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                 pr_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                 pr_err("%-15s", rname);
 293                 for (j = 0; j < 8; j++)
 294                         pr_cont(" %08x", regs[i*8+j]);
 295                 pr_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                 pr_info("Device Name     state            "
 326                         "trans_start      last_rx\n");
 327                 pr_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         pr_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         pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
 348         for (n = 0; n < adapter->num_tx_queues; n++) {
 349                 tx_ring = adapter->tx_ring[n];
 350                 tx_buffer_info =
 351                         &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
 352                 pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
 353                            n, tx_ring->next_to_use, tx_ring->next_to_clean,
 354                            (u64)tx_buffer_info->dma,
 355                            tx_buffer_info->length,
 356                            tx_buffer_info->next_to_watch,
 357                            (u64)tx_buffer_info->time_stamp);
 358         }
 359
 360         /* Print TX Rings */
 361         if (!netif_msg_tx_done(adapter))
 362                 goto rx_ring_summary;
 363
 364         dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
 365
 366         /* Transmit Descriptor Formats
 367          *
 368          * Advanced Transmit Descriptor
 369          *   +--------------------------------------------------------------+
 370          * 0 |         Buffer Address [63:0]                                |
 371          *   +--------------------------------------------------------------+
 372          * 8 |  PAYLEN  | PORTS  | IDX | STA | DCMD  |DTYP |  RSV |  DTALEN |
 373          *   +--------------------------------------------------------------+
 374          *   63       46 45    40 39 36 35 32 31   24 23 20 19              0
 375          */
 376
 377         for (n = 0; n < adapter->num_tx_queues; n++) {
 378                 tx_ring = adapter->tx_ring[n];
 379                 pr_info("------------------------------------\n");
 380                 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
 381                 pr_info("------------------------------------\n");
 382                 pr_info("T [desc]     [address 63:0  ] "
 383                         "[PlPOIdStDDt Ln] [bi->dma       ] "
 384                         "leng  ntw timestamp        bi->skb\n");
 385
 386                 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
 387                         tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
 388                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
 389                         u0 = (struct my_u0 *)tx_desc;
 390                         pr_info("T [0x%03X]    %016llX %016llX %016llX"
 391                                 " %04X  %3X %016llX %p", i,
 392                                 le64_to_cpu(u0->a),
 393                                 le64_to_cpu(u0->b),
 394                                 (u64)tx_buffer_info->dma,
 395                                 tx_buffer_info->length,
 396                                 tx_buffer_info->next_to_watch,
 397                                 (u64)tx_buffer_info->time_stamp,
 398                                 tx_buffer_info->skb);
 399                         if (i == tx_ring->next_to_use &&
 400                                 i == tx_ring->next_to_clean)
 401                                 pr_cont(" NTC/U\n");
 402                         else if (i == tx_ring->next_to_use)
 403                                 pr_cont(" NTU\n");
 404                         else if (i == tx_ring->next_to_clean)
 405                                 pr_cont(" NTC\n");
 406                         else
 407                                 pr_cont("\n");
 408
 409                         if (netif_msg_pktdata(adapter) &&
 410                                 tx_buffer_info->dma != 0)
 411                                 print_hex_dump(KERN_INFO, "",
 412                                         DUMP_PREFIX_ADDRESS, 16, 1,
 413                                         phys_to_virt(tx_buffer_info->dma),
 414                                         tx_buffer_info->length, true);
 415                 }
 416         }
 417
 418         /* Print RX Rings Summary */
 419 rx_ring_summary:
 420         dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
 421         pr_info("Queue [NTU] [NTC]\n");
 422         for (n = 0; n < adapter->num_rx_queues; n++) {
 423                 rx_ring = adapter->rx_ring[n];
 424                 pr_info("%5d %5X %5X\n",
 425                         n, rx_ring->next_to_use, rx_ring->next_to_clean);
 426         }
 427
 428         /* Print RX Rings */
 429         if (!netif_msg_rx_status(adapter))
 430                 goto exit;
 431
 432         dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
 433
 434         /* Advanced Receive Descriptor (Read) Format
 435          *    63                                           1        0
 436          *    +-----------------------------------------------------+
 437          *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
 438          *    +----------------------------------------------+------+
 439          *  8 |       Header Buffer Address [63:1]           |  DD  |
 440          *    +-----------------------------------------------------+
 441          *
 442          *
 443          * Advanced Receive Descriptor (Write-Back) Format
 444          *
 445          *   63       48 47    32 31  30      21 20 16 15   4 3     0
 446          *   +------------------------------------------------------+
 447          * 0 | Packet     IP     |SPH| HDR_LEN   | RSV|Packet|  RSS |
 448          *   | Checksum   Ident  |   |           |    | Type | Type |
