector[vector]);
1638
1639                 if(handler == &ixgbe_msix_clean_rx) {
1640                         sprintf(adapter->name[vector], "%s-%s-%d",
1641                                 netdev->name, "rx", ri++);
1642                 }
1643                 else if(handler == &ixgbe_msix_clean_tx) {
1644                         sprintf(adapter->name[vector], "%s-%s-%d",
1645                                 netdev->name, "tx", ti++);
1646                 }
1647                 else
1648                         sprintf(adapter->name[vector], "%s-%s-%d",
1649                                 netdev->name, "TxRx", vector);
1650
1651                 err = request_irq(adapter->msix_entries[vector].vector,
1652                                   handler, 0, adapter->name[vector],
1653                                   adapter->q_vector[vector]);
1654                 if (err) {
1655                         DPRINTK(PROBE, ERR,
1656                                 "request_irq failed for MSIX interrupt "
1657                                 "Error: %d\n", err);
1658                         goto free_queue_irqs;
1659                 }
1660         }
1661
1662         sprintf(adapter->name[vector], "%s:lsc", netdev->name);
1663         err = request_irq(adapter->msix_entries[vector].vector,
1664                           &ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
1665         if (err) {
1666                 DPRINTK(PROBE, ERR,
1667                         "request_irq for msix_lsc failed: %d\n", err);
1668                 goto free_queue_irqs;
1669         }
1670
1671         return 0;
1672
1673 free_queue_irqs:
1674         for (i = vector - 1; i >= 0; i--)
1675                 free_irq(adapter->msix_entries[--vector].vector,
1676                          adapter->q_vector[i]);
1677         adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
1678         pci_disable_msix(adapter->pdev);
1679         kfree(adapter->msix_entries);
1680         adapter->msix_entries = NULL;
1681 out:
1682         return err;
1683 }
1684
1685 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
1686 {
1687         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
1688         u8 current_itr;
1689         u32 new_itr = q_vector->eitr;
1690         struct ixgbe_ring *rx_ring = &adapter->rx_ring[0];
1691         struct ixgbe_ring *tx_ring = &adapter->tx_ring[0];
1692
1693         q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
1694                                             q_vector->tx_itr,
1695                                             tx_ring->total_packets,
1696                                             tx_ring->total_bytes);
1697         q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
1698                                             q_vector->rx_itr,
1699                                             rx_ring->total_packets,
1700                                             rx_ring->total_bytes);
1701
1702         current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1703
1704         switch (current_itr) {
1705         /* counts and packets in update_itr are dependent on these numbers */
1706         case lowest_latency:
1707                 new_itr = 100000;
1708                 break;
1709         case low_latency:
1710                 new_itr = 20000; /* aka hwitr = ~200 */
1711                 break;
1712         case bulk_latency:
1713                 new_itr = 8000;
1714                 break;
1715         default:
1716                 break;
1717         }
1718
1719         if (new_itr != q_vector->eitr) {
1720                 /* do an exponential smoothing */
1721                 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1722
1723                 /* save the algorithm value here, not the smoothed one */
1724                 q_vector->eitr = new_itr;
1725
1726                 ixgbe_write_eitr(q_vector);
1727         }
1728
1729         return;
1730 }
1731
1732 /**
1733  * ixgbe_irq_enable - Enable default interrupt generation settings
1734  * @adapter: board private structure
1735  **/
1736 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter)
1737 {
1738         u32 mask;
1739
1740         mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
1741         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
1742                 mask |= IXGBE_EIMS_GPI_SDP1;
1743         if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1744                 mask |= IXGBE_EIMS_ECC;
1745                 mask |= IXGBE_EIMS_GPI_SDP1;
1746                 mask |= IXGBE_EIMS_GPI_SDP2;
1747         }
1748         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
1749             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
1750                 mask |= IXGBE_EIMS_FLOW_DIR;
1751
1752         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1753         ixgbe_irq_enable_queues(adapter, ~0);
1754         IXGBE_WRITE_FLUSH(&adapter->hw);
1755 }
1756
1757 /**
1758  * ixgbe_intr - legacy mode Interrupt Handler
1759  * @irq: interrupt number
1760  * @data: pointer to a network interface device structure
1761  **/
1762 static irqreturn_t ixgbe_intr(int irq, void *data)
1763 {
1764         struct net_device *netdev = data;
1765         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1766         struct ixgbe_hw *hw = &adapter->hw;
1767         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
1768         u32 eicr;
1769
1770         /*
1771          * Workaround for silicon errata.  Mask the interrupts
1772          * before the read of EICR.
1773          */
1774         IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
1775
1776         /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
1777          * therefore no explict interrupt disable is necessary */
1778         eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
1779         if (!eicr) {
1780                 /* shared interrupt alert!
1781                  * make sure interrupts are enabled because the read will
1782                  * have disabled interrupts due to EIAM */
1783                 ixgbe_irq_enable(adapter);
1784                 return IRQ_NONE;        /* Not our interrupt */
1785         }
1786
1787         if (eicr & IXGBE_EICR_LSC)
1788                 ixgbe_check_lsc(adapter);
1789
1790         if (hw->mac.type == ixgbe_mac_82599EB)
1791                 ixgbe_check_sfp_event(adapter, eicr);
1792
1793         ixgbe_check_fan_failure(adapter, eicr);
1794
1795         if (napi_schedule_prep(&(q_vector->napi))) {
1796                 adapter->tx_ring[0].total_packets = 0;
1797                 adapter->tx_ring[0].total_bytes = 0;
1798                 adapter->rx_ring[0].total_packets = 0;
1799                 adapter->rx_ring[0].total_bytes = 0;
1800                 /* would disable interrupts here but EIAM disabled it */
1801                 __napi_schedule(&(q_vector->napi));
1802         }
1803
1804         return IRQ_HANDLED;
1805 }
1806
1807 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
1808 {
1809         int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1810
1811         for (i = 0; i < q_vectors; i++) {
1812                 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
1813                 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
1814                 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
1815                 q_vector->rxr_count = 0;
1816                 q_vector->txr_count = 0;
1817         }
1818 }
1819
1820 /**
1821  * ixgbe_request_irq - initialize interrupts
1822  * @adapter: board private structure
1823  *
1824  * Attempts to configure interrupts using the best available
1825  * capabilities of the hardware and kernel.
1826  **/
1827 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
1828 {
1829         struct net_device *netdev = adapter->netdev;
1830         int err;
1831
1832         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1833                 err = ixgbe_request_msix_irqs(adapter);
1834         } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
1835                 err = request_irq(adapter->pdev->irq, &ixgbe_intr, 0,
1836                                   netdev->name, netdev);
1837         } else {
1838                 err = request_irq(adapter->pdev->irq, &ixgbe_intr, IRQF_SHARED,
1839                                   netdev->name, netdev);
1840         }
1841
1842         if (err)
1843                 DPRINTK(PROBE, ERR, "request_irq failed, Error %d\n", err);
1844
1845         return err;
1846 }
1847
1848 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
1849 {
1850         struct net_device *netdev = adapter->netdev;
1851
1852         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1853                 int i, q_vectors;
1854
1855                 q_vectors = adapter->num_msix_vectors;
1856
1857                 i = q_vectors - 1;
1858                 free_irq(adapter->msix_entries[i].vector, netdev);
1859
1860                 i--;
1861                 for (; i >= 0; i--) {
1862                         free_irq(adapter->msix_entries[i].vector,
1863                                  adapter->q_vector[i]);
1864                 }
1865
1866                 ixgbe_reset_q_vectors(adapter);
1867         } else {
1868                 free_irq(adapter->pdev->irq, netdev);
1869         }
1870 }
1871
1872 /**
1873  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
1874  * @adapter: board private structure
1875  **/
1876 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
1877 {
1878         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1879                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
1880         } else {
1881                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
1882                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
1883                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
1884         }
1885         IXGBE_WRITE_FLUSH(&adapter->hw);
1886         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1887                 int i;
1888                 for (i = 0; i < adapter->num_msix_vectors; i++)
1889                         synchronize_irq(adapter->msix_entries[i].vector);
1890         } else {
1891                 synchronize_irq(adapter->pdev->irq);
1892         }
1893 }
1894
1895 /**
1896  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
1897  *
1898  **/
1899 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
1900 {
1901         struct ixgbe_hw *hw = &adapter->hw;
1902
1903         IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
1904                         EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
1905
1906         ixgbe_set_ivar(adapter, 0, 0, 0);
1907         ixgbe_set_ivar(adapter, 1, 0, 0);
1908
1909         map_vector_to_rxq(adapter, 0, 0);
1910         map_vector_to_txq(adapter, 0, 0);
1911
1912         DPRINTK(HW, INFO, "Legacy interrupt IVAR setup done\n");
1913 }
1914
1915 /**
1916  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
1917  * @adapter: board private structure
1918  *
1919  * Configure the Tx unit of the MAC after a reset.
1920  **/
1921 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
1922 {
1923         u64 tdba;
1924         struct ixgbe_hw *hw = &adapter->hw;
1925         u32 i, j, tdlen, txctrl;
1926
1927         /* Setup the HW Tx Head and Tail descriptor pointers */
1928         for (i = 0; i < adapter->num_tx_queues; i++) {
1929                 struct ixgbe_ring *ring = &adapter->tx_ring[i];
1930                 j = ring->reg_idx;
1931                 tdba = ring->dma;
1932                 tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
1933                 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(j),
1934                                 (tdba & DMA_BIT_MASK(32)));
1935                 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(j), (tdba >> 32));
1936                 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(j), tdlen);
1937                 IXGBE_WRITE_REG(hw, IXGBE_TDH(j), 0);
1938                 IXGBE_WRITE_REG(hw, IXGBE_TDT(j), 0);
1939                 adapter->tx_ring[i].head = IXGBE_TDH(j);
1940                 adapter->tx_ring[i].tail = IXGBE_TDT(j);
1941                 /*
1942                  * Disable Tx Head Writeback RO bit, since this hoses
1943                  * bookkeeping if things aren't delivered in order.
1944                  */
1945                 switch (hw->mac.type) {
1946                 case ixgbe_mac_82598EB:
1947                         txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(j));
1948                         break;
1949                 case ixgbe_mac_82599EB:
1950                 default:
1951                         txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(j));
1952                         break;
1953                 }
1954                 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
1955                 switch (hw->mac.type) {
1956                 case ixgbe_mac_82598EB:
1957                         IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(j), txctrl);
1958                         break;
1959                 case ixgbe_mac_82599EB:
1960                 default:
1961                         IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(j), txctrl);
1962                         break;
1963                 }
1964         }
1965
1966         if (hw->mac.type == ixgbe_mac_82599EB) {
1967                 u32 rttdcs;
1968
1969                 /* disable the arbiter while setting MTQC */
1970                 rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
1971                 rttdcs |= IXGBE_RTTDCS_ARBDIS;
1972                 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
1973
1974                 /* We enable 8 traffic classes, DCB only */
1975                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
1976                         IXGBE_WRITE_REG(hw, IXGBE_MTQC, (IXGBE_MTQC_RT_ENA |
1977                                         IXGBE_MTQC_8TC_8TQ));
1978                 else
1979                         IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
1980
1981                 /* re-eable the arbiter */
1982                 rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
1983                 IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
1984         }
1985 }
1986
1987 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1988
1989 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
1990                                    struct ixgbe_ring *rx_ring)
1991 {
1992         u32 srrctl;
1993         int index;
1994         struct ixgbe_ring_feature *feature = adapter->ring_feature;
1995
1996         index = rx_ring->reg_idx;
1997         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1998                 unsigned long mask;
1999                 mask = (unsigned long) feature[RING_F_RSS].mask;
2000                 index = index & mask;
2001         }
2002         srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
2003
2004         srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2005         srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2006
2007         srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2008                   IXGBE_SRRCTL_BSIZEHDR_MASK;
2009
2010         if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
2011 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2012                 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2013 #else
2014                 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2015 #endif
2016                 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2017         } else {
2018                 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2019                           IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2020                 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2021         }
2022
2023         IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
2024 }
2025
2026 static u32 ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2027 {
2028         u32 mrqc = 0;
2029         int mask;
2030
2031         if (!(adapter->hw.mac.type == ixgbe_mac_82599EB))
2032                 return mrqc;
2033
2034         mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2035 #ifdef CONFIG_IXGBE_DCB
2036                                  | IXGBE_FLAG_DCB_ENABLED
2037 #endif
2038                                 );
2039
2040         switch (mask) {
2041         case (IXGBE_FLAG_RSS_ENABLED):
2042                 mrqc = IXGBE_MRQC_RSSEN;
2043                 break;
2044 #ifdef CONFIG_IXGBE_DCB
2045         case (IXGBE_FLAG_DCB_ENABLED):
2046                 mrqc = IXGBE_MRQC_RT8TCEN;
2047                 break;
2048 #endif /* CONFIG_IXGBE_DCB */
2049         default:
2050                 break;
2051         }
2052
2053         return mrqc;
2054 }
2055
2056 /**
2057  * ixgbe_configure_rscctl - enable RSC for the indicated ring
2058  * @adapter:    address of board private structure
2059  * @index:      index of ring to set
2060  * @rx_buf_len: rx buffer length
2061  **/
2062 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter, int index,
2063                                    int rx_buf_len)
2064 {
2065         struct ixgbe_ring *rx_ring;
2066         struct ixgbe_hw *hw = &adapter->hw;
2067         int j;
2068         u32 rscctrl;
2069
2070         rx_ring = &adapter->rx_ring[index];
2071         j = rx_ring->reg_idx;
2072         rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(j));
2073         rscctrl |= IXGBE_RSCCTL_RSCEN;
2074         /*
2075          * we must limit the number of descriptors so that the
2076          * total size of max desc * buf_len is not greater
2077          * than 65535
2078          */
2079         if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
2080 #if (MAX_SKB_FRAGS > 16)
2081                 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2082 #elif (MAX_SKB_FRAGS > 8)
2083                 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2084 #elif (MAX_SKB_FRAGS > 4)
2085                 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2086 #else
2087                 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2088 #endif
2089         } else {
2090                 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2091                         rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2092                 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2093                         rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2094                 else
2095                         rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2096         }
2097         IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(j), rscctrl);
2098 }
2099
2100 /**
2101  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
2102  * @adapter: board private structure
2103  *
2104  * Configure the Rx unit of the MAC after a reset.
2105  **/
2106 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
2107 {
2108         u64 rdba;
2109         struct ixgbe_hw *hw = &adapter->hw;
2110         struct ixgbe_ring *rx_ring;
2111         struct net_device *netdev = adapter->netdev;
2112         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2113         int i, j;
2114         u32 rdlen, rxctrl, rxcsum;
2115         static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2116                           0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2117                           0x6A3E67EA, 0x14364D17, 0x3BED200D};
2118         u32 fctrl, hlreg0;
2119         u32 reta = 0, mrqc = 0;
2120         u32 rdrxctl;
2121         int rx_buf_len;
2122
2123         /* Decide whether to use packet split mode or not */
2124         adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
2125
2126         /* Set the RX buffer length according to the mode */
2127         if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2128                 rx_buf_len = IXGBE_RX_HDR_SIZE;
2129                 if (hw->mac.type == ixgbe_mac_82599EB) {
2130                         /* PSRTYPE must be initialized in 82599 */
2131                         u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2132                                       IXGBE_PSRTYPE_UDPHDR |
2133                                       IXGBE_PSRTYPE_IPV4HDR |
2134                                       IXGBE_PSRTYPE_IPV6HDR |
2135                                       IXGBE_PSRTYPE_L2HDR;
2136                         IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(0), psrtype);
2137                 }
2138         } else {
2139                 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
2140                     (netdev->mtu <= ETH_DATA_LEN))
2141                         rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2142                 else
2143                         rx_buf_len = ALIGN(max_frame, 1024);
2144         }
2145
2146         fctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_FCTRL);
2147         fctrl |= IXGBE_FCTRL_BAM;
2148         fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
2149         fctrl |= IXGBE_FCTRL_PMCF;
2150         IXGBE_WRITE_REG(&adapter->hw, IXGBE_FCTRL, fctrl);
2151
2152         hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2153         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2154                 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
2155         else
2156                 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2157 #ifdef IXGBE_FCOE
2158         if (netdev->features & NETIF_F_FCOE_MTU)
2159                 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2160 #endif
2161         IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
2162
2163         rdlen = adapter->rx_ring[0].count * sizeof(union ixgbe_adv_rx_desc);
2164         /* disable receives while setting up the descriptors */
2165         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2166         IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
2167
2168         /*
2169          * Setup the HW Rx Head and Tail Descriptor Pointers and
2170          * the Base and Length of the Rx Descriptor Ring
2171          */
2172         for (i = 0; i < adapter->num_rx_queues; i++) {
2173                 rx_ring = &adapter->rx_ring[i];
2174                 rdba = rx_ring->dma;
2175                 j = rx_ring->reg_idx;
2176                 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(j), (rdba & DMA_BIT_MASK(32)));
2177                 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(j), (rdba >> 32));
2178                 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(j), rdlen);
2179                 IXGBE_WRITE_REG(hw, IXGBE_RDH(j), 0);
2180                 IXGBE_WRITE_REG(hw, IXGBE_RDT(j), 0);
2181                 rx_ring->head = IXGBE_RDH(j);
2182                 rx_ring->tail = IXGBE_RDT(j);
2183                 rx_ring->rx_buf_len = rx_buf_len;
2184
2185                 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
2186                         rx_ring->flags |= IXGBE_RING_RX_PS_ENABLED;
2187                 else
2188                         rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2189
2190 #ifdef IXGBE_FCOE
2191                 if (netdev->features & NETIF_F_FCOE_MTU) {
2192                         struct ixgbe_ring_feature *f;
2193                         f = &adapter->ring_feature[RING_F_FCOE];
2194                         if ((i >= f->mask) && (i < f->mask + f->indices)) {
2195                                 rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2196                                 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
2197                                         rx_ring->rx_buf_len =
2198                                                 IXGBE_FCOE_JUMBO_FRAME_SIZE;
2199                         }
2200                 }
2201
2202 #endif /* IXGBE_FCOE */
2203                 ixgbe_configure_srrctl(adapter, rx_ring);
2204         }
2205
2206         if (hw->mac.type == ixgbe_mac_82598EB) {
2207                 /*
2208                  * For VMDq support of different descriptor types or
2209                  * buffer sizes through the use of multiple SRRCTL
2210                  * registers, RDRXCTL.MVMEN must be set to 1
2211                  *
2212                  * also, the manual doesn't mention it clearly but DCA hints
2213                  * will only use queue 0's tags unless this bit is set.  Side
2214                  * effects of setting this bit are only that SRRCTL must be
2215                  * fully programmed [0..15]
2216                  */
2217                 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2218                 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
2219                 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2220         }
2221
2222         /* Program MRQC for the distribution of queues */
2223         mrqc = ixgbe_setup_mrqc(adapter);
2224
2225         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
2226                 /* Fill out redirection table */
2227                 for (i = 0, j = 0; i < 128; i++, j++) {
2228                         if (j == adapter->ring_feature[RING_F_RSS].indices)
2229                                 j = 0;
2230                         /* reta = 4-byte sliding window of
2231                          * 0x00..(indices-1)(indices-1)00..etc. */
2232                         reta = (reta << 8) | (j * 0x11);
2233                         if ((i & 3) == 3)
2234                                 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2235                 }
2236
2237                 /* Fill out hash function seeds */
2238                 for (i = 0; i < 10; i++)
2239                         IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2240
2241                 if (hw->mac.type == ixgbe_mac_82598EB)
2242                         mrqc |= IXGBE_MRQC_RSSEN;
2243                     /* Perform hash on these packet types */
2244                 mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2245                       | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2246                       | IXGBE_MRQC_RSS_FIELD_IPV4_UDP
2247                       | IXGBE_MRQC_RSS_FIELD_IPV6
2248                       | IXGBE_MRQC_RSS_FIELD_IPV6_TCP
2249                       | IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
2250         }
2251         IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2252
2253         rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2254
2255         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED ||
2256             adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED) {
2257                 /* Disable indicating checksum in descriptor, enables
2258                  * RSS hash */
2259                 rxcsum |= IXGBE_RXCSUM_PCSD;
2260         }
2261         if (!(rxcsum & IXGBE_RXCSUM_PCSD)) {
2262                 /* Enable IPv4 payload checksum for UDP fragments
2263                  * if PCSD is not set */
2264                 rxcsum |= IXGBE_RXCSUM_IPPCSE;
2265         }
2266
2267         IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2268
2269         if (hw->mac.type == ixgbe_mac_82599EB) {
2270                 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2271                 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
2272                 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
2273                 IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2274         }
2275
2276         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
2277                 /* Enable 82599 HW-RSC */
2278                 for (i = 0; i < adapter->num_rx_queues; i++)
2279                         ixgbe_configure_rscctl(adapter, i, rx_buf_len);
2280
2281                 /* Disable RSC for ACK packets */
2282                 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
2283                    (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
2284         }
2285 }
2286
2287 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2288 {
2289         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2290         struct ixgbe_hw *hw = &adapter->hw;
2291
2292         /* add VID to filter table */
2293         hw->mac.ops.set_vfta(&adapter->hw, vid, 0, true);
2294 }
2295
2296 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2297 {
2298         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2299         struct ixgbe_hw *hw = &adapter->hw;
2300
2301         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2302                 ixgbe_irq_disable(adapter);
2303
2304         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2305
2306         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2307                 ixgbe_irq_enable(adapter);
2308
2309         /* remove VID from filter table */
2310         hw->mac.ops.set_vfta(&adapter->hw, vid, 0, false);
2311 }
2312
2313 static void ixgbe_vlan_rx_register(struct net_device *netdev,
2314                                    struct vlan_group *grp)
2315 {
2316         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2317         u32 ctrl;
2318         int i, j;
2319
2320         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2321                 ixgbe_irq_disable(adapter);
2322         adapter->vlgrp = grp;
2323
2324         /*
2325          * For a DCB driver, always enable VLAN tag stripping so we can
2326          * still receive traffic from a DCB-enabled host even if we're
2327          * not in DCB mode.
