drivers/net/ixgbe/ixgbe_main.c

   1 /*******************************************************************************
   2
   3   Intel 10 Gigabit PCI Express Linux driver
   4   Copyright(c) 1999 - 2010 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 <linux/slab.h>
  40 #include <net/checksum.h>
  41 #include <net/ip6_checksum.h>
  42 #include <linux/ethtool.h>
  43 #include <linux/if_vlan.h>
  44 #include <scsi/fc/fc_fcoe.h>
  45
  46 #include "ixgbe.h"
  47 #include "ixgbe_common.h"
  48 #include "ixgbe_dcb_82599.h"
  49 #include "ixgbe_sriov.h"
  50
  51 char ixgbe_driver_name[] = "ixgbe";
  52 static const char ixgbe_driver_string[] =
  53                               "Intel(R) 10 Gigabit PCI Express Network Driver";
  54
  55 #define DRV_VERSION "2.0.84-k2"
  56 const char ixgbe_driver_version[] = DRV_VERSION;
  57 static char ixgbe_copyright[] = "Copyright (c) 1999-2010 Intel Corporation.";
  58
  59 static const struct ixgbe_info *ixgbe_info_tbl[] = {
  60         [board_82598] = &ixgbe_82598_info,
  61         [board_82599] = &ixgbe_82599_info,
  62 };
  63
  64 /* ixgbe_pci_tbl - PCI Device ID Table
  65  *
  66  * Wildcard entries (PCI_ANY_ID) should come last
  67  * Last entry must be all 0s
  68  *
  69  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
  70  *   Class, Class Mask, private data (not used) }
  71  */
  72 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
  73         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598),
  74          board_82598 },
  75         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT),
  76          board_82598 },
  77         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT),
  78          board_82598 },
  79         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT),
  80          board_82598 },
  81         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2),
  82          board_82598 },
  83         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4),
  84          board_82598 },
  85         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT),
  86          board_82598 },
  87         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT),
  88          board_82598 },
  89         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM),
  90          board_82598 },
  91         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR),
  92          board_82598 },
  93         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM),
  94          board_82598 },
  95         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX),
  96          board_82598 },
  97         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4),
  98          board_82599 },
  99         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM),
 100          board_82599 },
 101         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR),
 102          board_82599 },
 103         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP),
 104          board_82599 },
 105         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM),
 106          board_82599 },
 107         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ),
 108          board_82599 },
 109         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4),
 110          board_82599 },
 111         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM),
 112          board_82599 },
 113         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE),
 114          board_82599 },
 115
 116         /* required last entry */
 117         {0, }
 118 };
 119 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
 120
 121 #ifdef CONFIG_IXGBE_DCA
 122 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
 123                             void *p);
 124 static struct notifier_block dca_notifier = {
 125         .notifier_call = ixgbe_notify_dca,
 126         .next          = NULL,
 127         .priority      = 0
 128 };
 129 #endif
 130
 131 #ifdef CONFIG_PCI_IOV
 132 static unsigned int max_vfs;
 133 module_param(max_vfs, uint, 0);
 134 MODULE_PARM_DESC(max_vfs,
 135                  "Maximum number of virtual functions to allocate per physical function");
 136 #endif /* CONFIG_PCI_IOV */
 137
 138 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
 139 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
 140 MODULE_LICENSE("GPL");
 141 MODULE_VERSION(DRV_VERSION);
 142
 143 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
 144
 145 static inline void ixgbe_disable_sriov(struct ixgbe_adapter *adapter)
 146 {
 147         struct ixgbe_hw *hw = &adapter->hw;
 148         u32 gcr;
 149         u32 gpie;
 150         u32 vmdctl;
 151
 152 #ifdef CONFIG_PCI_IOV
 153         /* disable iov and allow time for transactions to clear */
 154         pci_disable_sriov(adapter->pdev);
 155 #endif
 156
 157         /* turn off device IOV mode */
 158         gcr = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
 159         gcr &= ~(IXGBE_GCR_EXT_SRIOV);
 160         IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr);
 161         gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
 162         gpie &= ~IXGBE_GPIE_VTMODE_MASK;
 163         IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
 164
 165         /* set default pool back to 0 */
 166         vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
 167         vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
 168         IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
 169
 170         /* take a breather then clean up driver data */
 171         msleep(100);
 172
 173         kfree(adapter->vfinfo);
 174         adapter->vfinfo = NULL;
 175
 176         adapter->num_vfs = 0;
 177         adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
 178 }
 179
 180 struct ixgbe_reg_info {
 181         u32 ofs;
 182         char *name;
 183 };
 184
 185 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
 186
 187         /* General Registers */
 188         {IXGBE_CTRL, "CTRL"},
 189         {IXGBE_STATUS, "STATUS"},
 190         {IXGBE_CTRL_EXT, "CTRL_EXT"},
 191
 192         /* Interrupt Registers */
 193         {IXGBE_EICR, "EICR"},
 194
 195         /* RX Registers */
 196         {IXGBE_SRRCTL(0), "SRRCTL"},
 197         {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
 198         {IXGBE_RDLEN(0), "RDLEN"},
 199         {IXGBE_RDH(0), "RDH"},
 200         {IXGBE_RDT(0), "RDT"},
 201         {IXGBE_RXDCTL(0), "RXDCTL"},
 202         {IXGBE_RDBAL(0), "RDBAL"},
 203         {IXGBE_RDBAH(0), "RDBAH"},
 204
 205         /* TX Registers */
 206         {IXGBE_TDBAL(0), "TDBAL"},
 207         {IXGBE_TDBAH(0), "TDBAH"},
 208         {IXGBE_TDLEN(0), "TDLEN"},
 209         {IXGBE_TDH(0), "TDH"},
 210         {IXGBE_TDT(0), "TDT"},
 211         {IXGBE_TXDCTL(0), "TXDCTL"},
 212
 213         /* List Terminator */
 214         {}
 215 };
 216
 217
 218 /*
 219  * ixgbe_regdump - register printout routine
 220  */
 221 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
 222 {
 223         int i = 0, j = 0;
 224         char rname[16];
 225         u32 regs[64];
 226
 227         switch (reginfo->ofs) {
 228         case IXGBE_SRRCTL(0):
 229                 for (i = 0; i < 64; i++)
 230                         regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
 231                 break;
 232         case IXGBE_DCA_RXCTRL(0):
 233                 for (i = 0; i < 64; i++)
 234                         regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
 235                 break;
 236         case IXGBE_RDLEN(0):
 237                 for (i = 0; i < 64; i++)
 238                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
 239                 break;
 240         case IXGBE_RDH(0):
 241                 for (i = 0; i < 64; i++)
 242                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
 243                 break;
 244         case IXGBE_RDT(0):
 245                 for (i = 0; i < 64; i++)
 246                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
 247                 break;
 248         case IXGBE_RXDCTL(0):
 249                 for (i = 0; i < 64; i++)
 250                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
 251                 break;
 252         case IXGBE_RDBAL(0):
 253                 for (i = 0; i < 64; i++)
 254                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
 255                 break;
 256         case IXGBE_RDBAH(0):
 257                 for (i = 0; i < 64; i++)
 258                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
 259                 break;
 260         case IXGBE_TDBAL(0):
 261                 for (i = 0; i < 64; i++)
 262                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
 263                 break;
 264         case IXGBE_TDBAH(0):
 265                 for (i = 0; i < 64; i++)
 266                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
 267                 break;
 268         case IXGBE_TDLEN(0):
 269                 for (i = 0; i < 64; i++)
 270                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
 271                 break;
 272         case IXGBE_TDH(0):
 273                 for (i = 0; i < 64; i++)
 274                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
 275                 break;
 276         case IXGBE_TDT(0):
 277                 for (i = 0; i < 64; i++)
 278                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
 279                 break;
 280         case IXGBE_TXDCTL(0):
 281                 for (i = 0; i < 64; i++)
 282                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
 283                 break;
 284         default:
 285                 printk(KERN_INFO "%-15s %08x\n", reginfo->name,
 286                         IXGBE_READ_REG(hw, reginfo->ofs));
 287                 return;
 288         }
 289
 290         for (i = 0; i < 8; i++) {
 291                 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
 292                 printk(KERN_ERR "%-15s ", rname);
 293                 for (j = 0; j < 8; j++)
 294                         printk(KERN_CONT "%08x ", regs[i*8+j]);
 295                 printk(KERN_CONT "\n");
 296         }
 297
 298 }
 299
 300 /*
 301  * ixgbe_dump - Print registers, tx-rings and rx-rings
 302  */
 303 static void ixgbe_dump(struct ixgbe_adapter *adapter)
 304 {
 305         struct net_device *netdev = adapter->netdev;
 306         struct ixgbe_hw *hw = &adapter->hw;
 307         struct ixgbe_reg_info *reginfo;
 308         int n = 0;
 309         struct ixgbe_ring *tx_ring;
 310         struct ixgbe_tx_buffer *tx_buffer_info;
 311         union ixgbe_adv_tx_desc *tx_desc;
 312         struct my_u0 { u64 a; u64 b; } *u0;
 313         struct ixgbe_ring *rx_ring;
 314         union ixgbe_adv_rx_desc *rx_desc;
 315         struct ixgbe_rx_buffer *rx_buffer_info;
 316         u32 staterr;
 317         int i = 0;
 318
 319         if (!netif_msg_hw(adapter))
 320                 return;
 321
 322         /* Print netdevice Info */
 323         if (netdev) {
 324                 dev_info(&adapter->pdev->dev, "Net device Info\n");
 325                 printk(KERN_INFO "Device Name     state            "
 326                         "trans_start      last_rx\n");
 327                 printk(KERN_INFO "%-15s %016lX %016lX %016lX\n",
 328                 netdev->name,
 329                 netdev->state,
 330                 netdev->trans_start,
 331                 netdev->last_rx);
 332         }
 333
 334         /* Print Registers */
 335         dev_info(&adapter->pdev->dev, "Register Dump\n");
 336         printk(KERN_INFO " Register Name   Value\n");
 337         for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
 338              reginfo->name; reginfo++) {
 339                 ixgbe_regdump(hw, reginfo);
 340         }
 341
 342         /* Print TX Ring Summary */
 343         if (!netdev || !netif_running(netdev))
 344                 goto exit;
 345
 346         dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
 347         printk(KERN_INFO "Queue [NTU] [NTC] [bi(ntc)->dma  ] "
 348                 "leng ntw timestamp\n");
 349         for (n = 0; n < adapter->num_tx_queues; n++) {
 350                 tx_ring = adapter->tx_ring[n];
 351                 tx_buffer_info =
 352                         &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
 353                 printk(KERN_INFO " %5d %5X %5X %016llX %04X %3X %016llX\n",
 354                            n, tx_ring->next_to_use, tx_ring->next_to_clean,
 355                            (u64)tx_buffer_info->dma,
 356                            tx_buffer_info->length,
 357                            tx_buffer_info->next_to_watch,
 358                            (u64)tx_buffer_info->time_stamp);
 359         }
 360
 361         /* Print TX Rings */
 362         if (!netif_msg_tx_done(adapter))
 363                 goto rx_ring_summary;
 364
 365         dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
 366
 367         /* Transmit Descriptor Formats
 368          *
 369          * Advanced Transmit Descriptor
 370          *   +--------------------------------------------------------------+
 371          * 0 |         Buffer Address [63:0]                                |
 372          *   +--------------------------------------------------------------+
 373          * 8 |  PAYLEN  | PORTS  | IDX | STA | DCMD  |DTYP |  RSV |  DTALEN |
 374          *   +--------------------------------------------------------------+
 375          *   63       46 45    40 39 36 35 32 31   24 23 20 19              0
 376          */
 377
 378         for (n = 0; n < adapter->num_tx_queues; n++) {
 379                 tx_ring = adapter->tx_ring[n];
 380                 printk(KERN_INFO "------------------------------------\n");
 381                 printk(KERN_INFO "TX QUEUE INDEX = %d\n", tx_ring->queue_index);
 382                 printk(KERN_INFO "------------------------------------\n");
 383                 printk(KERN_INFO "T [desc]     [address 63:0  ] "
 384                         "[PlPOIdStDDt Ln] [bi->dma       ] "
 385                         "leng  ntw timestamp        bi->skb\n");
 386
 387                 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
 388                         tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
 389                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
 390                         u0 = (struct my_u0 *)tx_desc;
 391                         printk(KERN_INFO "T [0x%03X]    %016llX %016llX %016llX"
 392                                 " %04X  %3X %016llX %p", i,
 393                                 le64_to_cpu(u0->a),
 394                                 le64_to_cpu(u0->b),
 395                                 (u64)tx_buffer_info->dma,
 396                                 tx_buffer_info->length,
 397                                 tx_buffer_info->next_to_watch,
 398                                 (u64)tx_buffer_info->time_stamp,
 399                                 tx_buffer_info->skb);
 400                         if (i == tx_ring->next_to_use &&
 401                                 i == tx_ring->next_to_clean)
 402                                 printk(KERN_CONT " NTC/U\n");
 403                         else if (i == tx_ring->next_to_use)
 404                                 printk(KERN_CONT " NTU\n");
 405                         else if (i == tx_ring->next_to_clean)
 406                                 printk(KERN_CONT " NTC\n");
 407                         else
 408                                 printk(KERN_CONT "\n");
 409
 410                         if (netif_msg_pktdata(adapter) &&
 411                                 tx_buffer_info->dma != 0)
 412                                 print_hex_dump(KERN_INFO, "",
 413                                         DUMP_PREFIX_ADDRESS, 16, 1,
 414                                         phys_to_virt(tx_buffer_info->dma),
 415                                         tx_buffer_info->length, true);
 416                 }
 417         }
 418
 419         /* Print RX Rings Summary */
 420 rx_ring_summary:
 421         dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
 422         printk(KERN_INFO "Queue [NTU] [NTC]\n");
 423         for (n = 0; n < adapter->num_rx_queues; n++) {
 424                 rx_ring = adapter->rx_ring[n];
 425                 printk(KERN_INFO "%5d %5X %5X\n", n,
 426                            rx_ring->next_to_use, rx_ring->next_to_clean);
 427         }
 428
 429         /* Print RX Rings */
 430         if (!netif_msg_rx_status(adapter))
 431                 goto exit;
 432
 433         dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
 434
 435         /* Advanced Receive Descriptor (Read) Format
 436          *    63                                           1        0
 437          *    +-----------------------------------------------------+
 438          *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
 439          *    +----------------------------------------------+------+
 440          *  8 |       Header Buffer Address [63:1]           |  DD  |
 441          *    +-----------------------------------------------------+
 442          *
 443          *
 444          * Advanced Receive Descriptor (Write-Back) Format
 445          *
 446          *   63       48 47    32 31  30      21 20 16 15   4 3     0
 447          *   +------------------------------------------------------+
 448          * 0 | Packet     IP     |SPH| HDR_LEN   | RSV|Packet|  RSS |
 449          *   | Checksum   Ident  |   |           |    | Type | Type |
 450          *   +------------------------------------------------------+
 451          * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 452          *   +------------------------------------------------------+
 453          *   63       48 47    32 31            20 19               0
 454          */
 455         for (n = 0; n < adapter->num_rx_queues; n++) {
 456                 rx_ring = adapter->rx_ring[n];
 457                 printk(KERN_INFO "------------------------------------\n");
 458                 printk(KERN_INFO "RX QUEUE INDEX = %d\n", rx_ring->queue_index);
 459                 printk(KERN_INFO "------------------------------------\n");
 460                 printk(KERN_INFO "R  [desc]      [ PktBuf     A0] "
 461                         "[  HeadBuf   DD] [bi->dma       ] [bi->skb] "
 462                         "<-- Adv Rx Read format\n");
 463                 printk(KERN_INFO "RWB[desc]      [PcsmIpSHl PtRs] "
 464                         "[vl er S cks ln] ---------------- [bi->skb] "
 465                         "<-- Adv Rx Write-Back format\n");
 466
 467                 for (i = 0; i < rx_ring->count; i++) {
 468                         rx_buffer_info = &rx_ring->rx_buffer_info[i];
 469                         rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
 470                         u0 = (struct my_u0 *)rx_desc;
 471                         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 472                         if (staterr & IXGBE_RXD_STAT_DD) {
 473                                 /* Descriptor Done */
 474                                 printk(KERN_INFO "RWB[0x%03X]     %016llX "
 475                                         "%016llX ---------------- %p", i,
 476                                         le64_to_cpu(u0->a),
 477                                         le64_to_cpu(u0->b),
 478                                         rx_buffer_info->skb);
 479                         } else {
 480                                 printk(KERN_INFO "R  [0x%03X]     %016llX "
 481                                         "%016llX %016llX %p", i,
 482                                         le64_to_cpu(u0->a),
 483                                         le64_to_cpu(u0->b),
 484                                         (u64)rx_buffer_info->dma,
 485                                         rx_buffer_info->skb);
 486
 487                                 if (netif_msg_pktdata(adapter)) {
 488                                         print_hex_dump(KERN_INFO, "",
 489                                            DUMP_PREFIX_ADDRESS, 16, 1,
 490                                            phys_to_virt(rx_buffer_info->dma),
 491                                            rx_ring->rx_buf_len, true);
 492
 493                                         if (rx_ring->rx_buf_len
 494                                                 < IXGBE_RXBUFFER_2048)
 495                                                 print_hex_dump(KERN_INFO, "",
 496                                                   DUMP_PREFIX_ADDRESS, 16, 1,
 497                                                   phys_to_virt(
 498                                                     rx_buffer_info->page_dma +
 499                                                     rx_buffer_info->page_offset
 500                                                   ),
 501                                                   PAGE_SIZE/2, true);
 502                                 }
 503                         }
 504
 505                         if (i == rx_ring->next_to_use)
 506                                 printk(KERN_CONT " NTU\n");
 507                         else if (i == rx_ring->next_to_clean)
 508                                 printk(KERN_CONT " NTC\n");
 509                         else
 510                                 printk(KERN_CONT "\n");
 511
 512                 }
 513         }
 514
 515 exit:
 516         return;
 517 }
 518
 519 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
 520 {
 521         u32 ctrl_ext;
 522
 523         /* Let firmware take over control of h/w */
 524         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
 525         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
 526                         ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
 527 }
 528
 529 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
 530 {
 531         u32 ctrl_ext;
 532
 533         /* Let firmware know the driver has taken over */
 534         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
 535         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
 536                         ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
 537 }
 538
 539 /*
 540  * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
 541  * @adapter: pointer to adapter struct
 542  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
 543  * @queue: queue to map the corresponding interrupt to
 544  * @msix_vector: the vector to map to the corresponding queue
 545  *
 546  */
 547 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
 548                            u8 queue, u8 msix_vector)
 549 {
 550         u32 ivar, index;
 551         struct ixgbe_hw *hw = &adapter->hw;
 552         switch (hw->mac.type) {
 553         case ixgbe_mac_82598EB:
 554                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 555                 if (direction == -1)
 556                         direction = 0;
 557                 index = (((direction * 64) + queue) >> 2) & 0x1F;
 558                 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
 559                 ivar &= ~(0xFF << (8 * (queue & 0x3)));
 560                 ivar |= (msix_vector << (8 * (queue & 0x3)));
 561                 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
 562                 break;
 563         case ixgbe_mac_82599EB:
 564                 if (direction == -1) {
 565                         /* other causes */
 566                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 567                         index = ((queue & 1) * 8);
 568                         ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
 569                         ivar &= ~(0xFF << index);
 570                         ivar |= (msix_vector << index);
 571                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
 572                         break;
 573                 } else {
 574                         /* tx or rx causes */
 575                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 576                         index = ((16 * (queue & 1)) + (8 * direction));
 577                         ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
 578                         ivar &= ~(0xFF << index);
 579                         ivar |= (msix_vector << index);
 580                         IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
 581                         break;
 582                 }
 583         default:
 584                 break;
 585         }
 586 }
 587
 588 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
 589                                           u64 qmask)
 590 {
 591         u32 mask;
 592
 593         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
 594                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
 595                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
 596         } else {
 597                 mask = (qmask & 0xFFFFFFFF);
 598                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
 599                 mask = (qmask >> 32);
 600                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
 601         }
 602 }
 603
 604 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_adapter *adapter,
 605                                       struct ixgbe_tx_buffer
 606                                       *tx_buffer_info)
 607 {
 608         if (tx_buffer_info->dma) {
 609                 if (tx_buffer_info->mapped_as_page)
 610                         dma_unmap_page(&adapter->pdev->dev,
 611                                        tx_buffer_info->dma,
 612                                        tx_buffer_info->length,
 613                                        DMA_TO_DEVICE);
 614                 else
 615                         dma_unmap_single(&adapter->pdev->dev,
 616                                          tx_buffer_info->dma,
 617                                          tx_buffer_info->length,
 618                                          DMA_TO_DEVICE);
 619                 tx_buffer_info->dma = 0;
 620         }
 621         if (tx_buffer_info->skb) {
 622                 dev_kfree_skb_any(tx_buffer_info->skb);
 623                 tx_buffer_info->skb = NULL;
 624         }
 625         tx_buffer_info->time_stamp = 0;
 626         /* tx_buffer_info must be completely set up in the transmit path */
 627 }
 628
 629 /**
 630  * ixgbe_tx_xon_state - check the tx ring xon state
 631  * @adapter: the ixgbe adapter
 632  * @tx_ring: the corresponding tx_ring
 633  *
 634  * If not in DCB mode, checks TFCS.TXOFF, otherwise, find out the
 635  * corresponding TC of this tx_ring when checking TFCS.
 636  *
 637  * Returns : true if in xon state (currently not paused)
 638  */
 639 static inline bool ixgbe_tx_xon_state(struct ixgbe_adapter *adapter,
 640                                       struct ixgbe_ring *tx_ring)
 641 {
 642         u32 txoff = IXGBE_TFCS_TXOFF;
 643
 644 #ifdef CONFIG_IXGBE_DCB
 645         if (adapter->dcb_cfg.pfc_mode_enable) {
 646                 int tc;
 647                 int reg_idx = tx_ring->reg_idx;
 648                 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
 649
 650                 switch (adapter->hw.mac.type) {
 651                 case ixgbe_mac_82598EB:
 652                         tc = reg_idx >> 2;
 653                         txoff = IXGBE_TFCS_TXOFF0;
 654                         break;
 655                 case ixgbe_mac_82599EB:
 656                         tc = 0;
 657                         txoff = IXGBE_TFCS_TXOFF;
 658                         if (dcb_i == 8) {
 659                                 /* TC0, TC1 */
 660                                 tc = reg_idx >> 5;
 661                                 if (tc == 2) /* TC2, TC3 */
 662                                         tc += (reg_idx - 64) >> 4;
 663                                 else if (tc == 3) /* TC4, TC5, TC6, TC7 */
 664                                         tc += 1 + ((reg_idx - 96) >> 3);
 665                         } else if (dcb_i == 4) {
 666                                 /* TC0, TC1 */
 667                                 tc = reg_idx >> 6;
 668                                 if (tc == 1) {
 669                                         tc += (reg_idx - 64) >> 5;
 670                                         if (tc == 2) /* TC2, TC3 */
 671                                                 tc += (reg_idx - 96) >> 4;
 672                                 }
 673                         }
 674                         break;
 675                 default:
 676                         tc = 0;
 677                 }
 678                 txoff <<= tc;
 679         }
 680 #endif
 681         return IXGBE_READ_REG(&adapter->hw, IXGBE_TFCS) & txoff;
 682 }
 683
 684 static inline bool ixgbe_check_tx_hang(struct ixgbe_adapter *adapter,
 685                                        struct ixgbe_ring *tx_ring,
 686                                        unsigned int eop)
 687 {
 688         struct ixgbe_hw *hw = &adapter->hw;
 689
 690         /* Detect a transmit hang in hardware, this serializes the
 691          * check with the clearing of time_stamp and movement of eop */
 692         adapter->detect_tx_hung = false;
 693         if (tx_ring->tx_buffer_info[eop].time_stamp &&
 694             time_after(jiffies, tx_ring->tx_buffer_info[eop].time_stamp + HZ) &&
 695             ixgbe_tx_xon_state(adapter, tx_ring)) {
 696                 /* detected Tx unit hang */
 697                 union ixgbe_adv_tx_desc *tx_desc;
 698                 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
 699                 e_err(drv, "Detected Tx Unit Hang\n"
 700                       "  Tx Queue             <%d>\n"
 701                       "  TDH, TDT             <%x>, <%x>\n"
 702                       "  next_to_use          <%x>\n"
 703                       "  next_to_clean        <%x>\n"
 704                       "tx_buffer_info[next_to_clean]\n"
 705                       "  time_stamp           <%lx>\n"
 706                       "  jiffies              <%lx>\n",
 707                       tx_ring->queue_index,
 708                       IXGBE_READ_REG(hw, tx_ring->head),
 709                       IXGBE_READ_REG(hw, tx_ring->tail),
 710                       tx_ring->next_to_use, eop,
 711                       tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
 712                 return true;
 713         }
 714
 715         return false;
 716 }
 717
 718 #define IXGBE_MAX_TXD_PWR       14
 719 #define IXGBE_MAX_DATA_PER_TXD  (1 << IXGBE_MAX_TXD_PWR)
 720
 721 /* Tx Descriptors needed, worst case */
 722 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
 723                          (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
 724 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
 725         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
 726
 727 static void ixgbe_tx_timeout(struct net_device *netdev);
 728
 729 /**
 730  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
 731  * @q_vector: structure containing interrupt and ring information
 732  * @tx_ring: tx ring to clean
 733  **/
 734 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
 735                                struct ixgbe_ring *tx_ring)
 736 {
 737         struct ixgbe_adapter *adapter = q_vector->adapter;
 738         struct net_device *netdev = adapter->netdev;
 739         union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
 740         struct ixgbe_tx_buffer *tx_buffer_info;
 741         unsigned int i, eop, count = 0;
 742         unsigned int total_bytes = 0, total_packets = 0;
 743
 744         i = tx_ring->next_to_clean;
 745         eop = tx_ring->tx_buffer_info[i].next_to_watch;
 746         eop_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
 747
 748         while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
 749                (count < tx_ring->work_limit)) {
 750                 bool cleaned = false;
 751                 rmb(); /* read buffer_info after eop_desc */
 752                 for ( ; !cleaned; count++) {
 753                         struct sk_buff *skb;
 754                         tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
 755                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
 756                         cleaned = (i == eop);
 757                         skb = tx_buffer_info->skb;
 758
 759                         if (cleaned && skb) {
 760                                 unsigned int segs, bytecount;
 761                                 unsigned int hlen = skb_headlen(skb);
 762
 763                                 /* gso_segs is currently only valid for tcp */
 764                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 765 #ifdef IXGBE_FCOE
 766                                 /* adjust for FCoE Sequence Offload */
 767                                 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
 768                                     && (skb->protocol == htons(ETH_P_FCOE)) &&
 769                                     skb_is_gso(skb)) {
 770                                         hlen = skb_transport_offset(skb) +
 771                                                 sizeof(struct fc_frame_header) +
 772                                                 sizeof(struct fcoe_crc_eof);
 773                                         segs = DIV_ROUND_UP(skb->len - hlen,
 774                                                 skb_shinfo(skb)->gso_size);
 775                                 }
 776 #endif /* IXGBE_FCOE */
 777                                 /* multiply data chunks by size of headers */
 778                                 bytecount = ((segs - 1) * hlen) + skb->len;
 779                                 total_packets += segs;
 780                                 total_bytes += bytecount;
 781                         }
 782
 783                         ixgbe_unmap_and_free_tx_resource(adapter,
 784                                                          tx_buffer_info);
 785
 786                         tx_desc->wb.status = 0;
 787
 788                         i++;
 789                         if (i == tx_ring->count)
 790                                 i = 0;
 791                 }
 792
 793                 eop = tx_ring->tx_buffer_info[i].next_to_watch;
 794                 eop_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
 795         }
 796
 797         tx_ring->next_to_clean = i;
 798
 799 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
 800         if (unlikely(count && netif_carrier_ok(netdev) &&
 801                      (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
 802                 /* Make sure that anybody stopping the queue after this
 803                  * sees the new next_to_clean.