 449          *   +------------------------------------------------------+
 450          * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 451          *   +------------------------------------------------------+
 452          *   63       48 47    32 31            20 19               0
 453          */
 454         for (n = 0; n < adapter->num_rx_queues; n++) {
 455                 rx_ring = adapter->rx_ring[n];
 456                 pr_info("------------------------------------\n");
 457                 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
 458                 pr_info("------------------------------------\n");
 459                 pr_info("R  [desc]      [ PktBuf     A0] "
 460                         "[  HeadBuf   DD] [bi->dma       ] [bi->skb] "
 461                         "<-- Adv Rx Read format\n");
 462                 pr_info("RWB[desc]      [PcsmIpSHl PtRs] "
 463                         "[vl er S cks ln] ---------------- [bi->skb] "
 464                         "<-- Adv Rx Write-Back format\n");
 465
 466                 for (i = 0; i < rx_ring->count; i++) {
 467                         rx_buffer_info = &rx_ring->rx_buffer_info[i];
 468                         rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
 469                         u0 = (struct my_u0 *)rx_desc;
 470                         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 471                         if (staterr & IXGBE_RXD_STAT_DD) {
 472                                 /* Descriptor Done */
 473                                 pr_info("RWB[0x%03X]     %016llX "
 474                                         "%016llX ---------------- %p", i,
 475                                         le64_to_cpu(u0->a),
 476                                         le64_to_cpu(u0->b),
 477                                         rx_buffer_info->skb);
 478                         } else {
 479                                 pr_info("R  [0x%03X]     %016llX "
 480                                         "%016llX %016llX %p", i,
 481                                         le64_to_cpu(u0->a),
 482                                         le64_to_cpu(u0->b),
 483                                         (u64)rx_buffer_info->dma,
 484                                         rx_buffer_info->skb);
 485
 486                                 if (netif_msg_pktdata(adapter)) {
 487                                         print_hex_dump(KERN_INFO, "",
 488                                            DUMP_PREFIX_ADDRESS, 16, 1,
 489                                            phys_to_virt(rx_buffer_info->dma),
 490                                            rx_ring->rx_buf_len, true);
 491
 492                                         if (rx_ring->rx_buf_len
 493                                                 < IXGBE_RXBUFFER_2048)
 494                                                 print_hex_dump(KERN_INFO, "",
 495                                                   DUMP_PREFIX_ADDRESS, 16, 1,
 496                                                   phys_to_virt(
 497                                                     rx_buffer_info->page_dma +
 498                                                     rx_buffer_info->page_offset
 499                                                   ),
 500                                                   PAGE_SIZE/2, true);
 501                                 }
 502                         }
 503
 504                         if (i == rx_ring->next_to_use)
 505                                 pr_cont(" NTU\n");
 506                         else if (i == rx_ring->next_to_clean)
 507                                 pr_cont(" NTC\n");
 508                         else
 509                                 pr_cont("\n");
 510
 511                 }
 512         }
 513
 514 exit:
 515         return;
 516 }
 517
 518 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
 519 {
 520         u32 ctrl_ext;
 521
 522         /* Let firmware take over control of h/w */
 523         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
 524         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
 525                         ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
 526 }
 527
 528 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
 529 {
 530         u32 ctrl_ext;
 531
 532         /* Let firmware know the driver has taken over */
 533         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
 534         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
 535                         ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
 536 }
 537
 538 /*
 539  * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
 540  * @adapter: pointer to adapter struct
 541  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
 542  * @queue: queue to map the corresponding interrupt to
 543  * @msix_vector: the vector to map to the corresponding queue
 544  *
 545  */
 546 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
 547                            u8 queue, u8 msix_vector)
 548 {
 549         u32 ivar, index;
 550         struct ixgbe_hw *hw = &adapter->hw;
 551         switch (hw->mac.type) {
 552         case ixgbe_mac_82598EB:
 553                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 554                 if (direction == -1)
 555                         direction = 0;
 556                 index = (((direction * 64) + queue) >> 2) & 0x1F;
 557                 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
 558                 ivar &= ~(0xFF << (8 * (queue & 0x3)));
 559                 ivar |= (msix_vector << (8 * (queue & 0x3)));
 560                 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
 561                 break;
 562         case ixgbe_mac_82599EB:
 563                 if (direction == -1) {
 564                         /* other causes */
 565                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 566                         index = ((queue & 1) * 8);