2328          */
2329         ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_VLNCTRL);
2330         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2331                 ctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
2332                 ctrl &= ~IXGBE_VLNCTRL_CFIEN;
2333                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VLNCTRL, ctrl);
2334         } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
2335                 ctrl |= IXGBE_VLNCTRL_VFE;
2336                 /* enable VLAN tag insert/strip */
2337                 ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_VLNCTRL);
2338                 ctrl &= ~IXGBE_VLNCTRL_CFIEN;
2339                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VLNCTRL, ctrl);
2340                 for (i = 0; i < adapter->num_rx_queues; i++) {
2341                         j = adapter->rx_ring[i].reg_idx;
2342                         ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_RXDCTL(j));
2343                         ctrl |= IXGBE_RXDCTL_VME;
2344                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_RXDCTL(j), ctrl);
2345                 }
2346         }
2347         ixgbe_vlan_rx_add_vid(netdev, 0);
2348
2349         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2350                 ixgbe_irq_enable(adapter);
2351 }
2352
2353 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
2354 {
2355         ixgbe_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2356
2357         if (adapter->vlgrp) {
2358                 u16 vid;
2359                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2360                         if (!vlan_group_get_device(adapter->vlgrp, vid))
2361                                 continue;
2362                         ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
2363                 }
2364         }
2365 }
2366
2367 static u8 *ixgbe_addr_list_itr(struct ixgbe_hw *hw, u8 **mc_addr_ptr, u32 *vmdq)
2368 {
2369         struct dev_mc_list *mc_ptr;
2370         u8 *addr = *mc_addr_ptr;
2371         *vmdq = 0;
2372
2373         mc_ptr = container_of(addr, struct dev_mc_list, dmi_addr[0]);
2374         if (mc_ptr->next)
2375                 *mc_addr_ptr = mc_ptr->next->dmi_addr;
2376         else
2377                 *mc_addr_ptr = NULL;
2378
2379         return addr;
2380 }
2381
2382 /**
2383  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
2384  * @netdev: network interface device structure
2385  *
2386  * The set_rx_method entry point is called whenever the unicast/multicast
2387  * address list or the network interface flags are updated.  This routine is
2388  * responsible for configuring the hardware for proper unicast, multicast and
2389  * promiscuous mode.
2390  **/
2391 static void ixgbe_set_rx_mode(struct net_device *netdev)
2392 {
2393         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2394         struct ixgbe_hw *hw = &adapter->hw;
2395         u32 fctrl, vlnctrl;
2396         u8 *addr_list = NULL;
2397         int addr_count = 0;
2398
2399         /* Check for Promiscuous and All Multicast modes */
2400
2401         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
2402         vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2403
2404         if (netdev->flags & IFF_PROMISC) {
2405                 hw->addr_ctrl.user_set_promisc = 1;
2406                 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
2407                 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
2408         } else {
2409                 if (netdev->flags & IFF_ALLMULTI) {
2410                         fctrl |= IXGBE_FCTRL_MPE;
2411                         fctrl &= ~IXGBE_FCTRL_UPE;
2412                 } else {
2413                         fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
2414                 }
2415                 vlnctrl |= IXGBE_VLNCTRL_VFE;
2416                 hw->addr_ctrl.user_set_promisc = 0;
2417         }
2418
2419         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
2420         IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2421
2422         /* reprogram secondary unicast list */
2423         hw->mac.ops.update_uc_addr_list(hw, &netdev->uc.list);
2424
2425         /* reprogram multicast list */
2426         addr_count = netdev->mc_count;
2427         if (addr_count)
2428                 addr_list = netdev->mc_list->dmi_addr;
2429         hw->mac.ops.update_mc_addr_list(hw, addr_list, addr_count,
2430                                         ixgbe_addr_list_itr);
2431 }
2432
2433 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
2434 {
2435         int q_idx;
2436         struct ixgbe_q_vector *q_vector;
2437         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2438
2439         /* legacy and MSI only use one vector */
2440         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2441                 q_vectors = 1;
2442
2443         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2444                 struct napi_struct *napi;
2445                 q_vector = adapter->q_vector[q_idx];
2446                 napi = &q_vector->napi;
2447                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2448                         if (!q_vector->rxr_count || !q_vector->txr_count) {
2449                                 if (q_vector->txr_count == 1)
2450                                         napi->poll = &ixgbe_clean_txonly;
2451                                 else if (q_vector->rxr_count == 1)
2452                                         napi->poll = &ixgbe_clean_rxonly;
2453                         }
2454                 }
2455
2456                 napi_enable(napi);
2457         }
2458 }
2459
2460 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
2461 {
2462         int q_idx;
2463         struct ixgbe_q_vector *q_vector;
2464         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2465
2466         /* legacy and MSI only use one vector */
2467         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2468                 q_vectors = 1;
2469
2470         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2471                 q_vector = adapter->q_vector[q_idx];
2472                 napi_disable(&q_vector->napi);
2473         }
2474 }
2475
2476 #ifdef CONFIG_IXGBE_DCB
2477 /*
2478  * ixgbe_configure_dcb - Configure DCB hardware
2479  * @adapter: ixgbe adapter struct
2480  *
2481  * This is called by the driver on open to configure the DCB hardware.
2482  * This is also called by the gennetlink interface when reconfiguring
2483  * the DCB state.
2484  */
2485 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
2486 {
2487         struct ixgbe_hw *hw = &adapter->hw;
2488         u32 txdctl, vlnctrl;
2489         int i, j;
2490
2491         ixgbe_dcb_check_config(&adapter->dcb_cfg);
2492         ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_TX_CONFIG);
2493         ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_RX_CONFIG);
2494
2495         /* reconfigure the hardware */
2496         ixgbe_dcb_hw_config(&adapter->hw, &adapter->dcb_cfg);
2497
2498         for (i = 0; i < adapter->num_tx_queues; i++) {
2499                 j = adapter->tx_ring[i].reg_idx;
2500                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2501                 /* PThresh workaround for Tx hang with DFP enabled. */
2502                 txdctl |= 32;
2503                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2504         }
2505         /* Enable VLAN tag insert/strip */
2506         vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
2507         if (hw->mac.type == ixgbe_mac_82598EB) {
2508                 vlnctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
2509                 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2510                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2511         } else if (hw->mac.type == ixgbe_mac_82599EB) {
2512                 vlnctrl |= IXGBE_VLNCTRL_VFE;
2513                 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
2514                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
2515                 for (i = 0; i < adapter->num_rx_queues; i++) {
2516                         j = adapter->rx_ring[i].reg_idx;
2517                         vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2518                         vlnctrl |= IXGBE_RXDCTL_VME;
2519                         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
2520                 }
2521         }
2522         hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
2523 }
2524
2525 #endif
2526 static void ixgbe_configure(struct ixgbe_adapter *adapter)
2527 {
2528         struct net_device *netdev = adapter->netdev;
2529         struct ixgbe_hw *hw = &adapter->hw;
2530         int i;
2531
2532         ixgbe_set_rx_mode(netdev);
2533
2534         ixgbe_restore_vlan(adapter);
2535 #ifdef CONFIG_IXGBE_DCB
2536         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2537                 if (hw->mac.type == ixgbe_mac_82598EB)
2538                         netif_set_gso_max_size(netdev, 32768);
2539                 else
2540                         netif_set_gso_max_size(netdev, 65536);
2541                 ixgbe_configure_dcb(adapter);
2542         } else {
2543                 netif_set_gso_max_size(netdev, 65536);
2544         }
2545 #else
2546         netif_set_gso_max_size(netdev, 65536);
2547 #endif
2548
2549 #ifdef IXGBE_FCOE
2550         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
2551                 ixgbe_configure_fcoe(adapter);
2552
2553 #endif /* IXGBE_FCOE */
2554         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2555                 for (i = 0; i < adapter->num_tx_queues; i++)
2556                         adapter->tx_ring[i].atr_sample_rate =
2557                                                        adapter->atr_sample_rate;
2558                 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
2559         } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
2560                 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
2561         }
2562
2563         ixgbe_configure_tx(adapter);
2564         ixgbe_configure_rx(adapter);
2565         for (i = 0; i < adapter->num_rx_queues; i++)
2566                 ixgbe_alloc_rx_buffers(adapter, &adapter->rx_ring[i],
2567                                        (adapter->rx_ring[i].count - 1));
2568 }
2569
2570 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2571 {
2572         switch (hw->phy.type) {
2573         case ixgbe_phy_sfp_avago:
2574         case ixgbe_phy_sfp_ftl:
2575         case ixgbe_phy_sfp_intel:
2576         case ixgbe_phy_sfp_unknown:
2577         case ixgbe_phy_tw_tyco:
2578         case ixgbe_phy_tw_unknown:
2579                 return true;
2580         default:
2581                 return false;
2582         }
2583 }
2584
2585 /**
2586  * ixgbe_sfp_link_config - set up SFP+ link
2587  * @adapter: pointer to private adapter struct
2588  **/
2589 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
2590 {
2591         struct ixgbe_hw *hw = &adapter->hw;
2592
2593                 if (hw->phy.multispeed_fiber) {
2594                         /*
2595                          * In multispeed fiber setups, the device may not have
2596                          * had a physical connection when the driver loaded.
2597                          * If that's the case, the initial link configuration
2598                          * couldn't get the MAC into 10G or 1G mode, so we'll
2599                          * never have a link status change interrupt fire.
2600                          * We need to try and force an autonegotiation
2601                          * session, then bring up link.
2602                          */
2603                         hw->mac.ops.setup_sfp(hw);
2604                         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
2605                                 schedule_work(&adapter->multispeed_fiber_task);
2606                 } else {
2607                         /*
2608                          * Direct Attach Cu and non-multispeed fiber modules
2609                          * still need to be configured properly prior to
2610                          * attempting link.
2611                          */
2612                         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
2613                                 schedule_work(&adapter->sfp_config_module_task);
2614                 }
2615 }
2616
2617 /**
2618  * ixgbe_non_sfp_link_config - set up non-SFP+ link
2619  * @hw: pointer to private hardware struct
2620  *
2621  * Returns 0 on success, negative on failure
2622  **/
2623 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
2624 {
2625         u32 autoneg;
2626         bool negotiation, link_up = false;
2627         u32 ret = IXGBE_ERR_LINK_SETUP;
2628
2629         if (hw->mac.ops.check_link)
2630                 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
2631
2632         if (ret)
2633                 goto link_cfg_out;
2634
2635         if (hw->mac.ops.get_link_capabilities)
2636                 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
2637         if (ret)
2638                 goto link_cfg_out;
2639
2640         if (hw->mac.ops.setup_link)
2641                 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
2642 link_cfg_out:
2643         return ret;
2644 }
2645
2646 #define IXGBE_MAX_RX_DESC_POLL 10
2647 static inline void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2648                                               int rxr)
2649 {
2650         int j = adapter->rx_ring[rxr].reg_idx;
2651         int k;
2652
2653         for (k = 0; k < IXGBE_MAX_RX_DESC_POLL; k++) {
2654                 if (IXGBE_READ_REG(&adapter->hw,
2655                                    IXGBE_RXDCTL(j)) & IXGBE_RXDCTL_ENABLE)
2656                         break;
2657                 else
2658                         msleep(1);
2659         }
2660         if (k >= IXGBE_MAX_RX_DESC_POLL) {
2661                 DPRINTK(DRV, ERR, "RXDCTL.ENABLE on Rx queue %d "
2662                         "not set within the polling period\n", rxr);
2663         }
2664         ixgbe_release_rx_desc(&adapter->hw, &adapter->rx_ring[rxr],
2665                               (adapter->rx_ring[rxr].count - 1));
2666 }
2667
2668 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
2669 {
2670         struct net_device *netdev = adapter->netdev;
2671         struct ixgbe_hw *hw = &adapter->hw;
2672         int i, j = 0;
2673         int num_rx_rings = adapter->num_rx_queues;
2674         int err;
2675         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2676         u32 txdctl, rxdctl, mhadd;
2677         u32 dmatxctl;
2678         u32 gpie;
2679
2680         ixgbe_get_hw_control(adapter);
2681
2682         if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) ||
2683             (adapter->flags & IXGBE_FLAG_MSI_ENABLED)) {
2684                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2685                         gpie = (IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_EIAME |
2686                                 IXGBE_GPIE_PBA_SUPPORT | IXGBE_GPIE_OCD);
2687                 } else {
2688                         /* MSI only */
2689                         gpie = 0;
2690                 }
2691                 /* XXX: to interrupt immediately for EICS writes, enable this */
2692                 /* gpie |= IXGBE_GPIE_EIMEN; */
2693                 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2694         }
2695
2696         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
2697                 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
2698                  * specifically only auto mask tx and rx interrupts */
2699                 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
2700         }
2701
2702         /* Enable fan failure interrupt if media type is copper */
2703         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
2704                 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
2705                 gpie |= IXGBE_SDP1_GPIEN;
2706                 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2707         }
2708
2709         if (hw->mac.type == ixgbe_mac_82599EB) {
2710                 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
2711                 gpie |= IXGBE_SDP1_GPIEN;
2712                 gpie |= IXGBE_SDP2_GPIEN;
2713                 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
2714         }
2715
2716 #ifdef IXGBE_FCOE
2717         /* adjust max frame to be able to do baby jumbo for FCoE */
2718         if ((netdev->features & NETIF_F_FCOE_MTU) &&
2719             (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
2720                 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
2721
2722 #endif /* IXGBE_FCOE */
2723         mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
2724         if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
2725                 mhadd &= ~IXGBE_MHADD_MFS_MASK;
2726                 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
2727
2728                 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
2729         }
2730
2731         for (i = 0; i < adapter->num_tx_queues; i++) {
2732                 j = adapter->tx_ring[i].reg_idx;
2733                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2734                 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2735                 txdctl |= (8 << 16);
2736                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2737         }
2738
2739         if (hw->mac.type == ixgbe_mac_82599EB) {
2740                 /* DMATXCTL.EN must be set after all Tx queue config is done */
2741                 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2742                 dmatxctl |= IXGBE_DMATXCTL_TE;
2743                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2744         }
2745         for (i = 0; i < adapter->num_tx_queues; i++) {
2746                 j = adapter->tx_ring[i].reg_idx;
2747                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2748                 txdctl |= IXGBE_TXDCTL_ENABLE;
2749                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
2750         }
2751
2752         for (i = 0; i < num_rx_rings; i++) {
2753                 j = adapter->rx_ring[i].reg_idx;
2754                 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
2755                 /* enable PTHRESH=32 descriptors (half the internal cache)
2756                  * and HTHRESH=0 descriptors (to minimize latency on fetch),
2757                  * this also removes a pesky rx_no_buffer_count increment */
2758                 rxdctl |= 0x0020;
2759                 rxdctl |= IXGBE_RXDCTL_ENABLE;
2760                 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), rxdctl);
2761                 if (hw->mac.type == ixgbe_mac_82599EB)
2762                         ixgbe_rx_desc_queue_enable(adapter, i);
2763         }
2764         /* enable all receives */
2765         rxdctl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2766         if (hw->mac.type == ixgbe_mac_82598EB)
2767                 rxdctl |= (IXGBE_RXCTRL_DMBYPS | IXGBE_RXCTRL_RXEN);
2768         else
2769                 rxdctl |= IXGBE_RXCTRL_RXEN;
2770         hw->mac.ops.enable_rx_dma(hw, rxdctl);
2771
2772         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2773                 ixgbe_configure_msix(adapter);
2774         else
2775                 ixgbe_configure_msi_and_legacy(adapter);
2776
2777         clear_bit(__IXGBE_DOWN, &adapter->state);
2778         ixgbe_napi_enable_all(adapter);
2779
2780         /* clear any pending interrupts, may auto mask */
2781         IXGBE_READ_REG(hw, IXGBE_EICR);
2782
2783         ixgbe_irq_enable(adapter);
2784
2785         /*
2786          * If this adapter has a fan, check to see if we had a failure
2787          * before we enabled the interrupt.