 804                  */
 805                 smp_mb();
 806                 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
 807                     !test_bit(__IXGBE_DOWN, &adapter->state)) {
 808                         netif_wake_subqueue(netdev, tx_ring->queue_index);
 809                         ++tx_ring->restart_queue;
 810                 }
 811         }
 812
 813         if (adapter->detect_tx_hung) {
 814                 if (ixgbe_check_tx_hang(adapter, tx_ring, i)) {
 815                         /* schedule immediate reset if we believe we hung */
 816                         e_info(probe, "tx hang %d detected, resetting "
 817                                "adapter\n", adapter->tx_timeout_count + 1);
 818                         ixgbe_tx_timeout(adapter->netdev);
 819                 }
 820         }
 821
 822         /* re-arm the interrupt */
 823         if (count >= tx_ring->work_limit)
 824                 ixgbe_irq_rearm_queues(adapter, ((u64)1 << q_vector->v_idx));
 825
 826         tx_ring->total_bytes += total_bytes;
 827         tx_ring->total_packets += total_packets;
 828         tx_ring->stats.packets += total_packets;
 829         tx_ring->stats.bytes += total_bytes;
 830         return (count < tx_ring->work_limit);
 831 }
 832
 833 #ifdef CONFIG_IXGBE_DCA
 834 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
 835                                 struct ixgbe_ring *rx_ring)
 836 {
 837         u32 rxctrl;
 838         int cpu = get_cpu();
 839         int q = rx_ring->reg_idx;
 840
 841         if (rx_ring->cpu != cpu) {
 842                 rxctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q));
 843                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
 844                         rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
 845                         rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
 846                 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
 847                         rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
 848                         rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
 849                                    IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
 850                 }
 851                 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
 852                 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
 853                 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
 854                 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN |
 855                             IXGBE_DCA_RXCTRL_DESC_HSRO_EN);
 856                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q), rxctrl);
 857                 rx_ring->cpu = cpu;
 858         }
 859         put_cpu();
 860 }
 861
 862 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
 863                                 struct ixgbe_ring *tx_ring)
 864 {
 865         u32 txctrl;
 866         int cpu = get_cpu();
 867         int q = tx_ring->reg_idx;
 868         struct ixgbe_hw *hw = &adapter->hw;
 869
 870         if (tx_ring->cpu != cpu) {
 871                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
 872                         txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(q));
 873                         txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
 874                         txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
 875                         txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
 876                         IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(q), txctrl);
 877                 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
 878                         txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(q));
 879                         txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
 880                         txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
 881                                   IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
 882                         txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
 883                         IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(q), txctrl);
 884                 }
 885                 tx_ring->cpu = cpu;
 886         }
 887         put_cpu();
 888 }
 889
 890 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
 891 {
 892         int i;
 893
 894         if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
 895                 return;
 896
 897         /* always use CB2 mode, difference is masked in the CB driver */
 898         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
 899
 900         for (i = 0; i < adapter->num_tx_queues; i++) {
 901                 adapter->tx_ring[i]->cpu = -1;
 902                 ixgbe_update_tx_dca(adapter, adapter->tx_ring[i]);
 903         }
 904         for (i = 0; i < adapter->num_rx_queues; i++) {
 905                 adapter->rx_ring[i]->cpu = -1;
 906                 ixgbe_update_rx_dca(adapter, adapter->rx_ring[i]);
 907         }
 908 }
 909
 910 static int __ixgbe_notify_dca(struct device *dev, void *data)
 911 {
 912         struct net_device *netdev = dev_get_drvdata(dev);
 913         struct ixgbe_adapter *adapter = netdev_priv(netdev);
 914         unsigned long event = *(unsigned long *)data;
 915
 916         switch (event) {
 917         case DCA_PROVIDER_ADD:
 918                 /* if we're already enabled, don't do it again */
 919                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
 920                         break;
 921                 if (dca_add_requester(dev) == 0) {
 922                         adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
 923                         ixgbe_setup_dca(adapter);
 924                         break;
 925                 }
 926                 /* Fall Through since DCA is disabled. */
 927         case DCA_PROVIDER_REMOVE:
 928                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
 929                         dca_remove_requester(dev);
 930                         adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
 931                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
 932                 }
 933                 break;
 934         }
 935
 936         return 0;
 937 }
 938
 939 #endif /* CONFIG_IXGBE_DCA */
 940 /**
 941  * ixgbe_receive_skb - Send a completed packet up the stack
 942  * @adapter: board private structure
 943  * @skb: packet to send up
 944  * @status: hardware indication of status of receive
 945  * @rx_ring: rx descriptor ring (for a specific queue) to setup
 946  * @rx_desc: rx descriptor
 947  **/
 948 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
 949                               struct sk_buff *skb, u8 status,
 950                               struct ixgbe_ring *ring,
 951                               union ixgbe_adv_rx_desc *rx_desc)
 952 {
 953         struct ixgbe_adapter *adapter = q_vector->adapter;
 954         struct napi_struct *napi = &q_vector->napi;
 955         bool is_vlan = (status & IXGBE_RXD_STAT_VP);
 956         u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
 957
 958         if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL)) {
 959                 if (adapter->vlgrp && is_vlan && (tag & VLAN_VID_MASK))
 960                         vlan_gro_receive(napi, adapter->vlgrp, tag, skb);
 961                 else
 962                         napi_gro_receive(napi, skb);
 963         } else {
 964                 if (adapter->vlgrp && is_vlan && (tag & VLAN_VID_MASK))
 965                         vlan_hwaccel_rx(skb, adapter->vlgrp, tag);
 966                 else
 967                         netif_rx(skb);
 968         }
 969 }
 970
 971 /**
 972  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
 973  * @adapter: address of board private structure
 974  * @status_err: hardware indication of status of receive
 975  * @skb: skb currently being received and modified
 976  **/
 977 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
 978                                      union ixgbe_adv_rx_desc *rx_desc,
 979                                      struct sk_buff *skb)
 980 {
 981         u32 status_err = le32_to_cpu(rx_desc->wb.upper.status_error);
 982
 983         skb_checksum_none_assert(skb);
 984
 985         /* Rx csum disabled */
 986         if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
 987                 return;
 988
 989         /* if IP and error */
 990         if ((status_err & IXGBE_RXD_STAT_IPCS) &&
 991             (status_err & IXGBE_RXDADV_ERR_IPE)) {
 992                 adapter->hw_csum_rx_error++;
 993                 return;
 994         }
 995
 996         if (!(status_err & IXGBE_RXD_STAT_L4CS))
 997                 return;
 998
 999         if (status_err & IXGBE_RXDADV_ERR_TCPE) {
1000                 u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1001
1002                 /*
1003                  * 82599 errata, UDP frames with a 0 checksum can be marked as
1004                  * checksum errors.
1005                  */
1006                 if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
1007                     (adapter->hw.mac.type == ixgbe_mac_82599EB))
1008                         return;
1009
1010                 adapter->hw_csum_rx_error++;
1011                 return;
1012         }
1013
1014         /* It must be a TCP or UDP packet with a valid checksum */
1015         skb->ip_summed = CHECKSUM_UNNECESSARY;
1016 }
1017
1018 static inline void ixgbe_release_rx_desc(struct ixgbe_hw *hw,
1019                                          struct ixgbe_ring *rx_ring, u32 val)
1020 {
1021         /*
1022          * Force memory writes to complete before letting h/w
1023          * know there are new descriptors to fetch.  (Only
1024          * applicable for weak-ordered memory model archs,
1025          * such as IA-64).
1026          */
1027         wmb();
1028         IXGBE_WRITE_REG(hw, IXGBE_RDT(rx_ring->reg_idx), val);
1029 }
1030
1031 /**
1032  * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
1033  * @adapter: address of board private structure
1034  **/
1035 void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
1036                             struct ixgbe_ring *rx_ring,
1037                             int cleaned_count)
1038 {
1039         struct net_device *netdev = adapter->netdev;
1040         struct pci_dev *pdev = adapter->pdev;
1041         union ixgbe_adv_rx_desc *rx_desc;
1042         struct ixgbe_rx_buffer *bi;
1043         unsigned int i;
1044         unsigned int bufsz = rx_ring->rx_buf_len;
1045
1046         i = rx_ring->next_to_use;
1047         bi = &rx_ring->rx_buffer_info[i];
1048
1049         while (cleaned_count--) {
1050                 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1051
1052                 if (!bi->page_dma &&
1053                     (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED)) {
1054                         if (!bi->page) {
1055                                 bi->page = netdev_alloc_page(netdev);
1056                                 if (!bi->page) {
1057                                         adapter->alloc_rx_page_failed++;
1058                                         goto no_buffers;
1059                                 }
1060                                 bi->page_offset = 0;
1061                         } else {
1062                                 /* use a half page if we're re-using */
1063                                 bi->page_offset ^= (PAGE_SIZE / 2);
1064                         }
1065
1066                         bi->page_dma = dma_map_page(&pdev->dev, bi->page,
1067                                                     bi->page_offset,
1068                                                     (PAGE_SIZE / 2),
1069                                                     DMA_FROM_DEVICE);
1070                 }
1071
1072                 if (!bi->skb) {
1073                         struct sk_buff *skb = netdev_alloc_skb_ip_align(netdev,
1074                                                                         bufsz);
1075                         bi->skb = skb;
1076
1077                         if (!skb) {
1078                                 adapter->alloc_rx_buff_failed++;
1079                                 goto no_buffers;
1080                         }
1081                         /* initialize queue mapping */
1082                         skb_record_rx_queue(skb, rx_ring->queue_index);
1083                 }
1084
1085                 if (!bi->dma) {
1086                         bi->dma = dma_map_single(&pdev->dev,
1087                                                  bi->skb->data,
1088                                                  rx_ring->rx_buf_len,
1089                                                  DMA_FROM_DEVICE);
1090                 }
1091                 /* Refresh the desc even if buffer_addrs didn't change because
1092                  * each write-back erases this info. */
1093                 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
1094                         rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
1095                         rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
1096                 } else {
1097                         rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
1098                         rx_desc->read.hdr_addr = 0;
1099                 }
1100
1101                 i++;
1102                 if (i == rx_ring->count)
1103                         i = 0;
1104                 bi = &rx_ring->rx_buffer_info[i];
1105         }
1106
1107 no_buffers:
1108         if (rx_ring->next_to_use != i) {
1109                 rx_ring->next_to_use = i;
1110                 if (i-- == 0)
1111                         i = (rx_ring->count - 1);
1112
1113                 ixgbe_release_rx_desc(&adapter->hw, rx_ring, i);
1114         }
1115 }
1116
1117 static inline u16 ixgbe_get_hdr_info(union ixgbe_adv_rx_desc *rx_desc)
1118 {
1119         return rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
1120 }
1121
1122 static inline u16 ixgbe_get_pkt_info(union ixgbe_adv_rx_desc *rx_desc)
1123 {
1124         return rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1125 }
1126
1127 static inline u32 ixgbe_get_rsc_count(union ixgbe_adv_rx_desc *rx_desc)
1128 {
1129         return (le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
1130                 IXGBE_RXDADV_RSCCNT_MASK) >>
1131                 IXGBE_RXDADV_RSCCNT_SHIFT;
1132 }
1133
1134 /**
1135  * ixgbe_transform_rsc_queue - change rsc queue into a full packet
1136  * @skb: pointer to the last skb in the rsc queue
1137  * @count: pointer to number of packets coalesced in this context
1138  *
1139  * This function changes a queue full of hw rsc buffers into a completed
1140  * packet.  It uses the ->prev pointers to find the first packet and then
1141  * turns it into the frag list owner.
1142  **/
1143 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb,
1144                                                         u64 *count)
1145 {
1146         unsigned int frag_list_size = 0;
1147
1148         while (skb->prev) {
1149                 struct sk_buff *prev = skb->prev;
1150                 frag_list_size += skb->len;
1151                 skb->prev = NULL;
1152                 skb = prev;
1153                 *count += 1;
1154         }
1155
1156         skb_shinfo(skb)->frag_list = skb->next;
1157         skb->next = NULL;
1158         skb->len += frag_list_size;
1159         skb->data_len += frag_list_size;
1160         skb->truesize += frag_list_size;
1161         return skb;
1162 }
1163
1164 struct ixgbe_rsc_cb {
1165         dma_addr_t dma;
1166         bool delay_unmap;
1167 };
1168
1169 #define IXGBE_RSC_CB(skb) ((struct ixgbe_rsc_cb *)(skb)->cb)
1170
1171 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1172                                struct ixgbe_ring *rx_ring,
1173                                int *work_done, int work_to_do)
1174 {
1175         struct ixgbe_adapter *adapter = q_vector->adapter;
1176         struct net_device *netdev = adapter->netdev;
1177         struct pci_dev *pdev = adapter->pdev;
1178         union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
1179         struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
1180         struct sk_buff *skb;
1181         unsigned int i, rsc_count = 0;
1182         u32 len, staterr;
1183         u16 hdr_info;
1184         bool cleaned = false;
1185         int cleaned_count = 0;
1186         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1187 #ifdef IXGBE_FCOE
1188         int ddp_bytes = 0;
1189 #endif /* IXGBE_FCOE */
1190
1191         i = rx_ring->next_to_clean;
1192         rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1193         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1194         rx_buffer_info = &rx_ring->rx_buffer_info[i];
1195
1196         while (staterr & IXGBE_RXD_STAT_DD) {
1197                 u32 upper_len = 0;
1198                 if (*work_done >= work_to_do)
1199                         break;
1200                 (*work_done)++;
1201
1202                 rmb(); /* read descriptor and rx_buffer_info after status DD */
1203                 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
1204                         hdr_info = le16_to_cpu(ixgbe_get_hdr_info(rx_desc));
1205                         len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
1206                                IXGBE_RXDADV_HDRBUFLEN_SHIFT;
1207                         upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1208                         if ((len > IXGBE_RX_HDR_SIZE) ||
1209                             (upper_len && !(hdr_info & IXGBE_RXDADV_SPH)))
1210                                 len = IXGBE_RX_HDR_SIZE;
1211                 } else {
1212                         len = le16_to_cpu(rx_desc->wb.upper.length);
1213                 }
1214
1215                 cleaned = true;
1216                 skb = rx_buffer_info->skb;
1217                 prefetch(skb->data);
1218                 rx_buffer_info->skb = NULL;
1219
1220                 if (rx_buffer_info->dma) {
1221                         if ((adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
1222                             (!(staterr & IXGBE_RXD_STAT_EOP)) &&
1223                                  (!(skb->prev))) {
1224                                 /*
1225                                  * When HWRSC is enabled, delay unmapping
1226                                  * of the first packet. It carries the
1227                                  * header information, HW may still
1228                                  * access the header after the writeback.
1229                                  * Only unmap it when EOP is reached
1230                                  */
1231                                 IXGBE_RSC_CB(skb)->delay_unmap = true;
1232                                 IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
1233                         } else {
1234                                 dma_unmap_single(&pdev->dev,
1235                                                  rx_buffer_info->dma,
1236                                                  rx_ring->rx_buf_len,
1237                                                  DMA_FROM_DEVICE);
1238                         }
1239                         rx_buffer_info->dma = 0;
1240                         skb_put(skb, len);
1241                 }
1242
1243                 if (upper_len) {
1244                         dma_unmap_page(&pdev->dev, rx_buffer_info->page_dma,
1245                                        PAGE_SIZE / 2, DMA_FROM_DEVICE);
1246                         rx_buffer_info->page_dma = 0;
1247                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1248                                            rx_buffer_info->page,
1249                                            rx_buffer_info->page_offset,
1250                                            upper_len);
1251
1252                         if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
1253                             (page_count(rx_buffer_info->page) != 1))
1254                                 rx_buffer_info->page = NULL;
1255                         else
1256                                 get_page(rx_buffer_info->page);
1257
1258                         skb->len += upper_len;
1259                         skb->data_len += upper_len;
1260                         skb->truesize += upper_len;
1261                 }
1262
1263                 i++;
1264                 if (i == rx_ring->count)
1265                         i = 0;
1266
1267                 next_rxd = IXGBE_RX_DESC_ADV(rx_ring, i);
1268                 prefetch(next_rxd);
1269                 cleaned_count++;
1270
1271                 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
1272                         rsc_count = ixgbe_get_rsc_count(rx_desc);
1273
1274                 if (rsc_count) {
1275                         u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
1276                                      IXGBE_RXDADV_NEXTP_SHIFT;
1277                         next_buffer = &rx_ring->rx_buffer_info[nextp];
1278                 } else {
1279                         next_buffer = &rx_ring->rx_buffer_info[i];
1280                 }
1281
1282                 if (staterr & IXGBE_RXD_STAT_EOP) {
1283                         if (skb->prev)
1284                                 skb = ixgbe_transform_rsc_queue(skb,
1285                                                                 &(rx_ring->rsc_count));
1286                         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
1287                                 if (IXGBE_RSC_CB(skb)->delay_unmap) {
1288                                         dma_unmap_single(&pdev->dev,
1289                                                          IXGBE_RSC_CB(skb)->dma,
1290                                                          rx_ring->rx_buf_len,
1291                                                          DMA_FROM_DEVICE);
1292                                         IXGBE_RSC_CB(skb)->dma = 0;
1293                                         IXGBE_RSC_CB(skb)->delay_unmap = false;
1294                                 }
1295                                 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED)
1296                                         rx_ring->rsc_count +=
1297                                                 skb_shinfo(skb)->nr_frags;
1298                                 else
1299                                         rx_ring->rsc_count++;
1300                                 rx_ring->rsc_flush++;
1301                         }
1302                         rx_ring->stats.packets++;
1303                         rx_ring->stats.bytes += skb->len;
1304                 } else {
1305                         if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
1306                                 rx_buffer_info->skb = next_buffer->skb;
1307                                 rx_buffer_info->dma = next_buffer->dma;
1308                                 next_buffer->skb = skb;
1309                                 next_buffer->dma = 0;
1310                         } else {
1311                                 skb->next = next_buffer->skb;
1312                                 skb->next->prev = skb;
1313                         }
1314                         rx_ring->non_eop_descs++;
1315                         goto next_desc;
1316                 }
1317
1318                 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
1319                         dev_kfree_skb_irq(skb);
1320                         goto next_desc;
1321                 }
1322
1323                 ixgbe_rx_checksum(adapter, rx_desc, skb);
1324
1325                 /* probably a little skewed due to removing CRC */
1326                 total_rx_bytes += skb->len;
1327                 total_rx_packets++;
1328
1329                 skb->protocol = eth_type_trans(skb, adapter->netdev);
1330 #ifdef IXGBE_FCOE
1331                 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1332                 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
1333                         ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1334                         if (!ddp_bytes)
1335                                 goto next_desc;
1336                 }
1337 #endif /* IXGBE_FCOE */
1338                 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
1339
1340 next_desc:
1341                 rx_desc->wb.upper.status_error = 0;
1342
1343                 /* return some buffers to hardware, one at a time is too slow */
1344                 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1345                         ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
1346                         cleaned_count = 0;
1347                 }
1348
1349                 /* use prefetched values */
1350                 rx_desc = next_rxd;
1351                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1352
1353                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1354         }
1355
1356         rx_ring->next_to_clean = i;
1357         cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
1358
1359         if (cleaned_count)
1360                 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
1361
1362 #ifdef IXGBE_FCOE
1363         /* include DDPed FCoE data */
1364         if (ddp_bytes > 0) {
1365                 unsigned int mss;
1366
1367                 mss = adapter->netdev->mtu - sizeof(struct fcoe_hdr) -
1368                         sizeof(struct fc_frame_header) -
1369                         sizeof(struct fcoe_crc_eof);
1370                 if (mss > 512)
1371                         mss &= ~511;
1372                 total_rx_bytes += ddp_bytes;
1373                 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1374         }
1375 #endif /* IXGBE_FCOE */
1376
1377         rx_ring->total_packets += total_rx_packets;
1378         rx_ring->total_bytes += total_rx_bytes;
1379         netdev->stats.rx_bytes += total_rx_bytes;
1380         netdev->stats.rx_packets += total_rx_packets;
1381
1382         return cleaned;
1383 }
1384
1385 static int ixgbe_clean_rxonly(struct napi_struct *, int);
1386 /**
1387  * ixgbe_configure_msix - Configure MSI-X hardware
1388  * @adapter: board private structure
1389  *
1390  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1391  * interrupts.
1392  **/
1393 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1394 {
1395         struct ixgbe_q_vector *q_vector;
1396         int i, j, q_vectors, v_idx, r_idx;
1397         u32 mask;
1398
1399         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1400
1401         /*
1402          * Populate the IVAR table and set the ITR values to the
1403          * corresponding register.
1404          */
1405         for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1406                 q_vector = adapter->q_vector[v_idx];
1407                 /* XXX for_each_set_bit(...) */
1408                 r_idx = find_first_bit(q_vector->rxr_idx,
1409                                        adapter->num_rx_queues);
1410
1411                 for (i = 0; i < q_vector->rxr_count; i++) {
1412                         j = adapter->rx_ring[r_idx]->reg_idx;
1413                         ixgbe_set_ivar(adapter, 0, j, v_idx);
1414                         r_idx = find_next_bit(q_vector->rxr_idx,
1415                                               adapter->num_rx_queues,
1416                                               r_idx + 1);
1417                 }
1418                 r_idx = find_first_bit(q_vector->txr_idx,
1419                                        adapter->num_tx_queues);
1420
1421                 for (i = 0; i < q_vector->txr_count; i++) {
1422                         j = adapter->tx_ring[r_idx]->reg_idx;
1423                         ixgbe_set_ivar(adapter, 1, j, v_idx);
1424                         r_idx = find_next_bit(q_vector->txr_idx,
1425                                               adapter->num_tx_queues,
1426                                               r_idx + 1);
1427                 }
1428
1429                 if (q_vector->txr_count && !q_vector->rxr_count)
1430                         /* tx only */
1431                         q_vector->eitr = adapter->tx_eitr_param;
1432                 else if (q_vector->rxr_count)
1433                         /* rx or mixed */
1434                         q_vector->eitr = adapter->rx_eitr_param;
1435
1436                 ixgbe_write_eitr(q_vector);
1437         }
1438
1439         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
1440                 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1441                                v_idx);
1442         else if (adapter->hw.mac.type == ixgbe_mac_82599EB)
1443                 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1444         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1445
1446         /* set up to autoclear timer, and the vectors */
1447         mask = IXGBE_EIMS_ENABLE_MASK;
1448         if (adapter->num_vfs)
1449                 mask &= ~(IXGBE_EIMS_OTHER |
1450                           IXGBE_EIMS_MAILBOX |
1451                           IXGBE_EIMS_LSC);
1452         else
1453                 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
1454         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1455 }
1456
1457 enum latency_range {
1458         lowest_latency = 0,
1459         low_latency = 1,
1460         bulk_latency = 2,
1461         latency_invalid = 255
1462 };
1463
1464 /**
1465  * ixgbe_update_itr - update the dynamic ITR value based on statistics
1466  * @adapter: pointer to adapter
1467  * @eitr: eitr setting (ints per sec) to give last timeslice
1468  * @itr_setting: current throttle rate in ints/second
1469  * @packets: the number of packets during this measurement interval
1470  * @bytes: the number of bytes during this measurement interval
1471  *
1472  *      Stores a new ITR value based on packets and byte
1473  *      counts during the last interrupt.  The advantage of per interrupt
1474  *      computation is faster updates and more accurate ITR for the current
1475  *      traffic pattern.  Constants in this function were computed
1476  *      based on theoretical maximum wire speed and thresholds were set based
1477  *      on testing data as well as attempting to minimize response time
1478  *      while increasing bulk throughput.
1479  *      this functionality is controlled by the InterruptThrottleRate module
1480  *      parameter (see ixgbe_param.c)
1481  **/
1482 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
1483                            u32 eitr, u8 itr_setting,
1484                            int packets, int bytes)
1485 {
1486         unsigned int retval = itr_setting;
1487         u32 timepassed_us;
1488         u64 bytes_perint;
1489
1490         if (packets == 0)
1491                 goto update_itr_done;
1492
1493
1494         /* simple throttlerate management
1495          *    0-20MB/s lowest (100000 ints/s)
1496          *   20-100MB/s low   (20000 ints/s)
1497          *  100-1249MB/s bulk (8000 ints/s)
1498          */
1499         /* what was last interrupt timeslice? */
1500         timepassed_us = 1000000/eitr;
1501         bytes_perint = bytes / timepassed_us; /* bytes/usec */
1502
1503         switch (itr_setting) {
1504         case lowest_latency:
1505                 if (bytes_perint > adapter->eitr_low)
1506                         retval = low_latency;
1507                 break;
1508         case low_latency:
1509                 if (bytes_perint > adapter->eitr_high)
1510                         retval = bulk_latency;
1511                 else if (bytes_perint <= adapter->eitr_low)
1512                         retval = lowest_latency;
1513                 break;
1514         case bulk_latency:
1515                 if (bytes_perint <= adapter->eitr_high)
1516                         retval = low_latency;
1517                 break;
1518         }
1519
1520 update_itr_done:
1521         return retval;
1522 }
1523
1524 /**
1525  * ixgbe_write_eitr - write EITR register in hardware specific way
1526  * @q_vector: structure containing interrupt and ring information
1527  *
1528  * This function is made to be called by ethtool and by the driver
1529  * when it needs to update EITR registers at runtime.  Hardware
1530  * specific quirks/differences are taken care of here.
1531  */
1532 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1533 {
1534         struct ixgbe_adapter *adapter = q_vector->adapter;
1535         struct ixgbe_hw *hw = &adapter->hw;
1536         int v_idx = q_vector->v_idx;
1537         u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1538
1539         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1540                 /* must write high and low 16 bits to reset counter */
1541                 itr_reg |= (itr_reg << 16);
1542         } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1543                 /*
1544                  * 82599 can support a value of zero, so allow it for
1545                  * max interrupt rate, but there is an errata where it can
1546                  * not be zero with RSC
1547                  */
1548                 if (itr_reg == 8 &&
1549                     !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
1550                         itr_reg = 0;
1551
1552                 /*
1553                  * set the WDIS bit to not clear the timer bits and cause an
1554                  * immediate assertion of the interrupt
1555                  */
1556                 itr_reg |= IXGBE_EITR_CNT_WDIS;
1557         }
1558         IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1559 }
1560
1561 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1562 {
1563         struct ixgbe_adapter *adapter = q_vector->adapter;
1564         u32 new_itr;
1565         u8 current_itr, ret_itr;
1566         int i, r_idx;
1567         struct ixgbe_ring *rx_ring, *tx_ring;
1568
1569         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1570         for (i = 0; i < q_vector->txr_count; i++) {
1571                 tx_ring = adapter->tx_ring[r_idx];
1572                 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1573                                            q_vector->tx_itr,
1574                                            tx_ring->total_packets,
1575                                            tx_ring->total_bytes);
1576                 /* if the result for this queue would decrease interrupt
1577                  * rate for this vector then use that result */
1578                 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1579                                     q_vector->tx_itr - 1 : ret_itr);
1580                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1581                                       r_idx + 1);
1582         }
1583
1584         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1585         for (i = 0; i < q_vector->rxr_count; i++) {
1586                 rx_ring = adapter->rx_ring[r_idx];
1587                 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1588                                            q_vector->rx_itr,
1589                                            rx_ring->total_packets,
1590                                            rx_ring->total_bytes);
1591                 /* if the result for this queue would decrease interrupt
1592                  * rate for this vector then use that result */
1593                 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1594                                     q_vector->rx_itr - 1 : ret_itr);
1595                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1596                                       r_idx + 1);
1597         }
1598
1599         current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1600
1601         switch (current_itr) {
1602         /* counts and packets in update_itr are dependent on these numbers */
1603         case lowest_latency:
1604                 new_itr = 100000;
1605                 break;
1606         case low_latency:
1607                 new_itr = 20000; /* aka hwitr = ~200 */
1608                 break;
1609         case bulk_latency:
1610         default:
1611                 new_itr = 8000;
1612                 break;
1613         }
1614
1615         if (new_itr != q_vector->eitr) {
1616                 /* do an exponential smoothing */
1617                 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1618
1619                 /* save the algorithm value here, not the smoothed one */
1620                 q_vector->eitr = new_itr;
1621
1622                 ixgbe_write_eitr(q_vector);
1623         }
1624 }
1625
1626 /**
1627  * ixgbe_check_overtemp_task - worker thread to check over tempurature
1628  * @work: pointer to work_struct containing our data
1629  **/
1630 static void ixgbe_check_overtemp_task(struct work_struct *work)
1631 {
1632         struct ixgbe_adapter *adapter = container_of(work,
1633                                                      struct ixgbe_adapter,
1634                                                      check_overtemp_task);
1635         struct ixgbe_hw *hw = &adapter->hw;
1636         u32 eicr = adapter->interrupt_event;
1637
1638         if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
1639                 switch (hw->device_id) {
1640                 case IXGBE_DEV_ID_82599_T3_LOM: {
1641                         u32 autoneg;
1642                         bool link_up = false;
1643
1644                         if (hw->mac.ops.check_link)
1645                                 hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
1646
1647                         if (((eicr & IXGBE_EICR_GPI_SDP0) && (!link_up)) ||
1648                             (eicr & IXGBE_EICR_LSC))
1649                                 /* Check if this is due to overtemp */
1650                                 if (hw->phy.ops.check_overtemp(hw) == IXGBE_ERR_OVERTEMP)
1651                                         break;
1652                         }
1653                         return;
1654                 default:
1655                         if (!(eicr & IXGBE_EICR_GPI_SDP0))
1656                                 return;
1657                         break;
1658                 }
1659                 e_crit(drv, "Network adapter has been stopped because it has "
1660                        "over heated. Restart the computer. If the problem "
1661                        "persists, power off the system and replace the "
1662                        "adapter\n");
1663                 /* write to clear the interrupt */
1664                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP0);
1665         }
1666 }
1667
1668 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1669 {
1670         struct ixgbe_hw *hw = &adapter->hw;
1671
1672         if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1673             (eicr & IXGBE_EICR_GPI_SDP1)) {
1674                 e_crit(probe, "Fan has stopped, replace the adapter\n");
1675                 /* write to clear the interrupt */
1676                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1677         }
1678 }
1679
1680 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1681 {
1682         struct ixgbe_hw *hw = &adapter->hw;
1683
1684         if (eicr & IXGBE_EICR_GPI_SDP1) {
1685                 /* Clear the interrupt */
1686                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1687                 schedule_work(&adapter->multispeed_fiber_task);
1688         } else if (eicr & IXGBE_EICR_GPI_SDP2) {
1689                 /* Clear the interrupt */
1690                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1691                 schedule_work(&adapter->sfp_config_module_task);
1692         } else {
1693                 /* Interrupt isn't for us... */
1694                 return;
1695         }
1696 }
1697
1698 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1699 {
1700         struct ixgbe_hw *hw = &adapter->hw;
1701
1702         adapter->lsc_int++;
1703         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1704         adapter->link_check_timeout = jiffies;
1705         if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1706                 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1707                 IXGBE_WRITE_FLUSH(hw);
1708                 schedule_work(&adapter->watchdog_task);
1709         }
1710 }
1711
1712 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1713 {
1714         struct net_device *netdev = data;
1715         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1716         struct ixgbe_hw *hw = &adapter->hw;
1717         u32 eicr;
1718
1719         /*
1720          * Workaround for Silicon errata.  Use clear-by-write instead
1721          * of clear-by-read.  Reading with EICS will return the
1722          * interrupt causes without clearing, which later be done
1723          * with the write to EICR.