 567                         ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
 568                         ivar &= ~(0xFF << index);
 569                         ivar |= (msix_vector << index);
 570                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
 571                         break;
 572                 } else {
 573                         /* tx or rx causes */
 574                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 575                         index = ((16 * (queue & 1)) + (8 * direction));
 576                         ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
 577                         ivar &= ~(0xFF << index);
 578                         ivar |= (msix_vector << index);
 579                         IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
 580                         break;
 581                 }
 582         default:
 583                 break;
 584         }
 585 }
 586
 587 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
 588                                           u64 qmask)
 589 {
 590         u32 mask;
 591
 592         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
 593                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
 594                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
 595         } else {
 596                 mask = (qmask & 0xFFFFFFFF);
 597                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
 598                 mask = (qmask >> 32);
 599                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
 600         }
 601 }
 602
 603 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_adapter *adapter,
 604                                       struct ixgbe_tx_buffer
 605                                       *tx_buffer_info)
 606 {
 607         if (tx_buffer_info->dma) {
 608                 if (tx_buffer_info->mapped_as_page)
 609                         dma_unmap_page(&adapter->pdev->dev,
 610                                        tx_buffer_info->dma,
 611                                        tx_buffer_info->length,
 612                                        DMA_TO_DEVICE);
 613                 else
 614                         dma_unmap_single(&adapter->pdev->dev,
 615                                          tx_buffer_info->dma,
 616                                          tx_buffer_info->length,
 617                                          DMA_TO_DEVICE);
 618                 tx_buffer_info->dma = 0;
 619         }
 620         if (tx_buffer_info->skb) {
 621                 dev_kfree_skb_any(tx_buffer_info->skb);
 622                 tx_buffer_info->skb = NULL;
 623         }
 624         tx_buffer_info->time_stamp = 0;
 625         /* tx_buffer_info must be completely set up in the transmit path */
 626 }
 627
 628 /**
 629  * ixgbe_tx_xon_state - check the tx ring xon state
 630  * @adapter: the ixgbe adapter
 631  * @tx_ring: the corresponding tx_ring
 632  *
 633  * If not in DCB mode, checks TFCS.TXOFF, otherwise, find out the
 634  * corresponding TC of this tx_ring when checking TFCS.
 635  *
 636  * Returns : true if in xon state (currently not paused)
 637  */
 638 static inline bool ixgbe_tx_xon_state(struct ixgbe_adapter *adapter,
 639                                       struct ixgbe_ring *tx_ring)
 640 {
 641         u32 txoff = IXGBE_TFCS_TXOFF;
 642
 643 #ifdef CONFIG_IXGBE_DCB
 644         if (adapter->dcb_cfg.pfc_mode_enable) {
 645                 int tc;
 646                 int reg_idx = tx_ring->reg_idx;
 647                 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
 648
 649                 switch (adapter->hw.mac.type) {
 650                 case ixgbe_mac_82598EB:
 651                         tc = reg_idx >> 2;
 652                         txoff = IXGBE_TFCS_TXOFF0;
 653                         break;
 654                 case ixgbe_mac_82599EB:
 655                         tc = 0;
 656                         txoff = IXGBE_TFCS_TXOFF;
 657                         if (dcb_i == 8) {
 658                                 /* TC0, TC1 */
 659                                 tc = reg_idx >> 5;
 660                                 if (tc == 2) /* TC2, TC3 */
 661                                         tc += (reg_idx - 64) >> 4;
 662                                 else if (tc == 3) /* TC4, TC5, TC6, TC7 */
 663                                         tc += 1 + ((reg_idx - 96) >> 3);
 664                         } else if (dcb_i == 4) {
 665                                 /* TC0, TC1 */
 666                                 tc = reg_idx >> 6;
 667                                 if (tc == 1) {
 668                                         tc += (reg_idx - 64) >> 5;
 669                                         if (tc == 2) /* TC2, TC3 */
 670                                                 tc += (reg_idx - 96) >> 4;
 671                                 }
 672                         }
 673                         break;
 674                 default:
 675                         tc = 0;
 676                 }
 677                 txoff <<= tc;
 678         }
 679 #endif
 680         return IXGBE_READ_REG(&adapter->hw, IXGBE_TFCS) & txoff;
 681 }
 682
 683 static inline bool ixgbe_check_tx_hang(struct ixgbe_adapter *adapter,
 684                                        struct ixgbe_ring *tx_ring,
 685                                        unsigned int eop)
 686 {
 687         struct ixgbe_hw *hw = &adapter->hw;
 688
 689         /* Detect a transmit hang in hardware, this serializes the
 690          * check with the clearing of time_stamp and movement of eop */
 691         adapter->detect_tx_hung = false;
 692         if (tx_ring->tx_buffer_info[eop].time_stamp &&
 693             time_after(jiffies, tx_ring->tx_buffer_info[eop].time_stamp + HZ) &&
 694             ixgbe_tx_xon_state(adapter, tx_ring)) {
 695                 /* detected Tx unit hang */
 696                 union ixgbe_adv_tx_desc *tx_desc;