2788          */
2789         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
2790                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
2791                 if (esdp & IXGBE_ESDP_SDP1)
2792                         DPRINTK(DRV, CRIT,
2793                                 "Fan has stopped, replace the adapter\n");
2794         }
2795
2796         /*
2797          * For hot-pluggable SFP+ devices, a new SFP+ module may have
2798          * arrived before interrupts were enabled but after probe.  Such
2799          * devices wouldn't have their type identified yet. We need to
2800          * kick off the SFP+ module setup first, then try to bring up link.
2801          * If we're not hot-pluggable SFP+, we just need to configure link
2802          * and bring it up.
2803          */
2804         if (hw->phy.type == ixgbe_phy_unknown) {
2805                 err = hw->phy.ops.identify(hw);
2806                 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
2807                         /*
2808                          * Take the device down and schedule the sfp tasklet
2809                          * which will unregister_netdev and log it.
2810                          */
2811                         ixgbe_down(adapter);
2812                         schedule_work(&adapter->sfp_config_module_task);
2813                         return err;
2814                 }
2815         }
2816
2817         if (ixgbe_is_sfp(hw)) {
2818                 ixgbe_sfp_link_config(adapter);
2819         } else {
2820                 err = ixgbe_non_sfp_link_config(hw);
2821                 if (err)
2822                         DPRINTK(PROBE, ERR, "link_config FAILED %d\n", err);
2823         }
2824
2825         for (i = 0; i < adapter->num_tx_queues; i++)
2826                 set_bit(__IXGBE_FDIR_INIT_DONE,
2827                         &(adapter->tx_ring[i].reinit_state));
2828
2829         /* enable transmits */
2830         netif_tx_start_all_queues(netdev);
2831
2832         /* bring the link up in the watchdog, this could race with our first
2833          * link up interrupt but shouldn't be a problem */
2834         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2835         adapter->link_check_timeout = jiffies;
2836         mod_timer(&adapter->watchdog_timer, jiffies);
2837         return 0;
2838 }
2839
2840 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
2841 {
2842         WARN_ON(in_interrupt());
2843         while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
2844                 msleep(1);
2845         ixgbe_down(adapter);
2846         ixgbe_up(adapter);
2847         clear_bit(__IXGBE_RESETTING, &adapter->state);
2848 }
2849
2850 int ixgbe_up(struct ixgbe_adapter *adapter)
2851 {
2852         /* hardware has been reset, we need to reload some things */
2853         ixgbe_configure(adapter);
2854
2855         return ixgbe_up_complete(adapter);
2856 }
2857
2858 void ixgbe_reset(struct ixgbe_adapter *adapter)
2859 {
2860         struct ixgbe_hw *hw = &adapter->hw;
2861         int err;
2862
2863         err = hw->mac.ops.init_hw(hw);
2864         switch (err) {
2865         case 0:
2866         case IXGBE_ERR_SFP_NOT_PRESENT:
2867                 break;
2868         case IXGBE_ERR_MASTER_REQUESTS_PENDING:
2869                 dev_err(&adapter->pdev->dev, "master disable timed out\n");
2870                 break;
2871         case IXGBE_ERR_EEPROM_VERSION:
2872                 /* We are running on a pre-production device, log a warning */
2873                 dev_warn(&adapter->pdev->dev, "This device is a pre-production "
2874                          "adapter/LOM.  Please be aware there may be issues "
2875                          "associated with your hardware.  If you are "
2876                          "experiencing problems please contact your Intel or "
2877                          "hardware representative who provided you with this "
2878                          "hardware.\n");
2879                 break;
2880         default:
2881                 dev_err(&adapter->pdev->dev, "Hardware Error: %d\n", err);
2882         }
2883
2884         /* reprogram the RAR[0] in case user changed it. */
2885         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
2886 }
2887
2888 /**
2889  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
2890  * @adapter: board private structure
2891  * @rx_ring: ring to free buffers from
2892  **/
2893 static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
2894                                 struct ixgbe_ring *rx_ring)
2895 {
2896         struct pci_dev *pdev = adapter->pdev;
2897         unsigned long size;
2898         unsigned int i;
2899
2900         /* Free all the Rx ring sk_buffs */
2901
2902         for (i = 0; i < rx_ring->count; i++) {
2903                 struct ixgbe_rx_buffer *rx_buffer_info;
2904
2905                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
2906                 if (rx_buffer_info->dma) {
2907                         pci_unmap_single(pdev, rx_buffer_info->dma,
2908                                          rx_ring->rx_buf_len,
2909                                          PCI_DMA_FROMDEVICE);
2910                         rx_buffer_info->dma = 0;
2911                 }
2912                 if (rx_buffer_info->skb) {
2913                         struct sk_buff *skb = rx_buffer_info->skb;
2914                         rx_buffer_info->skb = NULL;
2915                         do {
2916                                 struct sk_buff *this = skb;
2917                                 skb = skb->prev;
2918                                 dev_kfree_skb(this);
2919                         } while (skb);
2920                 }
2921                 if (!rx_buffer_info->page)
2922                         continue;
2923                 if (rx_buffer_info->page_dma) {
2924                         pci_unmap_page(pdev, rx_buffer_info->page_dma,
2925                                        PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
2926                         rx_buffer_info->page_dma = 0;
2927                 }
2928                 put_page(rx_buffer_info->page);
2929                 rx_buffer_info->page = NULL;
2930                 rx_buffer_info->page_offset = 0;
2931         }
2932
2933         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
2934         memset(rx_ring->rx_buffer_info, 0, size);
2935
2936         /* Zero out the descriptor ring */
2937         memset(rx_ring->desc, 0, rx_ring->size);
2938
2939         rx_ring->next_to_clean = 0;
2940         rx_ring->next_to_use = 0;
2941
2942         if (rx_ring->head)
2943                 writel(0, adapter->hw.hw_addr + rx_ring->head);
2944         if (rx_ring->tail)
2945                 writel(0, adapter->hw.hw_addr + rx_ring->tail);
2946 }
2947
2948 /**
2949  * ixgbe_clean_tx_ring - Free Tx Buffers
2950  * @adapter: board private structure
2951  * @tx_ring: ring to be cleaned
2952  **/
2953 static void ixgbe_clean_tx_ring(struct ixgbe_adapter *adapter,
2954                                 struct ixgbe_ring *tx_ring)
2955 {
2956         struct ixgbe_tx_buffer *tx_buffer_info;
2957         unsigned long size;
2958         unsigned int i;
2959
2960         /* Free all the Tx ring sk_buffs */
2961
2962         for (i = 0; i < tx_ring->count; i++) {
2963                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2964                 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
2965         }
2966
2967         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
2968         memset(tx_ring->tx_buffer_info, 0, size);
2969
2970         /* Zero out the descriptor ring */
2971         memset(tx_ring->desc, 0, tx_ring->size);
2972
2973         tx_ring->next_to_use = 0;
2974         tx_ring->next_to_clean = 0;
2975
2976         if (tx_ring->head)
2977                 writel(0, adapter->hw.hw_addr + tx_ring->head);
2978         if (tx_ring->tail)
2979                 writel(0, adapter->hw.hw_addr + tx_ring->tail);
2980 }
2981
2982 /**
2983  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
2984  * @adapter: board private structure
2985  **/
2986 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
2987 {
2988         int i;
2989
2990         for (i = 0; i < adapter->num_rx_queues; i++)
2991                 ixgbe_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2992 }
2993
2994 /**
2995  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
2996  * @adapter: board private structure
2997  **/
2998 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
2999 {
3000         int i;
3001
3002         for (i = 0; i < adapter->num_tx_queues; i++)
3003                 ixgbe_clean_tx_ring(adapter, &adapter->tx_ring[i]);
3004 }
3005
3006 void ixgbe_down(struct ixgbe_adapter *adapter)
3007 {
3008         struct net_device *netdev = adapter->netdev;
3009         struct ixgbe_hw *hw = &adapter->hw;
3010         u32 rxctrl;
3011         u32 txdctl;
3012         int i, j;
3013
3014         /* signal that we are down to the interrupt handler */
3015         set_bit(__IXGBE_DOWN, &adapter->state);
3016
3017         /* disable receives */
3018         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3019         IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3020
3021         netif_tx_disable(netdev);
3022
3023         IXGBE_WRITE_FLUSH(hw);
3024         msleep(10);
3025
3026         netif_tx_stop_all_queues(netdev);
3027
3028         ixgbe_irq_disable(adapter);
3029
3030         ixgbe_napi_disable_all(adapter);
3031
3032         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3033         del_timer_sync(&adapter->sfp_timer);
3034         del_timer_sync(&adapter->watchdog_timer);
3035         cancel_work_sync(&adapter->watchdog_task);
3036
3037         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3038             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3039                 cancel_work_sync(&adapter->fdir_reinit_task);
3040
3041         /* disable transmits in the hardware now that interrupts are off */
3042         for (i = 0; i < adapter->num_tx_queues; i++) {
3043                 j = adapter->tx_ring[i].reg_idx;
3044                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3045                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
3046                                 (txdctl & ~IXGBE_TXDCTL_ENABLE));
3047         }
3048         /* Disable the Tx DMA engine on 82599 */
3049         if (hw->mac.type == ixgbe_mac_82599EB)
3050                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
3051                                 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
3052                                  ~IXGBE_DMATXCTL_TE));
3053
3054         netif_carrier_off(netdev);
3055
3056         if (!pci_channel_offline(adapter->pdev))
3057                 ixgbe_reset(adapter);
3058         ixgbe_clean_all_tx_rings(adapter);
3059         ixgbe_clean_all_rx_rings(adapter);
3060
3061 #ifdef CONFIG_IXGBE_DCA
3062         /* since we reset the hardware DCA settings were cleared */
3063         ixgbe_setup_dca(adapter);
3064 #endif
3065 }
3066
3067 /**
3068  * ixgbe_poll - NAPI Rx polling callback
3069  * @napi: structure for representing this polling device
3070  * @budget: how many packets driver is allowed to clean
3071  *
3072  * This function is used for legacy and MSI, NAPI mode
3073  **/
3074 static int ixgbe_poll(struct napi_struct *napi, int budget)
3075 {
3076         struct ixgbe_q_vector *q_vector =
3077                                 container_of(napi, struct ixgbe_q_vector, napi);
3078         struct ixgbe_adapter *adapter = q_vector->adapter;
3079         int tx_clean_complete, work_done = 0;
3080
3081 #ifdef CONFIG_IXGBE_DCA
3082         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
3083                 ixgbe_update_tx_dca(adapter, adapter->tx_ring);
3084                 ixgbe_update_rx_dca(adapter, adapter->rx_ring);
3085         }
3086 #endif
3087
3088         tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring);
3089         ixgbe_clean_rx_irq(q_vector, adapter->rx_ring, &work_done, budget);
3090
3091         if (!tx_clean_complete)
3092                 work_done = budget;
3093
3094         /* If budget not fully consumed, exit the polling mode */
3095         if (work_done < budget) {
3096                 napi_complete(napi);
3097                 if (adapter->rx_itr_setting & 1)
3098                         ixgbe_set_itr(adapter);
3099                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3100                         ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
3101         }
3102         return work_done;
3103 }
3104
3105 /**
3106  * ixgbe_tx_timeout - Respond to a Tx Hang
3107  * @netdev: network interface device structure
3108  **/
3109 static void ixgbe_tx_timeout(struct net_device *netdev)
3110 {
3111         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3112
3113         /* Do the reset outside of interrupt context */
3114         schedule_work(&adapter->reset_task);
3115 }
3116
3117 static void ixgbe_reset_task(struct work_struct *work)
3118 {
3119         struct ixgbe_adapter *adapter;
3120         adapter = container_of(work, struct ixgbe_adapter, reset_task);
3121
3122         /* If we're already down or resetting, just bail */
3123         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
3124             test_bit(__IXGBE_RESETTING, &adapter->state))
3125                 return;
3126
3127         adapter->tx_timeout_count++;
3128
3129         ixgbe_reinit_locked(adapter);
3130 }
3131
3132 #ifdef CONFIG_IXGBE_DCB
3133 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
3134 {
3135         bool ret = false;
3136         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
3137
3138         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
3139                 return ret;
3140
3141         f->mask = 0x7 << 3;
3142         adapter->num_rx_queues = f->indices;
3143         adapter->num_tx_queues = f->indices;
3144         ret = true;
3145
3146         return ret;
3147 }
3148 #endif
3149
3150 /**
3151  * ixgbe_set_rss_queues: Allocate queues for RSS
3152  * @adapter: board private structure to initialize
3153  *
3154  * This is our "base" multiqueue mode.  RSS (Receive Side Scaling) will try
3155  * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
3156  *
3157  **/
3158 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
3159 {
3160         bool ret = false;
3161         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
3162
3163         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3164                 f->mask = 0xF;
3165                 adapter->num_rx_queues = f->indices;
3166                 adapter->num_tx_queues = f->indices;
3167                 ret = true;
3168         } else {
3169                 ret = false;
3170         }
3171
3172         return ret;
3173 }
3174
3175 /**
3176  * ixgbe_set_fdir_queues: Allocate queues for Flow Director
3177  * @adapter: board private structure to initialize
3178  *
3179  * Flow Director is an advanced Rx filter, attempting to get Rx flows back
3180  * to the original CPU that initiated the Tx session.  This runs in addition
3181  * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
3182  * Rx load across CPUs using RSS.
3183  *
3184  **/
3185 static bool inline ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
3186 {
3187         bool ret = false;
3188         struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
3189
3190         f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
3191         f_fdir->mask = 0;
3192
3193         /* Flow Director must have RSS enabled */
3194         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
3195             ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3196              (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
3197                 adapter->num_tx_queues = f_fdir->indices;
3198                 adapter->num_rx_queues = f_fdir->indices;
3199                 ret = true;
3200         } else {
3201                 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
3202                 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
3203         }
3204         return ret;
3205 }
3206
3207 #ifdef IXGBE_FCOE
3208 /**
3209  * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
3210  * @adapter: board private structure to initialize
3211  *
3212  * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
3213  * The ring feature mask is not used as a mask for FCoE, as it can take any 8
3214  * rx queues out of the max number of rx queues, instead, it is used as the
3215  * index of the first rx queue used by FCoE.
3216  *
3217  **/
3218 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
3219 {
3220         bool ret = false;
3221         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
3222
3223         f->indices = min((int)num_online_cpus(), f->indices);
3224         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
3225                 adapter->num_rx_queues = 1;
3226                 adapter->num_tx_queues = 1;
3227 #ifdef CONFIG_IXGBE_DCB
3228                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3229                         DPRINTK(PROBE, INFO, "FCoE enabled with DCB \n");
3230                         ixgbe_set_dcb_queues(adapter);
3231                 }
3232 #endif
3233                 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3234                         DPRINTK(PROBE, INFO, "FCoE enabled with RSS \n");
3235                         if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3236                             (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
3237                                 ixgbe_set_fdir_queues(adapter);
3238                         else
3239                                 ixgbe_set_rss_queues(adapter);
3240                 }
3241                 /* adding FCoE rx rings to the end */
3242                 f->mask = adapter->num_rx_queues;
3243                 adapter->num_rx_queues += f->indices;
3244                 adapter->num_tx_queues += f->indices;
3245
3246                 ret = true;
3247         }
3248
3249         return ret;
3250 }
3251
3252 #endif /* IXGBE_FCOE */
3253 /*
3254  * ixgbe_set_num_queues: Allocate queues for device, feature dependant
3255  * @adapter: board private structure to initialize
3256  *
3257  * This is the top level queue allocation routine.  The order here is very
3258  * important, starting with the "most" number of features turned on at once,
3259  * and ending with the smallest set of features.  This way large combinations
3260  * can be allocated if they're turned on, and smaller combinations are the
3261  * fallthrough conditions.
3262  *
3263  **/
3264 static void ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
3265 {
3266 #ifdef IXGBE_FCOE
3267         if (ixgbe_set_fcoe_queues(adapter))
3268                 goto done;
3269
3270 #endif /* IXGBE_FCOE */
3271 #ifdef CONFIG_IXGBE_DCB
3272         if (ixgbe_set_dcb_queues(adapter))
3273                 goto done;
3274
3275 #endif
3276         if (ixgbe_set_fdir_queues(adapter))
3277                 goto done;
3278
3279         if (ixgbe_set_rss_queues(adapter))
3280                 goto done;
3281
3282         /* fallback to base case */
3283         adapter->num_rx_queues = 1;
3284         adapter->num_tx_queues = 1;
3285
3286 done:
3287         /* Notify the stack of the (possibly) reduced Tx Queue count. */
3288         adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
3289 }
3290
3291 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
3292                                        int vectors)
3293 {
3294         int err, vector_threshold;
3295
3296         /* We'll want at least 3 (vector_threshold):
3297          * 1) TxQ[0] Cleanup
3298          * 2) RxQ[0] Cleanup
3299          * 3) Other (Link Status Change, etc.)
3300          * 4) TCP Timer (optional)
3301          */
3302         vector_threshold = MIN_MSIX_COUNT;
3303
3304         /* The more we get, the more we will assign to Tx/Rx Cleanup
3305          * for the separate queues...where Rx Cleanup >= Tx Cleanup.
3306          * Right now, we simply care about how many we'll get; we'll
3307          * set them up later while requesting irq's.
3308          */
3309         while (vectors >= vector_threshold) {
3310                 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
3311                                       vectors);
3312                 if (!err) /* Success in acquiring all requested vectors. */
3313                         break;
3314                 else if (err < 0)
3315                         vectors = 0; /* Nasty failure, quit now */
3316                 else /* err == number of vectors we should try again with */
3317                         vectors = err;
3318         }
3319
3320         if (vectors < vector_threshold) {
3321                 /* Can't allocate enough MSI-X interrupts?  Oh well.
3322                  * This just means we'll go with either a single MSI
3323                  * vector or fall back to legacy interrupts.
3324                  */
3325                 DPRINTK(HW, DEBUG, "Unable to allocate MSI-X interrupts\n");
3326                 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3327                 kfree(adapter->msix_entries);
3328                 adapter->msix_entries = NULL;
3329         } else {
3330                 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
3331                 /*
3332                  * Adjust for only the vectors we'll use, which is minimum
3333                  * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
3334                  * vectors we were allocated.
3335                  */
3336                 adapter->num_msix_vectors = min(vectors,
3337                                    adapter->max_msix_q_vectors + NON_Q_VECTORS);
3338         }
3339 }
3340
3341 /**
3342  * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
3343  * @adapter: board private structure to initialize
3344  *
3345  * Cache the descriptor ring offsets for RSS to the assigned rings.