1724          */
1725         eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1726         IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1727
1728         if (eicr & IXGBE_EICR_LSC)
1729                 ixgbe_check_lsc(adapter);
1730
1731         if (eicr & IXGBE_EICR_MAILBOX)
1732                 ixgbe_msg_task(adapter);
1733
1734         if (hw->mac.type == ixgbe_mac_82598EB)
1735                 ixgbe_check_fan_failure(adapter, eicr);
1736
1737         if (hw->mac.type == ixgbe_mac_82599EB) {
1738                 ixgbe_check_sfp_event(adapter, eicr);
1739                 adapter->interrupt_event = eicr;
1740                 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
1741                     ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)))
1742                         schedule_work(&adapter->check_overtemp_task);
1743
1744                 /* Handle Flow Director Full threshold interrupt */
1745                 if (eicr & IXGBE_EICR_FLOW_DIR) {
1746                         int i;
1747                         IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1748                         /* Disable transmits before FDIR Re-initialization */
1749                         netif_tx_stop_all_queues(netdev);
1750                         for (i = 0; i < adapter->num_tx_queues; i++) {
1751                                 struct ixgbe_ring *tx_ring =
1752                                                             adapter->tx_ring[i];
1753                                 if (test_and_clear_bit(__IXGBE_FDIR_INIT_DONE,
1754                                                        &tx_ring->reinit_state))
1755                                         schedule_work(&adapter->fdir_reinit_task);
1756                         }
1757                 }
1758         }
1759         if (!test_bit(__IXGBE_DOWN, &adapter->state))
1760                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1761
1762         return IRQ_HANDLED;
1763 }
1764
1765 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1766                                            u64 qmask)
1767 {
1768         u32 mask;
1769
1770         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1771                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1772                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1773         } else {
1774                 mask = (qmask & 0xFFFFFFFF);
1775                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(0), mask);
1776                 mask = (qmask >> 32);
1777                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(1), mask);
1778         }
1779         /* skip the flush */
1780 }
1781
1782 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1783                                             u64 qmask)
1784 {
1785         u32 mask;
1786
1787         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1788                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1789                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, mask);
1790         } else {
1791                 mask = (qmask & 0xFFFFFFFF);
1792                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), mask);
1793                 mask = (qmask >> 32);
1794                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), mask);
1795         }
1796         /* skip the flush */
1797 }
1798
1799 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1800 {
1801         struct ixgbe_q_vector *q_vector = data;
1802         struct ixgbe_adapter  *adapter = q_vector->adapter;
1803         struct ixgbe_ring     *tx_ring;
1804         int i, r_idx;
1805
1806         if (!q_vector->txr_count)
1807                 return IRQ_HANDLED;
1808
1809         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1810         for (i = 0; i < q_vector->txr_count; i++) {
1811                 tx_ring = adapter->tx_ring[r_idx];
1812                 tx_ring->total_bytes = 0;
1813                 tx_ring->total_packets = 0;
1814                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1815                                       r_idx + 1);
1816         }
1817
1818         /* EIAM disabled interrupts (on this vector) for us */
1819         napi_schedule(&q_vector->napi);
1820
1821         return IRQ_HANDLED;
1822 }
1823
1824 /**
1825  * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
1826  * @irq: unused
1827  * @data: pointer to our q_vector struct for this interrupt vector
1828  **/
1829 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
1830 {
1831         struct ixgbe_q_vector *q_vector = data;
1832         struct ixgbe_adapter  *adapter = q_vector->adapter;
1833         struct ixgbe_ring  *rx_ring;
1834         int r_idx;
1835         int i;
1836
1837         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1838         for (i = 0;  i < q_vector->rxr_count; i++) {
1839                 rx_ring = adapter->rx_ring[r_idx];
1840                 rx_ring->total_bytes = 0;
1841                 rx_ring->total_packets = 0;
1842                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1843                                       r_idx + 1);
1844         }
1845
1846         if (!q_vector->rxr_count)
1847                 return IRQ_HANDLED;
1848
1849         /* disable interrupts on this vector only */
1850         /* EIAM disabled interrupts (on this vector) for us */
1851         napi_schedule(&q_vector->napi);
1852
1853         return IRQ_HANDLED;
1854 }
1855
1856 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1857 {
1858         struct ixgbe_q_vector *q_vector = data;
1859         struct ixgbe_adapter  *adapter = q_vector->adapter;
1860         struct ixgbe_ring  *ring;
1861         int r_idx;
1862         int i;
1863
1864         if (!q_vector->txr_count && !q_vector->rxr_count)
1865                 return IRQ_HANDLED;
1866
1867         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1868         for (i = 0; i < q_vector->txr_count; i++) {
1869                 ring = adapter->tx_ring[r_idx];
1870                 ring->total_bytes = 0;
1871                 ring->total_packets = 0;
1872                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1873                                       r_idx + 1);
1874         }
1875
1876         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1877         for (i = 0; i < q_vector->rxr_count; i++) {
1878                 ring = adapter->rx_ring[r_idx];
1879                 ring->total_bytes = 0;
1880                 ring->total_packets = 0;
1881                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1882                                       r_idx + 1);
1883         }
1884
1885         /* EIAM disabled interrupts (on this vector) for us */
1886         napi_schedule(&q_vector->napi);
1887
1888         return IRQ_HANDLED;
1889 }
1890
1891 /**
1892  * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1893  * @napi: napi struct with our devices info in it
1894  * @budget: amount of work driver is allowed to do this pass, in packets
1895  *
1896  * This function is optimized for cleaning one queue only on a single
1897  * q_vector!!!
1898  **/
1899 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1900 {
1901         struct ixgbe_q_vector *q_vector =
1902                                container_of(napi, struct ixgbe_q_vector, napi);
1903         struct ixgbe_adapter *adapter = q_vector->adapter;
1904         struct ixgbe_ring *rx_ring = NULL;
1905         int work_done = 0;
1906         long r_idx;
1907
1908         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1909         rx_ring = adapter->rx_ring[r_idx];
1910 #ifdef CONFIG_IXGBE_DCA
1911         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1912                 ixgbe_update_rx_dca(adapter, rx_ring);
1913 #endif
1914
1915         ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1916
1917         /* If all Rx work done, exit the polling mode */
1918         if (work_done < budget) {
1919                 napi_complete(napi);
1920                 if (adapter->rx_itr_setting & 1)
1921                         ixgbe_set_itr_msix(q_vector);
1922                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1923                         ixgbe_irq_enable_queues(adapter,
1924                                                 ((u64)1 << q_vector->v_idx));
1925         }
1926
1927         return work_done;
1928 }
1929
1930 /**
1931  * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
1932  * @napi: napi struct with our devices info in it
1933  * @budget: amount of work driver is allowed to do this pass, in packets
1934  *
1935  * This function will clean more than one rx queue associated with a
1936  * q_vector.
1937  **/
1938 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
1939 {
1940         struct ixgbe_q_vector *q_vector =
1941                                container_of(napi, struct ixgbe_q_vector, napi);
1942         struct ixgbe_adapter *adapter = q_vector->adapter;
1943         struct ixgbe_ring *ring = NULL;
1944         int work_done = 0, i;
1945         long r_idx;
1946         bool tx_clean_complete = true;
1947
1948         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1949         for (i = 0; i < q_vector->txr_count; i++) {
1950                 ring = adapter->tx_ring[r_idx];
1951 #ifdef CONFIG_IXGBE_DCA
1952                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1953                         ixgbe_update_tx_dca(adapter, ring);
1954 #endif
1955                 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
1956                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1957                                       r_idx + 1);
1958         }
1959
1960         /* attempt to distribute budget to each queue fairly, but don't allow
1961          * the budget to go below 1 because we'll exit polling */
1962         budget /= (q_vector->rxr_count ?: 1);
1963         budget = max(budget, 1);
1964         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1965         for (i = 0; i < q_vector->rxr_count; i++) {
1966                 ring = adapter->rx_ring[r_idx];
1967 #ifdef CONFIG_IXGBE_DCA
1968                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1969                         ixgbe_update_rx_dca(adapter, ring);
1970 #endif
1971                 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
1972                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1973                                       r_idx + 1);
1974         }
1975
1976         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1977         ring = adapter->rx_ring[r_idx];
1978         /* If all Rx work done, exit the polling mode */
1979         if (work_done < budget) {
1980                 napi_complete(napi);
1981                 if (adapter->rx_itr_setting & 1)
1982                         ixgbe_set_itr_msix(q_vector);
1983                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1984                         ixgbe_irq_enable_queues(adapter,
1985                                                 ((u64)1 << q_vector->v_idx));
1986                 return 0;
1987         }
1988
1989         return work_done;
1990 }
1991
1992 /**
1993  * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
1994  * @napi: napi struct with our devices info in it
1995  * @budget: amount of work driver is allowed to do this pass, in packets
1996  *
1997  * This function is optimized for cleaning one queue only on a single
1998  * q_vector!!!
1999  **/
2000 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
2001 {
2002         struct ixgbe_q_vector *q_vector =
2003                                container_of(napi, struct ixgbe_q_vector, napi);
2004         struct ixgbe_adapter *adapter = q_vector->adapter;
2005         struct ixgbe_ring *tx_ring = NULL;
2006         int work_done = 0;
2007         long r_idx;
2008
2009         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2010         tx_ring = adapter->tx_ring[r_idx];
2011 #ifdef CONFIG_IXGBE_DCA
2012         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2013                 ixgbe_update_tx_dca(adapter, tx_ring);
2014 #endif
2015
2016         if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
2017                 work_done = budget;
2018
2019         /* If all Tx work done, exit the polling mode */
2020         if (work_done < budget) {
2021                 napi_complete(napi);
2022                 if (adapter->tx_itr_setting & 1)
2023                         ixgbe_set_itr_msix(q_vector);
2024                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2025                         ixgbe_irq_enable_queues(adapter,
2026                                                 ((u64)1 << q_vector->v_idx));
2027         }
2028
2029         return work_done;
2030 }
2031
2032 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
2033                                      int r_idx)
2034 {
2035         struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2036
2037         set_bit(r_idx, q_vector->rxr_idx);
2038         q_vector->rxr_count++;
2039 }
2040
2041 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
2042                                      int t_idx)
2043 {
2044         struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2045
2046         set_bit(t_idx, q_vector->txr_idx);
2047         q_vector->txr_count++;
2048 }
2049
2050 /**
2051  * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
2052  * @adapter: board private structure to initialize
2053  * @vectors: allotted vector count for descriptor rings
2054  *
2055  * This function maps descriptor rings to the queue-specific vectors
2056  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
2057  * one vector per ring/queue, but on a constrained vector budget, we
2058  * group the rings as "efficiently" as possible.  You would add new
2059  * mapping configurations in here.
2060  **/
2061 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
2062                                       int vectors)
2063 {
2064         int v_start = 0;
2065         int rxr_idx = 0, txr_idx = 0;
2066         int rxr_remaining = adapter->num_rx_queues;
2067         int txr_remaining = adapter->num_tx_queues;
2068         int i, j;
2069         int rqpv, tqpv;
2070         int err = 0;
2071
2072         /* No mapping required if MSI-X is disabled. */
2073         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2074                 goto out;
2075
2076         /*
2077          * The ideal configuration...
2078          * We have enough vectors to map one per queue.
2079          */
2080         if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
2081                 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
2082                         map_vector_to_rxq(adapter, v_start, rxr_idx);
2083
2084                 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
2085                         map_vector_to_txq(adapter, v_start, txr_idx);
2086
2087                 goto out;
2088         }
2089
2090         /*
2091          * If we don't have enough vectors for a 1-to-1
2092          * mapping, we'll have to group them so there are
2093          * multiple queues per vector.
2094          */
2095         /* Re-adjusting *qpv takes care of the remainder. */
2096         for (i = v_start; i < vectors; i++) {
2097                 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
2098                 for (j = 0; j < rqpv; j++) {
2099                         map_vector_to_rxq(adapter, i, rxr_idx);
2100                         rxr_idx++;
2101                         rxr_remaining--;
2102                 }
2103         }
2104         for (i = v_start; i < vectors; i++) {
2105                 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
2106                 for (j = 0; j < tqpv; j++) {
2107                         map_vector_to_txq(adapter, i, txr_idx);
2108                         txr_idx++;
2109                         txr_remaining--;
2110                 }
2111         }
2112
2113 out:
2114         return err;
2115 }
2116
2117 /**
2118  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2119  * @adapter: board private structure
2120  *
2121  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2122  * interrupts from the kernel.
2123  **/
2124 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2125 {
2126         struct net_device *netdev = adapter->netdev;
2127         irqreturn_t (*handler)(int, void *);
2128         int i, vector, q_vectors, err;
2129         int ri = 0, ti = 0;
2130
2131         /* Decrement for Other and TCP Timer vectors */
2132         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2133
2134         /* Map the Tx/Rx rings to the vectors we were allotted. */
2135         err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
2136         if (err)
2137                 goto out;
2138
2139 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
2140                          (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
2141                          &ixgbe_msix_clean_many)
2142         for (vector = 0; vector < q_vectors; vector++) {
2143                 handler = SET_HANDLER(adapter->q_vector[vector]);
2144
2145                 if (handler == &ixgbe_msix_clean_rx) {
2146                         sprintf(adapter->name[vector], "%s-%s-%d",
2147                                 netdev->name, "rx", ri++);
2148                 } else if (handler == &ixgbe_msix_clean_tx) {
2149                         sprintf(adapter->name[vector], "%s-%s-%d",
2150                                 netdev->name, "tx", ti++);
2151                 } else
2152                         sprintf(adapter->name[vector], "%s-%s-%d",
2153                                 netdev->name, "TxRx", vector);
2154
2155                 err = request_irq(adapter->msix_entries[vector].vector,
2156                                   handler, 0, adapter->name[vector],
2157                                   adapter->q_vector[vector]);
2158                 if (err) {
2159                         e_err(probe, "request_irq failed for MSIX interrupt "
2160                               "Error: %d\n", err);
2161                         goto free_queue_irqs;
2162                 }
2163         }
2164
2165         sprintf(adapter->name[vector], "%s:lsc", netdev->name);
2166         err = request_irq(adapter->msix_entries[vector].vector,
2167                           ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
2168         if (err) {
2169                 e_err(probe, "request_irq for msix_lsc failed: %d\n", err);
2170                 goto free_queue_irqs;
2171         }
2172
2173         return 0;
2174
2175 free_queue_irqs:
2176         for (i = vector - 1; i >= 0; i--)
2177                 free_irq(adapter->msix_entries[--vector].vector,
2178                          adapter->q_vector[i]);
2179         adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2180         pci_disable_msix(adapter->pdev);
2181         kfree(adapter->msix_entries);
2182         adapter->msix_entries = NULL;
2183 out:
2184         return err;
2185 }
2186
2187 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
2188 {
2189         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2190         u8 current_itr;
2191         u32 new_itr = q_vector->eitr;
2192         struct ixgbe_ring *rx_ring = adapter->rx_ring[0];
2193         struct ixgbe_ring *tx_ring = adapter->tx_ring[0];
2194
2195         q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
2196                                             q_vector->tx_itr,
2197                                             tx_ring->total_packets,
2198                                             tx_ring->total_bytes);
2199         q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
2200                                             q_vector->rx_itr,
2201                                             rx_ring->total_packets,
2202                                             rx_ring->total_bytes);
2203
2204         current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
2205
2206         switch (current_itr) {
2207         /* counts and packets in update_itr are dependent on these numbers */
2208         case lowest_latency:
2209                 new_itr = 100000;
2210                 break;
2211         case low_latency:
2212                 new_itr = 20000; /* aka hwitr = ~200 */
2213                 break;
2214         case bulk_latency:
2215                 new_itr = 8000;
2216                 break;
2217         default:
2218                 break;
2219         }
2220
2221         if (new_itr != q_vector->eitr) {
2222                 /* do an exponential smoothing */
2223                 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
2224
2225                 /* save the algorithm value here, not the smoothed one */
2226                 q_vector->eitr = new_itr;
2227
2228                 ixgbe_write_eitr(q_vector);
2229         }
2230 }
2231
2232 /**
2233  * ixgbe_irq_enable - Enable default interrupt generation settings
2234  * @adapter: board private structure
2235  **/
2236 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter)
2237 {
2238         u32 mask;
2239
2240         mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2241         if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2242                 mask |= IXGBE_EIMS_GPI_SDP0;
2243         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2244                 mask |= IXGBE_EIMS_GPI_SDP1;
2245         if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
2246                 mask |= IXGBE_EIMS_ECC;
2247                 mask |= IXGBE_EIMS_GPI_SDP1;
2248                 mask |= IXGBE_EIMS_GPI_SDP2;
2249                 if (adapter->num_vfs)
2250                         mask |= IXGBE_EIMS_MAILBOX;
2251         }
2252         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
2253             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
2254                 mask |= IXGBE_EIMS_FLOW_DIR;
2255
2256         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2257         ixgbe_irq_enable_queues(adapter, ~0);
2258         IXGBE_WRITE_FLUSH(&adapter->hw);
2259
2260         if (adapter->num_vfs > 32) {
2261                 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2262                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2263         }
2264 }
2265
2266 /**
2267  * ixgbe_intr - legacy mode Interrupt Handler
2268  * @irq: interrupt number
2269  * @data: pointer to a network interface device structure
2270  **/
2271 static irqreturn_t ixgbe_intr(int irq, void *data)
2272 {
2273         struct net_device *netdev = data;
2274         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2275         struct ixgbe_hw *hw = &adapter->hw;
2276         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2277         u32 eicr;
2278
2279         /*
2280          * Workaround for silicon errata.  Mask the interrupts
2281          * before the read of EICR.
2282          */
2283         IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2284
2285         /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2286          * therefore no explict interrupt disable is necessary */
2287         eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2288         if (!eicr) {
2289                 /* shared interrupt alert!
2290                  * make sure interrupts are enabled because the read will
2291                  * have disabled interrupts due to EIAM */
2292                 ixgbe_irq_enable(adapter);
2293                 return IRQ_NONE;        /* Not our interrupt */
2294         }
2295
2296         if (eicr & IXGBE_EICR_LSC)
2297                 ixgbe_check_lsc(adapter);
2298
2299         if (hw->mac.type == ixgbe_mac_82599EB)
2300                 ixgbe_check_sfp_event(adapter, eicr);
2301
2302         ixgbe_check_fan_failure(adapter, eicr);
2303         if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2304             ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)))
2305                 schedule_work(&adapter->check_overtemp_task);
2306
2307         if (napi_schedule_prep(&(q_vector->napi))) {
2308                 adapter->tx_ring[0]->total_packets = 0;
2309                 adapter->tx_ring[0]->total_bytes = 0;
2310                 adapter->rx_ring[0]->total_packets = 0;
2311                 adapter->rx_ring[0]->total_bytes = 0;
2312                 /* would disable interrupts here but EIAM disabled it */
2313                 __napi_schedule(&(q_vector->napi));
2314         }
2315
2316         return IRQ_HANDLED;
2317 }
2318
2319 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
2320 {
2321         int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2322
2323         for (i = 0; i < q_vectors; i++) {
2324                 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
2325                 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
2326                 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
2327                 q_vector->rxr_count = 0;
2328                 q_vector->txr_count = 0;
2329         }
2330 }
2331
2332 /**
2333  * ixgbe_request_irq - initialize interrupts
2334  * @adapter: board private structure
2335  *
2336  * Attempts to configure interrupts using the best available
2337  * capabilities of the hardware and kernel.
2338  **/
2339 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2340 {
2341         struct net_device *netdev = adapter->netdev;
2342         int err;
2343
2344         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2345                 err = ixgbe_request_msix_irqs(adapter);
2346         } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
2347                 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2348                                   netdev->name, netdev);
2349         } else {
2350                 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2351                                   netdev->name, netdev);
2352         }
2353
2354         if (err)
2355                 e_err(probe, "request_irq failed, Error %d\n", err);
2356
2357         return err;
2358 }
2359
2360 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2361 {
2362         struct net_device *netdev = adapter->netdev;
2363
2364         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2365                 int i, q_vectors;
2366
2367                 q_vectors = adapter->num_msix_vectors;
2368
2369                 i = q_vectors - 1;
2370                 free_irq(adapter->msix_entries[i].vector, netdev);
2371
2372                 i--;
2373                 for (; i >= 0; i--) {
2374                         free_irq(adapter->msix_entries[i].vector,
2375                                  adapter->q_vector[i]);
2376                 }
2377
2378                 ixgbe_reset_q_vectors(adapter);
2379         } else {
2380                 free_irq(adapter->pdev->irq, netdev);
2381         }
2382 }
2383
2384 /**
2385  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2386  * @adapter: board private structure
2387  **/
2388 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2389 {
2390         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2391                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2392         } else {
2393                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2394                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2395                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2396                 if (adapter->num_vfs > 32)
2397                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
2398         }
2399         IXGBE_WRITE_FLUSH(&adapter->hw);
2400         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2401                 int i;
2402                 for (i = 0; i < adapter->num_msix_vectors; i++)
2403                         synchronize_irq(adapter->msix_entries[i].vector);
2404         } else {
2405                 synchronize_irq(adapter->pdev->irq);
2406         }
2407 }
2408
2409 /**
2410  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2411  *
2412  **/
2413 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2414 {
2415         struct ixgbe_hw *hw = &adapter->hw;
2416
2417         IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
2418                         EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
2419
2420         ixgbe_set_ivar(adapter, 0, 0, 0);
2421         ixgbe_set_ivar(adapter, 1, 0, 0);
2422
2423         map_vector_to_rxq(adapter, 0, 0);
2424         map_vector_to_txq(adapter, 0, 0);
2425
2426         e_info(hw, "Legacy interrupt IVAR setup done\n");
2427 }
2428
2429 /**
2430  * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2431  * @adapter: board private structure
2432  * @ring: structure containing ring specific data
2433  *
2434  * Configure the Tx descriptor ring after a reset.
2435  **/
2436 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2437                              struct ixgbe_ring *ring)
2438 {
2439         struct ixgbe_hw *hw = &adapter->hw;
2440         u64 tdba = ring->dma;
2441         int wait_loop = 10;
2442         u32 txdctl;
2443         u16 reg_idx = ring->reg_idx;
2444
2445         /* disable queue to avoid issues while updating state */
2446         txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2447         IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx),
2448                         txdctl & ~IXGBE_TXDCTL_ENABLE);
2449         IXGBE_WRITE_FLUSH(hw);
2450
2451         IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2452                         (tdba & DMA_BIT_MASK(32)));
2453         IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2454         IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2455                         ring->count * sizeof(union ixgbe_adv_tx_desc));
2456         IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2457         IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2458         ring->head = IXGBE_TDH(reg_idx);
2459         ring->tail = IXGBE_TDT(reg_idx);
2460
2461         /* configure fetching thresholds */
2462         if (adapter->rx_itr_setting == 0) {
2463                 /* cannot set wthresh when itr==0 */
2464                 txdctl &= ~0x007F0000;
2465         } else {
2466                 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2467                 txdctl |= (8 << 16);
2468         }
2469         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2470                 /* PThresh workaround for Tx hang with DFP enabled. */
2471                 txdctl |= 32;
2472         }
2473
2474         /* reinitialize flowdirector state */
2475         set_bit(__IXGBE_FDIR_INIT_DONE, &ring->reinit_state);
2476
2477         /* enable queue */
2478         txdctl |= IXGBE_TXDCTL_ENABLE;
2479         IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2480
2481         /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2482         if (hw->mac.type == ixgbe_mac_82598EB &&
2483             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2484                 return;
2485
2486         /* poll to verify queue is enabled */
2487         do {
2488                 msleep(1);
2489                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2490         } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2491         if (!wait_loop)
2492                 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2493 }
2494
2495 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2496 {
2497         struct ixgbe_hw *hw = &adapter->hw;
2498         u32 rttdcs;
2499         u32 mask;
2500
2501         if (hw->mac.type == ixgbe_mac_82598EB)
2502                 return;
2503
2504         /* disable the arbiter while setting MTQC */
2505         rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2506         rttdcs |= IXGBE_RTTDCS_ARBDIS;
2507         IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2508
2509         /* set transmit pool layout */
2510         mask = (IXGBE_FLAG_SRIOV_ENABLED | IXGBE_FLAG_DCB_ENABLED);
2511         switch (adapter->flags & mask) {
2512
2513         case (IXGBE_FLAG_SRIOV_ENABLED):
2514                 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2515                                 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2516                 break;
2517
2518         case (IXGBE_FLAG_DCB_ENABLED):
2519                 /* We enable 8 traffic classes, DCB only */
2520                 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2521                               (IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ));
2522                 break;
2523
2524         default:
2525                 IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
2526                 break;
2527         }
2528
2529         /* re-enable the arbiter */
2530         rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2531         IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2532 }
2533
2534 /**
2535  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2536  * @adapter: board private structure
2537  *
2538  * Configure the Tx unit of the MAC after a reset.
2539  **/
2540 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2541 {
2542         struct ixgbe_hw *hw = &adapter->hw;
2543         u32 dmatxctl;
2544         u32 i;
2545
2546         ixgbe_setup_mtqc(adapter);
2547
2548         if (hw->mac.type != ixgbe_mac_82598EB) {
2549                 /* DMATXCTL.EN must be before Tx queues are enabled */
2550                 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2551                 dmatxctl |= IXGBE_DMATXCTL_TE;
2552                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2553         }
2554
2555         /* Setup the HW Tx Head and Tail descriptor pointers */
2556         for (i = 0; i < adapter->num_tx_queues; i++)
2557                 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2558 }
2559
2560 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2561
2562 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2563                                    struct ixgbe_ring *rx_ring)
2564 {
2565         u32 srrctl;
2566         int index;
2567         struct ixgbe_ring_feature *feature = adapter->ring_feature;
2568
2569         index = rx_ring->reg_idx;
2570         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2571                 unsigned long mask;
2572                 mask = (unsigned long) feature[RING_F_RSS].mask;
2573                 index = index & mask;
2574         }
2575         srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
2576
2577         srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2578         srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2579         if (adapter->num_vfs)
2580                 srrctl |= IXGBE_SRRCTL_DROP_EN;
2581
2582         srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2583                   IXGBE_SRRCTL_BSIZEHDR_MASK;
2584
2585         if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
2586 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2587                 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2588 #else
2589                 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2590 #endif
2591                 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2592         } else {
2593                 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2594                           IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2595                 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2596         }
2597
2598         IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
2599 }
2600
2601 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2602 {
2603         struct ixgbe_hw *hw = &adapter->hw;
2604         static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2605                           0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2606                           0x6A3E67EA, 0x14364D17, 0x3BED200D};
2607         u32 mrqc = 0, reta = 0;
2608         u32 rxcsum;
2609         int i, j;
2610         int mask;
2611
2612         /* Fill out hash function seeds */
2613         for (i = 0; i < 10; i++)
2614                 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2615
2616         /* Fill out redirection table */
2617         for (i = 0, j = 0; i < 128; i++, j++) {
2618                 if (j == adapter->ring_feature[RING_F_RSS].indices)
2619                         j = 0;
2620                 /* reta = 4-byte sliding window of
2621                  * 0x00..(indices-1)(indices-1)00..etc. */
2622                 reta = (reta << 8) | (j * 0x11);
2623                 if ((i & 3) == 3)
2624                         IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2625         }
2626
2627         /* Disable indicating checksum in descriptor, enables RSS hash */
2628         rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2629         rxcsum |= IXGBE_RXCSUM_PCSD;
2630         IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2631
2632         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
2633                 mask = adapter->flags & IXGBE_FLAG_RSS_ENABLED;
2634         else
2635                 mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2636 #ifdef CONFIG_IXGBE_DCB
2637                                          | IXGBE_FLAG_DCB_ENABLED
2638 #endif
2639                                          | IXGBE_FLAG_SRIOV_ENABLED
2640                                         );
2641
2642         switch (mask) {
2643         case (IXGBE_FLAG_RSS_ENABLED):
2644                 mrqc = IXGBE_MRQC_RSSEN;
2645                 break;
2646         case (IXGBE_FLAG_SRIOV_ENABLED):
2647                 mrqc = IXGBE_MRQC_VMDQEN;
2648                 break;
2649 #ifdef CONFIG_IXGBE_DCB
2650         case (IXGBE_FLAG_DCB_ENABLED):
2651                 mrqc = IXGBE_MRQC_RT8TCEN;
2652                 break;
2653 #endif /* CONFIG_IXGBE_DCB */
2654         default:
2655                 break;
2656         }
2657
2658         /* Perform hash on these packet types */
2659         mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2660               | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2661               | IXGBE_MRQC_RSS_FIELD_IPV6
2662               | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2663
2664         IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2665 }
2666
2667 /**
2668  * ixgbe_configure_rscctl - enable RSC for the indicated ring
2669  * @adapter:    address of board private structure
2670  * @index:      index of ring to set
2671  **/
2672 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2673                                    struct ixgbe_ring *ring)
2674 {
2675         struct ixgbe_hw *hw = &adapter->hw;
2676         u32 rscctrl;
2677         int rx_buf_len;
2678         u16 reg_idx = ring->reg_idx;
2679
2680         if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
2681                 return;
2682
2683         rx_buf_len = ring->rx_buf_len;
2684         rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2685         rscctrl |= IXGBE_RSCCTL_RSCEN;
2686         /*
2687          * we must limit the number of descriptors so that the
2688          * total size of max desc * buf_len is not greater
2689          * than 65535
2690          */
2691         if (ring->flags & IXGBE_RING_RX_PS_ENABLED) {
2692 #if (MAX_SKB_FRAGS > 16)
2693                 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2694 #elif (MAX_SKB_FRAGS > 8)
2695                 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2696 #elif (MAX_SKB_FRAGS > 4)
2697                 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2698 #else
2699                 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2700 #endif
2701         } else {
2702                 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2703                         rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2704                 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2705                         rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2706                 else
2707                         rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2708         }
2709         IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2710 }
2711
2712 /**
2713  *  ixgbe_set_uta - Set unicast filter table address
2714  *  @adapter: board private structure
2715  *
2716  *  The unicast table address is a register array of 32-bit registers.