3346  *
3347  **/
3348 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
3349 {
3350         int i;
3351         bool ret = false;
3352
3353         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3354                 for (i = 0; i < adapter->num_rx_queues; i++)
3355                         adapter->rx_ring[i].reg_idx = i;
3356                 for (i = 0; i < adapter->num_tx_queues; i++)
3357                         adapter->tx_ring[i].reg_idx = i;
3358                 ret = true;
3359         } else {
3360                 ret = false;
3361         }
3362
3363         return ret;
3364 }
3365
3366 #ifdef CONFIG_IXGBE_DCB
3367 /**
3368  * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
3369  * @adapter: board private structure to initialize
3370  *
3371  * Cache the descriptor ring offsets for DCB to the assigned rings.
3372  *
3373  **/
3374 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
3375 {
3376         int i;
3377         bool ret = false;
3378         int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
3379
3380         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3381                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3382                         /* the number of queues is assumed to be symmetric */
3383                         for (i = 0; i < dcb_i; i++) {
3384                                 adapter->rx_ring[i].reg_idx = i << 3;
3385                                 adapter->tx_ring[i].reg_idx = i << 2;
3386                         }
3387                         ret = true;
3388                 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
3389                         if (dcb_i == 8) {
3390                                 /*
3391                                  * Tx TC0 starts at: descriptor queue 0
3392                                  * Tx TC1 starts at: descriptor queue 32
3393                                  * Tx TC2 starts at: descriptor queue 64
3394                                  * Tx TC3 starts at: descriptor queue 80
3395                                  * Tx TC4 starts at: descriptor queue 96
3396                                  * Tx TC5 starts at: descriptor queue 104
3397                                  * Tx TC6 starts at: descriptor queue 112
3398                                  * Tx TC7 starts at: descriptor queue 120
3399                                  *
3400                                  * Rx TC0-TC7 are offset by 16 queues each
3401                                  */
3402                                 for (i = 0; i < 3; i++) {
3403                                         adapter->tx_ring[i].reg_idx = i << 5;
3404                                         adapter->rx_ring[i].reg_idx = i << 4;
3405                                 }
3406                                 for ( ; i < 5; i++) {
3407                                         adapter->tx_ring[i].reg_idx =
3408                                                                  ((i + 2) << 4);
3409                                         adapter->rx_ring[i].reg_idx = i << 4;
3410                                 }
3411                                 for ( ; i < dcb_i; i++) {
3412                                         adapter->tx_ring[i].reg_idx =
3413                                                                  ((i + 8) << 3);
3414                                         adapter->rx_ring[i].reg_idx = i << 4;
3415                                 }
3416
3417                                 ret = true;
3418                         } else if (dcb_i == 4) {
3419                                 /*
3420                                  * Tx TC0 starts at: descriptor queue 0
3421                                  * Tx TC1 starts at: descriptor queue 64
3422                                  * Tx TC2 starts at: descriptor queue 96
3423                                  * Tx TC3 starts at: descriptor queue 112
3424                                  *
3425                                  * Rx TC0-TC3 are offset by 32 queues each
3426                                  */
3427                                 adapter->tx_ring[0].reg_idx = 0;
3428                                 adapter->tx_ring[1].reg_idx = 64;
3429                                 adapter->tx_ring[2].reg_idx = 96;
3430                                 adapter->tx_ring[3].reg_idx = 112;
3431                                 for (i = 0 ; i < dcb_i; i++)
3432                                         adapter->rx_ring[i].reg_idx = i << 5;
3433
3434                                 ret = true;
3435                         } else {
3436                                 ret = false;
3437                         }
3438                 } else {
3439                         ret = false;
3440                 }
3441         } else {
3442                 ret = false;
3443         }
3444
3445         return ret;
3446 }
3447 #endif
3448
3449 /**
3450  * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
3451  * @adapter: board private structure to initialize
3452  *
3453  * Cache the descriptor ring offsets for Flow Director to the assigned rings.
3454  *
3455  **/
3456 static bool inline ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
3457 {
3458         int i;
3459         bool ret = false;
3460
3461         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
3462             ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3463              (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
3464                 for (i = 0; i < adapter->num_rx_queues; i++)
3465                         adapter->rx_ring[i].reg_idx = i;
3466                 for (i = 0; i < adapter->num_tx_queues; i++)
3467                         adapter->tx_ring[i].reg_idx = i;
3468                 ret = true;
3469         }
3470
3471         return ret;
3472 }
3473
3474 #ifdef IXGBE_FCOE
3475 /**
3476  * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
3477  * @adapter: board private structure to initialize
3478  *
3479  * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
3480  *
3481  */
3482 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
3483 {
3484         int i, fcoe_rx_i = 0, fcoe_tx_i = 0;
3485         bool ret = false;
3486         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
3487
3488         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
3489 #ifdef CONFIG_IXGBE_DCB
3490                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
3491                         struct ixgbe_fcoe *fcoe = &adapter->fcoe;
3492
3493                         ixgbe_cache_ring_dcb(adapter);
3494                         /* find out queues in TC for FCoE */
3495                         fcoe_rx_i = adapter->rx_ring[fcoe->tc].reg_idx + 1;
3496                         fcoe_tx_i = adapter->tx_ring[fcoe->tc].reg_idx + 1;
3497                         /*
3498                          * In 82599, the number of Tx queues for each traffic
3499                          * class for both 8-TC and 4-TC modes are:
3500                          * TCs  : TC0 TC1 TC2 TC3 TC4 TC5 TC6 TC7
3501                          * 8 TCs:  32  32  16  16   8   8   8   8
3502                          * 4 TCs:  64  64  32  32
3503                          * We have max 8 queues for FCoE, where 8 the is
3504                          * FCoE redirection table size. If TC for FCoE is
3505                          * less than or equal to TC3, we have enough queues
3506                          * to add max of 8 queues for FCoE, so we start FCoE
3507                          * tx descriptor from the next one, i.e., reg_idx + 1.
3508                          * If TC for FCoE is above TC3, implying 8 TC mode,
3509                          * and we need 8 for FCoE, we have to take all queues
3510                          * in that traffic class for FCoE.
3511                          */
3512                         if ((f->indices == IXGBE_FCRETA_SIZE) && (fcoe->tc > 3))
3513                                 fcoe_tx_i--;
3514                 }
3515 #endif /* CONFIG_IXGBE_DCB */
3516                 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3517                         if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
3518                             (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
3519                                 ixgbe_cache_ring_fdir(adapter);
3520                         else
3521                                 ixgbe_cache_ring_rss(adapter);
3522
3523                         fcoe_rx_i = f->mask;
3524                         fcoe_tx_i = f->mask;
3525                 }
3526                 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
3527                         adapter->rx_ring[f->mask + i].reg_idx = fcoe_rx_i;
3528                         adapter->tx_ring[f->mask + i].reg_idx = fcoe_tx_i;
3529                 }
3530                 ret = true;
3531         }
3532         return ret;
3533 }
3534
3535 #endif /* IXGBE_FCOE */
3536 /**
3537  * ixgbe_cache_ring_register - Descriptor ring to register mapping
3538  * @adapter: board private structure to initialize
3539  *
3540  * Once we know the feature-set enabled for the device, we'll cache
3541  * the register offset the descriptor ring is assigned to.
3542  *
3543  * Note, the order the various feature calls is important.  It must start with
3544  * the "most" features enabled at the same time, then trickle down to the
3545  * least amount of features turned on at once.
3546  **/
3547 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
3548 {
3549         /* start with default case */
3550         adapter->rx_ring[0].reg_idx = 0;
3551         adapter->tx_ring[0].reg_idx = 0;
3552
3553 #ifdef IXGBE_FCOE
3554         if (ixgbe_cache_ring_fcoe(adapter))
3555                 return;
3556
3557 #endif /* IXGBE_FCOE */
3558 #ifdef CONFIG_IXGBE_DCB
3559         if (ixgbe_cache_ring_dcb(adapter))
3560                 return;
3561
3562 #endif
3563         if (ixgbe_cache_ring_fdir(adapter))
3564                 return;
3565
3566         if (ixgbe_cache_ring_rss(adapter))
3567                 return;
3568 }
3569
3570 /**
3571  * ixgbe_alloc_queues - Allocate memory for all rings
3572  * @adapter: board private structure to initialize
3573  *
3574  * We allocate one ring per queue at run-time since we don't know the
3575  * number of queues at compile-time.  The polling_netdev array is
3576  * intended for Multiqueue, but should work fine with a single queue.
3577  **/
3578 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
3579 {
3580         int i;
3581
3582         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
3583                                    sizeof(struct ixgbe_ring), GFP_KERNEL);
3584         if (!adapter->tx_ring)
3585                 goto err_tx_ring_allocation;
3586
3587         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
3588                                    sizeof(struct ixgbe_ring), GFP_KERNEL);
3589         if (!adapter->rx_ring)
3590                 goto err_rx_ring_allocation;
3591
3592         for (i = 0; i < adapter->num_tx_queues; i++) {
3593                 adapter->tx_ring[i].count = adapter->tx_ring_count;
3594                 adapter->tx_ring[i].queue_index = i;
3595         }
3596
3597         for (i = 0; i < adapter->num_rx_queues; i++) {
3598                 adapter->rx_ring[i].count = adapter->rx_ring_count;
3599                 adapter->rx_ring[i].queue_index = i;
3600         }
3601
3602         ixgbe_cache_ring_register(adapter);
3603
3604         return 0;
3605
3606 err_rx_ring_allocation:
3607         kfree(adapter->tx_ring);
3608 err_tx_ring_allocation:
3609         return -ENOMEM;
3610 }
3611
3612 /**
3613  * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
3614  * @adapter: board private structure to initialize
3615  *
3616  * Attempt to configure the interrupts using the best available
3617  * capabilities of the hardware and the kernel.
3618  **/
3619 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
3620 {
3621         struct ixgbe_hw *hw = &adapter->hw;
3622         int err = 0;
3623         int vector, v_budget;
3624
3625         /*
3626          * It's easy to be greedy for MSI-X vectors, but it really
3627          * doesn't do us much good if we have a lot more vectors
3628          * than CPU's.  So let's be conservative and only ask for
3629          * (roughly) the same number of vectors as there are CPU's.
3630          */
3631         v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
3632                        (int)num_online_cpus()) + NON_Q_VECTORS;
3633
3634         /*
3635          * At the same time, hardware can only support a maximum of
3636          * hw.mac->max_msix_vectors vectors.  With features
3637          * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
3638          * descriptor queues supported by our device.  Thus, we cap it off in
3639          * those rare cases where the cpu count also exceeds our vector limit.
3640          */
3641         v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
3642
3643         /* A failure in MSI-X entry allocation isn't fatal, but it does
3644          * mean we disable MSI-X capabilities of the adapter. */
3645         adapter->msix_entries = kcalloc(v_budget,
3646                                         sizeof(struct msix_entry), GFP_KERNEL);
3647         if (adapter->msix_entries) {
3648                 for (vector = 0; vector < v_budget; vector++)
3649                         adapter->msix_entries[vector].entry = vector;
3650
3651                 ixgbe_acquire_msix_vectors(adapter, v_budget);
3652
3653                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3654                         goto out;
3655         }
3656
3657         adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
3658         adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
3659         adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
3660         adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
3661         adapter->atr_sample_rate = 0;
3662         ixgbe_set_num_queues(adapter);
3663
3664         err = pci_enable_msi(adapter->pdev);
3665         if (!err) {
3666                 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
3667         } else {
3668                 DPRINTK(HW, DEBUG, "Unable to allocate MSI interrupt, "
3669                         "falling back to legacy.  Error: %d\n", err);
3670                 /* reset err */
3671                 err = 0;
3672         }
3673
3674 out:
3675         return err;
3676 }
3677
3678 /**
3679  * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
3680  * @adapter: board private structure to initialize
3681  *
3682  * We allocate one q_vector per queue interrupt.  If allocation fails we
3683  * return -ENOMEM.
3684  **/
3685 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
3686 {
3687         int q_idx, num_q_vectors;
3688         struct ixgbe_q_vector *q_vector;
3689         int napi_vectors;
3690         int (*poll)(struct napi_struct *, int);
3691
3692         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3693                 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3694                 napi_vectors = adapter->num_rx_queues;
3695                 poll = &ixgbe_clean_rxtx_many;
3696         } else {
3697                 num_q_vectors = 1;
3698                 napi_vectors = 1;
3699                 poll = &ixgbe_poll;
3700         }
3701
3702         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
3703                 q_vector = kzalloc(sizeof(struct ixgbe_q_vector), GFP_KERNEL);
3704                 if (!q_vector)
3705                         goto err_out;
3706                 q_vector->adapter = adapter;
3707                 if (q_vector->txr_count && !q_vector->rxr_count)
3708                         q_vector->eitr = adapter->tx_eitr_param;
3709                 else
3710                         q_vector->eitr = adapter->rx_eitr_param;
3711                 q_vector->v_idx = q_idx;
3712                 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
3713                 adapter->q_vector[q_idx] = q_vector;
3714         }
3715
3716         return 0;
3717
3718 err_out:
3719         while (q_idx) {
3720                 q_idx--;
3721                 q_vector = adapter->q_vector[q_idx];
3722                 netif_napi_del(&q_vector->napi);
3723                 kfree(q_vector);
3724                 adapter->q_vector[q_idx] = NULL;
3725         }
3726         return -ENOMEM;
3727 }
3728
3729 /**
3730  * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
3731  * @adapter: board private structure to initialize
3732  *
3733  * This function frees the memory allocated to the q_vectors.  In addition if
3734  * NAPI is enabled it will delete any references to the NAPI struct prior
3735  * to freeing the q_vector.
3736  **/
3737 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
3738 {
3739         int q_idx, num_q_vectors;
3740
3741         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3742                 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3743         else
3744                 num_q_vectors = 1;
3745
3746         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
3747                 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
3748                 adapter->q_vector[q_idx] = NULL;
3749                 netif_napi_del(&q_vector->napi);
3750                 kfree(q_vector);
3751         }
3752 }
3753
3754 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
3755 {
3756         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3757                 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3758                 pci_disable_msix(adapter->pdev);
3759                 kfree(adapter->msix_entries);
3760                 adapter->msix_entries = NULL;
3761         } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
3762                 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
3763                 pci_disable_msi(adapter->pdev);
3764         }
3765         return;
3766 }
3767
3768 /**
3769  * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
3770  * @adapter: board private structure to initialize
3771  *
3772  * We determine which interrupt scheme to use based on...
3773  * - Kernel support (MSI, MSI-X)
3774  *   - which can be user-defined (via MODULE_PARAM)
3775  * - Hardware queue count (num_*_queues)
3776  *   - defined by miscellaneous hardware support/features (RSS, etc.)
3777  **/
3778 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
3779 {
3780         int err;
3781
3782         /* Number of supported queues */
3783         ixgbe_set_num_queues(adapter);
3784
3785         err = ixgbe_set_interrupt_capability(adapter);
3786         if (err) {
3787                 DPRINTK(PROBE, ERR, "Unable to setup interrupt capabilities\n");
3788                 goto err_set_interrupt;
3789         }
3790
3791         err = ixgbe_alloc_q_vectors(adapter);
3792         if (err) {
3793                 DPRINTK(PROBE, ERR, "Unable to allocate memory for queue "
3794                         "vectors\n");
3795                 goto err_alloc_q_vectors;
3796         }
3797
3798         err = ixgbe_alloc_queues(adapter);
3799         if (err) {
3800                 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
3801                 goto err_alloc_queues;
3802         }
3803
3804         DPRINTK(DRV, INFO, "Multiqueue %s: Rx Queue count = %u, "
3805                 "Tx Queue count = %u\n",
3806                 (adapter->num_rx_queues > 1) ? "Enabled" :
3807                 "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
3808
3809         set_bit(__IXGBE_DOWN, &adapter->state);
3810
3811         return 0;
3812
3813 err_alloc_queues:
3814         ixgbe_free_q_vectors(adapter);
3815 err_alloc_q_vectors:
3816         ixgbe_reset_interrupt_capability(adapter);
3817 err_set_interrupt:
3818         return err;
3819 }
3820
3821 /**
3822  * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
3823  * @adapter: board private structure to clear interrupt scheme on
3824  *
3825  * We go through and clear interrupt specific resources and reset the structure
3826  * to pre-load conditions
3827  **/
3828 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
3829 {
3830         kfree(adapter->tx_ring);
3831         kfree(adapter->rx_ring);
3832         adapter->tx_ring = NULL;
3833         adapter->rx_ring = NULL;
3834
3835         ixgbe_free_q_vectors(adapter);
3836         ixgbe_reset_interrupt_capability(adapter);
3837 }
3838
3839 /**
3840  * ixgbe_sfp_timer - worker thread to find a missing module
3841  * @data: pointer to our adapter struct
3842  **/
3843 static void ixgbe_sfp_timer(unsigned long data)
3844 {
3845         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
3846
3847         /*
3848          * Do the sfp_timer outside of interrupt context due to the
3849          * delays that sfp+ detection requires
3850          */
3851         schedule_work(&adapter->sfp_task);
3852 }
3853
3854 /**
3855  * ixgbe_sfp_task - worker thread to find a missing module
3856  * @work: pointer to work_struct containing our data
3857  **/
3858 static void ixgbe_sfp_task(struct work_struct *work)
3859 {
3860         struct ixgbe_adapter *adapter = container_of(work,
3861                                                      struct ixgbe_adapter,
3862                                                      sfp_task);
3863         struct ixgbe_hw *hw = &adapter->hw;
3864
3865         if ((hw->phy.type == ixgbe_phy_nl) &&
3866             (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
3867                 s32 ret = hw->phy.ops.identify_sfp(hw);
3868                 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
3869                         goto reschedule;
3870                 ret = hw->phy.ops.reset(hw);
3871                 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
3872                         dev_err(&adapter->pdev->dev, "failed to initialize "
3873                                 "because an unsupported SFP+ module type "
3874                                 "was detected.\n"
3875                                 "Reload the driver after installing a "
3876                                 "supported module.\n");
3877                         unregister_netdev(adapter->netdev);
3878                 } else {
3879                         DPRINTK(PROBE, INFO, "detected SFP+: %d\n",
3880                                 hw->phy.sfp_type);
3881                 }
3882                 /* don't need this routine any more */
3883                 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3884         }
3885         return;
3886 reschedule:
3887         if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
3888                 mod_timer(&adapter->sfp_timer,
3889                           round_jiffies(jiffies + (2 * HZ)));
3890 }
3891
3892 /**
3893  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
3894  * @adapter: board private structure to initialize
3895  *
3896  * ixgbe_sw_init initializes the Adapter private data structure.
3897  * Fields are initialized based on PCI device information and
3898  * OS network device settings (MTU size).
3899  **/
3900 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
3901 {
3902         struct ixgbe_hw *hw = &adapter->hw;
3903         struct pci_dev *pdev = adapter->pdev;
3904         unsigned int rss;
3905 #ifdef CONFIG_IXGBE_DCB
3906         int j;
3907         struct tc_configuration *tc;
3908 #endif
3909
3910         /* PCI config space info */
3911
3912         hw->vendor_id = pdev->vendor;
3913         hw->device_id = pdev->device;
3914         hw->revision_id = pdev->revision;
3915         hw->subsystem_vendor_id = pdev->subsystem_vendor;
3916         hw->subsystem_device_id = pdev->subsystem_device;
3917
3918         /* Set capability flags */
3919         rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
3920         adapter->ring_feature[RING_F_RSS].indices = rss;
3921         adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
3922         adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
3923         if (hw->mac.type == ixgbe_mac_82598EB) {
3924                 if (hw->device_id == IXGBE_DEV_ID_82598AT)
3925                         adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
3926                 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
3927         } else if (hw->mac.type == ixgbe_mac_82599EB) {
3928                 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
3929                 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
3930                 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
3931                 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
3932                 adapter->ring_feature[RING_F_FDIR].indices =
3933                                                          IXGBE_MAX_FDIR_INDICES;
3934                 adapter->atr_sample_rate = 20;
3935                 adapter->fdir_pballoc = 0;
3936 #ifdef IXGBE_FCOE
3937                 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
3938                 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
3939                 adapter->ring_feature[RING_F_FCOE].indices = 0;
3940                 /* Default traffic class to use for FCoE */
3941                 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
3942 #endif /* IXGBE_FCOE */
3943         }
3944
3945 #ifdef CONFIG_IXGBE_DCB
3946         /* Configure DCB traffic classes */
3947         for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
3948                 tc = &adapter->dcb_cfg.tc_config[j];
3949                 tc->path[DCB_TX_CONFIG].bwg_id = 0;
3950                 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
3951                 tc->path[DCB_RX_CONFIG].bwg_id = 0;
3952                 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
3953                 tc->dcb_pfc = pfc_disabled;
3954         }
3955         adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
3956         adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
3957         adapter->dcb_cfg.rx_pba_cfg = pba_equal;
3958         adapter->dcb_cfg.pfc_mode_enable = false;
3959         adapter->dcb_cfg.round_robin_enable = false;
3960         adapter->dcb_set_bitmap = 0x00;
3961         ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
3962                            adapter->ring_feature[RING_F_DCB].indices);
3963
3964 #endif
3965
3966         /* default flow control settings */
3967         hw->fc.requested_mode = ixgbe_fc_full;
3968         hw->fc.current_mode = ixgbe_fc_full;    /* init for ethtool output */
3969 #ifdef CONFIG_DCB
3970         adapter->last_lfc_mode = hw->fc.current_mode;
3971 #endif
3972         hw->fc.high_water = IXGBE_DEFAULT_FCRTH;
3973         hw->fc.low_water = IXGBE_DEFAULT_FCRTL;
3974         hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
3975         hw->fc.send_xon = true;
3976         hw->fc.disable_fc_autoneg = false;
3977
3978         /* enable itr by default in dynamic mode */
3979         adapter->rx_itr_setting = 1;
3980         adapter->rx_eitr_param = 20000;
3981         adapter->tx_itr_setting = 1;
3982         adapter->tx_eitr_param = 10000;
3983
3984         /* set defaults for eitr in MegaBytes */
3985         adapter->eitr_low = 10;
3986         adapter->eitr_high = 20;
3987
3988         /* set default ring sizes */
3989         adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
3990         adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
3991
3992         /* initialize eeprom parameters */
3993         if (ixgbe_init_eeprom_params_generic(hw)) {
3994                 dev_err(&pdev->dev, "EEPROM initialization failed\n");
3995                 return -EIO;
3996         }
3997
3998         /* enable rx csum by default */
3999         adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
4000
4001         set_bit(__IXGBE_DOWN, &adapter->state);
4002
4003         return 0;
4004 }
4005
4006 /**
4007  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4008  * @adapter: board private structure
4009  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
4010  *
4011  * Return 0 on success, negative on failure
4012  **/
4013 int ixgbe_setup_tx_resources(struct ixgbe_adapter *adapter,
4014                              struct ixgbe_ring *tx_ring)
4015 {
4016         struct pci_dev *pdev = adapter->pdev;
4017         int size;
4018
4019         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4020         tx_ring->tx_buffer_info = vmalloc(size);
4021         if (!tx_ring->tx_buffer_info)
4022                 goto err;
4023         memset(tx_ring->tx_buffer_info, 0, size);
4024
4025         /* round up to nearest 4K */
4026         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4027         tx_ring->size = ALIGN(tx_ring->size, 4096);
4028
4029         tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
4030                                              &tx_ring->dma);
4031         if (!tx_ring->desc)
4032                 goto err;
4033
4034         tx_ring->next_to_use = 0;
4035         tx_ring->next_to_clean = 0;
4036         tx_ring->work_limit = tx_ring->count;
4037         return 0;
4038
4039 err:
4040         vfree(tx_ring->tx_buffer_info);
4041         tx_ring->tx_buffer_info = NULL;
4042         DPRINTK(PROBE, ERR, "Unable to allocate memory for the transmit "
4043                             "descriptor ring\n");
4044         return -ENOMEM;
4045 }
4046
4047 /**
4048  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4049  * @adapter: board private structure
4050  *
4051  * If this function returns with an error, then it's possible one or
4052  * more of the rings is populated (while the rest are not).  It is the
4053  * callers duty to clean those orphaned rings.
4054  *
4055  * Return 0 on success, negative on failure
4056  **/
4057 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4058 {
4059         int i, err = 0;
4060
4061         for (i = 0; i < adapter->num_tx_queues; i++) {
4062                 err = ixgbe_setup_tx_resources(adapter, &adapter->tx_ring[i]);
4063                 if (!err)
4064                         continue;
4065                 DPRINTK(PROBE, ERR, "Allocation for Tx Queue %u failed\n", i);
4066                 break;
4067         }
4068
4069         return err;
4070 }
4071
4072 /**
4073  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4074  * @adapter: board private structure
4075  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
4076  *
4077  * Returns 0 on success, negative on failure
4078  **/
4079 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
4080                              struct ixgbe_ring *rx_ring)
4081 {
4082         struct pci_dev *pdev = adapter->pdev;
4083         int size;
4084
4085         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4086         rx_ring->rx_buffer_info = vmalloc(size);
4087         if (!rx_ring->rx_buffer_info) {
4088                 DPRINTK(PROBE, ERR,
4089                         "vmalloc allocation failed for the rx desc ring\n");
4090                 goto alloc_failed;
4091         }
4092         memset(rx_ring->rx_buffer_info, 0, size);
4093
4094         /* Round up to nearest 4K */
4095         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
4096         rx_ring->size = ALIGN(rx_ring->size, 4096);
4097
4098         rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size, &rx_ring->dma);
4099
4100         if (!rx_ring->desc) {
4101                 DPRINTK(PROBE, ERR,
4102                         "Memory allocation failed for the rx desc ring\n");
4103                 vfree(rx_ring->rx_buffer_info);
4104                 goto alloc_failed;
4105         }
4106
4107         rx_ring->next_to_clean = 0;
4108         rx_ring->next_to_use = 0;
4109
4110         return 0;
4111
4112 alloc_failed:
4113         return -ENOMEM;
4114 }
4115
4116 /**
4117  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
4118  * @adapter: board private structure
4119  *
4120  * If this function returns with an error, then it's possible one or
4121  * more of the rings is populated (while the rest are not).  It is the
4122  * callers duty to clean those orphaned rings.
4123  *
4124  * Return 0 on success, negative on failure
4125  **/
4126
4127 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
4128 {
4129         int i, err = 0;
4130
4131         for (i = 0; i < adapter->num_rx_queues; i++) {
4132                 err = ixgbe_setup_rx_resources(adapter, &adapter->rx_ring[i]);
4133                 if (!err)
4134                         continue;
4135                 DPRINTK(PROBE, ERR, "Allocation for Rx Queue %u failed\n", i);
4136                 break;
4137         }
4138
4139         return err;
4140 }
4141
4142 /**
4143  * ixgbe_free_tx_resources - Free Tx Resources per Queue
4144  * @adapter: board private structure
4145  * @tx_ring: Tx descriptor ring for a specific queue
4146  *
4147  * Free all transmit software resources
4148  **/
4149 void ixgbe_free_tx_resources(struct ixgbe_adapter *adapter,
4150                              struct ixgbe_ring *tx_ring)
4151 {
4152         struct pci_dev *pdev = adapter->pdev;
4153
4154         ixgbe_clean_tx_ring(adapter, tx_ring);
4155
4156         vfree(tx_ring->tx_buffer_info);
4157         tx_ring->tx_buffer_info = NULL;
4158
4159         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
4160
4161         tx_ring->desc = NULL;
4162 }
4163
4164 /**
4165  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
4166  * @adapter: board private structure
4167  *
4168  * Free all transmit software resources
4169  **/
4170 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
4171 {
4172         int i;
4173
4174         for (i = 0; i < adapter->num_tx_queues; i++)
4175                 if (adapter->tx_ring[i].desc)
4176                         ixgbe_free_tx_resources(adapter, &adapter->tx_ring[i]);
4177 }
4178
4179 /**
4180  * ixgbe_free_rx_resources - Free Rx Resources
4181  * @adapter: board private structure
4182  * @rx_ring: ring to clean the resources from
4183  *
4184  * Free all receive software resources
4185  **/
4186 void ixgbe_free_rx_resources(struct ixgbe_adapter *adapter,
4187                              struct ixgbe_ring *rx_ring)
4188 {
4189         struct pci_dev *pdev = adapter->pdev;
4190
4191         ixgbe_clean_rx_ring(adapter, rx_ring);
4192
4193         vfree(rx_ring->rx_buffer_info);
4194         rx_ring->rx_buffer_info = NULL;
4195
4196         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
4197
4198         rx_ring->desc = NULL;
4199 }
4200
4201 /**
4202  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
4203  * @adapter: board private structure
4204  *
4205  * Free all receive software resources
4206  **/
4207 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
4208 {
4209         int i;
4210
4211         for (i = 0; i < adapter->num_rx_queues; i++)
4212                 if (adapter->rx_ring[i].desc)
4213                         ixgbe_free_rx_resources(adapter, &adapter->rx_ring[i]);
4214 }
4215
4216 /**
4217  * ixgbe_change_mtu - Change the Maximum Transfer Unit
4218  * @netdev: network interface device structure
4219  * @new_mtu: new value for maximum frame size
4220  *
4221  * Returns 0 on success, negative on failure
4222  **/
4223 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
4224 {
4225         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4226         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4227
4228         /* MTU < 68 is an error and causes problems on some kernels */
4229         if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
4230                 return -EINVAL;
4231
4232         DPRINTK(PROBE, INFO, "changing MTU from %d to %d\n",
4233                 netdev->mtu, new_mtu);
4234         /* must set new MTU before calling down or up */
4235         netdev->mtu = new_mtu;
4236
4237         if (netif_running(netdev))
4238                 ixgbe_reinit_locked(adapter);
4239
4240         return 0;
4241 }
4242
4243 /**
4244  * ixgbe_open - Called when a network interface is made active
4245  * @netdev: network interface device structure
4246  *
4247  * Returns 0 on success, negative value on failure
4248  *
4249  * The open entry point is called when a network interface is made
4250  * active by the system (IFF_UP).  At this point all resources needed
4251  * for transmit and receive operations are allocated, the interrupt
4252  * handler is registered with the OS, the watchdog timer is started,
4253  * and the stack is notified that the interface is ready.
4254  **/
4255 static int ixgbe_open(struct net_device *netdev)
4256 {
4257         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4258         int err;
4259
4260         /* disallow open during test */
4261         if (test_bit(__IXGBE_TESTING, &adapter->state))
4262                 return -EBUSY;
4263
4264         netif_carrier_off(netdev);
4265
4266         /* allocate transmit descriptors */
4267         err = ixgbe_setup_all_tx_resources(adapter);
4268         if (err)
4269                 goto err_setup_tx;
4270
4271         /* allocate receive descriptors */
4272         err = ixgbe_setup_all_rx_resources(adapter);
4273         if (err)
4274                 goto err_setup_rx;
4275
4276         ixgbe_configure(adapter);
4277
4278         err = ixgbe_request_irq(adapter);
4279         if (err)
4280                 goto err_req_irq;
4281
4282         err = ixgbe_up_complete(adapter);
4283         if (err)
4284                 goto err_up;
4285
4286         netif_tx_start_all_queues(netdev);
4287
4288         return 0;
4289
4290 err_up:
4291         ixgbe_release_hw_control(adapter);
4292         ixgbe_free_irq(adapter);
4293 err_req_irq:
4294 err_setup_rx:
4295         ixgbe_free_all_rx_resources(adapter);
4296 err_setup_tx:
4297         ixgbe_free_all_tx_resources(adapter);
4298         ixgbe_reset(adapter);
4299
4300         return err;
4301 }
4302
4303 /**
4304  * ixgbe_close - Disables a network interface
4305  * @netdev: network interface device structure
4306  *
4307  * Returns 0, this is not allowed to fail
4308  *
4309  * The close entry point is called when an interface is de-activated
4310  * by the OS.  The hardware is still under the drivers control, but
4311  * needs to be disabled.  A global MAC reset is issued to stop the
4312  * hardware, and all transmit and receive resources are freed.
4313  **/
4314 static int ixgbe_close(struct net_device *netdev)
4315 {
4316         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4317
4318         ixgbe_down(adapter);
4319         ixgbe_free_irq(adapter);
4320
4321         ixgbe_free_all_tx_resources(adapter);
4322         ixgbe_free_all_rx_resources(adapter);
4323
4324         ixgbe_release_hw_control(adapter);
4325
4326         return 0;
4327 }
4328
4329 #ifdef CONFIG_PM
4330 static int ixgbe_resume(struct pci_dev *pdev)
4331 {
4332         struct net_device *netdev = pci_get_drvdata(pdev);
4333         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4334         u32 err;
4335
4336         pci_set_power_state(pdev, PCI_D0);
4337         pci_restore_state(pdev);
4338
4339         err = pci_enable_device_mem(pdev);
4340         if (err) {
4341                 printk(KERN_ERR "ixgbe: Cannot enable PCI device from "
4342                                 "suspend\n");
4343                 return err;
4344         }
4345         pci_set_master(pdev);
4346
4347         pci_wake_from_d3(pdev, false);
4348
4349         err = ixgbe_init_interrupt_scheme(adapter);
4350         if (err) {
4351                 printk(KERN_ERR "ixgbe: Cannot initialize interrupts for "
4352                                 "device\n");
4353                 return err;
4354         }
4355
4356         ixgbe_reset(adapter);
4357
4358         IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
4359
4360         if (netif_running(netdev)) {
4361                 err = ixgbe_open(adapter->netdev);
4362                 if (err)
4363                         return err;
4364         }
4365
4366         netif_device_attach(netdev);
4367
4368         return 0;
4369 }
4370 #endif /* CONFIG_PM */
4371
4372 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
4373 {
4374         struct net_device *netdev = pci_get_drvdata(pdev);
4375         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4376         struct ixgbe_hw *hw = &adapter->hw;
4377         u32 ctrl, fctrl;
4378         u32 wufc = adapter->wol;
4379 #ifdef CONFIG_PM
4380         int retval = 0;
4381 #endif
4382
4383         netif_device_detach(netdev);
4384
4385         if (netif_running(netdev)) {
4386                 ixgbe_down(adapter);
4387                 ixgbe_free_irq(adapter);
4388                 ixgbe_free_all_tx_resources(adapter);
4389                 ixgbe_free_all_rx_resources(adapter);
4390         }
4391         ixgbe_clear_interrupt_scheme(adapter);
4392
4393 #ifdef CONFIG_PM
4394         retval = pci_save_state(pdev);
4395         if (retval)
4396                 return retval;
4397
4398 #endif
4399         if (wufc) {
4400                 ixgbe_set_rx_mode(netdev);
4401
4402                 /* turn on all-multi mode if wake on multicast is enabled */
4403                 if (wufc & IXGBE_WUFC_MC) {
4404                         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4405                         fctrl |= IXGBE_FCTRL_MPE;
4406                         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4407                 }
4408
4409                 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
4410                 ctrl |= IXGBE_CTRL_GIO_DIS;
4411                 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
4412
4413                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
4414         } else {
4415                 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
4416                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
4417         }
4418
4419         if (wufc && hw->mac.type == ixgbe_mac_82599EB)
4420                 pci_wake_from_d3(pdev, true);
4421         else
4422                 pci_wake_from_d3(pdev, false);
4423
4424         *enable_wake = !!wufc;
4425
4426         ixgbe_release_hw_control(adapter);
4427
4428         pci_disable_device(pdev);
4429
4430         return 0;
4431 }
4432
4433 #ifdef CONFIG_PM
4434 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
4435 {
4436         int retval;
4437         bool wake;
4438
4439         retval = __ixgbe_shutdown(pdev, &wake);
4440         if (retval)
4441                 return retval;
4442
4443         if (wake) {
4444                 pci_prepare_to_sleep(pdev);
4445         } else {
4446                 pci_wake_from_d3(pdev, false);
4447                 pci_set_power_state(pdev, PCI_D3hot);
4448         }
4449
4450         return 0;
4451 }
4452 #endif /* CONFIG_PM */
4453
4454 static void ixgbe_shutdown(struct pci_dev *pdev)
4455 {
4456         bool wake;
4457
4458         __ixgbe_shutdown(pdev, &wake);
4459
4460         if (system_state == SYSTEM_POWER_OFF) {
4461                 pci_wake_from_d3(pdev, wake);
4462                 pci_set_power_state(pdev, PCI_D3hot);
4463         }
4464 }
4465
4466 /**
4467  * ixgbe_update_stats - Update the board statistics counters.