2717  *  The table is meant to be used in a way similar to how the MTA is used
2718  *  however due to certain limitations in the hardware it is necessary to
2719  *  set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
2720  *  enable bit to allow vlan tag stripping when promiscuous mode is enabled
2721  **/
2722 static void ixgbe_set_uta(struct ixgbe_adapter *adapter)
2723 {
2724         struct ixgbe_hw *hw = &adapter->hw;
2725         int i;
2726
2727         /* The UTA table only exists on 82599 hardware and newer */
2728         if (hw->mac.type < ixgbe_mac_82599EB)
2729                 return;
2730
2731         /* we only need to do this if VMDq is enabled */
2732         if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2733                 return;
2734
2735         for (i = 0; i < 128; i++)
2736                 IXGBE_WRITE_REG(hw, IXGBE_UTA(i), ~0);
2737 }
2738
2739 #define IXGBE_MAX_RX_DESC_POLL 10
2740 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2741                                        struct ixgbe_ring *ring)
2742 {
2743         struct ixgbe_hw *hw = &adapter->hw;
2744         int reg_idx = ring->reg_idx;
2745         int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2746         u32 rxdctl;
2747
2748         /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2749         if (hw->mac.type == ixgbe_mac_82598EB &&
2750             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2751                 return;
2752
2753         do {
2754                 msleep(1);
2755                 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2756         } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
2757
2758         if (!wait_loop) {
2759                 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
2760                       "the polling period\n", reg_idx);
2761         }
2762 }
2763
2764 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
2765                              struct ixgbe_ring *ring)
2766 {
2767         struct ixgbe_hw *hw = &adapter->hw;
2768         u64 rdba = ring->dma;
2769         u32 rxdctl;
2770         u16 reg_idx = ring->reg_idx;
2771
2772         /* disable queue to avoid issues while updating state */
2773         rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2774         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx),
2775                         rxdctl & ~IXGBE_RXDCTL_ENABLE);
2776         IXGBE_WRITE_FLUSH(hw);
2777
2778         IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
2779         IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
2780         IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
2781                         ring->count * sizeof(union ixgbe_adv_rx_desc));
2782         IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
2783         IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
2784         ring->head = IXGBE_RDH(reg_idx);
2785         ring->tail = IXGBE_RDT(reg_idx);
2786
2787         ixgbe_configure_srrctl(adapter, ring);
2788         ixgbe_configure_rscctl(adapter, ring);
2789
2790         if (hw->mac.type == ixgbe_mac_82598EB) {
2791                 /*
2792                  * enable cache line friendly hardware writes:
2793                  * PTHRESH=32 descriptors (half the internal cache),
2794                  * this also removes ugly rx_no_buffer_count increment
2795                  * HTHRESH=4 descriptors (to minimize latency on fetch)
2796                  * WTHRESH=8 burst writeback up to two cache lines
2797                  */
2798                 rxdctl &= ~0x3FFFFF;
2799                 rxdctl |=  0x080420;
2800         }
2801
2802         /* enable receive descriptor ring */
2803         rxdctl |= IXGBE_RXDCTL_ENABLE;
2804         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
2805
2806         ixgbe_rx_desc_queue_enable(adapter, ring);
2807         ixgbe_alloc_rx_buffers(adapter, ring, IXGBE_DESC_UNUSED(ring));
2808 }
2809
2810 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
2811 {
2812         struct ixgbe_hw *hw = &adapter->hw;
2813         int p;
2814
2815         /* PSRTYPE must be initialized in non 82598 adapters */
2816         u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2817                       IXGBE_PSRTYPE_UDPHDR |
2818                       IXGBE_PSRTYPE_IPV4HDR |
2819                       IXGBE_PSRTYPE_L2HDR |
2820                       IXGBE_PSRTYPE_IPV6HDR;
2821
2822         if (hw->mac.type == ixgbe_mac_82598EB)
2823                 return;
2824
2825         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
2826                 psrtype |= (adapter->num_rx_queues_per_pool << 29);
2827
2828         for (p = 0; p < adapter->num_rx_pools; p++)
2829                 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
2830                                 psrtype);
2831 }
2832
2833 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
2834 {
2835         struct ixgbe_hw *hw = &adapter->hw;
2836         u32 gcr_ext;
2837         u32 vt_reg_bits;
2838         u32 reg_offset, vf_shift;
2839         u32 vmdctl;
2840
2841         if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2842                 return;
2843
2844         vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
2845         vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN | IXGBE_VT_CTL_REPLEN;
2846         vt_reg_bits |= (adapter->num_vfs << IXGBE_VT_CTL_POOL_SHIFT);
2847         IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
2848
2849         vf_shift = adapter->num_vfs % 32;
2850         reg_offset = (adapter->num_vfs > 32) ? 1 : 0;
2851
2852         /* Enable only the PF's pool for Tx/Rx */
2853         IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
2854         IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), 0);
2855         IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
2856         IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), 0);
2857         IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2858
2859         /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
2860         hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
2861
2862         /*
2863          * Set up VF register offsets for selected VT Mode,
2864          * i.e. 32 or 64 VFs for SR-IOV
2865          */
2866         gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
2867         gcr_ext |= IXGBE_GCR_EXT_MSIX_EN;
2868         gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
2869         IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
2870
2871         /* enable Tx loopback for VF/PF communication */
2872         IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2873 }
2874
2875 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
2876 {
2877         struct ixgbe_hw *hw = &adapter->hw;
2878         struct net_device *netdev = adapter->netdev;
2879         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2880         int rx_buf_len;
2881         struct ixgbe_ring *rx_ring;
2882         int i;
2883         u32 mhadd, hlreg0;
2884
2885         /* Decide whether to use packet split mode or not */
2886         /* Do not use packet split if we're in SR-IOV Mode */
2887         if (!adapter->num_vfs)
2888                 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
2889
2890         /* Set the RX buffer length according to the mode */
2891         if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2892                 rx_buf_len = IXGBE_RX_HDR_SIZE;
2893         } else {
2894                 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
2895                     (netdev->mtu <= ETH_DATA_LEN))
2896                         rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2897                 else
2898                         rx_buf_len = ALIGN(max_frame + VLAN_HLEN, 1024);
2899         }
2900
2901 #ifdef IXGBE_FCOE
2902         /* adjust max frame to be able to do baby jumbo for FCoE */
2903         if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
2904             (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
2905                 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
2906
2907 #endif /* IXGBE_FCOE */
2908         mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
2909         if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
2910                 mhadd &= ~IXGBE_MHADD_MFS_MASK;
2911                 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
2912
2913                 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
2914         }
2915
2916         hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2917         /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
2918         hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2919         IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
2920
2921         /*
2922          * Setup the HW Rx Head and Tail Descriptor Pointers and
2923          * the Base and Length of the Rx Descriptor Ring
2924          */
2925         for (i = 0; i < adapter->num_rx_queues; i++) {
2926                 rx_ring = adapter->rx_ring[i];
2927                 rx_ring->rx_buf_len = rx_buf_len;
2928
2929                 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
2930                         rx_ring->flags |= IXGBE_RING_RX_PS_ENABLED;
2931                 else
2932                         rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2933
2934 #ifdef IXGBE_FCOE
2935                 if (netdev->features & NETIF_F_FCOE_MTU) {
2936                         struct ixgbe_ring_feature *f;
2937                         f = &adapter->ring_feature[RING_F_FCOE];
2938                         if ((i >= f->mask) && (i < f->mask + f->indices)) {
2939                                 rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2940                                 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
2941                                         rx_ring->rx_buf_len =
2942                                                 IXGBE_FCOE_JUMBO_FRAME_SIZE;
2943                         }
2944                 }
2945 #endif /* IXGBE_FCOE */
2946         }
2947
2948 }
2949
2950 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
2951 {
2952         struct ixgbe_hw *hw = &adapter->hw;
2953         u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2954
2955         switch (hw->mac.type) {
2956         case ixgbe_mac_82598EB:
2957                 /*
2958                  * For VMDq support of different descriptor types or
2959                  * buffer sizes through the use of multiple SRRCTL
2960                  * registers, RDRXCTL.MVMEN must be set to 1
2961                  *
2962                  * also, the manual doesn't mention it clearly but DCA hints
2963                  * will only use queue 0's tags unless this bit is set.  Side
2964                  * effects of setting this bit are only that SRRCTL must be
2965                  * fully programmed [0..15]
2966                  */
2967                 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
2968                 break;
2969         case ixgbe_mac_82599EB:
2970                 /* Disable RSC for ACK packets */
2971                 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
2972                    (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
2973                 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
2974                 /* hardware requires some bits to be set by default */
2975                 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
2976                 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
2977                 break;
2978         default:
2979                 /* We should do nothing since we don't know this hardware */
2980                 return;
2981         }
2982
2983         IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2984 }
2985
2986 /**
2987  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
2988  * @adapter: board private structure
2989  *
2990  * Configure the Rx unit of the MAC after a reset.
2991  **/
2992 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
2993 {
2994         struct ixgbe_hw *hw = &adapter->hw;
2995         int i;
2996         u32 rxctrl;
2997
2998         /* disable receives while setting up the descriptors */
2999         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3000         IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3001
3002         ixgbe_setup_psrtype(adapter);
3003         ixgbe_setup_rdrxctl(adapter);
3004
3005         /* Program registers for the distribution of queues */
3006         ixgbe_setup_mrqc(adapter);
3007
3008         ixgbe_set_uta(adapter);
3009
3010         /* set_rx_buffer_len must be called before ring initialization */
3011         ixgbe_set_rx_buffer_len(adapter);
3012
3013         /*
3014          * Setup the HW Rx Head and Tail Descriptor Pointers and
3015          * the Base and Length of the Rx Descriptor Ring
3016          */
3017         for (i = 0; i < adapter->num_rx_queues; i++)
3018                 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3019
3020         /* disable drop enable for 82598 parts */
3021         if (hw->mac.type == ixgbe_mac_82598EB)
3022                 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3023
3024         /* enable all receives */
3025         rxctrl |= IXGBE_RXCTRL_RXEN;
3026         hw->mac.ops.enable_rx_dma(hw, rxctrl);
3027 }
3028
3029 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3030 {
3031         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3032         struct ixgbe_hw *hw = &adapter->hw;
3033         int pool_ndx = adapter->num_vfs;
3034
3035         /* add VID to filter table */
3036         hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3037 }
3038
3039 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3040 {
3041         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3042         struct ixgbe_hw *hw = &adapter->hw;
3043         int pool_ndx = adapter->num_vfs;
3044
3045         if (!test_bit(__IXGBE_DOWN, &adapter->state))
3046                 ixgbe_irq_disable(adapter);
3047
3048         vlan_group_set_device(adapter->vlgrp, vid, NULL);
3049
3050         if (!test_bit(__IXGBE_DOWN, &adapter->state))
3051                 ixgbe_irq_enable(adapter);
3052
3053         /* remove VID from filter table */
3054         hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3055 }
3056
3057 /**
3058  * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3059  * @adapter: driver data
3060  */
3061 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3062 {
3063         struct ixgbe_hw *hw = &adapter->hw;
3064         u32 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3065         int i, j;
3066
3067         switch (hw->mac.type) {
3068         case ixgbe_mac_82598EB:
3069                 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
3070 #ifdef CONFIG_IXGBE_DCB
3071                 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
3072                         vlnctrl &= ~IXGBE_VLNCTRL_VME;
3073 #endif
3074                 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3075                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3076                 break;
3077         case ixgbe_mac_82599EB:
3078                 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
3079                 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3080                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3081 #ifdef CONFIG_IXGBE_DCB
3082                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
3083                         break;
3084 #endif
3085                 for (i = 0; i < adapter->num_rx_queues; i++) {
3086                         j = adapter->rx_ring[i]->reg_idx;
3087                         vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3088                         vlnctrl &= ~IXGBE_RXDCTL_VME;
3089                         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3090                 }
3091                 break;
3092         default:
3093                 break;
3094         }
3095 }
3096
3097 /**
3098  * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3099  * @adapter: driver data
3100  */
3101 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3102 {
3103         struct ixgbe_hw *hw = &adapter->hw;
3104         u32 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3105         int i, j;
3106
3107         switch (hw->mac.type) {
3108         case ixgbe_mac_82598EB:
3109                 vlnctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
3110                 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3111                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3112                 break;
3113         case ixgbe_mac_82599EB:
3114                 vlnctrl |= IXGBE_VLNCTRL_VFE;
3115                 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3116                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3117                 for (i = 0; i < adapter->num_rx_queues; i++) {
3118                         j = adapter->rx_ring[i]->reg_idx;
3119                         vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3120                         vlnctrl |= IXGBE_RXDCTL_VME;
3121                         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3122                 }
3123                 break;
3124         default:
3125                 break;
3126         }
3127 }
3128
3129 static void ixgbe_vlan_rx_register(struct net_device *netdev,
3130                                    struct vlan_group *grp)
3131 {
3132         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3133
3134         if (!test_bit(__IXGBE_DOWN, &adapter->state))
3135                 ixgbe_irq_disable(adapter);
3136         adapter->vlgrp = grp;
3137
3138         /*
3139          * For a DCB driver, always enable VLAN tag stripping so we can
3140          * still receive traffic from a DCB-enabled host even if we're
3141          * not in DCB mode.
3142          */
3143         ixgbe_vlan_filter_enable(adapter);
3144
3145         ixgbe_vlan_rx_add_vid(netdev, 0);
3146
3147         if (!test_bit(__IXGBE_DOWN, &adapter->state))
3148                 ixgbe_irq_enable(adapter);
3149 }
3150
3151 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3152 {
3153         ixgbe_vlan_rx_register(adapter->netdev, adapter->vlgrp);
3154
3155         if (adapter->vlgrp) {
3156                 u16 vid;
3157                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
3158                         if (!vlan_group_get_device(adapter->vlgrp, vid))
3159                                 continue;
3160                         ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3161                 }
3162         }
3163 }
3164
3165 /**
3166  * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3167  * @netdev: network interface device structure
3168  *
3169  * Writes unicast address list to the RAR table.
3170  * Returns: -ENOMEM on failure/insufficient address space
3171  *                0 on no addresses written
3172  *                X on writing X addresses to the RAR table
3173  **/
3174 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3175 {
3176         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3177         struct ixgbe_hw *hw = &adapter->hw;
3178         unsigned int vfn = adapter->num_vfs;
3179         unsigned int rar_entries = hw->mac.num_rar_entries - (vfn + 1);
3180         int count = 0;
3181
3182         /* return ENOMEM indicating insufficient memory for addresses */
3183         if (netdev_uc_count(netdev) > rar_entries)
3184                 return -ENOMEM;
3185
3186         if (!netdev_uc_empty(netdev) && rar_entries) {
3187                 struct netdev_hw_addr *ha;
3188                 /* return error if we do not support writing to RAR table */
3189                 if (!hw->mac.ops.set_rar)
3190                         return -ENOMEM;
3191
3192                 netdev_for_each_uc_addr(ha, netdev) {
3193                         if (!rar_entries)
3194                                 break;
3195                         hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3196                                             vfn, IXGBE_RAH_AV);
3197                         count++;
3198                 }
3199         }
3200         /* write the addresses in reverse order to avoid write combining */
3201         for (; rar_entries > 0 ; rar_entries--)
3202                 hw->mac.ops.clear_rar(hw, rar_entries);
3203
3204         return count;
3205 }
3206
3207 /**
3208  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3209  * @netdev: network interface device structure
3210  *
3211  * The set_rx_method entry point is called whenever the unicast/multicast
3212  * address list or the network interface flags are updated.  This routine is
3213  * responsible for configuring the hardware for proper unicast, multicast and
3214  * promiscuous mode.
3215  **/
3216 void ixgbe_set_rx_mode(struct net_device *netdev)
3217 {
3218         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3219         struct ixgbe_hw *hw = &adapter->hw;
3220         u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3221         int count;
3222
3223         /* Check for Promiscuous and All Multicast modes */
3224
3225         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3226
3227         /* set all bits that we expect to always be set */
3228         fctrl |= IXGBE_FCTRL_BAM;
3229         fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3230         fctrl |= IXGBE_FCTRL_PMCF;
3231
3232         /* clear the bits we are changing the status of */
3233         fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3234
3235         if (netdev->flags & IFF_PROMISC) {
3236                 hw->addr_ctrl.user_set_promisc = true;
3237                 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3238                 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3239                 /* don't hardware filter vlans in promisc mode */
3240                 ixgbe_vlan_filter_disable(adapter);
3241         } else {
3242                 if (netdev->flags & IFF_ALLMULTI) {
3243                         fctrl |= IXGBE_FCTRL_MPE;
3244                         vmolr |= IXGBE_VMOLR_MPE;
3245                 } else {
3246                         /*
3247                          * Write addresses to the MTA, if the attempt fails
3248                          * then we should just turn on promiscous mode so
3249                          * that we can at least receive multicast traffic
3250                          */
3251                         hw->mac.ops.update_mc_addr_list(hw, netdev);
3252                         vmolr |= IXGBE_VMOLR_ROMPE;
3253                 }
3254                 ixgbe_vlan_filter_enable(adapter);
3255                 hw->addr_ctrl.user_set_promisc = false;
3256                 /*
3257                  * Write addresses to available RAR registers, if there is not
3258                  * sufficient space to store all the addresses then enable
3259                  * unicast promiscous mode
3260                  */
3261                 count = ixgbe_write_uc_addr_list(netdev);
3262                 if (count < 0) {
3263                         fctrl |= IXGBE_FCTRL_UPE;
3264                         vmolr |= IXGBE_VMOLR_ROPE;
3265                 }
3266         }
3267
3268         if (adapter->num_vfs) {
3269                 ixgbe_restore_vf_multicasts(adapter);
3270                 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3271                          ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3272                            IXGBE_VMOLR_ROPE);
3273                 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3274         }
3275
3276         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3277 }
3278
3279 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3280 {
3281         int q_idx;
3282         struct ixgbe_q_vector *q_vector;
3283         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3284
3285         /* legacy and MSI only use one vector */
3286         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3287                 q_vectors = 1;
3288
3289         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3290                 struct napi_struct *napi;
3291                 q_vector = adapter->q_vector[q_idx];
3292                 napi = &q_vector->napi;
3293                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3294                         if (!q_vector->rxr_count || !q_vector->txr_count) {
3295                                 if (q_vector->txr_count == 1)
3296                                         napi->poll = &ixgbe_clean_txonly;
3297                                 else if (q_vector->rxr_count == 1)
3298                                         napi->poll = &ixgbe_clean_rxonly;
3299                         }
3300                 }
3301
3302                 napi_enable(napi);
3303         }
3304 }
3305
3306 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3307 {
3308         int q_idx;
3309         struct ixgbe_q_vector *q_vector;
3310         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3311
3312         /* legacy and MSI only use one vector */
3313         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3314                 q_vectors = 1;
3315
3316         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3317                 q_vector = adapter->q_vector[q_idx];
3318                 napi_disable(&q_vector->napi);
3319         }
3320 }
3321
3322 #ifdef CONFIG_IXGBE_DCB
3323 /*
3324  * ixgbe_configure_dcb - Configure DCB hardware
3325  * @adapter: ixgbe adapter struct
3326  *
3327  * This is called by the driver on open to configure the DCB hardware.
3328  * This is also called by the gennetlink interface when reconfiguring
3329  * the DCB state.
3330  */
3331 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3332 {
3333         struct ixgbe_hw *hw = &adapter->hw;
3334         u32 txdctl;
3335         int i, j;
3336
3337         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3338                 if (hw->mac.type == ixgbe_mac_82598EB)
3339                         netif_set_gso_max_size(adapter->netdev, 65536);
3340                 return;
3341         }
3342
3343         if (hw->mac.type == ixgbe_mac_82598EB)
3344                 netif_set_gso_max_size(adapter->netdev, 32768);
3345
3346         ixgbe_dcb_check_config(&adapter->dcb_cfg);
3347         ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_TX_CONFIG);
3348         ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_RX_CONFIG);
3349
3350         /* reconfigure the hardware */
3351         ixgbe_dcb_hw_config(&adapter->hw, &adapter->dcb_cfg);
3352
3353         for (i = 0; i < adapter->num_tx_queues; i++) {
3354                 j = adapter->tx_ring[i]->reg_idx;
3355                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3356                 /* PThresh workaround for Tx hang with DFP enabled. */
3357                 txdctl |= 32;
3358                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
3359         }
3360         /* Enable VLAN tag insert/strip */
3361         ixgbe_vlan_filter_enable(adapter);
3362
3363         hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3364 }
3365
3366 #endif
3367 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3368 {
3369         struct net_device *netdev = adapter->netdev;
3370         struct ixgbe_hw *hw = &adapter->hw;
3371         int i;
3372
3373         ixgbe_set_rx_mode(netdev);
3374
3375         ixgbe_restore_vlan(adapter);
3376 #ifdef CONFIG_IXGBE_DCB
3377         ixgbe_configure_dcb(adapter);
3378 #endif
3379
3380 #ifdef IXGBE_FCOE
3381         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3382                 ixgbe_configure_fcoe(adapter);
3383
3384 #endif /* IXGBE_FCOE */
3385         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3386                 for (i = 0; i < adapter->num_tx_queues; i++)
3387                         adapter->tx_ring[i]->atr_sample_rate =
3388                                                        adapter->atr_sample_rate;
3389                 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
3390         } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3391                 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
3392         }
3393         ixgbe_configure_virtualization(adapter);
3394
3395         ixgbe_configure_tx(adapter);
3396         ixgbe_configure_rx(adapter);
3397 }
3398
3399 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3400 {
3401         switch (hw->phy.type) {
3402         case ixgbe_phy_sfp_avago:
3403         case ixgbe_phy_sfp_ftl:
3404         case ixgbe_phy_sfp_intel:
3405         case ixgbe_phy_sfp_unknown:
3406         case ixgbe_phy_sfp_passive_tyco:
3407         case ixgbe_phy_sfp_passive_unknown:
3408         case ixgbe_phy_sfp_active_unknown:
3409         case ixgbe_phy_sfp_ftl_active:
3410                 return true;
3411         default:
3412                 return false;
3413         }
3414 }
3415
3416 /**
3417  * ixgbe_sfp_link_config - set up SFP+ link
3418  * @adapter: pointer to private adapter struct
3419  **/
3420 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3421 {
3422         struct ixgbe_hw *hw = &adapter->hw;
3423
3424                 if (hw->phy.multispeed_fiber) {
3425                         /*
3426                          * In multispeed fiber setups, the device may not have
3427                          * had a physical connection when the driver loaded.
3428                          * If that's the case, the initial link configuration
3429                          * couldn't get the MAC into 10G or 1G mode, so we'll
3430                          * never have a link status change interrupt fire.
3431                          * We need to try and force an autonegotiation
3432                          * session, then bring up link.
3433                          */
3434                         hw->mac.ops.setup_sfp(hw);
3435                         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
3436                                 schedule_work(&adapter->multispeed_fiber_task);
3437                 } else {
3438                         /*
3439                          * Direct Attach Cu and non-multispeed fiber modules
3440                          * still need to be configured properly prior to
3441                          * attempting link.
3442                          */
3443                         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
3444                                 schedule_work(&adapter->sfp_config_module_task);
3445                 }
3446 }
3447
3448 /**
3449  * ixgbe_non_sfp_link_config - set up non-SFP+ link
3450  * @hw: pointer to private hardware struct
3451  *
3452  * Returns 0 on success, negative on failure
3453  **/
3454 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3455 {
3456         u32 autoneg;
3457         bool negotiation, link_up = false;
3458         u32 ret = IXGBE_ERR_LINK_SETUP;
3459
3460         if (hw->mac.ops.check_link)
3461                 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3462
3463         if (ret)
3464                 goto link_cfg_out;
3465
3466         if (hw->mac.ops.get_link_capabilities)
3467                 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3468                                                         &negotiation);
3469         if (ret)
3470                 goto link_cfg_out;
3471
3472         if (hw->mac.ops.setup_link)
3473                 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3474 link_cfg_out:
3475         return ret;
3476 }
3477
3478 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3479 {
3480         struct ixgbe_hw *hw = &adapter->hw;
3481         u32 gpie = 0;
3482
3483         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3484                 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3485                        IXGBE_GPIE_OCD;
3486                 gpie |= IXGBE_GPIE_EIAME;
3487                 /*
3488                  * use EIAM to auto-mask when MSI-X interrupt is asserted
3489                  * this saves a register write for every interrupt
3490                  */
3491                 switch (hw->mac.type) {
3492                 case ixgbe_mac_82598EB:
3493                         IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3494                         break;
3495                 default:
3496                 case ixgbe_mac_82599EB:
3497                         IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3498                         IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3499                         break;
3500                 }
3501         } else {
3502                 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3503                  * specifically only auto mask tx and rx interrupts */
3504                 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3505         }
3506
3507         /* XXX: to interrupt immediately for EICS writes, enable this */
3508         /* gpie |= IXGBE_GPIE_EIMEN; */
3509
3510         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3511                 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3512                 gpie |= IXGBE_GPIE_VTMODE_64;
3513         }
3514
3515         /* Enable fan failure interrupt */
3516         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3517                 gpie |= IXGBE_SDP1_GPIEN;
3518
3519         if (hw->mac.type == ixgbe_mac_82599EB)
3520                 gpie |= IXGBE_SDP1_GPIEN;
3521                 gpie |= IXGBE_SDP2_GPIEN;
3522
3523         IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3524 }
3525
3526 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
3527 {
3528         struct ixgbe_hw *hw = &adapter->hw;
3529         int err;
3530         u32 ctrl_ext;
3531
3532         ixgbe_get_hw_control(adapter);
3533         ixgbe_setup_gpie(adapter);
3534
3535         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3536                 ixgbe_configure_msix(adapter);
3537         else
3538                 ixgbe_configure_msi_and_legacy(adapter);
3539
3540         /* enable the optics */
3541         if (hw->phy.multispeed_fiber)
3542                 hw->mac.ops.enable_tx_laser(hw);
3543
3544         clear_bit(__IXGBE_DOWN, &adapter->state);
3545         ixgbe_napi_enable_all(adapter);
3546
3547         /* clear any pending interrupts, may auto mask */
3548         IXGBE_READ_REG(hw, IXGBE_EICR);
3549         ixgbe_irq_enable(adapter);
3550
3551         /*
3552          * If this adapter has a fan, check to see if we had a failure
3553          * before we enabled the interrupt.
3554          */
3555         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3556                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3557                 if (esdp & IXGBE_ESDP_SDP1)
3558                         e_crit(drv, "Fan has stopped, replace the adapter\n");
3559         }
3560
3561         /*
3562          * For hot-pluggable SFP+ devices, a new SFP+ module may have
3563          * arrived before interrupts were enabled but after probe.  Such
3564          * devices wouldn't have their type identified yet. We need to
3565          * kick off the SFP+ module setup first, then try to bring up link.
3566          * If we're not hot-pluggable SFP+, we just need to configure link
3567          * and bring it up.
3568          */
3569         if (hw->phy.type == ixgbe_phy_unknown) {
3570                 err = hw->phy.ops.identify(hw);
3571                 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
3572                         /*
3573                          * Take the device down and schedule the sfp tasklet
3574                          * which will unregister_netdev and log it.
3575                          */
3576                         ixgbe_down(adapter);
3577                         schedule_work(&adapter->sfp_config_module_task);
3578                         return err;
3579                 }
3580         }
3581
3582         if (ixgbe_is_sfp(hw)) {
3583                 ixgbe_sfp_link_config(adapter);
3584         } else {
3585                 err = ixgbe_non_sfp_link_config(hw);
3586                 if (err)
3587                         e_err(probe, "link_config FAILED %d\n", err);
3588         }
3589
3590         /* enable transmits */
3591         netif_tx_start_all_queues(adapter->netdev);
3592
3593         /* bring the link up in the watchdog, this could race with our first
3594          * link up interrupt but shouldn't be a problem */
3595         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3596         adapter->link_check_timeout = jiffies;
3597         mod_timer(&adapter->watchdog_timer, jiffies);
3598
3599         /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3600         ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3601         ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3602         IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3603
3604         return 0;
3605 }
3606
3607 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3608 {
3609         WARN_ON(in_interrupt());
3610         while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3611                 msleep(1);
3612         ixgbe_down(adapter);
3613         /*
3614          * If SR-IOV enabled then wait a bit before bringing the adapter
3615          * back up to give the VFs time to respond to the reset.  The
3616          * two second wait is based upon the watchdog timer cycle in
3617          * the VF driver.