4468  * @adapter: board private structure
4469  **/
4470 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
4471 {
4472         struct net_device *netdev = adapter->netdev;
4473         struct ixgbe_hw *hw = &adapter->hw;
4474         u64 total_mpc = 0;
4475         u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
4476
4477         if (hw->mac.type == ixgbe_mac_82599EB) {
4478                 u64 rsc_count = 0;
4479                 for (i = 0; i < 16; i++)
4480                         adapter->hw_rx_no_dma_resources +=
4481                                              IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
4482                 for (i = 0; i < adapter->num_rx_queues; i++)
4483                         rsc_count += adapter->rx_ring[i].rsc_count;
4484                 adapter->rsc_count = rsc_count;
4485         }
4486
4487         adapter->stats.crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
4488         for (i = 0; i < 8; i++) {
4489                 /* for packet buffers not used, the register should read 0 */
4490                 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
4491                 missed_rx += mpc;
4492                 adapter->stats.mpc[i] += mpc;
4493                 total_mpc += adapter->stats.mpc[i];
4494                 if (hw->mac.type == ixgbe_mac_82598EB)
4495                         adapter->stats.rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
4496                 adapter->stats.qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
4497                 adapter->stats.qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
4498                 adapter->stats.qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
4499                 adapter->stats.qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
4500                 if (hw->mac.type == ixgbe_mac_82599EB) {
4501                         adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
4502                                                             IXGBE_PXONRXCNT(i));
4503                         adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
4504                                                            IXGBE_PXOFFRXCNT(i));
4505                         adapter->stats.qprdc[i] += IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
4506                 } else {
4507                         adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
4508                                                               IXGBE_PXONRXC(i));
4509                         adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
4510                                                              IXGBE_PXOFFRXC(i));
4511                 }
4512                 adapter->stats.pxontxc[i] += IXGBE_READ_REG(hw,
4513                                                             IXGBE_PXONTXC(i));
4514                 adapter->stats.pxofftxc[i] += IXGBE_READ_REG(hw,
4515                                                              IXGBE_PXOFFTXC(i));
4516         }
4517         adapter->stats.gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
4518         /* work around hardware counting issue */
4519         adapter->stats.gprc -= missed_rx;
4520
4521         /* 82598 hardware only has a 32 bit counter in the high register */
4522         if (hw->mac.type == ixgbe_mac_82599EB) {
4523                 u64 tmp;
4524                 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
4525                 tmp = IXGBE_READ_REG(hw, IXGBE_GORCH) & 0xF; /* 4 high bits of GORC */
4526                 adapter->stats.gorc += (tmp << 32);
4527                 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
4528                 tmp = IXGBE_READ_REG(hw, IXGBE_GOTCH) & 0xF; /* 4 high bits of GOTC */
4529                 adapter->stats.gotc += (tmp << 32);
4530                 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORL);
4531                 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
4532                 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
4533                 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
4534                 adapter->stats.fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
4535                 adapter->stats.fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
4536 #ifdef IXGBE_FCOE
4537                 adapter->stats.fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
4538                 adapter->stats.fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
4539                 adapter->stats.fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
4540                 adapter->stats.fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
4541                 adapter->stats.fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
4542                 adapter->stats.fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
4543 #endif /* IXGBE_FCOE */
4544         } else {
4545                 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
4546                 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
4547                 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
4548                 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
4549                 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORH);
4550         }
4551         bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
4552         adapter->stats.bprc += bprc;
4553         adapter->stats.mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
4554         if (hw->mac.type == ixgbe_mac_82598EB)
4555                 adapter->stats.mprc -= bprc;
4556         adapter->stats.roc += IXGBE_READ_REG(hw, IXGBE_ROC);
4557         adapter->stats.prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
4558         adapter->stats.prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
4559         adapter->stats.prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
4560         adapter->stats.prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
4561         adapter->stats.prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
4562         adapter->stats.prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
4563         adapter->stats.rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
4564         lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
4565         adapter->stats.lxontxc += lxon;
4566         lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
4567         adapter->stats.lxofftxc += lxoff;
4568         adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
4569         adapter->stats.gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
4570         adapter->stats.mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
4571         /*
4572          * 82598 errata - tx of flow control packets is included in tx counters
4573          */
4574         xon_off_tot = lxon + lxoff;
4575         adapter->stats.gptc -= xon_off_tot;
4576         adapter->stats.mptc -= xon_off_tot;
4577         adapter->stats.gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
4578         adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
4579         adapter->stats.rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
4580         adapter->stats.rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
4581         adapter->stats.tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
4582         adapter->stats.ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
4583         adapter->stats.ptc64 -= xon_off_tot;
4584         adapter->stats.ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
4585         adapter->stats.ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
4586         adapter->stats.ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
4587         adapter->stats.ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
4588         adapter->stats.ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
4589         adapter->stats.bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
4590
4591         /* Fill out the OS statistics structure */
4592         netdev->stats.multicast = adapter->stats.mprc;
4593
4594         /* Rx Errors */
4595         netdev->stats.rx_errors = adapter->stats.crcerrs +
4596                                        adapter->stats.rlec;
4597         netdev->stats.rx_dropped = 0;
4598         netdev->stats.rx_length_errors = adapter->stats.rlec;
4599         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4600         netdev->stats.rx_missed_errors = total_mpc;
4601 }
4602
4603 /**
4604  * ixgbe_watchdog - Timer Call-back
4605  * @data: pointer to adapter cast into an unsigned long
4606  **/
4607 static void ixgbe_watchdog(unsigned long data)
4608 {
4609         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4610         struct ixgbe_hw *hw = &adapter->hw;
4611         u64 eics = 0;
4612         int i;
4613
4614         /*
4615          *  Do the watchdog outside of interrupt context due to the lovely
4616          * delays that some of the newer hardware requires
4617          */
4618
4619         if (test_bit(__IXGBE_DOWN, &adapter->state))
4620                 goto watchdog_short_circuit;
4621
4622         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
4623                 /*
4624                  * for legacy and MSI interrupts don't set any bits
4625                  * that are enabled for EIAM, because this operation
4626                  * would set *both* EIMS and EICS for any bit in EIAM
4627                  */
4628                 IXGBE_WRITE_REG(hw, IXGBE_EICS,
4629                         (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
4630                 goto watchdog_reschedule;
4631         }
4632
4633         /* get one bit for every active tx/rx interrupt vector */
4634         for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
4635                 struct ixgbe_q_vector *qv = adapter->q_vector[i];
4636                 if (qv->rxr_count || qv->txr_count)
4637                         eics |= ((u64)1 << i);
4638         }
4639
4640         /* Cause software interrupt to ensure rx rings are cleaned */
4641         ixgbe_irq_rearm_queues(adapter, eics);
4642
4643 watchdog_reschedule:
4644         /* Reset the timer */
4645         mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
4646
4647 watchdog_short_circuit:
4648         schedule_work(&adapter->watchdog_task);
4649 }
4650
4651 /**
4652  * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
4653  * @work: pointer to work_struct containing our data
4654  **/
4655 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
4656 {
4657         struct ixgbe_adapter *adapter = container_of(work,
4658                                                      struct ixgbe_adapter,
4659                                                      multispeed_fiber_task);
4660         struct ixgbe_hw *hw = &adapter->hw;
4661         u32 autoneg;
4662         bool negotiation;
4663
4664         adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
4665         autoneg = hw->phy.autoneg_advertised;
4666         if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
4667                 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
4668         if (hw->mac.ops.setup_link)
4669                 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
4670         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
4671         adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
4672 }
4673
4674 /**
4675  * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
4676  * @work: pointer to work_struct containing our data
4677  **/
4678 static void ixgbe_sfp_config_module_task(struct work_struct *work)
4679 {
4680         struct ixgbe_adapter *adapter = container_of(work,
4681                                                      struct ixgbe_adapter,
4682                                                      sfp_config_module_task);
4683         struct ixgbe_hw *hw = &adapter->hw;
4684         u32 err;
4685
4686         adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
4687
4688         /* Time for electrical oscillations to settle down */
4689         msleep(100);
4690         err = hw->phy.ops.identify_sfp(hw);
4691
4692         if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4693                 dev_err(&adapter->pdev->dev, "failed to initialize because "
4694                         "an unsupported SFP+ module type was detected.\n"
4695                         "Reload the driver after installing a supported "
4696                         "module.\n");
4697                 unregister_netdev(adapter->netdev);
4698                 return;
4699         }
4700         hw->mac.ops.setup_sfp(hw);
4701
4702         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
4703                 /* This will also work for DA Twinax connections */
4704                 schedule_work(&adapter->multispeed_fiber_task);
4705         adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
4706 }
4707
4708 /**
4709  * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
4710  * @work: pointer to work_struct containing our data
4711  **/
4712 static void ixgbe_fdir_reinit_task(struct work_struct *work)
4713 {
4714         struct ixgbe_adapter *adapter = container_of(work,
4715                                                      struct ixgbe_adapter,
4716                                                      fdir_reinit_task);
4717         struct ixgbe_hw *hw = &adapter->hw;
4718         int i;
4719
4720         if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
4721                 for (i = 0; i < adapter->num_tx_queues; i++)
4722                         set_bit(__IXGBE_FDIR_INIT_DONE,
4723                                 &(adapter->tx_ring[i].reinit_state));
4724         } else {
4725                 DPRINTK(PROBE, ERR, "failed to finish FDIR re-initialization, "
4726                         "ignored adding FDIR ATR filters \n");
4727         }
4728         /* Done FDIR Re-initialization, enable transmits */
4729         netif_tx_start_all_queues(adapter->netdev);
4730 }
4731
4732 /**
4733  * ixgbe_watchdog_task - worker thread to bring link up
4734  * @work: pointer to work_struct containing our data
4735  **/
4736 static void ixgbe_watchdog_task(struct work_struct *work)
4737 {
4738         struct ixgbe_adapter *adapter = container_of(work,
4739                                                      struct ixgbe_adapter,
4740                                                      watchdog_task);
4741         struct net_device *netdev = adapter->netdev;
4742         struct ixgbe_hw *hw = &adapter->hw;
4743         u32 link_speed = adapter->link_speed;
4744         bool link_up = adapter->link_up;
4745         int i;
4746         struct ixgbe_ring *tx_ring;
4747         int some_tx_pending = 0;
4748
4749         adapter->flags |= IXGBE_FLAG_IN_WATCHDOG_TASK;
4750
4751         if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
4752                 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
4753                 if (link_up) {
4754 #ifdef CONFIG_DCB
4755                         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4756                                 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
4757                                         hw->mac.ops.fc_enable(hw, i);
4758                         } else {
4759                                 hw->mac.ops.fc_enable(hw, 0);
4760                         }
4761 #else
4762                         hw->mac.ops.fc_enable(hw, 0);
4763 #endif
4764                 }
4765
4766                 if (link_up ||
4767                     time_after(jiffies, (adapter->link_check_timeout +
4768                                          IXGBE_TRY_LINK_TIMEOUT))) {
4769                         adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
4770                         IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
4771                 }
4772                 adapter->link_up = link_up;
4773                 adapter->link_speed = link_speed;
4774         }
4775
4776         if (link_up) {
4777                 if (!netif_carrier_ok(netdev)) {
4778                         bool flow_rx, flow_tx;
4779
4780                         if (hw->mac.type == ixgbe_mac_82599EB) {
4781                                 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
4782                                 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
4783                                 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
4784                                 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
4785                         } else {
4786                                 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4787                                 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
4788                                 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
4789                                 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
4790                         }
4791
4792                         printk(KERN_INFO "ixgbe: %s NIC Link is Up %s, "
4793                                "Flow Control: %s\n",
4794                                netdev->name,
4795                                (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
4796                                 "10 Gbps" :
4797                                 (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
4798                                  "1 Gbps" : "unknown speed")),
4799                                ((flow_rx && flow_tx) ? "RX/TX" :
4800                                 (flow_rx ? "RX" :
4801                                 (flow_tx ? "TX" : "None"))));
4802
4803                         netif_carrier_on(netdev);
4804                 } else {
4805                         /* Force detection of hung controller */
4806                         adapter->detect_tx_hung = true;
4807                 }
4808         } else {
4809                 adapter->link_up = false;
4810                 adapter->link_speed = 0;
4811                 if (netif_carrier_ok(netdev)) {
4812                         printk(KERN_INFO "ixgbe: %s NIC Link is Down\n",
4813                                netdev->name);
4814                         netif_carrier_off(netdev);
4815                 }
4816         }
4817
4818         if (!netif_carrier_ok(netdev)) {
4819                 for (i = 0; i < adapter->num_tx_queues; i++) {
4820                         tx_ring = &adapter->tx_ring[i];
4821                         if (tx_ring->next_to_use != tx_ring->next_to_clean) {
4822                                 some_tx_pending = 1;
4823                                 break;
4824                         }
4825                 }
4826
4827                 if (some_tx_pending) {
4828                         /* We've lost link, so the controller stops DMA,
4829                          * but we've got queued Tx work that's never going
4830                          * to get done, so reset controller to flush Tx.
4831                          * (Do the reset outside of interrupt context).
4832                          */
4833                          schedule_work(&adapter->reset_task);
4834                 }
4835         }
4836
4837         ixgbe_update_stats(adapter);
4838         adapter->flags &= ~IXGBE_FLAG_IN_WATCHDOG_TASK;
4839 }
4840
4841 static int ixgbe_tso(struct ixgbe_adapter *adapter,
4842                      struct ixgbe_ring *tx_ring, struct sk_buff *skb,
4843                      u32 tx_flags, u8 *hdr_len)
4844 {
4845         struct ixgbe_adv_tx_context_desc *context_desc;
4846         unsigned int i;
4847         int err;
4848         struct ixgbe_tx_buffer *tx_buffer_info;
4849         u32 vlan_macip_lens = 0, type_tucmd_mlhl;
4850         u32 mss_l4len_idx, l4len;
4851
4852         if (skb_is_gso(skb)) {
4853                 if (skb_header_cloned(skb)) {
4854                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4855                         if (err)
4856                                 return err;
4857                 }
4858                 l4len = tcp_hdrlen(skb);
4859                 *hdr_len += l4len;
4860
4861                 if (skb->protocol == htons(ETH_P_IP)) {
4862                         struct iphdr *iph = ip_hdr(skb);
4863                         iph->tot_len = 0;
4864                         iph->check = 0;
4865                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
4866                                                                  iph->daddr, 0,
4867                                                                  IPPROTO_TCP,
4868                                                                  0);
4869                         adapter->hw_tso_ctxt++;
4870                 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
4871                         ipv6_hdr(skb)->payload_len = 0;
4872                         tcp_hdr(skb)->check =
4873                             ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4874                                              &ipv6_hdr(skb)->daddr,
4875                                              0, IPPROTO_TCP, 0);
4876                         adapter->hw_tso6_ctxt++;
4877                 }
4878
4879                 i = tx_ring->next_to_use;
4880
4881                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4882                 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
4883
4884                 /* VLAN MACLEN IPLEN */
4885                 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
4886                         vlan_macip_lens |=
4887                             (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
4888                 vlan_macip_lens |= ((skb_network_offset(skb)) <<
4889                                     IXGBE_ADVTXD_MACLEN_SHIFT);
4890                 *hdr_len += skb_network_offset(skb);
4891                 vlan_macip_lens |=
4892                     (skb_transport_header(skb) - skb_network_header(skb));
4893                 *hdr_len +=
4894                     (skb_transport_header(skb) - skb_network_header(skb));
4895                 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
4896                 context_desc->seqnum_seed = 0;
4897
4898                 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
4899                 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
4900                                    IXGBE_ADVTXD_DTYP_CTXT);
4901
4902                 if (skb->protocol == htons(ETH_P_IP))
4903                         type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
4904                 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
4905                 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
4906
4907                 /* MSS L4LEN IDX */
4908                 mss_l4len_idx =
4909                     (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
4910                 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
4911                 /* use index 1 for TSO */
4912                 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
4913                 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
4914
4915                 tx_buffer_info->time_stamp = jiffies;
4916                 tx_buffer_info->next_to_watch = i;
4917
4918                 i++;
4919                 if (i == tx_ring->count)
4920                         i = 0;
4921                 tx_ring->next_to_use = i;
4922
4923                 return true;
4924         }
4925         return false;
4926 }
4927
4928 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
4929                           struct ixgbe_ring *tx_ring,
4930                           struct sk_buff *skb, u32 tx_flags)
4931 {
4932         struct ixgbe_adv_tx_context_desc *context_desc;
4933         unsigned int i;
4934         struct ixgbe_tx_buffer *tx_buffer_info;
4935         u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
4936
4937         if (skb->ip_summed == CHECKSUM_PARTIAL ||
4938             (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
4939                 i = tx_ring->next_to_use;
4940                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
4941                 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
4942
4943                 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
4944                         vlan_macip_lens |=
4945                             (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
4946                 vlan_macip_lens |= (skb_network_offset(skb) <<
4947                                     IXGBE_ADVTXD_MACLEN_SHIFT);
4948                 if (skb->ip_summed == CHECKSUM_PARTIAL)
4949                         vlan_macip_lens |= (skb_transport_header(skb) -
4950                                             skb_network_header(skb));
4951
4952                 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
4953                 context_desc->seqnum_seed = 0;
4954
4955                 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
4956                                     IXGBE_ADVTXD_DTYP_CTXT);
4957
4958                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
4959                         switch (skb->protocol) {
4960                         case cpu_to_be16(ETH_P_IP):
4961                                 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
4962                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4963                                         type_tucmd_mlhl |=
4964                                                 IXGBE_ADVTXD_TUCMD_L4T_TCP;
4965                                 else if (ip_hdr(skb)->protocol == IPPROTO_SCTP)
4966                                         type_tucmd_mlhl |=
4967                                                 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
4968                                 break;
4969                         case cpu_to_be16(ETH_P_IPV6):
4970                                 /* XXX what about other V6 headers?? */
4971                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4972                                         type_tucmd_mlhl |=
4973                                                 IXGBE_ADVTXD_TUCMD_L4T_TCP;
4974                                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_SCTP)
4975                                         type_tucmd_mlhl |=
4976                                                 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
4977                                 break;
4978                         default:
4979                                 if (unlikely(net_ratelimit())) {
4980                                         DPRINTK(PROBE, WARNING,
4981                                          "partial checksum but proto=%x!\n",
4982                                          skb->protocol);
4983                                 }
4984                                 break;
4985                         }
4986                 }
4987
4988                 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
4989                 /* use index zero for tx checksum offload */
4990                 context_desc->mss_l4len_idx = 0;
4991
4992                 tx_buffer_info->time_stamp = jiffies;
4993                 tx_buffer_info->next_to_watch = i;
4994
4995                 adapter->hw_csum_tx_good++;
4996                 i++;
4997                 if (i == tx_ring->count)
4998                         i = 0;
4999                 tx_ring->next_to_use = i;
5000
5001                 return true;
5002         }
5003
5004         return false;
5005 }
5006
5007 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
5008                         struct ixgbe_ring *tx_ring,
5009                         struct sk_buff *skb, u32 tx_flags,
5010                         unsigned int first)
5011 {
5012         struct ixgbe_tx_buffer *tx_buffer_info;
5013         unsigned int len;
5014         unsigned int total = skb->len;
5015         unsigned int offset = 0, size, count = 0, i;
5016         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
5017         unsigned int f;
5018         dma_addr_t *map;
5019
5020         i = tx_ring->next_to_use;
5021
5022         if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
5023                 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
5024                 return 0;
5025         }
5026
5027         map = skb_shinfo(skb)->dma_maps;
5028
5029         if (tx_flags & IXGBE_TX_FLAGS_FCOE)
5030                 /* excluding fcoe_crc_eof for FCoE */
5031                 total -= sizeof(struct fcoe_crc_eof);
5032
5033         len = min(skb_headlen(skb), total);
5034         while (len) {
5035                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5036                 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
5037
5038                 tx_buffer_info->length = size;
5039                 tx_buffer_info->dma = skb_shinfo(skb)->dma_head + offset;
5040                 tx_buffer_info->time_stamp = jiffies;
5041                 tx_buffer_info->next_to_watch = i;
5042
5043                 len -= size;
5044                 total -= size;
5045                 offset += size;
5046                 count++;
5047
5048                 if (len) {
5049                         i++;
5050                         if (i == tx_ring->count)
5051                                 i = 0;
5052                 }
5053         }
5054
5055         for (f = 0; f < nr_frags; f++) {
5056                 struct skb_frag_struct *frag;
5057
5058                 frag = &skb_shinfo(skb)->frags[f];
5059                 len = min((unsigned int)frag->size, total);
5060                 offset = 0;
5061
5062                 while (len) {
5063                         i++;
5064                         if (i == tx_ring->count)
5065                                 i = 0;
5066
5067                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
5068                         size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
5069
5070                         tx_buffer_info->length = size;
5071                         tx_buffer_info->dma = map[f] + offset;
5072                         tx_buffer_info->time_stamp = jiffies;
5073                         tx_buffer_info->next_to_watch = i;
5074
5075                         len -= size;
5076                         total -= size;
5077                         offset += size;
5078                         count++;
5079                 }
5080                 if (total == 0)
5081                         break;
5082         }
5083
5084         tx_ring->tx_buffer_info[i].skb = skb;
5085         tx_ring->tx_buffer_info[first].next_to_watch = i;
5086
5087         return count;
5088 }
5089
5090 static void ixgbe_tx_queue(struct ixgbe_adapter *adapter,
5091                            struct ixgbe_ring *tx_ring,
5092                            int tx_flags, int count, u32 paylen, u8 hdr_len)
5093 {
5094         union ixgbe_adv_tx_desc *tx_desc = NULL;
5095         struct ixgbe_tx_buffer *tx_buffer_info;
5096         u32 olinfo_status = 0, cmd_type_len = 0;
5097         unsigned int i;
5098         u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
5099
5100         cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
5101
5102         cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
5103
5104         if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5105                 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
5106
5107         if (tx_flags & IXGBE_TX_FLAGS_TSO) {
5108                 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
5109
5110                 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
5111                                  IXGBE_ADVTXD_POPTS_SHIFT;
5112
5113                 /* use index 1 context for tso */
5114                 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5115                 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
5116                         olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
5117                                          IXGBE_ADVTXD_POPTS_SHIFT;
5118
5119         } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
5120                 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
5121                                  IXGBE_ADVTXD_POPTS_SHIFT;
5122
5123         if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
5124                 olinfo_status |= IXGBE_ADVTXD_CC;
5125                 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5126                 if (tx_flags & IXGBE_TX_FLAGS_FSO)
5127                         cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
5128         }
5129
5130         olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
5131
5132         i = tx_ring->next_to_use;
5133         while (count--) {
5134                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5135                 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
5136                 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
5137                 tx_desc->read.cmd_type_len =
5138                         cpu_to_le32(cmd_type_len | tx_buffer_info->length);
5139                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
5140                 i++;
5141                 if (i == tx_ring->count)
5142                         i = 0;
5143         }
5144
5145         tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
5146
5147         /*
5148          * Force memory writes to complete before letting h/w
5149          * know there are new descriptors to fetch.  (Only
5150          * applicable for weak-ordered memory model archs,
5151          * such as IA-64).