3618          */
3619         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3620                 msleep(2000);
3621         ixgbe_up(adapter);
3622         clear_bit(__IXGBE_RESETTING, &adapter->state);
3623 }
3624
3625 int ixgbe_up(struct ixgbe_adapter *adapter)
3626 {
3627         /* hardware has been reset, we need to reload some things */
3628         ixgbe_configure(adapter);
3629
3630         return ixgbe_up_complete(adapter);
3631 }
3632
3633 void ixgbe_reset(struct ixgbe_adapter *adapter)
3634 {
3635         struct ixgbe_hw *hw = &adapter->hw;
3636         int err;
3637
3638         err = hw->mac.ops.init_hw(hw);
3639         switch (err) {
3640         case 0:
3641         case IXGBE_ERR_SFP_NOT_PRESENT:
3642                 break;
3643         case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3644                 e_dev_err("master disable timed out\n");
3645                 break;
3646         case IXGBE_ERR_EEPROM_VERSION:
3647                 /* We are running on a pre-production device, log a warning */
3648                 e_dev_warn("This device is a pre-production adapter/LOM. "
3649                            "Please be aware there may be issuesassociated with "
3650                            "your hardware.  If you are experiencing problems "
3651                            "please contact your Intel or hardware "
3652                            "representative who provided you with this "
3653                            "hardware.\n");
3654                 break;
3655         default:
3656                 e_dev_err("Hardware Error: %d\n", err);
3657         }
3658
3659         /* reprogram the RAR[0] in case user changed it. */
3660         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3661                             IXGBE_RAH_AV);
3662 }
3663
3664 /**
3665  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3666  * @adapter: board private structure
3667  * @rx_ring: ring to free buffers from
3668  **/
3669 static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
3670                                 struct ixgbe_ring *rx_ring)
3671 {
3672         struct pci_dev *pdev = adapter->pdev;
3673         unsigned long size;
3674         unsigned int i;
3675
3676         /* ring already cleared, nothing to do */
3677         if (!rx_ring->rx_buffer_info)
3678                 return;
3679
3680         /* Free all the Rx ring sk_buffs */
3681         for (i = 0; i < rx_ring->count; i++) {
3682                 struct ixgbe_rx_buffer *rx_buffer_info;
3683
3684                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
3685                 if (rx_buffer_info->dma) {
3686                         dma_unmap_single(&pdev->dev, rx_buffer_info->dma,
3687                                          rx_ring->rx_buf_len,
3688                                          DMA_FROM_DEVICE);
3689                         rx_buffer_info->dma = 0;
3690                 }
3691                 if (rx_buffer_info->skb) {
3692                         struct sk_buff *skb = rx_buffer_info->skb;
3693                         rx_buffer_info->skb = NULL;
3694                         do {
3695                                 struct sk_buff *this = skb;
3696                                 if (IXGBE_RSC_CB(this)->delay_unmap) {
3697                                         dma_unmap_single(&pdev->dev,
3698                                                          IXGBE_RSC_CB(this)->dma,
3699                                                          rx_ring->rx_buf_len,
3700                                                          DMA_FROM_DEVICE);
3701                                         IXGBE_RSC_CB(this)->dma = 0;
3702                                         IXGBE_RSC_CB(skb)->delay_unmap = false;
3703                                 }
3704                                 skb = skb->prev;
3705                                 dev_kfree_skb(this);
3706                         } while (skb);
3707                 }
3708                 if (!rx_buffer_info->page)
3709                         continue;
3710                 if (rx_buffer_info->page_dma) {
3711                         dma_unmap_page(&pdev->dev, rx_buffer_info->page_dma,
3712                                        PAGE_SIZE / 2, DMA_FROM_DEVICE);
3713                         rx_buffer_info->page_dma = 0;
3714                 }
3715                 put_page(rx_buffer_info->page);
3716                 rx_buffer_info->page = NULL;
3717                 rx_buffer_info->page_offset = 0;
3718         }
3719
3720         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3721         memset(rx_ring->rx_buffer_info, 0, size);
3722
3723         /* Zero out the descriptor ring */
3724         memset(rx_ring->desc, 0, rx_ring->size);
3725
3726         rx_ring->next_to_clean = 0;
3727         rx_ring->next_to_use = 0;
3728
3729         if (rx_ring->head)
3730                 writel(0, adapter->hw.hw_addr + rx_ring->head);
3731         if (rx_ring->tail)
3732                 writel(0, adapter->hw.hw_addr + rx_ring->tail);
3733 }
3734
3735 /**
3736  * ixgbe_clean_tx_ring - Free Tx Buffers
3737  * @adapter: board private structure
3738  * @tx_ring: ring to be cleaned
3739  **/
3740 static void ixgbe_clean_tx_ring(struct ixgbe_adapter *adapter,
3741                                 struct ixgbe_ring *tx_ring)
3742 {
3743         struct ixgbe_tx_buffer *tx_buffer_info;
3744         unsigned long size;
3745         unsigned int i;
3746
3747         /* ring already cleared, nothing to do */
3748         if (!tx_ring->tx_buffer_info)
3749                 return;
3750
3751         /* Free all the Tx ring sk_buffs */
3752         for (i = 0; i < tx_ring->count; i++) {
3753                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3754                 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
3755         }
3756
3757         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3758         memset(tx_ring->tx_buffer_info, 0, size);
3759
3760         /* Zero out the descriptor ring */
3761         memset(tx_ring->desc, 0, tx_ring->size);
3762
3763         tx_ring->next_to_use = 0;
3764         tx_ring->next_to_clean = 0;
3765
3766         if (tx_ring->head)
3767                 writel(0, adapter->hw.hw_addr + tx_ring->head);
3768         if (tx_ring->tail)
3769                 writel(0, adapter->hw.hw_addr + tx_ring->tail);
3770 }
3771
3772 /**
3773  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
3774  * @adapter: board private structure
3775  **/
3776 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
3777 {
3778         int i;
3779
3780         for (i = 0; i < adapter->num_rx_queues; i++)
3781                 ixgbe_clean_rx_ring(adapter, adapter->rx_ring[i]);
3782 }
3783
3784 /**
3785  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
3786  * @adapter: board private structure
3787  **/
3788 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
3789 {
3790         int i;
3791
3792         for (i = 0; i < adapter->num_tx_queues; i++)
3793                 ixgbe_clean_tx_ring(adapter, adapter->tx_ring[i]);
3794 }
3795
3796 void ixgbe_down(struct ixgbe_adapter *adapter)
3797 {
3798         struct net_device *netdev = adapter->netdev;
3799         struct ixgbe_hw *hw = &adapter->hw;
3800         u32 rxctrl;
3801         u32 txdctl;
3802         int i, j;
3803
3804         /* signal that we are down to the interrupt handler */
3805         set_bit(__IXGBE_DOWN, &adapter->state);
3806
3807         /* disable receive for all VFs and wait one second */
3808         if (adapter->num_vfs) {
3809                 /* ping all the active vfs to let them know we are going down */
3810                 ixgbe_ping_all_vfs(adapter);
3811
3812                 /* Disable all VFTE/VFRE TX/RX */
3813                 ixgbe_disable_tx_rx(adapter);
3814
3815                 /* Mark all the VFs as inactive */
3816                 for (i = 0 ; i < adapter->num_vfs; i++)
3817                         adapter->vfinfo[i].clear_to_send = 0;
3818         }
3819
3820         /* disable receives */
3821         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3822         IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3823
3824         IXGBE_WRITE_FLUSH(hw);
3825         msleep(10);
3826
3827         netif_tx_stop_all_queues(netdev);
3828
3829         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3830         del_timer_sync(&adapter->sfp_timer);
3831         del_timer_sync(&adapter->watchdog_timer);
3832         cancel_work_sync(&adapter->watchdog_task);
3833
3834         netif_carrier_off(netdev);
3835         netif_tx_disable(netdev);
3836
3837         ixgbe_irq_disable(adapter);
3838
3839         ixgbe_napi_disable_all(adapter);
3840
3841         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3842             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3843                 cancel_work_sync(&adapter->fdir_reinit_task);
3844
3845         if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
3846                 cancel_work_sync(&adapter->check_overtemp_task);
3847
3848         /* disable transmits in the hardware now that interrupts are off */
3849         for (i = 0; i < adapter->num_tx_queues; i++) {
3850                 j = adapter->tx_ring[i]->reg_idx;
3851                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3852                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
3853                                 (txdctl & ~IXGBE_TXDCTL_ENABLE));
3854         }
3855         /* Disable the Tx DMA engine on 82599 */
3856         if (hw->mac.type == ixgbe_mac_82599EB)
3857                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
3858                                 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
3859                                  ~IXGBE_DMATXCTL_TE));
3860
3861         /* power down the optics */
3862         if (hw->phy.multispeed_fiber)
3863                 hw->mac.ops.disable_tx_laser(hw);
3864
3865         /* clear n-tuple filters that are cached */
3866         ethtool_ntuple_flush(netdev);
3867
3868         if (!pci_channel_offline(adapter->pdev))
3869                 ixgbe_reset(adapter);
3870         ixgbe_clean_all_tx_rings(adapter);
3871         ixgbe_clean_all_rx_rings(adapter);
3872
3873 #ifdef CONFIG_IXGBE_DCA
3874         /* since we reset the hardware DCA settings were cleared */
3875         ixgbe_setup_dca(adapter);
3876 #endif
3877 }
3878
3879 /**
3880  * ixgbe_poll - NAPI Rx polling callback
3881  * @napi: structure for representing this polling device
3882  * @budget: how many packets driver is allowed to clean
3883  *
3884  * This function is used for legacy and MSI, NAPI mode
3885  **/
3886 static int ixgbe_poll(struct napi_struct *napi, int budget)
3887 {
3888         struct ixgbe_q_vector *q_vector =
3889                                 container_of(napi, struct ixgbe_q_vector, napi);
3890         struct ixgbe_adapter *adapter = q_vector->adapter;
3891         int tx_clean_complete, work_done = 0;
3892
3893 #ifdef CONFIG_IXGBE_DCA
3894         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
3895                 ixgbe_update_tx_dca(adapter, adapter->tx_ring[0]);
3896                 ixgbe_update_rx_dca(adapter, adapter->rx_ring[0]);
3897         }
3898 #endif
3899
3900         tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring[0]);
3901         ixgbe_clean_rx_irq(q_vector, adapter->rx_ring[0], &work_done, budget);
3902
3903         if (!tx_clean_complete)
3904                 work_done = budget;
3905
3906         /* If budget not fully consumed, exit the polling mode */
3907         if (work_done < budget) {
3908                 napi_complete(napi);
3909                 if (adapter->rx_itr_setting & 1)
3910                         ixgbe_set_itr(adapter);
3911                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3912                         ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
3913         }
3914         return work_done;
3915 }
3916
3917 /**
3918  * ixgbe_tx_timeout - Respond to a Tx Hang
3919  * @netdev: network interface device structure
3920  **/
3921 static void ixgbe_tx_timeout(struct net_device *netdev)
3922 {
3923         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3924
3925         /* Do the reset outside of interrupt context */
3926         schedule_work(&adapter->reset_task);
3927 }
3928
3929 static void ixgbe_reset_task(struct work_struct *work)
3930 {
3931         struct ixgbe_adapter *adapter;
3932         adapter = container_of(work, struct ixgbe_adapter, reset_task);
3933
3934         /* If we're already down or resetting, just bail */
3935         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
3936             test_bit(__IXGBE_RESETTING, &adapter->state))
3937                 return;
3938
3939         adapter->tx_timeout_count++;
3940
3941         ixgbe_dump(adapter);
3942         netdev_err(adapter->netdev, "Reset adapter\n");
3943         ixgbe_reinit_locked(adapter);
3944 }
3945
3946 #ifdef CONFIG_IXGBE_DCB
3947 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
3948 {
3949         bool ret = false;
3950         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
3951
3952         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
3953                 return ret;
3954
3955         f->mask = 0x7 << 3;
3956         adapter->num_rx_queues = f->indices;
3957         adapter->num_tx_queues = f->indices;
3958         ret = true;
3959
3960         return ret;
3961 }
3962 #endif
3963
3964 /**
3965  * ixgbe_set_rss_queues: Allocate queues for RSS
3966  * @adapter: board private structure to initialize
3967  *
3968  * This is our "base" multiqueue mode.  RSS (Receive Side Scaling) will try
3969  * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
3970  *
3971  **/
3972 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
3973 {
3974         bool ret = false;
3975         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
3976
3977         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3978                 f->mask = 0xF;
3979                 adapter->num_rx_queues = f->indices;
3980                 adapter->num_tx_queues = f->indices;
3981                 ret = true;
3982         } else {
3983                 ret = false;
3984         }
3985
3986         return ret;
3987 }
3988
3989 /**
3990  * ixgbe_set_fdir_queues: Allocate queues for Flow Director
3991  * @adapter: board private structure to initialize
3992  *
3993  * Flow Director is an advanced Rx filter, attempting to get Rx flows back
3994  * to the original CPU that initiated the Tx session.  This runs in addition
3995  * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
3996  * Rx load across CPUs using RSS.
3997  *
3998  **/
3999 static inline bool ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
4000 {
4001         bool ret = false;
4002         struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
4003
4004         f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
4005         f_fdir->mask = 0;
4006
4007         /* Flow Director must have RSS enabled */
4008         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4009             ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4010              (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
4011                 adapter->num_tx_queues = f_fdir->indices;
4012                 adapter->num_rx_queues = f_fdir->indices;
4013                 ret = true;
4014         } else {
4015                 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4016                 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4017         }
4018         return ret;
4019 }
4020
4021 #ifdef IXGBE_FCOE
4022 /**
4023  * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
4024  * @adapter: board private structure to initialize
4025  *
4026  * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
4027  * The ring feature mask is not used as a mask for FCoE, as it can take any 8
4028  * rx queues out of the max number of rx queues, instead, it is used as the
4029  * index of the first rx queue used by FCoE.
4030  *
4031  **/
4032 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
4033 {
4034         bool ret = false;
4035         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4036
4037         f->indices = min((int)num_online_cpus(), f->indices);
4038         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4039                 adapter->num_rx_queues = 1;
4040                 adapter->num_tx_queues = 1;
4041 #ifdef CONFIG_IXGBE_DCB
4042                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4043                         e_info(probe, "FCoE enabled with DCB\n");
4044                         ixgbe_set_dcb_queues(adapter);
4045                 }
4046 #endif
4047                 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4048                         e_info(probe, "FCoE enabled with RSS\n");
4049                         if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4050                             (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4051                                 ixgbe_set_fdir_queues(adapter);
4052                         else
4053                                 ixgbe_set_rss_queues(adapter);
4054                 }
4055                 /* adding FCoE rx rings to the end */
4056                 f->mask = adapter->num_rx_queues;
4057                 adapter->num_rx_queues += f->indices;
4058                 adapter->num_tx_queues += f->indices;
4059
4060                 ret = true;
4061         }
4062
4063         return ret;
4064 }
4065
4066 #endif /* IXGBE_FCOE */
4067 /**
4068  * ixgbe_set_sriov_queues: Allocate queues for IOV use
4069  * @adapter: board private structure to initialize
4070  *
4071  * IOV doesn't actually use anything, so just NAK the
4072  * request for now and let the other queue routines
4073  * figure out what to do.
4074  */
4075 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
4076 {
4077         return false;
4078 }
4079
4080 /*
4081  * ixgbe_set_num_queues: Allocate queues for device, feature dependant
4082  * @adapter: board private structure to initialize
4083  *
4084  * This is the top level queue allocation routine.  The order here is very
4085  * important, starting with the "most" number of features turned on at once,
4086  * and ending with the smallest set of features.  This way large combinations
4087  * can be allocated if they're turned on, and smaller combinations are the
4088  * fallthrough conditions.
4089  *
4090  **/
4091 static void ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
4092 {
4093         /* Start with base case */
4094         adapter->num_rx_queues = 1;
4095         adapter->num_tx_queues = 1;
4096         adapter->num_rx_pools = adapter->num_rx_queues;
4097         adapter->num_rx_queues_per_pool = 1;
4098
4099         if (ixgbe_set_sriov_queues(adapter))
4100                 return;
4101
4102 #ifdef IXGBE_FCOE
4103         if (ixgbe_set_fcoe_queues(adapter))
4104                 goto done;
4105
4106 #endif /* IXGBE_FCOE */
4107 #ifdef CONFIG_IXGBE_DCB
4108         if (ixgbe_set_dcb_queues(adapter))
4109                 goto done;
4110
4111 #endif
4112         if (ixgbe_set_fdir_queues(adapter))
4113                 goto done;
4114
4115         if (ixgbe_set_rss_queues(adapter))
4116                 goto done;
4117
4118         /* fallback to base case */
4119         adapter->num_rx_queues = 1;
4120         adapter->num_tx_queues = 1;
4121
4122 done:
4123         /* Notify the stack of the (possibly) reduced Tx Queue count. */
4124         netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
4125 }
4126
4127 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4128                                        int vectors)
4129 {
4130         int err, vector_threshold;
4131
4132         /* We'll want at least 3 (vector_threshold):
4133          * 1) TxQ[0] Cleanup
4134          * 2) RxQ[0] Cleanup
4135          * 3) Other (Link Status Change, etc.)
4136          * 4) TCP Timer (optional)
4137          */
4138         vector_threshold = MIN_MSIX_COUNT;
4139
4140         /* The more we get, the more we will assign to Tx/Rx Cleanup
4141          * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4142          * Right now, we simply care about how many we'll get; we'll
4143          * set them up later while requesting irq's.
4144          */
4145         while (vectors >= vector_threshold) {
4146                 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4147                                       vectors);
4148                 if (!err) /* Success in acquiring all requested vectors. */
4149                         break;
4150                 else if (err < 0)
4151                         vectors = 0; /* Nasty failure, quit now */
4152                 else /* err == number of vectors we should try again with */
4153                         vectors = err;
4154         }
4155
4156         if (vectors < vector_threshold) {
4157                 /* Can't allocate enough MSI-X interrupts?  Oh well.
4158                  * This just means we'll go with either a single MSI
4159                  * vector or fall back to legacy interrupts.
4160                  */
4161                 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4162                              "Unable to allocate MSI-X interrupts\n");
4163                 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4164                 kfree(adapter->msix_entries);
4165                 adapter->msix_entries = NULL;
4166         } else {
4167                 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4168                 /*
4169                  * Adjust for only the vectors we'll use, which is minimum
4170                  * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4171                  * vectors we were allocated.
4172                  */
4173                 adapter->num_msix_vectors = min(vectors,
4174                                    adapter->max_msix_q_vectors + NON_Q_VECTORS);
4175         }
4176 }
4177
4178 /**
4179  * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4180  * @adapter: board private structure to initialize
4181  *
4182  * Cache the descriptor ring offsets for RSS to the assigned rings.
4183  *
4184  **/
4185 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4186 {
4187         int i;
4188         bool ret = false;
4189
4190         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4191                 for (i = 0; i < adapter->num_rx_queues; i++)
4192                         adapter->rx_ring[i]->reg_idx = i;
4193                 for (i = 0; i < adapter->num_tx_queues; i++)
4194                         adapter->tx_ring[i]->reg_idx = i;
4195                 ret = true;
4196         } else {
4197                 ret = false;
4198         }
4199
4200         return ret;
4201 }
4202
4203 #ifdef CONFIG_IXGBE_DCB
4204 /**
4205  * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4206  * @adapter: board private structure to initialize
4207  *
4208  * Cache the descriptor ring offsets for DCB to the assigned rings.
4209  *
4210  **/
4211 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4212 {
4213         int i;
4214         bool ret = false;
4215         int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
4216
4217         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4218                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
4219                         /* the number of queues is assumed to be symmetric */
4220                         for (i = 0; i < dcb_i; i++) {
4221                                 adapter->rx_ring[i]->reg_idx = i << 3;
4222                                 adapter->tx_ring[i]->reg_idx = i << 2;
4223                         }
4224                         ret = true;
4225                 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
4226                         if (dcb_i == 8) {
4227                                 /*
4228                                  * Tx TC0 starts at: descriptor queue 0
4229                                  * Tx TC1 starts at: descriptor queue 32
4230                                  * Tx TC2 starts at: descriptor queue 64
4231                                  * Tx TC3 starts at: descriptor queue 80
4232                                  * Tx TC4 starts at: descriptor queue 96
4233                                  * Tx TC5 starts at: descriptor queue 104
4234                                  * Tx TC6 starts at: descriptor queue 112
4235                                  * Tx TC7 starts at: descriptor queue 120
4236                                  *
4237                                  * Rx TC0-TC7 are offset by 16 queues each
4238                                  */
4239                                 for (i = 0; i < 3; i++) {
4240                                         adapter->tx_ring[i]->reg_idx = i << 5;
4241                                         adapter->rx_ring[i]->reg_idx = i << 4;
4242                                 }
4243                                 for ( ; i < 5; i++) {
4244                                         adapter->tx_ring[i]->reg_idx =
4245                                                                  ((i + 2) << 4);
4246                                         adapter->rx_ring[i]->reg_idx = i << 4;
4247                                 }
4248                                 for ( ; i < dcb_i; i++) {
4249                                         adapter->tx_ring[i]->reg_idx =
4250                                                                  ((i + 8) << 3);
4251                                         adapter->rx_ring[i]->reg_idx = i << 4;
4252                                 }
4253
4254                                 ret = true;
4255                         } else if (dcb_i == 4) {
4256                                 /*
4257                                  * Tx TC0 starts at: descriptor queue 0
4258                                  * Tx TC1 starts at: descriptor queue 64
4259                                  * Tx TC2 starts at: descriptor queue 96
4260                                  * Tx TC3 starts at: descriptor queue 112
4261                                  *
4262                                  * Rx TC0-TC3 are offset by 32 queues each
4263                                  */
4264                                 adapter->tx_ring[0]->reg_idx = 0;
4265                                 adapter->tx_ring[1]->reg_idx = 64;
4266                                 adapter->tx_ring[2]->reg_idx = 96;
4267                                 adapter->tx_ring[3]->reg_idx = 112;
4268                                 for (i = 0 ; i < dcb_i; i++)
4269                                         adapter->rx_ring[i]->reg_idx = i << 5;
4270
4271                                 ret = true;
4272                         } else {
4273                                 ret = false;
4274                         }
4275                 } else {
4276                         ret = false;
4277                 }
4278         } else {
4279                 ret = false;
4280         }
4281
4282         return ret;
4283 }
4284 #endif
4285
4286 /**
4287  * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4288  * @adapter: board private structure to initialize
4289  *
4290  * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4291  *
4292  **/
4293 static inline bool ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4294 {
4295         int i;
4296         bool ret = false;
4297
4298         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4299             ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4300              (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
4301                 for (i = 0; i < adapter->num_rx_queues; i++)
4302                         adapter->rx_ring[i]->reg_idx = i;
4303                 for (i = 0; i < adapter->num_tx_queues; i++)
4304                         adapter->tx_ring[i]->reg_idx = i;
4305                 ret = true;
4306         }
4307
4308         return ret;
4309 }
4310
4311 #ifdef IXGBE_FCOE
4312 /**
4313  * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4314  * @adapter: board private structure to initialize
4315  *
4316  * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4317  *
4318  */
4319 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4320 {
4321         int i, fcoe_rx_i = 0, fcoe_tx_i = 0;
4322         bool ret = false;
4323         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4324
4325         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4326 #ifdef CONFIG_IXGBE_DCB
4327                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4328                         struct ixgbe_fcoe *fcoe = &adapter->fcoe;
4329
4330                         ixgbe_cache_ring_dcb(adapter);
4331                         /* find out queues in TC for FCoE */
4332                         fcoe_rx_i = adapter->rx_ring[fcoe->tc]->reg_idx + 1;
4333                         fcoe_tx_i = adapter->tx_ring[fcoe->tc]->reg_idx + 1;
4334                         /*
4335                          * In 82599, the number of Tx queues for each traffic
4336                          * class for both 8-TC and 4-TC modes are:
4337                          * TCs  : TC0 TC1 TC2 TC3 TC4 TC5 TC6 TC7
4338                          * 8 TCs:  32  32  16  16   8   8   8   8
4339                          * 4 TCs:  64  64  32  32
4340                          * We have max 8 queues for FCoE, where 8 the is
4341                          * FCoE redirection table size. If TC for FCoE is
4342                          * less than or equal to TC3, we have enough queues
4343                          * to add max of 8 queues for FCoE, so we start FCoE
4344                          * tx descriptor from the next one, i.e., reg_idx + 1.
4345                          * If TC for FCoE is above TC3, implying 8 TC mode,
4346                          * and we need 8 for FCoE, we have to take all queues
4347                          * in that traffic class for FCoE.
4348                          */
4349                         if ((f->indices == IXGBE_FCRETA_SIZE) && (fcoe->tc > 3))
4350                                 fcoe_tx_i--;
4351                 }
4352 #endif /* CONFIG_IXGBE_DCB */
4353                 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4354                         if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4355                             (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4356                                 ixgbe_cache_ring_fdir(adapter);
4357                         else
4358                                 ixgbe_cache_ring_rss(adapter);
4359
4360                         fcoe_rx_i = f->mask;
4361                         fcoe_tx_i = f->mask;
4362                 }
4363                 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4364                         adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4365                         adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4366                 }
4367                 ret = true;
4368         }
4369         return ret;
4370 }
4371
4372 #endif /* IXGBE_FCOE */
4373 /**
4374  * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4375  * @adapter: board private structure to initialize
4376  *
4377  * SR-IOV doesn't use any descriptor rings but changes the default if
4378  * no other mapping is used.
4379  *
4380  */
4381 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4382 {
4383         adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4384         adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4385         if (adapter->num_vfs)
4386                 return true;
4387         else
4388                 return false;
4389 }
4390
4391 /**
4392  * ixgbe_cache_ring_register - Descriptor ring to register mapping
4393  * @adapter: board private structure to initialize
4394  *
4395  * Once we know the feature-set enabled for the device, we'll cache
4396  * the register offset the descriptor ring is assigned to.
4397  *
4398  * Note, the order the various feature calls is important.  It must start with
4399  * the "most" features enabled at the same time, then trickle down to the
4400  * least amount of features turned on at once.
4401  **/
4402 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4403 {
4404         /* start with default case */
4405         adapter->rx_ring[0]->reg_idx = 0;
4406         adapter->tx_ring[0]->reg_idx = 0;
4407
4408         if (ixgbe_cache_ring_sriov(adapter))
4409                 return;
4410
4411 #ifdef IXGBE_FCOE
4412         if (ixgbe_cache_ring_fcoe(adapter))
4413                 return;
4414
4415 #endif /* IXGBE_FCOE */
4416 #ifdef CONFIG_IXGBE_DCB
4417         if (ixgbe_cache_ring_dcb(adapter))
4418                 return;
4419
4420 #endif
4421         if (ixgbe_cache_ring_fdir(adapter))
4422                 return;
4423
4424         if (ixgbe_cache_ring_rss(adapter))
4425                 return;
4426 }
4427
4428 /**
4429  * ixgbe_alloc_queues - Allocate memory for all rings
4430  * @adapter: board private structure to initialize
4431  *
4432  * We allocate one ring per queue at run-time since we don't know the
4433  * number of queues at compile-time.  The polling_netdev array is
4434  * intended for Multiqueue, but should work fine with a single queue.
4435  **/
4436 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
4437 {
4438         int i;
4439         int orig_node = adapter->node;
4440
4441         for (i = 0; i < adapter->num_tx_queues; i++) {
4442                 struct ixgbe_ring *ring = adapter->tx_ring[i];
4443                 if (orig_node == -1) {
4444                         int cur_node = next_online_node(adapter->node);
4445                         if (cur_node == MAX_NUMNODES)
4446                                 cur_node = first_online_node;
4447                         adapter->node = cur_node;
4448                 }
4449                 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4450                                     adapter->node);
4451                 if (!ring)
4452                         ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4453                 if (!ring)
4454                         goto err_tx_ring_allocation;
4455                 ring->count = adapter->tx_ring_count;
4456                 ring->queue_index = i;
4457                 ring->numa_node = adapter->node;
4458
4459                 adapter->tx_ring[i] = ring;
4460         }
4461
4462         /* Restore the adapter's original node */
4463         adapter->node = orig_node;
4464
4465         for (i = 0; i < adapter->num_rx_queues; i++) {
4466                 struct ixgbe_ring *ring = adapter->rx_ring[i];
4467                 if (orig_node == -1) {
4468                         int cur_node = next_online_node(adapter->node);
4469                         if (cur_node == MAX_NUMNODES)
4470                                 cur_node = first_online_node;
4471                         adapter->node = cur_node;
4472                 }
4473                 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4474                                     adapter->node);
4475                 if (!ring)
4476                         ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4477                 if (!ring)
4478                         goto err_rx_ring_allocation;
4479                 ring->count = adapter->rx_ring_count;
4480                 ring->queue_index = i;
4481                 ring->numa_node = adapter->node;
4482
4483                 adapter->rx_ring[i] = ring;
4484         }
4485
4486         /* Restore the adapter's original node */
4487         adapter->node = orig_node;
4488
4489         ixgbe_cache_ring_register(adapter);
4490
4491         return 0;
4492
4493 err_rx_ring_allocation:
4494         for (i = 0; i < adapter->num_tx_queues; i++)
4495                 kfree(adapter->tx_ring[i]);
4496 err_tx_ring_allocation:
4497         return -ENOMEM;
4498 }
4499
4500 /**
4501  * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4502  * @adapter: board private structure to initialize
4503  *
4504  * Attempt to configure the interrupts using the best available
4505  * capabilities of the hardware and the kernel.
4506  **/
4507 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4508 {
4509         struct ixgbe_hw *hw = &adapter->hw;
4510         int err = 0;
4511         int vector, v_budget;
4512
4513         /*
4514          * It's easy to be greedy for MSI-X vectors, but it really
4515          * doesn't do us much good if we have a lot more vectors
4516          * than CPU's.  So let's be conservative and only ask for
4517          * (roughly) the same number of vectors as there are CPU's.
4518          */
4519         v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
4520                        (int)num_online_cpus()) + NON_Q_VECTORS;
4521
4522         /*
4523          * At the same time, hardware can only support a maximum of
4524          * hw.mac->max_msix_vectors vectors.  With features
4525          * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4526          * descriptor queues supported by our device.  Thus, we cap it off in
4527          * those rare cases where the cpu count also exceeds our vector limit.
4528          */
4529         v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
4530
4531         /* A failure in MSI-X entry allocation isn't fatal, but it does
4532          * mean we disable MSI-X capabilities of the adapter. */
4533         adapter->msix_entries = kcalloc(v_budget,
4534                                         sizeof(struct msix_entry), GFP_KERNEL);
4535         if (adapter->msix_entries) {
4536                 for (vector = 0; vector < v_budget; vector++)
4537                         adapter->msix_entries[vector].entry = vector;
4538
4539                 ixgbe_acquire_msix_vectors(adapter, v_budget);
4540
4541                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4542                         goto out;
4543         }
4544
4545         adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4546         adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4547         adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4548         adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4549         adapter->atr_sample_rate = 0;
4550         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4551                 ixgbe_disable_sriov(adapter);
4552
4553         ixgbe_set_num_queues(adapter);
4554
4555         err = pci_enable_msi(adapter->pdev);
4556         if (!err) {
4557                 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4558         } else {
4559                 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4560                              "Unable to allocate MSI interrupt, "
4561                              "falling back to legacy.  Error: %d\n", err);
4562                 /* reset err */
4563                 err = 0;
4564         }
4565
4566 out:
4567         return err;
4568 }
4569
4570 /**
4571  * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4572  * @adapter: board private structure to initialize
4573  *
4574  * We allocate one q_vector per queue interrupt.  If allocation fails we
4575  * return -ENOMEM.
4576  **/
4577 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4578 {
4579         int q_idx, num_q_vectors;
4580         struct ixgbe_q_vector *q_vector;
4581         int napi_vectors;
4582         int (*poll)(struct napi_struct *, int);
4583
4584         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4585                 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4586                 napi_vectors = adapter->num_rx_queues;
4587                 poll = &ixgbe_clean_rxtx_many;
4588         } else {
4589                 num_q_vectors = 1;
4590                 napi_vectors = 1;
4591                 poll = &ixgbe_poll;
4592         }
4593
4594         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4595                 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4596                                         GFP_KERNEL, adapter->node);
4597                 if (!q_vector)
4598                         q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4599                                            GFP_KERNEL);
4600                 if (!q_vector)
4601                         goto err_out;
4602                 q_vector->adapter = adapter;
4603                 if (q_vector->txr_count && !q_vector->rxr_count)
4604                         q_vector->eitr = adapter->tx_eitr_param;
4605                 else
4606                         q_vector->eitr = adapter->rx_eitr_param;
4607                 q_vector->v_idx = q_idx;
4608                 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
4609                 adapter->q_vector[q_idx] = q_vector;
4610         }
4611
4612         return 0;
4613
4614 err_out:
4615         while (q_idx) {
4616                 q_idx--;
4617                 q_vector = adapter->q_vector[q_idx];
4618                 netif_napi_del(&q_vector->napi);
4619                 kfree(q_vector);
4620                 adapter->q_vector[q_idx] = NULL;
4621         }
4622         return -ENOMEM;
4623 }
4624
4625 /**
4626  * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4627  * @adapter: board private structure to initialize
4628  *
4629  * This function frees the memory allocated to the q_vectors.  In addition if
4630  * NAPI is enabled it will delete any references to the NAPI struct prior
4631  * to freeing the q_vector.
4632  **/
4633 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4634 {
4635         int q_idx, num_q_vectors;
4636
4637         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4638                 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4639         else
4640                 num_q_vectors = 1;
4641
4642         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4643                 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
4644                 adapter->q_vector[q_idx] = NULL;
4645                 netif_napi_del(&q_vector->napi);
4646                 kfree(q_vector);
4647         }
4648 }
4649
4650 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4651 {
4652         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4653                 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4654                 pci_disable_msix(adapter->pdev);
4655                 kfree(adapter->msix_entries);
4656                 adapter->msix_entries = NULL;
4657         } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4658                 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4659                 pci_disable_msi(adapter->pdev);
4660         }
4661 }
4662
4663 /**
4664  * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4665  * @adapter: board private structure to initialize
4666  *
4667  * We determine which interrupt scheme to use based on...
4668  * - Kernel support (MSI, MSI-X)
4669  *   - which can be user-defined (via MODULE_PARAM)
4670  * - Hardware queue count (num_*_queues)
4671  *   - defined by miscellaneous hardware support/features (RSS, etc.)
4672  **/
4673 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4674 {
4675         int err;
4676
4677         /* Number of supported queues */
4678         ixgbe_set_num_queues(adapter);
4679
4680         err = ixgbe_set_interrupt_capability(adapter);
4681         if (err) {
4682                 e_dev_err("Unable to setup interrupt capabilities\n");
4683                 goto err_set_interrupt;
4684         }
4685
4686         err = ixgbe_alloc_q_vectors(adapter);
4687         if (err) {
4688                 e_dev_err("Unable to allocate memory for queue vectors\n");
4689                 goto err_alloc_q_vectors;
4690         }
4691
4692         err = ixgbe_alloc_queues(adapter);
4693         if (err) {
4694                 e_dev_err("Unable to allocate memory for queues\n");
4695                 goto err_alloc_queues;
4696         }
4697
4698         e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
4699                    (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
4700                    adapter->num_rx_queues, adapter->num_tx_queues);
4701
4702         set_bit(__IXGBE_DOWN, &adapter->state);
4703
4704         return 0;
4705
4706 err_alloc_queues:
4707         ixgbe_free_q_vectors(adapter);
4708 err_alloc_q_vectors:
4709         ixgbe_reset_interrupt_capability(adapter);
4710 err_set_interrupt:
4711         return err;
4712 }
4713
4714 /**
4715  * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
4716  * @adapter: board private structure to clear interrupt scheme on
4717  *
4718  * We go through and clear interrupt specific resources and reset the structure
4719  * to pre-load conditions
4720  **/
4721 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
4722 {
4723         int i;
4724
4725         for (i = 0; i < adapter->num_tx_queues; i++) {
4726                 kfree(adapter->tx_ring[i]);
4727                 adapter->tx_ring[i] = NULL;
4728         }
4729         for (i = 0; i < adapter->num_rx_queues; i++) {
4730                 kfree(adapter->rx_ring[i]);
4731                 adapter->rx_ring[i] = NULL;
4732         }
4733
4734         ixgbe_free_q_vectors(adapter);
4735         ixgbe_reset_interrupt_capability(adapter);
4736 }
4737
4738 /**
4739  * ixgbe_sfp_timer - worker thread to find a missing module
4740  * @data: pointer to our adapter struct
4741  **/
4742 static void ixgbe_sfp_timer(unsigned long data)
4743 {
4744         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4745
4746         /*
4747          * Do the sfp_timer outside of interrupt context due to the
4748          * delays that sfp+ detection requires
4749          */
4750         schedule_work(&adapter->sfp_task);
4751 }
4752
4753 /**
4754  * ixgbe_sfp_task - worker thread to find a missing module
4755  * @work: pointer to work_struct containing our data
4756  **/
4757 static void ixgbe_sfp_task(struct work_struct *work)
4758 {
4759         struct ixgbe_adapter *adapter = container_of(work,
4760                                                      struct ixgbe_adapter,
4761                                                      sfp_task);
4762         struct ixgbe_hw *hw = &adapter->hw;
4763
4764         if ((hw->phy.type == ixgbe_phy_nl) &&
4765             (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
4766                 s32 ret = hw->phy.ops.identify_sfp(hw);
4767                 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
4768                         goto reschedule;
4769                 ret = hw->phy.ops.reset(hw);
4770                 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4771                         e_dev_err("failed to initialize because an unsupported "
4772                                   "SFP+ module type was detected.\n");
4773                         e_dev_err("Reload the driver after installing a "
4774                                   "supported module.\n");
4775                         unregister_netdev(adapter->netdev);
4776                 } else {
4777                         e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
4778                 }
4779                 /* don't need this routine any more */
4780                 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4781         }
4782         return;
4783 reschedule:
4784         if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
4785                 mod_timer(&adapter->sfp_timer,
4786                           round_jiffies(jiffies + (2 * HZ)));
4787 }
4788
4789 /**
4790  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4791  * @adapter: board private structure to initialize
4792  *
4793  * ixgbe_sw_init initializes the Adapter private data structure.
4794  * Fields are initialized based on PCI device information and
4795  * OS network device settings (MTU size).
4796  **/
4797 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4798 {
4799         struct ixgbe_hw *hw = &adapter->hw;
4800         struct pci_dev *pdev = adapter->pdev;
4801         struct net_device *dev = adapter->netdev;
4802         unsigned int rss;
4803 #ifdef CONFIG_IXGBE_DCB
4804         int j;
4805         struct tc_configuration *tc;
4806 #endif
4807
4808         /* PCI config space info */
4809
4810         hw->vendor_id = pdev->vendor;
4811         hw->device_id = pdev->device;
4812         hw->revision_id = pdev->revision;
4813         hw->subsystem_vendor_id = pdev->subsystem_vendor;
4814         hw->subsystem_device_id = pdev->subsystem_device;
4815
4816         /* Set capability flags */
4817         rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
4818         adapter->ring_feature[RING_F_RSS].indices = rss;
4819         adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4820         adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
4821         if (hw->mac.type == ixgbe_mac_82598EB) {
4822                 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4823                         adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4824                 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
4825         } else if (hw->mac.type == ixgbe_mac_82599EB) {
4826                 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
4827                 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4828                 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4829                 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4830                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4831                 if (dev->features & NETIF_F_NTUPLE) {
4832                         /* Flow Director perfect filter enabled */
4833                         adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4834                         adapter->atr_sample_rate = 0;
4835                         spin_lock_init(&adapter->fdir_perfect_lock);
4836                 } else {
4837                         /* Flow Director hash filters enabled */
4838                         adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4839                         adapter->atr_sample_rate = 20;
4840                 }
4841                 adapter->ring_feature[RING_F_FDIR].indices =
4842                                                          IXGBE_MAX_FDIR_INDICES;
4843                 adapter->fdir_pballoc = 0;
4844 #ifdef IXGBE_FCOE
4845                 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4846                 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4847                 adapter->ring_feature[RING_F_FCOE].indices = 0;
4848 #ifdef CONFIG_IXGBE_DCB
4849                 /* Default traffic class to use for FCoE */
4850                 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
4851                 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4852 #endif
4853 #endif /* IXGBE_FCOE */
4854         }
4855
4856 #ifdef CONFIG_IXGBE_DCB
4857         /* Configure DCB traffic classes */
4858         for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4859                 tc = &adapter->dcb_cfg.tc_config[j];
4860                 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4861                 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4862                 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4863                 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4864                 tc->dcb_pfc = pfc_disabled;
4865         }
4866         adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4867         adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4868         adapter->dcb_cfg.rx_pba_cfg = pba_equal;
4869         adapter->dcb_cfg.pfc_mode_enable = false;
4870         adapter->dcb_cfg.round_robin_enable = false;
4871         adapter->dcb_set_bitmap = 0x00;
4872         ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
4873                            adapter->ring_feature[RING_F_DCB].indices);
4874
4875 #endif
4876
4877         /* default flow control settings */
4878         hw->fc.requested_mode = ixgbe_fc_full;
4879         hw->fc.current_mode = ixgbe_fc_full;    /* init for ethtool output */
4880 #ifdef CONFIG_DCB
4881         adapter->last_lfc_mode = hw->fc.current_mode;
4882 #endif
4883         hw->fc.high_water = IXGBE_DEFAULT_FCRTH;
4884         hw->fc.low_water = IXGBE_DEFAULT_FCRTL;
4885         hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4886         hw->fc.send_xon = true;
4887         hw->fc.disable_fc_autoneg = false;
4888
4889         /* enable itr by default in dynamic mode */
4890         adapter->rx_itr_setting = 1;
4891         adapter->rx_eitr_param = 20000;
4892         adapter->tx_itr_setting = 1;
4893         adapter->tx_eitr_param = 10000;
4894
4895         /* set defaults for eitr in MegaBytes */
4896         adapter->eitr_low = 10;
4897         adapter->eitr_high = 20;
4898
4899         /* set default ring sizes */
4900         adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4901         adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4902
4903         /* initialize eeprom parameters */
4904         if (ixgbe_init_eeprom_params_generic(hw)) {
4905                 e_dev_err("EEPROM initialization failed\n");
4906                 return -EIO;
4907         }
4908
4909         /* enable rx csum by default */
4910         adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
4911
4912         /* get assigned NUMA node */
4913         adapter->node = dev_to_node(&pdev->dev);
4914
4915         set_bit(__IXGBE_DOWN, &adapter->state);
4916
4917         return 0;
4918 }
4919
4920 /**
4921  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4922  * @adapter: board private structure
4923  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
4924  *
4925  * Return 0 on success, negative on failure
4926  **/
4927 int ixgbe_setup_tx_resources(struct ixgbe_adapter *adapter,
4928                              struct ixgbe_ring *tx_ring)
4929 {
4930         struct pci_dev *pdev = adapter->pdev;
4931         int size;
4932
4933         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4934         tx_ring->tx_buffer_info = vmalloc_node(size, tx_ring->numa_node);
4935         if (!tx_ring->tx_buffer_info)
4936                 tx_ring->tx_buffer_info = vmalloc(size);
4937         if (!tx_ring->tx_buffer_info)
4938                 goto err;
4939         memset(tx_ring->tx_buffer_info, 0, size);
4940
4941         /* round up to nearest 4K */
4942         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4943         tx_ring->size = ALIGN(tx_ring->size, 4096);
4944
4945         tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
4946                                            &tx_ring->dma, GFP_KERNEL);
4947         if (!tx_ring->desc)
4948                 goto err;
4949
4950         tx_ring->next_to_use = 0;
4951         tx_ring->next_to_clean = 0;
4952         tx_ring->work_limit = tx_ring->count;
4953         return 0;
4954
4955 err:
4956         vfree(tx_ring->tx_buffer_info);
4957         tx_ring->tx_buffer_info = NULL;
4958         e_err(probe, "Unable to allocate memory for the Tx descriptor ring\n");
4959         return -ENOMEM;
4960 }
4961
4962 /**
4963  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4964  * @adapter: board private structure
4965  *
4966  * If this function returns with an error, then it's possible one or
4967  * more of the rings is populated (while the rest are not).  It is the
4968  * callers duty to clean those orphaned rings.
4969  *
4970  * Return 0 on success, negative on failure
4971  **/
4972 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4973 {
4974         int i, err = 0;
4975
4976         for (i = 0; i < adapter->num_tx_queues; i++) {
4977                 err = ixgbe_setup_tx_resources(adapter, adapter->tx_ring[i]);
4978                 if (!err)
4979                         continue;
4980                 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
4981                 break;
4982         }
4983
4984         return err;
4985 }
4986
4987 /**
4988  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
4989  * @adapter: board private structure
4990  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
4991  *
4992  * Returns 0 on success, negative on failure
4993  **/
4994 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
4995                              struct ixgbe_ring *rx_ring)
4996 {
4997         struct pci_dev *pdev = adapter->pdev;
4998         int size;
4999
5000         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5001         rx_ring->rx_buffer_info = vmalloc_node(size, adapter->node);
5002         if (!rx_ring->rx_buffer_info)
5003                 rx_ring->rx_buffer_info = vmalloc(size);
5004         if (!rx_ring->rx_buffer_info) {
5005                 e_err(probe, "vmalloc allocation failed for the Rx "
5006                       "descriptor ring\n");
5007                 goto alloc_failed;
5008         }
5009         memset(rx_ring->rx_buffer_info, 0, size);
5010
5011         /* Round up to nearest 4K */
5012         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5013         rx_ring->size = ALIGN(rx_ring->size, 4096);
5014
5015         rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
5016                                            &rx_ring->dma, GFP_KERNEL);
5017
5018         if (!rx_ring->desc) {
5019                 e_err(probe, "Memory allocation failed for the Rx "
5020                       "descriptor ring\n");
5021                 vfree(rx_ring->rx_buffer_info);
5022                 goto alloc_failed;
5023         }
5024
5025         rx_ring->next_to_clean = 0;
5026         rx_ring->next_to_use = 0;
5027
5028         return 0;
5029
5030 alloc_failed:
5031         return -ENOMEM;
5032 }
5033
5034 /**
5035  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5036  * @adapter: board private structure
5037  *
5038  * If this function returns with an error, then it's possible one or
5039  * more of the rings is populated (while the rest are not).  It is the
5040  * callers duty to clean those orphaned rings.
5041  *
5042  * Return 0 on success, negative on failure
5043  **/
5044
5045 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5046 {
5047         int i, err = 0;
5048
5049         for (i = 0; i < adapter->num_rx_queues; i++) {
5050                 err = ixgbe_setup_rx_resources(adapter, adapter->rx_ring[i]);
5051                 if (!err)
5052                         continue;
5053                 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5054                 break;
5055         }
5056
5057         return err;
5058 }
5059
5060 /**
5061  * ixgbe_free_tx_resources - Free Tx Resources per Queue
5062  * @adapter: board private structure
5063  * @tx_ring: Tx descriptor ring for a specific queue
5064  *
5065  * Free all transmit software resources
5066  **/
5067 void ixgbe_free_tx_resources(struct ixgbe_adapter *adapter,
5068                              struct ixgbe_ring *tx_ring)
5069 {
5070         struct pci_dev *pdev = adapter->pdev;
5071
5072         ixgbe_clean_tx_ring(adapter, tx_ring);
5073
5074         vfree(tx_ring->tx_buffer_info);
5075         tx_ring->tx_buffer_info = NULL;
5076
5077         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
5078                           tx_ring->dma);
5079
5080         tx_ring->desc = NULL;
5081 }
5082
5083 /**
5084  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5085  * @adapter: board private structure
5086  *
5087  * Free all transmit software resources
5088  **/
5089 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5090 {
5091         int i;
5092
5093         for (i = 0; i < adapter->num_tx_queues; i++)
5094                 if (adapter->tx_ring[i]->desc)
5095                         ixgbe_free_tx_resources(adapter, adapter->tx_ring[i]);
5096 }
5097
5098 /**
5099  * ixgbe_free_rx_resources - Free Rx Resources
5100  * @adapter: board private structure
5101  * @rx_ring: ring to clean the resources from
5102  *
5103  * Free all receive software resources
5104  **/
5105 void ixgbe_free_rx_resources(struct ixgbe_adapter *adapter,
5106                              struct ixgbe_ring *rx_ring)
5107 {
5108         struct pci_dev *pdev = adapter->pdev;
5109
5110         ixgbe_clean_rx_ring(adapter, rx_ring);
5111
5112         vfree(rx_ring->rx_buffer_info);
5113         rx_ring->rx_buffer_info = NULL;
5114
5115         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
5116                           rx_ring->dma);
5117
5118         rx_ring->desc = NULL;
5119 }
5120
5121 /**
5122  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5123  * @adapter: board private structure
5124  *
5125  * Free all receive software resources
5126  **/
5127 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5128 {
5129         int i;
5130
5131         for (i = 0; i < adapter->num_rx_queues; i++)
5132                 if (adapter->rx_ring[i]->desc)
5133                         ixgbe_free_rx_resources(adapter, adapter->rx_ring[i]);
5134 }
5135
5136 /**
5137  * ixgbe_change_mtu - Change the Maximum Transfer Unit
5138  * @netdev: network interface device structure
5139  * @new_mtu: new value for maximum frame size
5140  *
5141  * Returns 0 on success, negative on failure
5142  **/
5143 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5144 {
5145         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5146         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5147
5148         /* MTU < 68 is an error and causes problems on some kernels */
5149         if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5150                 return -EINVAL;
5151
5152         e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5153         /* must set new MTU before calling down or up */
5154         netdev->mtu = new_mtu;
5155
5156         if (netif_running(netdev))
5157                 ixgbe_reinit_locked(adapter);
5158
5159         return 0;
5160 }
5161
5162 /**
5163  * ixgbe_open - Called when a network interface is made active
5164  * @netdev: network interface device structure
5165  *
5166  * Returns 0 on success, negative value on failure
5167  *
5168  * The open entry point is called when a network interface is made
5169  * active by the system (IFF_UP).  At this point all resources needed
5170  * for transmit and receive operations are allocated, the interrupt
5171  * handler is registered with the OS, the watchdog timer is started,
5172  * and the stack is notified that the interface is ready.
5173  **/
5174 static int ixgbe_open(struct net_device *netdev)
5175 {
5176         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5177         int err;
5178
5179         /* disallow open during test */
5180         if (test_bit(__IXGBE_TESTING, &adapter->state))
5181                 return -EBUSY;
5182
5183         netif_carrier_off(netdev);
5184
5185         /* allocate transmit descriptors */
5186         err = ixgbe_setup_all_tx_resources(adapter);
5187         if (err)
5188                 goto err_setup_tx;
5189
5190         /* allocate receive descriptors */
5191         err = ixgbe_setup_all_rx_resources(adapter);
5192         if (err)
5193                 goto err_setup_rx;
5194
5195         ixgbe_configure(adapter);
5196
5197         err = ixgbe_request_irq(adapter);
5198         if (err)
5199                 goto err_req_irq;
5200
5201         err = ixgbe_up_complete(adapter);
5202         if (err)
5203                 goto err_up;
5204
5205         netif_tx_start_all_queues(netdev);
5206
5207         return 0;
5208
5209 err_up:
5210         ixgbe_release_hw_control(adapter);
5211         ixgbe_free_irq(adapter);
5212 err_req_irq:
5213 err_setup_rx:
5214         ixgbe_free_all_rx_resources(adapter);
5215 err_setup_tx:
5216         ixgbe_free_all_tx_resources(adapter);
5217         ixgbe_reset(adapter);
5218
5219         return err;
5220 }
5221
5222 /**
5223  * ixgbe_close - Disables a network interface
5224  * @netdev: network interface device structure
5225  *
5226  * Returns 0, this is not allowed to fail
5227  *
5228  * The close entry point is called when an interface is de-activated
5229  * by the OS.  The hardware is still under the drivers control, but
5230  * needs to be disabled.  A global MAC reset is issued to stop the
5231  * hardware, and all transmit and receive resources are freed.
5232  **/
5233 static int ixgbe_close(struct net_device *netdev)
5234 {
5235         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5236
5237         ixgbe_down(adapter);
5238         ixgbe_free_irq(adapter);
5239
5240         ixgbe_free_all_tx_resources(adapter);
5241         ixgbe_free_all_rx_resources(adapter);
5242
5243         ixgbe_release_hw_control(adapter);
5244
5245         return 0;
5246 }
5247
5248 #ifdef CONFIG_PM
5249 static int ixgbe_resume(struct pci_dev *pdev)
5250 {
5251         struct net_device *netdev = pci_get_drvdata(pdev);
5252         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5253         u32 err;
5254
5255         pci_set_power_state(pdev, PCI_D0);
5256         pci_restore_state(pdev);
5257         /*
5258          * pci_restore_state clears dev->state_saved so call
5259          * pci_save_state to restore it.
5260          */
5261         pci_save_state(pdev);
5262
5263         err = pci_enable_device_mem(pdev);
5264         if (err) {
5265                 e_dev_err("Cannot enable PCI device from suspend\n");
5266                 return err;
5267         }
5268         pci_set_master(pdev);
5269
5270         pci_wake_from_d3(pdev, false);
5271
5272         err = ixgbe_init_interrupt_scheme(adapter);
5273         if (err) {
5274                 e_dev_err("Cannot initialize interrupts for device\n");
5275                 return err;
5276         }
5277
5278         ixgbe_reset(adapter);
5279
5280         IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5281
5282         if (netif_running(netdev)) {
5283                 err = ixgbe_open(adapter->netdev);
5284                 if (err)
5285                         return err;
5286         }
5287
5288         netif_device_attach(netdev);
5289
5290         return 0;
5291 }
5292 #endif /* CONFIG_PM */
5293
5294 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5295 {
5296         struct net_device *netdev = pci_get_drvdata(pdev);
5297         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5298         struct ixgbe_hw *hw = &adapter->hw;
5299         u32 ctrl, fctrl;
5300         u32 wufc = adapter->wol;
5301 #ifdef CONFIG_PM
5302         int retval = 0;
5303 #endif
5304
5305         netif_device_detach(netdev);
5306
5307         if (netif_running(netdev)) {
5308                 ixgbe_down(adapter);
5309                 ixgbe_free_irq(adapter);
5310                 ixgbe_free_all_tx_resources(adapter);
5311                 ixgbe_free_all_rx_resources(adapter);
5312         }
5313
5314 #ifdef CONFIG_PM
5315         retval = pci_save_state(pdev);
5316         if (retval)
5317                 return retval;
5318
5319 #endif
5320         if (wufc) {
5321                 ixgbe_set_rx_mode(netdev);
5322
5323                 /* turn on all-multi mode if wake on multicast is enabled */
5324                 if (wufc & IXGBE_WUFC_MC) {
5325                         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5326                         fctrl |= IXGBE_FCTRL_MPE;
5327                         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5328                 }
5329
5330                 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5331                 ctrl |= IXGBE_CTRL_GIO_DIS;
5332                 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5333
5334                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5335         } else {
5336                 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5337                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5338         }
5339
5340         if (wufc && hw->mac.type == ixgbe_mac_82599EB)
5341                 pci_wake_from_d3(pdev, true);
5342         else
5343                 pci_wake_from_d3(pdev, false);
5344
5345         *enable_wake = !!wufc;
5346
5347         ixgbe_clear_interrupt_scheme(adapter);
5348
5349         ixgbe_release_hw_control(adapter);
5350
5351         pci_disable_device(pdev);
5352
5353         return 0;
5354 }
5355
5356 #ifdef CONFIG_PM
5357 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5358 {
5359         int retval;
5360         bool wake;
5361
5362         retval = __ixgbe_shutdown(pdev, &wake);
5363         if (retval)
5364                 return retval;
5365
5366         if (wake) {
5367                 pci_prepare_to_sleep(pdev);
5368         } else {
5369                 pci_wake_from_d3(pdev, false);
5370                 pci_set_power_state(pdev, PCI_D3hot);
5371         }
5372
5373         return 0;
5374 }
5375 #endif /* CONFIG_PM */
5376
5377 static void ixgbe_shutdown(struct pci_dev *pdev)
5378 {
5379         bool wake;
5380
5381         __ixgbe_shutdown(pdev, &wake);
5382
5383         if (system_state == SYSTEM_POWER_OFF) {
5384                 pci_wake_from_d3(pdev, wake);
5385                 pci_set_power_state(pdev, PCI_D3hot);
5386         }
5387 }
5388
5389 /**
5390  * ixgbe_update_stats - Update the board statistics counters.