5152          */
5153         wmb();
5154
5155         tx_ring->next_to_use = i;
5156         writel(i, adapter->hw.hw_addr + tx_ring->tail);
5157 }
5158
5159 static void ixgbe_atr(struct ixgbe_adapter *adapter, struct sk_buff *skb,
5160                       int queue, u32 tx_flags)
5161 {
5162         /* Right now, we support IPv4 only */
5163         struct ixgbe_atr_input atr_input;
5164         struct tcphdr *th;
5165         struct iphdr *iph = ip_hdr(skb);
5166         struct ethhdr *eth = (struct ethhdr *)skb->data;
5167         u16 vlan_id, src_port, dst_port, flex_bytes;
5168         u32 src_ipv4_addr, dst_ipv4_addr;
5169         u8 l4type = 0;
5170
5171         /* check if we're UDP or TCP */
5172         if (iph->protocol == IPPROTO_TCP) {
5173                 th = tcp_hdr(skb);
5174                 src_port = th->source;
5175                 dst_port = th->dest;
5176                 l4type |= IXGBE_ATR_L4TYPE_TCP;
5177                 /* l4type IPv4 type is 0, no need to assign */
5178         } else {
5179                 /* Unsupported L4 header, just bail here */
5180                 return;
5181         }
5182
5183         memset(&atr_input, 0, sizeof(struct ixgbe_atr_input));
5184
5185         vlan_id = (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK) >>
5186                    IXGBE_TX_FLAGS_VLAN_SHIFT;
5187         src_ipv4_addr = iph->saddr;
5188         dst_ipv4_addr = iph->daddr;
5189         flex_bytes = eth->h_proto;
5190
5191         ixgbe_atr_set_vlan_id_82599(&atr_input, vlan_id);
5192         ixgbe_atr_set_src_port_82599(&atr_input, dst_port);
5193         ixgbe_atr_set_dst_port_82599(&atr_input, src_port);
5194         ixgbe_atr_set_flex_byte_82599(&atr_input, flex_bytes);
5195         ixgbe_atr_set_l4type_82599(&atr_input, l4type);
5196         /* src and dst are inverted, think how the receiver sees them */
5197         ixgbe_atr_set_src_ipv4_82599(&atr_input, dst_ipv4_addr);
5198         ixgbe_atr_set_dst_ipv4_82599(&atr_input, src_ipv4_addr);
5199
5200         /* This assumes the Rx queue and Tx queue are bound to the same CPU */
5201         ixgbe_fdir_add_signature_filter_82599(&adapter->hw, &atr_input, queue);
5202 }
5203
5204 static int __ixgbe_maybe_stop_tx(struct net_device *netdev,
5205                                  struct ixgbe_ring *tx_ring, int size)
5206 {
5207         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5208
5209         netif_stop_subqueue(netdev, tx_ring->queue_index);
5210         /* Herbert's original patch had:
5211          *  smp_mb__after_netif_stop_queue();
5212          * but since that doesn't exist yet, just open code it. */
5213         smp_mb();
5214
5215         /* We need to check again in a case another CPU has just
5216          * made room available. */
5217         if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
5218                 return -EBUSY;
5219
5220         /* A reprieve! - use start_queue because it doesn't call schedule */
5221         netif_start_subqueue(netdev, tx_ring->queue_index);
5222         ++adapter->restart_queue;
5223         return 0;
5224 }
5225
5226 static int ixgbe_maybe_stop_tx(struct net_device *netdev,
5227                               struct ixgbe_ring *tx_ring, int size)
5228 {
5229         if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
5230                 return 0;
5231         return __ixgbe_maybe_stop_tx(netdev, tx_ring, size);
5232 }
5233
5234 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
5235 {
5236         struct ixgbe_adapter *adapter = netdev_priv(dev);
5237
5238         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE)
5239                 return smp_processor_id();
5240
5241         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
5242                 return (skb->vlan_tci & IXGBE_TX_FLAGS_VLAN_PRIO_MASK) >> 13;
5243
5244         return skb_tx_hash(dev, skb);
5245 }
5246
5247 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
5248                                     struct net_device *netdev)
5249 {
5250         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5251         struct ixgbe_ring *tx_ring;
5252         unsigned int first;
5253         unsigned int tx_flags = 0;
5254         u8 hdr_len = 0;
5255         int r_idx = 0, tso;
5256         int count = 0;
5257         unsigned int f;
5258
5259         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
5260                 tx_flags |= vlan_tx_tag_get(skb);
5261                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5262                         tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
5263                         tx_flags |= (skb->queue_mapping << 13);
5264                 }
5265                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
5266                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
5267         } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5268                 if (skb->priority != TC_PRIO_CONTROL) {
5269                         tx_flags |= (skb->queue_mapping << 13);
5270                         tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
5271                         tx_flags |= IXGBE_TX_FLAGS_VLAN;
5272                 } else {
5273                         skb->queue_mapping =
5274                                 adapter->ring_feature[RING_F_DCB].indices-1;
5275                 }
5276         }
5277
5278         r_idx = skb->queue_mapping;
5279         tx_ring = &adapter->tx_ring[r_idx];
5280
5281         if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
5282             (skb->protocol == htons(ETH_P_FCOE))) {
5283                 tx_flags |= IXGBE_TX_FLAGS_FCOE;
5284 #ifdef IXGBE_FCOE
5285                 r_idx = smp_processor_id();
5286                 r_idx &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
5287                 r_idx += adapter->ring_feature[RING_F_FCOE].mask;
5288                 tx_ring = &adapter->tx_ring[r_idx];
5289 #endif
5290         }
5291         /* four things can cause us to need a context descriptor */
5292         if (skb_is_gso(skb) ||
5293             (skb->ip_summed == CHECKSUM_PARTIAL) ||
5294             (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
5295             (tx_flags & IXGBE_TX_FLAGS_FCOE))
5296                 count++;
5297
5298         count += TXD_USE_COUNT(skb_headlen(skb));
5299         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
5300                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
5301
5302         if (ixgbe_maybe_stop_tx(netdev, tx_ring, count)) {
5303                 adapter->tx_busy++;
5304                 return NETDEV_TX_BUSY;
5305         }
5306
5307         first = tx_ring->next_to_use;
5308         if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
5309 #ifdef IXGBE_FCOE
5310                 /* setup tx offload for FCoE */
5311                 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
5312                 if (tso < 0) {
5313                         dev_kfree_skb_any(skb);
5314                         return NETDEV_TX_OK;
5315                 }
5316                 if (tso)
5317                         tx_flags |= IXGBE_TX_FLAGS_FSO;
5318 #endif /* IXGBE_FCOE */
5319         } else {
5320                 if (skb->protocol == htons(ETH_P_IP))
5321                         tx_flags |= IXGBE_TX_FLAGS_IPV4;
5322                 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
5323                 if (tso < 0) {
5324                         dev_kfree_skb_any(skb);
5325                         return NETDEV_TX_OK;
5326                 }
5327
5328                 if (tso)
5329                         tx_flags |= IXGBE_TX_FLAGS_TSO;
5330                 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags) &&
5331                          (skb->ip_summed == CHECKSUM_PARTIAL))
5332                         tx_flags |= IXGBE_TX_FLAGS_CSUM;
5333         }
5334
5335         count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first);
5336         if (count) {
5337                 /* add the ATR filter if ATR is on */
5338                 if (tx_ring->atr_sample_rate) {
5339                         ++tx_ring->atr_count;
5340                         if ((tx_ring->atr_count >= tx_ring->atr_sample_rate) &&
5341                              test_bit(__IXGBE_FDIR_INIT_DONE,
5342                                       &tx_ring->reinit_state)) {
5343                                 ixgbe_atr(adapter, skb, tx_ring->queue_index,
5344                                           tx_flags);
5345                                 tx_ring->atr_count = 0;
5346                         }
5347                 }
5348                 ixgbe_tx_queue(adapter, tx_ring, tx_flags, count, skb->len,
5349                                hdr_len);
5350                 ixgbe_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
5351
5352         } else {
5353                 dev_kfree_skb_any(skb);
5354                 tx_ring->tx_buffer_info[first].time_stamp = 0;
5355                 tx_ring->next_to_use = first;
5356         }
5357
5358         return NETDEV_TX_OK;
5359 }
5360
5361 /**
5362  * ixgbe_get_stats - Get System Network Statistics
5363  * @netdev: network interface device structure
5364  *
5365  * Returns the address of the device statistics structure.
5366  * The statistics are actually updated from the timer callback.
5367  **/
5368 static struct net_device_stats *ixgbe_get_stats(struct net_device *netdev)
5369 {
5370         /* only return the current stats */
5371         return &netdev->stats;
5372 }
5373
5374 /**
5375  * ixgbe_set_mac - Change the Ethernet Address of the NIC
5376  * @netdev: network interface device structure
5377  * @p: pointer to an address structure
5378  *
5379  * Returns 0 on success, negative on failure
5380  **/
5381 static int ixgbe_set_mac(struct net_device *netdev, void *p)
5382 {
5383         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5384         struct ixgbe_hw *hw = &adapter->hw;
5385         struct sockaddr *addr = p;
5386
5387         if (!is_valid_ether_addr(addr->sa_data))
5388                 return -EADDRNOTAVAIL;
5389
5390         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
5391         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
5392
5393         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
5394
5395         return 0;
5396 }
5397
5398 static int
5399 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
5400 {
5401         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5402         struct ixgbe_hw *hw = &adapter->hw;
5403         u16 value;
5404         int rc;
5405
5406         if (prtad != hw->phy.mdio.prtad)
5407                 return -EINVAL;
5408         rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
5409         if (!rc)
5410                 rc = value;
5411         return rc;
5412 }
5413
5414 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
5415                             u16 addr, u16 value)
5416 {
5417         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5418         struct ixgbe_hw *hw = &adapter->hw;
5419
5420         if (prtad != hw->phy.mdio.prtad)
5421                 return -EINVAL;
5422         return hw->phy.ops.write_reg(hw, addr, devad, value);
5423 }
5424
5425 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
5426 {
5427         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5428
5429         return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
5430 }
5431
5432 /**
5433  * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
5434  * netdev->dev_addrs
5435  * @netdev: network interface device structure
5436  *
5437  * Returns non-zero on failure
5438  **/
5439 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
5440 {
5441         int err = 0;
5442         struct ixgbe_adapter *adapter = netdev_priv(dev);
5443         struct ixgbe_mac_info *mac = &adapter->hw.mac;
5444
5445         if (is_valid_ether_addr(mac->san_addr)) {
5446                 rtnl_lock();
5447                 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
5448                 rtnl_unlock();
5449         }
5450         return err;
5451 }
5452
5453 /**
5454  * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
5455  * netdev->dev_addrs
5456  * @netdev: network interface device structure
5457  *
5458  * Returns non-zero on failure
5459  **/
5460 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
5461 {
5462         int err = 0;
5463         struct ixgbe_adapter *adapter = netdev_priv(dev);
5464         struct ixgbe_mac_info *mac = &adapter->hw.mac;
5465
5466         if (is_valid_ether_addr(mac->san_addr)) {
5467                 rtnl_lock();
5468                 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
5469                 rtnl_unlock();
5470         }
5471         return err;
5472 }
5473
5474 #ifdef CONFIG_NET_POLL_CONTROLLER
5475 /*
5476  * Polling 'interrupt' - used by things like netconsole to send skbs
5477  * without having to re-enable interrupts. It's not called while
5478  * the interrupt routine is executing.
5479  */
5480 static void ixgbe_netpoll(struct net_device *netdev)
5481 {
5482         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5483         int i;
5484
5485         adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
5486         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5487                 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
5488                 for (i = 0; i < num_q_vectors; i++) {
5489                         struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
5490                         ixgbe_msix_clean_many(0, q_vector);
5491                 }
5492         } else {
5493                 ixgbe_intr(adapter->pdev->irq, netdev);
5494         }
5495         adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
5496 }
5497 #endif
5498
5499 static const struct net_device_ops ixgbe_netdev_ops = {
5500         .ndo_open               = ixgbe_open,
5501         .ndo_stop               = ixgbe_close,
5502         .ndo_start_xmit         = ixgbe_xmit_frame,
5503         .ndo_select_queue       = ixgbe_select_queue,
5504         .ndo_get_stats          = ixgbe_get_stats,
5505         .ndo_set_rx_mode        = ixgbe_set_rx_mode,
5506         .ndo_set_multicast_list = ixgbe_set_rx_mode,
5507         .ndo_validate_addr      = eth_validate_addr,
5508         .ndo_set_mac_address    = ixgbe_set_mac,
5509         .ndo_change_mtu         = ixgbe_change_mtu,
5510         .ndo_tx_timeout         = ixgbe_tx_timeout,
5511         .ndo_vlan_rx_register   = ixgbe_vlan_rx_register,
5512         .ndo_vlan_rx_add_vid    = ixgbe_vlan_rx_add_vid,
5513         .ndo_vlan_rx_kill_vid   = ixgbe_vlan_rx_kill_vid,
5514         .ndo_do_ioctl           = ixgbe_ioctl,
5515 #ifdef CONFIG_NET_POLL_CONTROLLER
5516         .ndo_poll_controller    = ixgbe_netpoll,
5517 #endif
5518 #ifdef IXGBE_FCOE
5519         .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
5520         .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
5521         .ndo_fcoe_enable = ixgbe_fcoe_enable,
5522         .ndo_fcoe_disable = ixgbe_fcoe_disable,
5523         .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
5524 #endif /* IXGBE_FCOE */
5525 };
5526
5527 /**
5528  * ixgbe_probe - Device Initialization Routine
5529  * @pdev: PCI device information struct
5530  * @ent: entry in ixgbe_pci_tbl
5531  *
5532  * Returns 0 on success, negative on failure
5533  *
5534  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
5535  * The OS initialization, configuring of the adapter private structure,
5536  * and a hardware reset occur.