5391  * @adapter: board private structure
5392  **/
5393 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5394 {
5395         struct net_device *netdev = adapter->netdev;
5396         struct ixgbe_hw *hw = &adapter->hw;
5397         u64 total_mpc = 0;
5398         u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5399         u64 non_eop_descs = 0, restart_queue = 0;
5400
5401         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5402             test_bit(__IXGBE_RESETTING, &adapter->state))
5403                 return;
5404
5405         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5406                 u64 rsc_count = 0;
5407                 u64 rsc_flush = 0;
5408                 for (i = 0; i < 16; i++)
5409                         adapter->hw_rx_no_dma_resources +=
5410                                              IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5411                 for (i = 0; i < adapter->num_rx_queues; i++) {
5412                         rsc_count += adapter->rx_ring[i]->rsc_count;
5413                         rsc_flush += adapter->rx_ring[i]->rsc_flush;
5414                 }
5415                 adapter->rsc_total_count = rsc_count;
5416                 adapter->rsc_total_flush = rsc_flush;
5417         }
5418
5419         /* gather some stats to the adapter struct that are per queue */
5420         for (i = 0; i < adapter->num_tx_queues; i++)
5421                 restart_queue += adapter->tx_ring[i]->restart_queue;
5422         adapter->restart_queue = restart_queue;
5423
5424         for (i = 0; i < adapter->num_rx_queues; i++)
5425                 non_eop_descs += adapter->rx_ring[i]->non_eop_descs;
5426         adapter->non_eop_descs = non_eop_descs;
5427
5428         adapter->stats.crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5429         for (i = 0; i < 8; i++) {
5430                 /* for packet buffers not used, the register should read 0 */
5431                 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5432                 missed_rx += mpc;
5433                 adapter->stats.mpc[i] += mpc;
5434                 total_mpc += adapter->stats.mpc[i];
5435                 if (hw->mac.type == ixgbe_mac_82598EB)
5436                         adapter->stats.rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5437                 adapter->stats.qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5438                 adapter->stats.qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5439                 adapter->stats.qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5440                 adapter->stats.qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5441                 if (hw->mac.type == ixgbe_mac_82599EB) {
5442                         adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
5443                                                             IXGBE_PXONRXCNT(i));
5444                         adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
5445                                                            IXGBE_PXOFFRXCNT(i));
5446                         adapter->stats.qprdc[i] += IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5447                 } else {
5448                         adapter->stats.pxonrxc[i] += IXGBE_READ_REG(hw,
5449                                                               IXGBE_PXONRXC(i));
5450                         adapter->stats.pxoffrxc[i] += IXGBE_READ_REG(hw,
5451                                                              IXGBE_PXOFFRXC(i));
5452                 }
5453                 adapter->stats.pxontxc[i] += IXGBE_READ_REG(hw,
5454                                                             IXGBE_PXONTXC(i));
5455                 adapter->stats.pxofftxc[i] += IXGBE_READ_REG(hw,
5456                                                              IXGBE_PXOFFTXC(i));
5457         }
5458         adapter->stats.gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5459         /* work around hardware counting issue */
5460         adapter->stats.gprc -= missed_rx;
5461
5462         /* 82598 hardware only has a 32 bit counter in the high register */
5463         if (hw->mac.type == ixgbe_mac_82599EB) {
5464                 u64 tmp;
5465                 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5466                 tmp = IXGBE_READ_REG(hw, IXGBE_GORCH) & 0xF;
5467                                                 /* 4 high bits of GORC */
5468                 adapter->stats.gorc += (tmp << 32);
5469                 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5470                 tmp = IXGBE_READ_REG(hw, IXGBE_GOTCH) & 0xF;
5471                                                 /* 4 high bits of GOTC */
5472                 adapter->stats.gotc += (tmp << 32);
5473                 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5474                 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5475                 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5476                 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
5477                 adapter->stats.fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5478                 adapter->stats.fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5479 #ifdef IXGBE_FCOE
5480                 adapter->stats.fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5481                 adapter->stats.fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5482                 adapter->stats.fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5483                 adapter->stats.fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5484                 adapter->stats.fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5485                 adapter->stats.fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5486 #endif /* IXGBE_FCOE */
5487         } else {
5488                 adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5489                 adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
5490                 adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5491                 adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5492                 adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5493         }
5494         bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5495         adapter->stats.bprc += bprc;
5496         adapter->stats.mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5497         if (hw->mac.type == ixgbe_mac_82598EB)
5498                 adapter->stats.mprc -= bprc;
5499         adapter->stats.roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5500         adapter->stats.prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5501         adapter->stats.prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5502         adapter->stats.prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5503         adapter->stats.prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5504         adapter->stats.prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5505         adapter->stats.prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5506         adapter->stats.rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5507         lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5508         adapter->stats.lxontxc += lxon;
5509         lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5510         adapter->stats.lxofftxc += lxoff;
5511         adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5512         adapter->stats.gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5513         adapter->stats.mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5514         /*
5515          * 82598 errata - tx of flow control packets is included in tx counters
5516          */
5517         xon_off_tot = lxon + lxoff;
5518         adapter->stats.gptc -= xon_off_tot;
5519         adapter->stats.mptc -= xon_off_tot;
5520         adapter->stats.gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5521         adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5522         adapter->stats.rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5523         adapter->stats.rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5524         adapter->stats.tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5525         adapter->stats.ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5526         adapter->stats.ptc64 -= xon_off_tot;
5527         adapter->stats.ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5528         adapter->stats.ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5529         adapter->stats.ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5530         adapter->stats.ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5531         adapter->stats.ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5532         adapter->stats.bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5533
5534         /* Fill out the OS statistics structure */
5535         netdev->stats.multicast = adapter->stats.mprc;
5536
5537         /* Rx Errors */
5538         netdev->stats.rx_errors = adapter->stats.crcerrs +
5539                                        adapter->stats.rlec;
5540         netdev->stats.rx_dropped = 0;
5541         netdev->stats.rx_length_errors = adapter->stats.rlec;
5542         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
5543         netdev->stats.rx_missed_errors = total_mpc;
5544 }
5545
5546 /**
5547  * ixgbe_watchdog - Timer Call-back
5548  * @data: pointer to adapter cast into an unsigned long
5549  **/
5550 static void ixgbe_watchdog(unsigned long data)
5551 {
5552         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5553         struct ixgbe_hw *hw = &adapter->hw;
5554         u64 eics = 0;
5555         int i;
5556
5557         /*
5558          *  Do the watchdog outside of interrupt context due to the lovely
5559          * delays that some of the newer hardware requires
5560          */
5561
5562         if (test_bit(__IXGBE_DOWN, &adapter->state))
5563                 goto watchdog_short_circuit;
5564
5565         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5566                 /*
5567                  * for legacy and MSI interrupts don't set any bits
5568                  * that are enabled for EIAM, because this operation
5569                  * would set *both* EIMS and EICS for any bit in EIAM
5570                  */
5571                 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5572                         (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5573                 goto watchdog_reschedule;
5574         }
5575
5576         /* get one bit for every active tx/rx interrupt vector */
5577         for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5578                 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5579                 if (qv->rxr_count || qv->txr_count)
5580                         eics |= ((u64)1 << i);
5581         }
5582
5583         /* Cause software interrupt to ensure rx rings are cleaned */
5584         ixgbe_irq_rearm_queues(adapter, eics);
5585
5586 watchdog_reschedule:
5587         /* Reset the timer */
5588         mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
5589
5590 watchdog_short_circuit:
5591         schedule_work(&adapter->watchdog_task);
5592 }
5593
5594 /**
5595  * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
5596  * @work: pointer to work_struct containing our data
5597  **/
5598 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
5599 {
5600         struct ixgbe_adapter *adapter = container_of(work,
5601                                                      struct ixgbe_adapter,
5602                                                      multispeed_fiber_task);
5603         struct ixgbe_hw *hw = &adapter->hw;
5604         u32 autoneg;
5605         bool negotiation;
5606
5607         adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
5608         autoneg = hw->phy.autoneg_advertised;
5609         if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5610                 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5611         hw->mac.autotry_restart = false;
5612         if (hw->mac.ops.setup_link)
5613                 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5614         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5615         adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
5616 }
5617
5618 /**
5619  * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
5620  * @work: pointer to work_struct containing our data
5621  **/
5622 static void ixgbe_sfp_config_module_task(struct work_struct *work)
5623 {
5624         struct ixgbe_adapter *adapter = container_of(work,
5625                                                      struct ixgbe_adapter,
5626                                                      sfp_config_module_task);
5627         struct ixgbe_hw *hw = &adapter->hw;
5628         u32 err;
5629
5630         adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
5631
5632         /* Time for electrical oscillations to settle down */
5633         msleep(100);
5634         err = hw->phy.ops.identify_sfp(hw);
5635
5636         if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5637                 e_dev_err("failed to initialize because an unsupported SFP+ "
5638                           "module type was detected.\n");
5639                 e_dev_err("Reload the driver after installing a supported "
5640                           "module.\n");
5641                 unregister_netdev(adapter->netdev);
5642                 return;
5643         }
5644         hw->mac.ops.setup_sfp(hw);
5645
5646         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
5647                 /* This will also work for DA Twinax connections */
5648                 schedule_work(&adapter->multispeed_fiber_task);
5649         adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
5650 }
5651
5652 /**
5653  * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
5654  * @work: pointer to work_struct containing our data
5655  **/
5656 static void ixgbe_fdir_reinit_task(struct work_struct *work)
5657 {
5658         struct ixgbe_adapter *adapter = container_of(work,
5659                                                      struct ixgbe_adapter,
5660                                                      fdir_reinit_task);
5661         struct ixgbe_hw *hw = &adapter->hw;
5662         int i;
5663
5664         if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5665                 for (i = 0; i < adapter->num_tx_queues; i++)
5666                         set_bit(__IXGBE_FDIR_INIT_DONE,
5667                                 &(adapter->tx_ring[i]->reinit_state));
5668         } else {
5669                 e_err(probe, "failed to finish FDIR re-initialization, "
5670                       "ignored adding FDIR ATR filters\n");
5671         }
5672         /* Done FDIR Re-initialization, enable transmits */
5673         netif_tx_start_all_queues(adapter->netdev);
5674 }
5675
5676 static DEFINE_MUTEX(ixgbe_watchdog_lock);
5677
5678 /**
5679  * ixgbe_watchdog_task - worker thread to bring link up
5680  * @work: pointer to work_struct containing our data
5681  **/
5682 static void ixgbe_watchdog_task(struct work_struct *work)
5683 {
5684         struct ixgbe_adapter *adapter = container_of(work,
5685                                                      struct ixgbe_adapter,
5686                                                      watchdog_task);
5687         struct net_device *netdev = adapter->netdev;
5688         struct ixgbe_hw *hw = &adapter->hw;
5689         u32 link_speed;
5690         bool link_up;
5691         int i;
5692         struct ixgbe_ring *tx_ring;
5693         int some_tx_pending = 0;
5694
5695         mutex_lock(&ixgbe_watchdog_lock);
5696
5697         link_up = adapter->link_up;
5698         link_speed = adapter->link_speed;
5699
5700         if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
5701                 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5702                 if (link_up) {
5703 #ifdef CONFIG_DCB
5704                         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5705                                 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5706                                         hw->mac.ops.fc_enable(hw, i);
5707                         } else {
5708                                 hw->mac.ops.fc_enable(hw, 0);
5709                         }
5710 #else
5711                         hw->mac.ops.fc_enable(hw, 0);
5712 #endif
5713                 }
5714
5715                 if (link_up ||
5716                     time_after(jiffies, (adapter->link_check_timeout +
5717                                          IXGBE_TRY_LINK_TIMEOUT))) {
5718                         adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5719                         IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5720                 }
5721                 adapter->link_up = link_up;
5722                 adapter->link_speed = link_speed;
5723         }
5724
5725         if (link_up) {
5726                 if (!netif_carrier_ok(netdev)) {
5727                         bool flow_rx, flow_tx;
5728
5729                         if (hw->mac.type == ixgbe_mac_82599EB) {
5730                                 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5731                                 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5732                                 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5733                                 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5734                         } else {
5735                                 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5736                                 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5737                                 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5738                                 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5739                         }
5740
5741                         e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5742                                (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5743                                "10 Gbps" :
5744                                (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5745                                "1 Gbps" : "unknown speed")),
5746                                ((flow_rx && flow_tx) ? "RX/TX" :
5747                                (flow_rx ? "RX" :
5748                                (flow_tx ? "TX" : "None"))));
5749
5750                         netif_carrier_on(netdev);
5751                 } else {
5752                         /* Force detection of hung controller */
5753                         adapter->detect_tx_hung = true;
5754                 }
5755         } else {
5756                 adapter->link_up = false;
5757                 adapter->link_speed = 0;
5758                 if (netif_carrier_ok(netdev)) {
5759                         e_info(drv, "NIC Link is Down\n");
5760                         netif_carrier_off(netdev);
5761                 }
5762         }
5763
5764         if (!netif_carrier_ok(netdev)) {
5765                 for (i = 0; i < adapter->num_tx_queues; i++) {
5766                         tx_ring = adapter->tx_ring[i];
5767                         if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5768                                 some_tx_pending = 1;
5769                                 break;
5770                         }
5771                 }
5772
5773                 if (some_tx_pending) {
5774                         /* We've lost link, so the controller stops DMA,
5775                          * but we've got queued Tx work that's never going
5776                          * to get done, so reset controller to flush Tx.
5777                          * (Do the reset outside of interrupt context).
5778                          */
5779                          schedule_work(&adapter->reset_task);
5780                 }
5781         }
5782
5783         ixgbe_update_stats(adapter);
5784         mutex_unlock(&ixgbe_watchdog_lock);
5785 }
5786
5787 static int ixgbe_tso(struct ixgbe_adapter *adapter,
5788                      struct ixgbe_ring *tx_ring, struct sk_buff *skb,
5789                      u32 tx_flags, u8 *hdr_len)
5790 {
5791         struct ixgbe_adv_tx_context_desc *context_desc;
5792         unsigned int i;
5793         int err;
5794         struct ixgbe_tx_buffer *tx_buffer_info;
5795         u32 vlan_macip_lens = 0, type_tucmd_mlhl;
5796         u32 mss_l4len_idx, l4len;
5797
5798         if (skb_is_gso(skb)) {
5799                 if (skb_header_cloned(skb)) {
5800                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5801                         if (err)
5802                                 return err;
5803                 }
5804                 l4len = tcp_hdrlen(skb);
5805                 *hdr_len += l4len;
5806
5807                 if (skb->protocol == htons(ETH_P_IP)) {
5808                         struct iphdr *iph = ip_hdr(skb);
5809                         iph->tot_len = 0;
5810                         iph->check = 0;
5811                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5812                                                                  iph->daddr, 0,
5813                                                                  IPPROTO_TCP,
5814                                                                  0);
5815                 } else if (skb_is_gso_v6(skb)) {
5816                         ipv6_hdr(skb)->payload_len = 0;
5817                         tcp_hdr(skb)->check =
5818                             ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5819                                              &ipv6_hdr(skb)->daddr,
5820                                              0, IPPROTO_TCP, 0);
5821                 }
5822
5823                 i = tx_ring->next_to_use;
5824
5825                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5826                 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
5827
5828                 /* VLAN MACLEN IPLEN */
5829                 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5830                         vlan_macip_lens |=
5831                             (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5832                 vlan_macip_lens |= ((skb_network_offset(skb)) <<
5833                                     IXGBE_ADVTXD_MACLEN_SHIFT);
5834                 *hdr_len += skb_network_offset(skb);
5835                 vlan_macip_lens |=
5836                     (skb_transport_header(skb) - skb_network_header(skb));
5837                 *hdr_len +=
5838                     (skb_transport_header(skb) - skb_network_header(skb));
5839                 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5840                 context_desc->seqnum_seed = 0;
5841
5842                 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5843                 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
5844                                    IXGBE_ADVTXD_DTYP_CTXT);
5845
5846                 if (skb->protocol == htons(ETH_P_IP))
5847                         type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5848                 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5849                 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5850
5851                 /* MSS L4LEN IDX */
5852                 mss_l4len_idx =
5853                     (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
5854                 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
5855                 /* use index 1 for TSO */
5856                 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5857                 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
5858
5859                 tx_buffer_info->time_stamp = jiffies;
5860                 tx_buffer_info->next_to_watch = i;
5861
5862                 i++;
5863                 if (i == tx_ring->count)
5864                         i = 0;
5865                 tx_ring->next_to_use = i;
5866
5867                 return true;
5868         }
5869         return false;
5870 }
5871
5872 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
5873                           struct ixgbe_ring *tx_ring,
5874                           struct sk_buff *skb, u32 tx_flags)
5875 {
5876         struct ixgbe_adv_tx_context_desc *context_desc;
5877         unsigned int i;
5878         struct ixgbe_tx_buffer *tx_buffer_info;
5879         u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
5880
5881         if (skb->ip_summed == CHECKSUM_PARTIAL ||
5882             (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
5883                 i = tx_ring->next_to_use;
5884                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5885                 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
5886
5887                 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5888                         vlan_macip_lens |=
5889                             (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5890                 vlan_macip_lens |= (skb_network_offset(skb) <<
5891                                     IXGBE_ADVTXD_MACLEN_SHIFT);
5892                 if (skb->ip_summed == CHECKSUM_PARTIAL)
5893                         vlan_macip_lens |= (skb_transport_header(skb) -
5894                                             skb_network_header(skb));
5895
5896                 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5897                 context_desc->seqnum_seed = 0;
5898
5899                 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
5900                                     IXGBE_ADVTXD_DTYP_CTXT);
5901
5902                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
5903                         __be16 protocol;
5904
5905                         if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
5906                                 const struct vlan_ethhdr *vhdr =
5907                                         (const struct vlan_ethhdr *)skb->data;
5908
5909                                 protocol = vhdr->h_vlan_encapsulated_proto;
5910                         } else {
5911                                 protocol = skb->protocol;
5912                         }
5913
5914                         switch (protocol) {
5915                         case cpu_to_be16(ETH_P_IP):
5916                                 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5917                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
5918                                         type_tucmd_mlhl |=
5919                                                 IXGBE_ADVTXD_TUCMD_L4T_TCP;
5920                                 else if (ip_hdr(skb)->protocol == IPPROTO_SCTP)
5921                                         type_tucmd_mlhl |=
5922                                                 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5923                                 break;
5924                         case cpu_to_be16(ETH_P_IPV6):
5925                                 /* XXX what about other V6 headers?? */
5926                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
5927                                         type_tucmd_mlhl |=
5928                                                 IXGBE_ADVTXD_TUCMD_L4T_TCP;
5929                                 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_SCTP)
5930                                         type_tucmd_mlhl |=
5931                                                 IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5932                                 break;
5933                         default:
5934                                 if (unlikely(net_ratelimit())) {
5935                                         e_warn(probe, "partial checksum "
5936                                                "but proto=%x!\n",
5937                                                skb->protocol);
5938                                 }
5939                                 break;
5940                         }
5941                 }
5942
5943                 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5944                 /* use index zero for tx checksum offload */
5945                 context_desc->mss_l4len_idx = 0;
5946
5947                 tx_buffer_info->time_stamp = jiffies;
5948                 tx_buffer_info->next_to_watch = i;
5949
5950                 i++;
5951                 if (i == tx_ring->count)
5952                         i = 0;
5953                 tx_ring->next_to_use = i;
5954
5955                 return true;
5956         }
5957
5958         return false;
5959 }
5960
5961 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
5962                         struct ixgbe_ring *tx_ring,
5963                         struct sk_buff *skb, u32 tx_flags,
5964                         unsigned int first)
5965 {
5966         struct pci_dev *pdev = adapter->pdev;
5967         struct ixgbe_tx_buffer *tx_buffer_info;
5968         unsigned int len;
5969         unsigned int total = skb->len;
5970         unsigned int offset = 0, size, count = 0, i;
5971         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
5972         unsigned int f;
5973
5974         i = tx_ring->next_to_use;
5975
5976         if (tx_flags & IXGBE_TX_FLAGS_FCOE)
5977                 /* excluding fcoe_crc_eof for FCoE */
5978                 total -= sizeof(struct fcoe_crc_eof);
5979
5980         len = min(skb_headlen(skb), total);
5981         while (len) {
5982                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5983                 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
5984
5985                 tx_buffer_info->length = size;
5986                 tx_buffer_info->mapped_as_page = false;
5987                 tx_buffer_info->dma = dma_map_single(&pdev->dev,
5988                                                      skb->data + offset,
5989                                                      size, DMA_TO_DEVICE);
5990                 if (dma_mapping_error(&pdev->dev, tx_buffer_info->dma))
5991                         goto dma_error;
5992                 tx_buffer_info->time_stamp = jiffies;
5993                 tx_buffer_info->next_to_watch = i;
5994
5995                 len -= size;
5996                 total -= size;
5997                 offset += size;
5998                 count++;
5999
6000                 if (len) {
6001                         i++;
6002                         if (i == tx_ring->count)
6003                                 i = 0;
6004                 }
6005         }
6006
6007         for (f = 0; f < nr_frags; f++) {
6008                 struct skb_frag_struct *frag;
6009
6010                 frag = &skb_shinfo(skb)->frags[f];
6011                 len = min((unsigned int)frag->size, total);
6012                 offset = frag->page_offset;
6013
6014                 while (len) {
6015                         i++;
6016                         if (i == tx_ring->count)
6017                                 i = 0;
6018
6019                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
6020                         size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6021
6022                         tx_buffer_info->length = size;
6023                         tx_buffer_info->dma = dma_map_page(&adapter->pdev->dev,
6024                                                            frag->page,
6025                                                            offset, size,
6026                                                            DMA_TO_DEVICE);
6027                         tx_buffer_info->mapped_as_page = true;
6028                         if (dma_mapping_error(&pdev->dev, tx_buffer_info->dma))
6029                                 goto dma_error;
6030                         tx_buffer_info->time_stamp = jiffies;
6031                         tx_buffer_info->next_to_watch = i;
6032
6033                         len -= size;
6034                         total -= size;
6035                         offset += size;
6036                         count++;
6037                 }
6038                 if (total == 0)
6039                         break;
6040         }
6041
6042         tx_ring->tx_buffer_info[i].skb = skb;
6043         tx_ring->tx_buffer_info[first].next_to_watch = i;
6044
6045         return count;
6046
6047 dma_error:
6048         e_dev_err("TX DMA map failed\n");
6049
6050         /* clear timestamp and dma mappings for failed tx_buffer_info map */
6051         tx_buffer_info->dma = 0;
6052         tx_buffer_info->time_stamp = 0;
6053         tx_buffer_info->next_to_watch = 0;
6054         if (count)
6055                 count--;
6056
6057         /* clear timestamp and dma mappings for remaining portion of packet */
6058         while (count--) {
6059                 if (i == 0)
6060                         i += tx_ring->count;
6061                 i--;
6062                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6063                 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
6064         }
6065
6066         return 0;
6067 }
6068
6069 static void ixgbe_tx_queue(struct ixgbe_adapter *adapter,
6070                            struct ixgbe_ring *tx_ring,
6071                            int tx_flags, int count, u32 paylen, u8 hdr_len)
6072 {
6073         union ixgbe_adv_tx_desc *tx_desc = NULL;
6074         struct ixgbe_tx_buffer *tx_buffer_info;
6075         u32 olinfo_status = 0, cmd_type_len = 0;
6076         unsigned int i;
6077         u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
6078
6079         cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
6080
6081         cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
6082
6083         if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6084                 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
6085
6086         if (tx_flags & IXGBE_TX_FLAGS_TSO) {
6087                 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6088
6089                 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6090                                  IXGBE_ADVTXD_POPTS_SHIFT;
6091
6092                 /* use index 1 context for tso */
6093                 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6094                 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6095                         olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
6096                                          IXGBE_ADVTXD_POPTS_SHIFT;
6097
6098         } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6099                 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6100                                  IXGBE_ADVTXD_POPTS_SHIFT;
6101
6102         if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6103                 olinfo_status |= IXGBE_ADVTXD_CC;
6104                 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6105                 if (tx_flags & IXGBE_TX_FLAGS_FSO)
6106                         cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6107         }
6108
6109         olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
6110
6111         i = tx_ring->next_to_use;
6112         while (count--) {
6113                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6114                 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
6115                 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
6116                 tx_desc->read.cmd_type_len =
6117                         cpu_to_le32(cmd_type_len | tx_buffer_info->length);
6118                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6119                 i++;
6120                 if (i == tx_ring->count)
6121                         i = 0;
6122         }
6123
6124         tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
6125
6126         /*
6127          * Force memory writes to complete before letting h/w
6128          * know there are new descriptors to fetch.  (Only
6129          * applicable for weak-ordered memory model archs,
6130          * such as IA-64).
6131          */
6132         wmb();
6133
6134         tx_ring->next_to_use = i;
6135         writel(i, adapter->hw.hw_addr + tx_ring->tail);
6136 }
6137
6138 static void ixgbe_atr(struct ixgbe_adapter *adapter, struct sk_buff *skb,
6139                       int queue, u32 tx_flags)
6140 {
6141         struct ixgbe_atr_input atr_input;
6142         struct tcphdr *th;
6143         struct iphdr *iph = ip_hdr(skb);
6144         struct ethhdr *eth = (struct ethhdr *)skb->data;
6145         u16 vlan_id, src_port, dst_port, flex_bytes;
6146         u32 src_ipv4_addr, dst_ipv4_addr;
6147         u8 l4type = 0;
6148
6149         /* Right now, we support IPv4 only */
6150         if (skb->protocol != htons(ETH_P_IP))
6151                 return;
6152         /* check if we're UDP or TCP */
6153         if (iph->protocol == IPPROTO_TCP) {
6154                 th = tcp_hdr(skb);
6155                 src_port = th->source;
6156                 dst_port = th->dest;
6157                 l4type |= IXGBE_ATR_L4TYPE_TCP;
6158                 /* l4type IPv4 type is 0, no need to assign */
6159         } else {
6160                 /* Unsupported L4 header, just bail here */
6161                 return;
6162         }
6163
6164         memset(&atr_input, 0, sizeof(struct ixgbe_atr_input));
6165
6166         vlan_id = (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK) >>
6167                    IXGBE_TX_FLAGS_VLAN_SHIFT;
6168         src_ipv4_addr = iph->saddr;
6169         dst_ipv4_addr = iph->daddr;
6170         flex_bytes = eth->h_proto;
6171
6172         ixgbe_atr_set_vlan_id_82599(&atr_input, vlan_id);
6173         ixgbe_atr_set_src_port_82599(&atr_input, dst_port);
6174         ixgbe_atr_set_dst_port_82599(&atr_input, src_port);
6175         ixgbe_atr_set_flex_byte_82599(&atr_input, flex_bytes);
6176         ixgbe_atr_set_l4type_82599(&atr_input, l4type);
6177         /* src and dst are inverted, think how the receiver sees them */
6178         ixgbe_atr_set_src_ipv4_82599(&atr_input, dst_ipv4_addr);
6179         ixgbe_atr_set_dst_ipv4_82599(&atr_input, src_ipv4_addr);
6180
6181         /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6182         ixgbe_fdir_add_signature_filter_82599(&adapter->hw, &atr_input, queue);
6183 }
6184
6185 static int __ixgbe_maybe_stop_tx(struct net_device *netdev,
6186                                  struct ixgbe_ring *tx_ring, int size)
6187 {
6188         netif_stop_subqueue(netdev, tx_ring->queue_index);
6189         /* Herbert's original patch had:
6190          *  smp_mb__after_netif_stop_queue();
6191          * but since that doesn't exist yet, just open code it. */
6192         smp_mb();
6193
6194         /* We need to check again in a case another CPU has just
6195          * made room available. */
6196         if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
6197                 return -EBUSY;
6198
6199         /* A reprieve! - use start_queue because it doesn't call schedule */
6200         netif_start_subqueue(netdev, tx_ring->queue_index);
6201         ++tx_ring->restart_queue;
6202         return 0;
6203 }
6204
6205 static int ixgbe_maybe_stop_tx(struct net_device *netdev,
6206                               struct ixgbe_ring *tx_ring, int size)
6207 {
6208         if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
6209                 return 0;
6210         return __ixgbe_maybe_stop_tx(netdev, tx_ring, size);
6211 }
6212
6213 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6214 {
6215         struct ixgbe_adapter *adapter = netdev_priv(dev);
6216         int txq = smp_processor_id();
6217
6218 #ifdef IXGBE_FCOE
6219         if ((skb->protocol == htons(ETH_P_FCOE)) ||
6220             (skb->protocol == htons(ETH_P_FIP))) {
6221                 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
6222                         txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6223                         txq += adapter->ring_feature[RING_F_FCOE].mask;
6224                         return txq;
6225 #ifdef CONFIG_IXGBE_DCB
6226                 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6227                         txq = adapter->fcoe.up;
6228                         return txq;
6229 #endif
6230                 }
6231         }
6232 #endif
6233
6234         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6235                 while (unlikely(txq >= dev->real_num_tx_queues))
6236                         txq -= dev->real_num_tx_queues;
6237                 return txq;
6238         }
6239
6240         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6241                 if (skb->priority == TC_PRIO_CONTROL)
6242                         txq = adapter->ring_feature[RING_F_DCB].indices-1;
6243                 else
6244                         txq = (skb->vlan_tci & IXGBE_TX_FLAGS_VLAN_PRIO_MASK)
6245                                >> 13;
6246                 return txq;
6247         }
6248
6249         return skb_tx_hash(dev, skb);
6250 }
6251
6252 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb, struct net_device *netdev,
6253                           struct ixgbe_adapter *adapter,
6254                           struct ixgbe_ring *tx_ring)
6255 {
6256         struct netdev_queue *txq;
6257         unsigned int first;
6258         unsigned int tx_flags = 0;
6259         u8 hdr_len = 0;
6260         int tso;
6261         int count = 0;
6262         unsigned int f;
6263
6264         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
6265                 tx_flags |= vlan_tx_tag_get(skb);
6266                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6267                         tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6268                         tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6269                 }
6270                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6271                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6272         } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED &&
6273                    skb->priority != TC_PRIO_CONTROL) {
6274                 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6275                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6276                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6277         }
6278
6279 #ifdef IXGBE_FCOE
6280         /* for FCoE with DCB, we force the priority to what
6281          * was specified by the switch */
6282         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED &&
6283             (skb->protocol == htons(ETH_P_FCOE) ||
6284              skb->protocol == htons(ETH_P_FIP))) {
6285 #ifdef CONFIG_IXGBE_DCB
6286                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6287                         tx_flags &= ~(IXGBE_TX_FLAGS_VLAN_PRIO_MASK
6288                                       << IXGBE_TX_FLAGS_VLAN_SHIFT);
6289                         tx_flags |= ((adapter->fcoe.up << 13)
6290                                       << IXGBE_TX_FLAGS_VLAN_SHIFT);
6291                 }
6292 #endif
6293                 /* flag for FCoE offloads */
6294                 if (skb->protocol == htons(ETH_P_FCOE))
6295                         tx_flags |= IXGBE_TX_FLAGS_FCOE;
6296         }
6297 #endif
6298
6299         /* four things can cause us to need a context descriptor */
6300         if (skb_is_gso(skb) ||
6301             (skb->ip_summed == CHECKSUM_PARTIAL) ||
6302             (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
6303             (tx_flags & IXGBE_TX_FLAGS_FCOE))
6304                 count++;
6305
6306         count += TXD_USE_COUNT(skb_headlen(skb));
6307         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6308                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6309
6310         if (ixgbe_maybe_stop_tx(netdev, tx_ring, count)) {
6311                 adapter->tx_busy++;
6312                 return NETDEV_TX_BUSY;
6313         }
6314
6315         first = tx_ring->next_to_use;
6316         if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6317 #ifdef IXGBE_FCOE
6318                 /* setup tx offload for FCoE */
6319                 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
6320                 if (tso < 0) {
6321                         dev_kfree_skb_any(skb);
6322                         return NETDEV_TX_OK;
6323                 }
6324                 if (tso)
6325                         tx_flags |= IXGBE_TX_FLAGS_FSO;
6326 #endif /* IXGBE_FCOE */
6327         } else {
6328                 if (skb->protocol == htons(ETH_P_IP))
6329                         tx_flags |= IXGBE_TX_FLAGS_IPV4;
6330                 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
6331                 if (tso < 0) {
6332                         dev_kfree_skb_any(skb);
6333                         return NETDEV_TX_OK;
6334                 }
6335
6336                 if (tso)
6337                         tx_flags |= IXGBE_TX_FLAGS_TSO;
6338                 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags) &&
6339                          (skb->ip_summed == CHECKSUM_PARTIAL))
6340                         tx_flags |= IXGBE_TX_FLAGS_CSUM;
6341         }
6342
6343         count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first);
6344         if (count) {
6345                 /* add the ATR filter if ATR is on */
6346                 if (tx_ring->atr_sample_rate) {
6347                         ++tx_ring->atr_count;
6348                         if ((tx_ring->atr_count >= tx_ring->atr_sample_rate) &&
6349                              test_bit(__IXGBE_FDIR_INIT_DONE,
6350                                       &tx_ring->reinit_state)) {
6351                                 ixgbe_atr(adapter, skb, tx_ring->queue_index,
6352                                           tx_flags);
6353                                 tx_ring->atr_count = 0;
6354                         }
6355                 }
6356                 txq = netdev_get_tx_queue(netdev, tx_ring->queue_index);
6357                 txq->tx_bytes += skb->len;
6358                 txq->tx_packets++;
6359                 ixgbe_tx_queue(adapter, tx_ring, tx_flags, count, skb->len,
6360                                hdr_len);
6361                 ixgbe_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
6362
6363         } else {
6364                 dev_kfree_skb_any(skb);
6365                 tx_ring->tx_buffer_info[first].time_stamp = 0;
6366                 tx_ring->next_to_use = first;
6367         }
6368
6369         return NETDEV_TX_OK;
6370 }
6371
6372 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
6373 {
6374         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6375         struct ixgbe_ring *tx_ring;
6376
6377         tx_ring = adapter->tx_ring[skb->queue_mapping];
6378         return ixgbe_xmit_frame_ring(skb, netdev, adapter, tx_ring);
6379 }
6380
6381 /**
6382  * ixgbe_set_mac - Change the Ethernet Address of the NIC
6383  * @netdev: network interface device structure
6384  * @p: pointer to an address structure
6385  *
6386  * Returns 0 on success, negative on failure
6387  **/
6388 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6389 {
6390         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6391         struct ixgbe_hw *hw = &adapter->hw;
6392         struct sockaddr *addr = p;
6393
6394         if (!is_valid_ether_addr(addr->sa_data))
6395                 return -EADDRNOTAVAIL;
6396
6397         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6398         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6399
6400         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6401                             IXGBE_RAH_AV);
6402
6403         return 0;
6404 }
6405
6406 static int
6407 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6408 {
6409         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6410         struct ixgbe_hw *hw = &adapter->hw;
6411         u16 value;
6412         int rc;
6413
6414         if (prtad != hw->phy.mdio.prtad)
6415                 return -EINVAL;
6416         rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6417         if (!rc)
6418                 rc = value;
6419         return rc;
6420 }
6421
6422 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6423                             u16 addr, u16 value)
6424 {
6425         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6426         struct ixgbe_hw *hw = &adapter->hw;
6427
6428         if (prtad != hw->phy.mdio.prtad)
6429                 return -EINVAL;
6430         return hw->phy.ops.write_reg(hw, addr, devad, value);
6431 }
6432
6433 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6434 {
6435         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6436
6437         return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6438 }
6439
6440 /**
6441  * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6442  * netdev->dev_addrs
6443  * @netdev: network interface device structure
6444  *
6445  * Returns non-zero on failure
6446  **/
6447 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6448 {
6449         int err = 0;
6450         struct ixgbe_adapter *adapter = netdev_priv(dev);
6451         struct ixgbe_mac_info *mac = &adapter->hw.mac;
6452
6453         if (is_valid_ether_addr(mac->san_addr)) {
6454                 rtnl_lock();
6455                 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6456                 rtnl_unlock();
6457         }
6458         return err;
6459 }
6460
6461 /**
6462  * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6463  * netdev->dev_addrs
6464  * @netdev: network interface device structure
6465  *
6466  * Returns non-zero on failure
6467  **/
6468 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6469 {
6470         int err = 0;
6471         struct ixgbe_adapter *adapter = netdev_priv(dev);
6472         struct ixgbe_mac_info *mac = &adapter->hw.mac;
6473
6474         if (is_valid_ether_addr(mac->san_addr)) {
6475                 rtnl_lock();
6476                 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6477                 rtnl_unlock();
6478         }
6479         return err;
6480 }
6481
6482 #ifdef CONFIG_NET_POLL_CONTROLLER
6483 /*
6484  * Polling 'interrupt' - used by things like netconsole to send skbs
6485  * without having to re-enable interrupts. It's not called while
6486  * the interrupt routine is executing.