5537  **/
5538 static int __devinit ixgbe_probe(struct pci_dev *pdev,
5539                                  const struct pci_device_id *ent)
5540 {
5541         struct net_device *netdev;
5542         struct ixgbe_adapter *adapter = NULL;
5543         struct ixgbe_hw *hw;
5544         const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
5545         static int cards_found;
5546         int i, err, pci_using_dac;
5547 #ifdef IXGBE_FCOE
5548         u16 device_caps;
5549 #endif
5550         u32 part_num, eec;
5551
5552         err = pci_enable_device_mem(pdev);
5553         if (err)
5554                 return err;
5555
5556         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) &&
5557             !pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64))) {
5558                 pci_using_dac = 1;
5559         } else {
5560                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
5561                 if (err) {
5562                         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
5563                         if (err) {
5564                                 dev_err(&pdev->dev, "No usable DMA "
5565                                         "configuration, aborting\n");
5566                                 goto err_dma;
5567                         }
5568                 }
5569                 pci_using_dac = 0;
5570         }
5571
5572         err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
5573                                            IORESOURCE_MEM), ixgbe_driver_name);
5574         if (err) {
5575                 dev_err(&pdev->dev,
5576                         "pci_request_selected_regions failed 0x%x\n", err);
5577                 goto err_pci_reg;
5578         }
5579
5580         pci_enable_pcie_error_reporting(pdev);
5581
5582         pci_set_master(pdev);
5583         pci_save_state(pdev);
5584
5585         netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), MAX_TX_QUEUES);
5586         if (!netdev) {
5587                 err = -ENOMEM;
5588                 goto err_alloc_etherdev;
5589         }
5590
5591         SET_NETDEV_DEV(netdev, &pdev->dev);
5592
5593         pci_set_drvdata(pdev, netdev);
5594         adapter = netdev_priv(netdev);
5595
5596         adapter->netdev = netdev;
5597         adapter->pdev = pdev;
5598         hw = &adapter->hw;
5599         hw->back = adapter;
5600         adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
5601
5602         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
5603                               pci_resource_len(pdev, 0));
5604         if (!hw->hw_addr) {
5605                 err = -EIO;
5606                 goto err_ioremap;
5607         }
5608
5609         for (i = 1; i <= 5; i++) {
5610                 if (pci_resource_len(pdev, i) == 0)
5611                         continue;
5612         }
5613
5614         netdev->netdev_ops = &ixgbe_netdev_ops;
5615         ixgbe_set_ethtool_ops(netdev);
5616         netdev->watchdog_timeo = 5 * HZ;
5617         strcpy(netdev->name, pci_name(pdev));
5618
5619         adapter->bd_number = cards_found;
5620
5621         /* Setup hw api */
5622         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
5623         hw->mac.type  = ii->mac;
5624
5625         /* EEPROM */
5626         memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
5627         eec = IXGBE_READ_REG(hw, IXGBE_EEC);
5628         /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
5629         if (!(eec & (1 << 8)))
5630                 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
5631
5632         /* PHY */
5633         memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
5634         hw->phy.sfp_type = ixgbe_sfp_type_unknown;
5635         /* ixgbe_identify_phy_generic will set prtad and mmds properly */
5636         hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
5637         hw->phy.mdio.mmds = 0;
5638         hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
5639         hw->phy.mdio.dev = netdev;
5640         hw->phy.mdio.mdio_read = ixgbe_mdio_read;
5641         hw->phy.mdio.mdio_write = ixgbe_mdio_write;
5642
5643         /* set up this timer and work struct before calling get_invariants
5644          * which might start the timer
5645          */
5646         init_timer(&adapter->sfp_timer);
5647         adapter->sfp_timer.function = &ixgbe_sfp_timer;
5648         adapter->sfp_timer.data = (unsigned long) adapter;
5649
5650         INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
5651
5652         /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
5653         INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
5654
5655         /* a new SFP+ module arrival, called from GPI SDP2 context */
5656         INIT_WORK(&adapter->sfp_config_module_task,
5657                   ixgbe_sfp_config_module_task);
5658
5659         ii->get_invariants(hw);
5660
5661         /* setup the private structure */
5662         err = ixgbe_sw_init(adapter);
5663         if (err)
5664                 goto err_sw_init;
5665
5666         /*
5667          * If there is a fan on this device and it has failed log the
5668          * failure.
5669          */
5670         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
5671                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
5672                 if (esdp & IXGBE_ESDP_SDP1)
5673                         DPRINTK(PROBE, CRIT,
5674                                 "Fan has stopped, replace the adapter\n");
5675         }
5676
5677         /* reset_hw fills in the perm_addr as well */
5678         err = hw->mac.ops.reset_hw(hw);
5679         if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
5680             hw->mac.type == ixgbe_mac_82598EB) {
5681                 /*
5682                  * Start a kernel thread to watch for a module to arrive.
5683                  * Only do this for 82598, since 82599 will generate
5684                  * interrupts on module arrival.
5685                  */
5686                 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
5687                 mod_timer(&adapter->sfp_timer,
5688                           round_jiffies(jiffies + (2 * HZ)));
5689                 err = 0;
5690         } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5691                 dev_err(&adapter->pdev->dev, "failed to initialize because "
5692                         "an unsupported SFP+ module type was detected.\n"
5693                         "Reload the driver after installing a supported "
5694                         "module.\n");
5695                 goto err_sw_init;
5696         } else if (err) {
5697                 dev_err(&adapter->pdev->dev, "HW Init failed: %d\n", err);
5698                 goto err_sw_init;
5699         }
5700
5701         netdev->features = NETIF_F_SG |
5702                            NETIF_F_IP_CSUM |
5703                            NETIF_F_HW_VLAN_TX |
5704                            NETIF_F_HW_VLAN_RX |
5705                            NETIF_F_HW_VLAN_FILTER;
5706
5707         netdev->features |= NETIF_F_IPV6_CSUM;
5708         netdev->features |= NETIF_F_TSO;
5709         netdev->features |= NETIF_F_TSO6;
5710         netdev->features |= NETIF_F_GRO;
5711
5712         if (adapter->hw.mac.type == ixgbe_mac_82599EB)
5713                 netdev->features |= NETIF_F_SCTP_CSUM;
5714
5715         netdev->vlan_features |= NETIF_F_TSO;
5716         netdev->vlan_features |= NETIF_F_TSO6;
5717         netdev->vlan_features |= NETIF_F_IP_CSUM;
5718         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
5719         netdev->vlan_features |= NETIF_F_SG;
5720
5721         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
5722                 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
5723
5724 #ifdef CONFIG_IXGBE_DCB
5725         netdev->dcbnl_ops = &dcbnl_ops;
5726 #endif
5727
5728 #ifdef IXGBE_FCOE
5729         if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
5730                 if (hw->mac.ops.get_device_caps) {
5731                         hw->mac.ops.get_device_caps(hw, &device_caps);
5732                         if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
5733                                 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
5734                 }
5735         }
5736 #endif /* IXGBE_FCOE */
5737         if (pci_using_dac)
5738                 netdev->features |= NETIF_F_HIGHDMA;
5739
5740         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
5741                 netdev->features |= NETIF_F_LRO;
5742
5743         /* make sure the EEPROM is good */
5744         if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
5745                 dev_err(&pdev->dev, "The EEPROM Checksum Is Not Valid\n");
5746                 err = -EIO;
5747                 goto err_eeprom;
5748         }
5749
5750         memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
5751         memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
5752
5753         if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
5754                 dev_err(&pdev->dev, "invalid MAC address\n");
5755                 err = -EIO;
5756                 goto err_eeprom;
5757         }
5758
5759         init_timer(&adapter->watchdog_timer);
5760         adapter->watchdog_timer.function = &ixgbe_watchdog;
5761         adapter->watchdog_timer.data = (unsigned long)adapter;
5762
5763         INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
5764         INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
5765
5766         err = ixgbe_init_interrupt_scheme(adapter);
5767         if (err)
5768                 goto err_sw_init;
5769
5770         switch (pdev->device) {
5771         case IXGBE_DEV_ID_82599_KX4:
5772                 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
5773                                 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
5774                 /* Enable ACPI wakeup in GRC */
5775                 IXGBE_WRITE_REG(hw, IXGBE_GRC,
5776                              (IXGBE_READ_REG(hw, IXGBE_GRC) & ~IXGBE_GRC_APME));
5777                 break;
5778         default:
5779                 adapter->wol = 0;
5780                 break;
5781         }
5782         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5783
5784         /* pick up the PCI bus settings for reporting later */
5785         hw->mac.ops.get_bus_info(hw);
5786
5787         /* print bus type/speed/width info */
5788         dev_info(&pdev->dev, "(PCI Express:%s:%s) %pM\n",
5789                 ((hw->bus.speed == ixgbe_bus_speed_5000) ? "5.0Gb/s":
5790                  (hw->bus.speed == ixgbe_bus_speed_2500) ? "2.5Gb/s":"Unknown"),
5791                 ((hw->bus.width == ixgbe_bus_width_pcie_x8) ? "Width x8" :
5792                  (hw->bus.width == ixgbe_bus_width_pcie_x4) ? "Width x4" :
5793                  (hw->bus.width == ixgbe_bus_width_pcie_x1) ? "Width x1" :
5794                  "Unknown"),
5795                 netdev->dev_addr);
5796         ixgbe_read_pba_num_generic(hw, &part_num);
5797         if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
5798                 dev_info(&pdev->dev, "MAC: %d, PHY: %d, SFP+: %d, PBA No: %06x-%03x\n",
5799                          hw->mac.type, hw->phy.type, hw->phy.sfp_type,
5800                          (part_num >> 8), (part_num & 0xff));
5801         else
5802                 dev_info(&pdev->dev, "MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
5803                          hw->mac.type, hw->phy.type,
5804                          (part_num >> 8), (part_num & 0xff));
5805
5806         if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
5807                 dev_warn(&pdev->dev, "PCI-Express bandwidth available for "
5808                          "this card is not sufficient for optimal "
5809                          "performance.\n");
5810                 dev_warn(&pdev->dev, "For optimal performance a x8 "
5811                          "PCI-Express slot is required.\n");
5812         }
5813
5814         /* save off EEPROM version number */
5815         hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
5816
5817         /* reset the hardware with the new settings */
5818         err = hw->mac.ops.start_hw(hw);
5819
5820         if (err == IXGBE_ERR_EEPROM_VERSION) {
5821                 /* We are running on a pre-production device, log a warning */
5822                 dev_warn(&pdev->dev, "This device is a pre-production "
5823                          "adapter/LOM.  Please be aware there may be issues "
5824                          "associated with your hardware.  If you are "
5825                          "experiencing problems please contact your Intel or "
5826                          "hardware representative who provided you with this "
5827                          "hardware.\n");
5828         }
5829         strcpy(netdev->name, "eth%d");
5830         err = register_netdev(netdev);
5831         if (err)
5832                 goto err_register;
5833
5834         /* carrier off reporting is important to ethtool even BEFORE open */
5835         netif_carrier_off(netdev);
5836
5837         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5838             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5839                 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
5840
5841 #ifdef CONFIG_IXGBE_DCA
5842         if (dca_add_requester(&pdev->dev) == 0) {
5843                 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
5844                 ixgbe_setup_dca(adapter);
5845         }
5846 #endif
5847         /* add san mac addr to netdev */
5848         ixgbe_add_sanmac_netdev(netdev);
5849
5850         dev_info(&pdev->dev, "Intel(R) 10 Gigabit Network Connection\n");
5851         cards_found++;
5852         return 0;
5853
5854 err_register:
5855         ixgbe_release_hw_control(adapter);
5856         ixgbe_clear_interrupt_scheme(adapter);
5857 err_sw_init:
5858 err_eeprom:
5859         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
5860         del_timer_sync(&adapter->sfp_timer);
5861         cancel_work_sync(&adapter->sfp_task);
5862         cancel_work_sync(&adapter->multispeed_fiber_task);
5863         cancel_work_sync(&adapter->sfp_config_module_task);
5864         iounmap(hw->hw_addr);
5865 err_ioremap:
5866         free_netdev(netdev);
5867 err_alloc_etherdev:
5868         pci_release_selected_regions(pdev, pci_select_bars(pdev,
5869                                      IORESOURCE_MEM));
5870 err_pci_reg:
5871 err_dma:
5872         pci_disable_device(pdev);
5873         return err;
5874 }
5875
5876 /**
5877  * ixgbe_remove - Device Removal Routine
5878  * @pdev: PCI device information struct
5879  *
5880  * ixgbe_remove is called by the PCI subsystem to alert the driver
5881  * that it should release a PCI device.  The could be caused by a
5882  * Hot-Plug event, or because the driver is going to be removed from
5883  * memory.
5884  **/
5885 static void __devexit ixgbe_remove(struct pci_dev *pdev)
5886 {
5887         struct net_device *netdev = pci_get_drvdata(pdev);
5888         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5889
5890         set_bit(__IXGBE_DOWN, &adapter->state);
5891         /* clear the module not found bit to make sure the worker won't
5892          * reschedule
5893          */
5894         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
5895         del_timer_sync(&adapter->watchdog_timer);
5896
5897         del_timer_sync(&adapter->sfp_timer);
5898         cancel_work_sync(&adapter->watchdog_task);
5899         cancel_work_sync(&adapter->sfp_task);
5900         cancel_work_sync(&adapter->multispeed_fiber_task);
5901         cancel_work_sync(&adapter->sfp_config_module_task);
5902         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5903             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5904                 cancel_work_sync(&adapter->fdir_reinit_task);
5905         flush_scheduled_work();
5906
5907 #ifdef CONFIG_IXGBE_DCA
5908         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
5909                 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
5910                 dca_remove_requester(&pdev->dev);
5911                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
5912         }
5913
5914 #endif
5915 #ifdef IXGBE_FCOE
5916         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
5917                 ixgbe_cleanup_fcoe(adapter);
5918
5919 #endif /* IXGBE_FCOE */
5920
5921         /* remove the added san mac */
5922         ixgbe_del_sanmac_netdev(netdev);
5923
5924         if (netdev->reg_state == NETREG_REGISTERED)
5925                 unregister_netdev(netdev);
5926
5927         ixgbe_clear_interrupt_scheme(adapter);
5928
5929         ixgbe_release_hw_control(adapter);
5930
5931         iounmap(adapter->hw.hw_addr);
5932         pci_release_selected_regions(pdev, pci_select_bars(pdev,
5933                                      IORESOURCE_MEM));
5934
5935         DPRINTK(PROBE, INFO, "complete\n");
5936
5937         free_netdev(netdev);
5938
5939         pci_disable_pcie_error_reporting(pdev);
5940
5941         pci_disable_device(pdev);
5942 }
5943
5944 /**
5945  * ixgbe_io_error_detected - called when PCI error is detected
5946  * @pdev: Pointer to PCI device
5947  * @state: The current pci connection state
5948  *
5949  * This function is called after a PCI bus error affecting
5950  * this device has been detected.
5951  */
5952 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
5953                                                 pci_channel_state_t state)
5954 {
5955         struct net_device *netdev = pci_get_drvdata(pdev);
5956         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5957
5958         netif_device_detach(netdev);
5959
5960         if (state == pci_channel_io_perm_failure)
5961                 return PCI_ERS_RESULT_DISCONNECT;
5962
5963         if (netif_running(netdev))
5964                 ixgbe_down(adapter);
5965         pci_disable_device(pdev);
5966
5967         /* Request a slot reset. */
5968         return PCI_ERS_RESULT_NEED_RESET;
5969 }
5970
5971 /**
5972  * ixgbe_io_slot_reset - called after the pci bus has been reset.
5973  * @pdev: Pointer to PCI device
5974  *
5975  * Restart the card from scratch, as if from a cold-boot.
5976  */
5977 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
5978 {
5979         struct net_device *netdev = pci_get_drvdata(pdev);
5980         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5981         pci_ers_result_t result;
5982         int err;
5983
5984         if (pci_enable_device_mem(pdev)) {
5985                 DPRINTK(PROBE, ERR,
5986                         "Cannot re-enable PCI device after reset.\n");
5987                 result = PCI_ERS_RESULT_DISCONNECT;
5988         } else {
5989                 pci_set_master(pdev);
5990                 pci_restore_state(pdev);
5991                 pci_save_state(pdev);
5992
5993                 pci_wake_from_d3(pdev, false);
5994
5995                 ixgbe_reset(adapter);
5996                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5997                 result = PCI_ERS_RESULT_RECOVERED;
5998         }
5999
6000         err = pci_cleanup_aer_uncorrect_error_status(pdev);
6001         if (err) {
6002                 dev_err(&pdev->dev,
6003                   "pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n", err);
6004                 /* non-fatal, continue */
6005         }
6006
6007         return result;
6008 }
6009
6010 /**
6011  * ixgbe_io_resume - called when traffic can start flowing again.
6012  * @pdev: Pointer to PCI device
6013  *
6014  * This callback is called when the error recovery driver tells us that
6015  * its OK to resume normal operation.
6016  */
6017 static void ixgbe_io_resume(struct pci_dev *pdev)
6018 {
6019         struct net_device *netdev = pci_get_drvdata(pdev);
6020         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6021
6022         if (netif_running(netdev)) {
6023                 if (ixgbe_up(adapter)) {
6024                         DPRINTK(PROBE, INFO, "ixgbe_up failed after reset\n");
6025                         return;
6026                 }
6027         }
6028
6029         netif_device_attach(netdev);
6030 }
6031
6032 static struct pci_error_handlers ixgbe_err_handler = {
6033         .error_detected = ixgbe_io_error_detected,
6034         .slot_reset = ixgbe_io_slot_reset,
6035         .resume = ixgbe_io_resume,
6036 };
6037
6038 static struct pci_driver ixgbe_driver = {
6039         .name     = ixgbe_driver_name,
6040         .id_table = ixgbe_pci_tbl,
6041         .probe    = ixgbe_probe,
6042         .remove   = __devexit_p(ixgbe_remove),
6043 #ifdef CONFIG_PM
6044         .suspend  = ixgbe_suspend,
6045         .resume   = ixgbe_resume,
6046 #endif
6047         .shutdown = ixgbe_shutdown,
6048         .err_handler = &ixgbe_err_handler
6049 };
6050
6051 /**
6052  * ixgbe_init_module - Driver Registration Routine
6053  *
6054  * ixgbe_init_module is the first routine called when the driver is
6055  * loaded. All it does is register with the PCI subsystem.
6056  **/
6057 static int __init ixgbe_init_module(void)
6058 {
6059         int ret;
6060         printk(KERN_INFO "%s: %s - version %s\n", ixgbe_driver_name,
6061                ixgbe_driver_string, ixgbe_driver_version);
6062
6063         printk(KERN_INFO "%s: %s\n", ixgbe_driver_name, ixgbe_copyright);
6064
6065 #ifdef CONFIG_IXGBE_DCA
6066         dca_register_notify(&dca_notifier);
6067 #endif
6068
6069         ret = pci_register_driver(&ixgbe_driver);
6070         return ret;
6071 }
6072
6073 module_init(ixgbe_init_module);
6074
6075 /**
6076  * ixgbe_exit_module - Driver Exit Cleanup Routine
6077  *
6078  * ixgbe_exit_module is called just before the driver is removed
6079  * from memory.
6080  **/
6081 static void __exit ixgbe_exit_module(void)
6082 {
6083 #ifdef CONFIG_IXGBE_DCA
6084         dca_unregister_notify(&dca_notifier);
6085 #endif
6086         pci_unregister_driver(&ixgbe_driver);
6087 }
6088
6089 #ifdef CONFIG_IXGBE_DCA
6090 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
6091                             void *p)
6092 {
6093         int ret_val;
6094
6095         ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
6096                                          __ixgbe_notify_dca);
6097
6098         return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
6099 }
6100
6101 #endif /* CONFIG_IXGBE_DCA */
6102 #ifdef DEBUG
6103 /**
6104  * ixgbe_get_hw_dev_name - return device name string
6105  * used by hardware layer to print debugging information
6106  **/
6107 char *ixgbe_get_hw_dev_name(struct ixgbe_hw *hw)
6108 {
6109         struct ixgbe_adapter *adapter = hw->back;
6110         return adapter->netdev->name;
6111 }
6112
6113 #endif
6114 module_exit(ixgbe_exit_module);
6115
6116 /* ixgbe_main.c */