6487  */
6488 static void ixgbe_netpoll(struct net_device *netdev)
6489 {
6490         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6491         int i;
6492
6493         /* if interface is down do nothing */
6494         if (test_bit(__IXGBE_DOWN, &adapter->state))
6495                 return;
6496
6497         adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6498         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6499                 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
6500                 for (i = 0; i < num_q_vectors; i++) {
6501                         struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
6502                         ixgbe_msix_clean_many(0, q_vector);
6503                 }
6504         } else {
6505                 ixgbe_intr(adapter->pdev->irq, netdev);
6506         }
6507         adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6508 }
6509 #endif
6510
6511 static const struct net_device_ops ixgbe_netdev_ops = {
6512         .ndo_open               = ixgbe_open,
6513         .ndo_stop               = ixgbe_close,
6514         .ndo_start_xmit         = ixgbe_xmit_frame,
6515         .ndo_select_queue       = ixgbe_select_queue,
6516         .ndo_set_rx_mode        = ixgbe_set_rx_mode,
6517         .ndo_set_multicast_list = ixgbe_set_rx_mode,
6518         .ndo_validate_addr      = eth_validate_addr,
6519         .ndo_set_mac_address    = ixgbe_set_mac,
6520         .ndo_change_mtu         = ixgbe_change_mtu,
6521         .ndo_tx_timeout         = ixgbe_tx_timeout,
6522         .ndo_vlan_rx_register   = ixgbe_vlan_rx_register,
6523         .ndo_vlan_rx_add_vid    = ixgbe_vlan_rx_add_vid,
6524         .ndo_vlan_rx_kill_vid   = ixgbe_vlan_rx_kill_vid,
6525         .ndo_do_ioctl           = ixgbe_ioctl,
6526         .ndo_set_vf_mac         = ixgbe_ndo_set_vf_mac,
6527         .ndo_set_vf_vlan        = ixgbe_ndo_set_vf_vlan,
6528         .ndo_set_vf_tx_rate     = ixgbe_ndo_set_vf_bw,
6529         .ndo_get_vf_config      = ixgbe_ndo_get_vf_config,
6530 #ifdef CONFIG_NET_POLL_CONTROLLER
6531         .ndo_poll_controller    = ixgbe_netpoll,
6532 #endif
6533 #ifdef IXGBE_FCOE
6534         .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
6535         .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
6536         .ndo_fcoe_enable = ixgbe_fcoe_enable,
6537         .ndo_fcoe_disable = ixgbe_fcoe_disable,
6538         .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
6539 #endif /* IXGBE_FCOE */
6540 };
6541
6542 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
6543                            const struct ixgbe_info *ii)
6544 {
6545 #ifdef CONFIG_PCI_IOV
6546         struct ixgbe_hw *hw = &adapter->hw;
6547         int err;
6548
6549         if (hw->mac.type != ixgbe_mac_82599EB || !max_vfs)
6550                 return;
6551
6552         /* The 82599 supports up to 64 VFs per physical function
6553          * but this implementation limits allocation to 63 so that
6554          * basic networking resources are still available to the
6555          * physical function
6556          */
6557         adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
6558         adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
6559         err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
6560         if (err) {
6561                 e_err(probe, "Failed to enable PCI sriov: %d\n", err);
6562                 goto err_novfs;
6563         }
6564         /* If call to enable VFs succeeded then allocate memory
6565          * for per VF control structures.
6566          */
6567         adapter->vfinfo =
6568                 kcalloc(adapter->num_vfs,
6569                         sizeof(struct vf_data_storage), GFP_KERNEL);
6570         if (adapter->vfinfo) {
6571                 /* Now that we're sure SR-IOV is enabled
6572                  * and memory allocated set up the mailbox parameters
6573                  */
6574                 ixgbe_init_mbx_params_pf(hw);
6575                 memcpy(&hw->mbx.ops, ii->mbx_ops,
6576                        sizeof(hw->mbx.ops));
6577
6578                 /* Disable RSC when in SR-IOV mode */
6579                 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
6580                                      IXGBE_FLAG2_RSC_ENABLED);
6581                 return;
6582         }
6583
6584         /* Oh oh */
6585         e_err(probe, "Unable to allocate memory for VF Data Storage - "
6586               "SRIOV disabled\n");
6587         pci_disable_sriov(adapter->pdev);
6588
6589 err_novfs:
6590         adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
6591         adapter->num_vfs = 0;
6592 #endif /* CONFIG_PCI_IOV */
6593 }
6594
6595 /**
6596  * ixgbe_probe - Device Initialization Routine
6597  * @pdev: PCI device information struct
6598  * @ent: entry in ixgbe_pci_tbl
6599  *
6600  * Returns 0 on success, negative on failure
6601  *
6602  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
6603  * The OS initialization, configuring of the adapter private structure,
6604  * and a hardware reset occur.
6605  **/
6606 static int __devinit ixgbe_probe(struct pci_dev *pdev,
6607                                  const struct pci_device_id *ent)
6608 {
6609         struct net_device *netdev;
6610         struct ixgbe_adapter *adapter = NULL;
6611         struct ixgbe_hw *hw;
6612         const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
6613         static int cards_found;
6614         int i, err, pci_using_dac;
6615         unsigned int indices = num_possible_cpus();
6616 #ifdef IXGBE_FCOE
6617         u16 device_caps;
6618 #endif
6619         u32 part_num, eec;
6620
6621         /* Catch broken hardware that put the wrong VF device ID in
6622          * the PCIe SR-IOV capability.
6623          */
6624         if (pdev->is_virtfn) {
6625                 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
6626                      pci_name(pdev), pdev->vendor, pdev->device);
6627                 return -EINVAL;
6628         }
6629
6630         err = pci_enable_device_mem(pdev);
6631         if (err)
6632                 return err;
6633
6634         if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
6635             !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
6636                 pci_using_dac = 1;
6637         } else {
6638                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6639                 if (err) {
6640                         err = dma_set_coherent_mask(&pdev->dev,
6641                                                     DMA_BIT_MASK(32));
6642                         if (err) {
6643                                 dev_err(&pdev->dev,
6644                                         "No usable DMA configuration, aborting\n");
6645                                 goto err_dma;
6646                         }
6647                 }
6648                 pci_using_dac = 0;
6649         }
6650
6651         err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
6652                                            IORESOURCE_MEM), ixgbe_driver_name);
6653         if (err) {
6654                 dev_err(&pdev->dev,
6655                         "pci_request_selected_regions failed 0x%x\n", err);
6656                 goto err_pci_reg;
6657         }
6658
6659         pci_enable_pcie_error_reporting(pdev);
6660
6661         pci_set_master(pdev);
6662         pci_save_state(pdev);
6663
6664         if (ii->mac == ixgbe_mac_82598EB)
6665                 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
6666         else
6667                 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
6668
6669         indices = max_t(unsigned int, indices, IXGBE_MAX_DCB_INDICES);
6670 #ifdef IXGBE_FCOE
6671         indices += min_t(unsigned int, num_possible_cpus(),
6672                          IXGBE_MAX_FCOE_INDICES);
6673 #endif
6674         netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
6675         if (!netdev) {
6676                 err = -ENOMEM;
6677                 goto err_alloc_etherdev;
6678         }
6679
6680         SET_NETDEV_DEV(netdev, &pdev->dev);
6681
6682         pci_set_drvdata(pdev, netdev);
6683         adapter = netdev_priv(netdev);
6684
6685         adapter->netdev = netdev;
6686         adapter->pdev = pdev;
6687         hw = &adapter->hw;
6688         hw->back = adapter;
6689         adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
6690
6691         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
6692                               pci_resource_len(pdev, 0));
6693         if (!hw->hw_addr) {
6694                 err = -EIO;
6695                 goto err_ioremap;
6696         }
6697
6698         for (i = 1; i <= 5; i++) {
6699                 if (pci_resource_len(pdev, i) == 0)
6700                         continue;
6701         }
6702
6703         netdev->netdev_ops = &ixgbe_netdev_ops;
6704         ixgbe_set_ethtool_ops(netdev);
6705         netdev->watchdog_timeo = 5 * HZ;
6706         strcpy(netdev->name, pci_name(pdev));
6707
6708         adapter->bd_number = cards_found;
6709
6710         /* Setup hw api */
6711         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
6712         hw->mac.type  = ii->mac;
6713
6714         /* EEPROM */
6715         memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
6716         eec = IXGBE_READ_REG(hw, IXGBE_EEC);
6717         /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
6718         if (!(eec & (1 << 8)))
6719                 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
6720
6721         /* PHY */
6722         memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
6723         hw->phy.sfp_type = ixgbe_sfp_type_unknown;
6724         /* ixgbe_identify_phy_generic will set prtad and mmds properly */
6725         hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
6726         hw->phy.mdio.mmds = 0;
6727         hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
6728         hw->phy.mdio.dev = netdev;
6729         hw->phy.mdio.mdio_read = ixgbe_mdio_read;
6730         hw->phy.mdio.mdio_write = ixgbe_mdio_write;
6731
6732         /* set up this timer and work struct before calling get_invariants
6733          * which might start the timer
6734          */
6735         init_timer(&adapter->sfp_timer);
6736         adapter->sfp_timer.function = ixgbe_sfp_timer;
6737         adapter->sfp_timer.data = (unsigned long) adapter;
6738
6739         INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
6740
6741         /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
6742         INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
6743
6744         /* a new SFP+ module arrival, called from GPI SDP2 context */
6745         INIT_WORK(&adapter->sfp_config_module_task,
6746                   ixgbe_sfp_config_module_task);
6747
6748         ii->get_invariants(hw);
6749
6750         /* setup the private structure */
6751         err = ixgbe_sw_init(adapter);
6752         if (err)
6753                 goto err_sw_init;
6754
6755         /* Make it possible the adapter to be woken up via WOL */
6756         if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6757                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6758
6759         /*
6760          * If there is a fan on this device and it has failed log the
6761          * failure.
6762          */
6763         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
6764                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
6765                 if (esdp & IXGBE_ESDP_SDP1)
6766                         e_crit(probe, "Fan has stopped, replace the adapter\n");
6767         }
6768
6769         /* reset_hw fills in the perm_addr as well */
6770         hw->phy.reset_if_overtemp = true;
6771         err = hw->mac.ops.reset_hw(hw);
6772         hw->phy.reset_if_overtemp = false;
6773         if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
6774             hw->mac.type == ixgbe_mac_82598EB) {
6775                 /*
6776                  * Start a kernel thread to watch for a module to arrive.
6777                  * Only do this for 82598, since 82599 will generate
6778                  * interrupts on module arrival.
6779                  */
6780                 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6781                 mod_timer(&adapter->sfp_timer,
6782                           round_jiffies(jiffies + (2 * HZ)));
6783                 err = 0;
6784         } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
6785                 e_dev_err("failed to initialize because an unsupported SFP+ "
6786                           "module type was detected.\n");
6787                 e_dev_err("Reload the driver after installing a supported "
6788                           "module.\n");
6789                 goto err_sw_init;
6790         } else if (err) {
6791                 e_dev_err("HW Init failed: %d\n", err);
6792                 goto err_sw_init;
6793         }
6794
6795         ixgbe_probe_vf(adapter, ii);
6796
6797         netdev->features = NETIF_F_SG |
6798                            NETIF_F_IP_CSUM |
6799                            NETIF_F_HW_VLAN_TX |
6800                            NETIF_F_HW_VLAN_RX |
6801                            NETIF_F_HW_VLAN_FILTER;
6802
6803         netdev->features |= NETIF_F_IPV6_CSUM;
6804         netdev->features |= NETIF_F_TSO;
6805         netdev->features |= NETIF_F_TSO6;
6806         netdev->features |= NETIF_F_GRO;
6807
6808         if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6809                 netdev->features |= NETIF_F_SCTP_CSUM;
6810
6811         netdev->vlan_features |= NETIF_F_TSO;
6812         netdev->vlan_features |= NETIF_F_TSO6;
6813         netdev->vlan_features |= NETIF_F_IP_CSUM;
6814         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
6815         netdev->vlan_features |= NETIF_F_SG;
6816
6817         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6818                 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
6819                                     IXGBE_FLAG_DCB_ENABLED);
6820         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
6821                 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
6822
6823 #ifdef CONFIG_IXGBE_DCB
6824         netdev->dcbnl_ops = &dcbnl_ops;
6825 #endif
6826
6827 #ifdef IXGBE_FCOE
6828         if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6829                 if (hw->mac.ops.get_device_caps) {
6830                         hw->mac.ops.get_device_caps(hw, &device_caps);
6831                         if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
6832                                 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6833                 }
6834         }
6835         if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6836                 netdev->vlan_features |= NETIF_F_FCOE_CRC;
6837                 netdev->vlan_features |= NETIF_F_FSO;
6838                 netdev->vlan_features |= NETIF_F_FCOE_MTU;
6839         }
6840 #endif /* IXGBE_FCOE */
6841         if (pci_using_dac)
6842                 netdev->features |= NETIF_F_HIGHDMA;
6843
6844         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6845                 netdev->features |= NETIF_F_LRO;
6846
6847         /* make sure the EEPROM is good */
6848         if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
6849                 e_dev_err("The EEPROM Checksum Is Not Valid\n");
6850                 err = -EIO;
6851                 goto err_eeprom;
6852         }
6853
6854         memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
6855         memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
6856
6857         if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
6858                 e_dev_err("invalid MAC address\n");
6859                 err = -EIO;
6860                 goto err_eeprom;
6861         }
6862
6863         /* power down the optics */
6864         if (hw->phy.multispeed_fiber)
6865                 hw->mac.ops.disable_tx_laser(hw);
6866
6867         init_timer(&adapter->watchdog_timer);
6868         adapter->watchdog_timer.function = ixgbe_watchdog;
6869         adapter->watchdog_timer.data = (unsigned long)adapter;
6870
6871         INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
6872         INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
6873
6874         err = ixgbe_init_interrupt_scheme(adapter);
6875         if (err)
6876                 goto err_sw_init;
6877
6878         switch (pdev->device) {
6879         case IXGBE_DEV_ID_82599_KX4:
6880                 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
6881                                 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
6882                 break;
6883         default:
6884                 adapter->wol = 0;
6885                 break;
6886         }
6887         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6888
6889         /* pick up the PCI bus settings for reporting later */
6890         hw->mac.ops.get_bus_info(hw);
6891
6892         /* print bus type/speed/width info */
6893         e_dev_info("(PCI Express:%s:%s) %pM\n",
6894                    (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0Gb/s" :
6895                     hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5Gb/s" :
6896                     "Unknown"),
6897                    (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
6898                     hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
6899                     hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
6900                     "Unknown"),
6901                    netdev->dev_addr);
6902         ixgbe_read_pba_num_generic(hw, &part_num);
6903         if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
6904                 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, "
6905                            "PBA No: %06x-%03x\n",
6906                            hw->mac.type, hw->phy.type, hw->phy.sfp_type,
6907                            (part_num >> 8), (part_num & 0xff));
6908         else
6909                 e_dev_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
6910                            hw->mac.type, hw->phy.type,
6911                            (part_num >> 8), (part_num & 0xff));
6912
6913         if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
6914                 e_dev_warn("PCI-Express bandwidth available for this card is "
6915                            "not sufficient for optimal performance.\n");
6916                 e_dev_warn("For optimal performance a x8 PCI-Express slot "
6917                            "is required.\n");
6918         }
6919
6920         /* save off EEPROM version number */
6921         hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
6922
6923         /* reset the hardware with the new settings */
6924         err = hw->mac.ops.start_hw(hw);
6925
6926         if (err == IXGBE_ERR_EEPROM_VERSION) {
6927                 /* We are running on a pre-production device, log a warning */
6928                 e_dev_warn("This device is a pre-production adapter/LOM. "
6929                            "Please be aware there may be issues associated "
6930                            "with your hardware.  If you are experiencing "
6931                            "problems please contact your Intel or hardware "
6932                            "representative who provided you with this "
6933                            "hardware.\n");
6934         }
6935         strcpy(netdev->name, "eth%d");
6936         err = register_netdev(netdev);
6937         if (err)
6938                 goto err_register;
6939
6940         /* carrier off reporting is important to ethtool even BEFORE open */
6941         netif_carrier_off(netdev);
6942
6943         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
6944             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
6945                 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
6946
6947         if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
6948                 INIT_WORK(&adapter->check_overtemp_task,
6949                           ixgbe_check_overtemp_task);
6950 #ifdef CONFIG_IXGBE_DCA
6951         if (dca_add_requester(&pdev->dev) == 0) {
6952                 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
6953                 ixgbe_setup_dca(adapter);
6954         }
6955 #endif
6956         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
6957                 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
6958                 for (i = 0; i < adapter->num_vfs; i++)
6959                         ixgbe_vf_configuration(pdev, (i | 0x10000000));
6960         }
6961
6962         /* add san mac addr to netdev */
6963         ixgbe_add_sanmac_netdev(netdev);
6964
6965         e_dev_info("Intel(R) 10 Gigabit Network Connection\n");
6966         cards_found++;
6967         return 0;
6968
6969 err_register:
6970         ixgbe_release_hw_control(adapter);
6971         ixgbe_clear_interrupt_scheme(adapter);
6972 err_sw_init:
6973 err_eeprom:
6974         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6975                 ixgbe_disable_sriov(adapter);
6976         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6977         del_timer_sync(&adapter->sfp_timer);
6978         cancel_work_sync(&adapter->sfp_task);
6979         cancel_work_sync(&adapter->multispeed_fiber_task);
6980         cancel_work_sync(&adapter->sfp_config_module_task);
6981         iounmap(hw->hw_addr);
6982 err_ioremap:
6983         free_netdev(netdev);
6984 err_alloc_etherdev:
6985         pci_release_selected_regions(pdev,
6986                                      pci_select_bars(pdev, IORESOURCE_MEM));
6987 err_pci_reg:
6988 err_dma:
6989         pci_disable_device(pdev);
6990         return err;
6991 }
6992
6993 /**
6994  * ixgbe_remove - Device Removal Routine
6995  * @pdev: PCI device information struct
6996  *
6997  * ixgbe_remove is called by the PCI subsystem to alert the driver
6998  * that it should release a PCI device.  The could be caused by a
6999  * Hot-Plug event, or because the driver is going to be removed from
7000  * memory.
7001  **/
7002 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7003 {
7004         struct net_device *netdev = pci_get_drvdata(pdev);
7005         struct ixgbe_adapter *adapter = netdev_priv(netdev);
7006
7007         set_bit(__IXGBE_DOWN, &adapter->state);
7008         /* clear the module not found bit to make sure the worker won't
7009          * reschedule
7010          */
7011         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7012         del_timer_sync(&adapter->watchdog_timer);
7013
7014         del_timer_sync(&adapter->sfp_timer);
7015         cancel_work_sync(&adapter->watchdog_task);
7016         cancel_work_sync(&adapter->sfp_task);
7017         cancel_work_sync(&adapter->multispeed_fiber_task);
7018         cancel_work_sync(&adapter->sfp_config_module_task);
7019         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7020             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7021                 cancel_work_sync(&adapter->fdir_reinit_task);
7022         flush_scheduled_work();
7023
7024 #ifdef CONFIG_IXGBE_DCA
7025         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7026                 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7027                 dca_remove_requester(&pdev->dev);
7028                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7029         }
7030
7031 #endif
7032 #ifdef IXGBE_FCOE
7033         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7034                 ixgbe_cleanup_fcoe(adapter);
7035
7036 #endif /* IXGBE_FCOE */
7037
7038         /* remove the added san mac */
7039         ixgbe_del_sanmac_netdev(netdev);
7040
7041         if (netdev->reg_state == NETREG_REGISTERED)
7042                 unregister_netdev(netdev);
7043
7044         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7045                 ixgbe_disable_sriov(adapter);
7046
7047         ixgbe_clear_interrupt_scheme(adapter);
7048
7049         ixgbe_release_hw_control(adapter);
7050
7051         iounmap(adapter->hw.hw_addr);
7052         pci_release_selected_regions(pdev, pci_select_bars(pdev,
7053                                      IORESOURCE_MEM));
7054
7055         e_dev_info("complete\n");
7056
7057         free_netdev(netdev);
7058
7059         pci_disable_pcie_error_reporting(pdev);
7060
7061         pci_disable_device(pdev);
7062 }
7063
7064 /**
7065  * ixgbe_io_error_detected - called when PCI error is detected
7066  * @pdev: Pointer to PCI device
7067  * @state: The current pci connection state
7068  *
7069  * This function is called after a PCI bus error affecting
7070  * this device has been detected.
7071  */
7072 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7073                                                 pci_channel_state_t state)
7074 {
7075         struct net_device *netdev = pci_get_drvdata(pdev);
7076         struct ixgbe_adapter *adapter = netdev_priv(netdev);
7077
7078         netif_device_detach(netdev);
7079
7080         if (state == pci_channel_io_perm_failure)
7081                 return PCI_ERS_RESULT_DISCONNECT;
7082
7083         if (netif_running(netdev))
7084                 ixgbe_down(adapter);
7085         pci_disable_device(pdev);
7086
7087         /* Request a slot reset. */
7088         return PCI_ERS_RESULT_NEED_RESET;
7089 }
7090
7091 /**
7092  * ixgbe_io_slot_reset - called after the pci bus has been reset.
7093  * @pdev: Pointer to PCI device
7094  *
7095  * Restart the card from scratch, as if from a cold-boot.
7096  */
7097 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7098 {
7099         struct net_device *netdev = pci_get_drvdata(pdev);
7100         struct ixgbe_adapter *adapter = netdev_priv(netdev);
7101         pci_ers_result_t result;
7102         int err;
7103
7104         if (pci_enable_device_mem(pdev)) {
7105                 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7106                 result = PCI_ERS_RESULT_DISCONNECT;
7107         } else {
7108                 pci_set_master(pdev);
7109                 pci_restore_state(pdev);
7110                 pci_save_state(pdev);
7111
7112                 pci_wake_from_d3(pdev, false);
7113
7114                 ixgbe_reset(adapter);
7115                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7116                 result = PCI_ERS_RESULT_RECOVERED;
7117         }
7118
7119         err = pci_cleanup_aer_uncorrect_error_status(pdev);
7120         if (err) {
7121                 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7122                           "failed 0x%0x\n", err);
7123                 /* non-fatal, continue */
7124         }
7125
7126         return result;
7127 }
7128
7129 /**
7130  * ixgbe_io_resume - called when traffic can start flowing again.
7131  * @pdev: Pointer to PCI device
7132  *
7133  * This callback is called when the error recovery driver tells us that
7134  * its OK to resume normal operation.
7135  */
7136 static void ixgbe_io_resume(struct pci_dev *pdev)
7137 {
7138         struct net_device *netdev = pci_get_drvdata(pdev);
7139         struct ixgbe_adapter *adapter = netdev_priv(netdev);
7140
7141         if (netif_running(netdev)) {
7142                 if (ixgbe_up(adapter)) {
7143                         e_info(probe, "ixgbe_up failed after reset\n");
7144                         return;
7145                 }
7146         }
7147
7148         netif_device_attach(netdev);
7149 }
7150
7151 static struct pci_error_handlers ixgbe_err_handler = {
7152         .error_detected = ixgbe_io_error_detected,
7153         .slot_reset = ixgbe_io_slot_reset,
7154         .resume = ixgbe_io_resume,
7155 };
7156
7157 static struct pci_driver ixgbe_driver = {
7158         .name     = ixgbe_driver_name,
7159         .id_table = ixgbe_pci_tbl,
7160         .probe    = ixgbe_probe,
7161         .remove   = __devexit_p(ixgbe_remove),
7162 #ifdef CONFIG_PM
7163         .suspend  = ixgbe_suspend,
7164         .resume   = ixgbe_resume,
7165 #endif
7166         .shutdown = ixgbe_shutdown,
7167         .err_handler = &ixgbe_err_handler
7168 };
7169
7170 /**
7171  * ixgbe_init_module - Driver Registration Routine
7172  *
7173  * ixgbe_init_module is the first routine called when the driver is
7174  * loaded. All it does is register with the PCI subsystem.
7175  **/
7176 static int __init ixgbe_init_module(void)
7177 {
7178         int ret;
7179         pr_info("%s - version %s\n", ixgbe_driver_string,
7180                    ixgbe_driver_version);
7181         pr_info("%s\n", ixgbe_copyright);
7182
7183 #ifdef CONFIG_IXGBE_DCA
7184         dca_register_notify(&dca_notifier);
7185 #endif
7186
7187         ret = pci_register_driver(&ixgbe_driver);
7188         return ret;
7189 }
7190
7191 module_init(ixgbe_init_module);
7192
7193 /**
7194  * ixgbe_exit_module - Driver Exit Cleanup Routine
7195  *
7196  * ixgbe_exit_module is called just before the driver is removed
7197  * from memory.
7198  **/
7199 static void __exit ixgbe_exit_module(void)
7200 {
7201 #ifdef CONFIG_IXGBE_DCA
7202         dca_unregister_notify(&dca_notifier);
7203 #endif
7204         pci_unregister_driver(&ixgbe_driver);
7205 }
7206
7207 #ifdef CONFIG_IXGBE_DCA
7208 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7209                             void *p)
7210 {
7211         int ret_val;
7212
7213         ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7214                                          __ixgbe_notify_dca);
7215
7216         return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7217 }
7218
7219 #endif /* CONFIG_IXGBE_DCA */
7220
7221 /**
7222  * ixgbe_get_hw_dev return device
7223  * used by hardware layer to print debugging information
7224  **/
7225 struct net_device *ixgbe_get_hw_dev(struct ixgbe_hw *hw)
7226 {
7227         struct ixgbe_adapter *adapter = hw->back;
7228         return adapter->netdev;
7229 }
7230
7231 module_exit(ixgbe_exit_module);
7232
7233 /* ixgbe_main.c */