 697                 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
 698                 e_err(drv, "Detected Tx Unit Hang\n"
 699                       "  Tx Queue             <%d>\n"
 700                       "  TDH, TDT             <%x>, <%x>\n"
 701                       "  next_to_use          <%x>\n"
 702                       "  next_to_clean        <%x>\n"
 703                       "tx_buffer_info[next_to_clean]\n"
 704                       "  time_stamp           <%lx>\n"
 705                       "  jiffies              <%lx>\n",
 706                       tx_ring->queue_index,
 707                       IXGBE_READ_REG(hw, tx_ring->head),
 708                       IXGBE_READ_REG(hw, tx_ring->tail),
 709                       tx_ring->next_to_use, eop,
 710                       tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
 711                 return true;
 712         }
 713
 714         return false;
 715 }
 716
 717 #define IXGBE_MAX_TXD_PWR       14
 718 #define IXGBE_MAX_DATA_PER_TXD  (1 << IXGBE_MAX_TXD_PWR)
 719
 720 /* Tx Descriptors needed, worst case */
 721 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
 722                          (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
 723 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
 724         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
 725
 726 static void ixgbe_tx_timeout(struct net_device *netdev);
 727
 728 /**
 729  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
 730  * @q_vector: structure containing interrupt and ring information
 731  * @tx_ring: tx ring to clean
 732  **/
 733 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
 734                                struct ixgbe_ring *tx_ring)
 735 {
 736         struct ixgbe_adapter *adapter = q_vector->adapter;
 737         struct net_device *netdev = adapter->netdev;
 738         union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
 739         struct ixgbe_tx_buffer *tx_buffer_info;
 740         unsigned int i, eop, count = 0;
 741         unsigned int total_bytes = 0, total_packets = 0;
 742
 743         i = tx_ring->next_to_clean;
 744         eop = tx_ring->tx_buffer_info[i].next_to_watch;
 745         eop_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
 746
 747         while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
 748                (count < tx_ring->work_limit)) {
 749                 bool cleaned = false;
 750                 rmb(); /* read buffer_info after eop_desc */
 751                 for ( ; !cleaned; count++) {
 752                         struct sk_buff *skb;
 753                         tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
 754                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
 755                         cleaned = (i == eop);
 756                         skb = tx_buffer_info->skb;
 757
 758                         if (cleaned && skb) {
 759                                 unsigned int segs, bytecount;
 760                                 unsigned int hlen = skb_headlen(skb);
 761
 762                                 /* gso_segs is currently only valid for tcp */
 763                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 764 #ifdef IXGBE_FCOE
 765                                 /* adjust for FCoE Sequence Offload */
 766                                 if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
 767                                     && (skb->protocol == htons(ETH_P_FCOE)) &&
 768                                     skb_is_gso(skb)) {
 769                                         hlen = skb_transport_offset(skb) +
 770                                                 sizeof(struct fc_frame_header) +
 771                                                 sizeof(struct fcoe_crc_eof);
 772                                         segs = DIV_ROUND_UP(skb->len - hlen,
 773                                                 skb_shinfo(skb)->gso_size);
 774                                 }
 775 #endif /* IXGBE_FCOE */
 776                                 /* multiply data chunks by size of headers */
 777                                 bytecount = ((segs - 1) * hlen) + skb->len;
 778                                 total_packets += segs;
 779                                 total_bytes += bytecount;
 780                         }
 781
 782                         ixgbe_unmap_and_free_tx_resource(adapter,
 783                                                          tx_buffer_info);
 784
 785                         tx_desc->wb.status = 0;
 786
 787                         i++;
 788                         if (i == tx_ring->count)
 789                                 i = 0;
 790                 }
 791
 792                 eop = tx_ring->tx_buffer_info[i].next_to_watch;
 793                 eop_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
 794         }
 795
 796         tx_ring->next_to_clean = i;
 797
 798 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
 799         if (unlikely(count && netif_carrier_ok(netdev) &&
 800                      (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
 801                 /* Make sure that anybody stopping the queue after this
 802                  * sees the new next_to_clean.
 803                  */
 804                 smp_mb();
 805                 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
 806                     !test_bit(__IXGBE_DOWN, &adapter->state)) {
 807                         netif_wake_subqueue(netdev, tx_ring->queue_index);
 808                         ++tx_ring->restart_queue;
 809                 }
 810         }
 811
 812         if (adapter->detect_tx_hung) {
 813                 if (ixgbe_check_tx_hang(adapter, tx_ring, i)) {
 814                         /* schedule immediate reset if we believe we hung */
 815                         e_info(probe, "tx hang %d detected, resetting "
 816                                "adapter\n", adapter->tx_timeout_count + 1);
 817                         ixgbe_tx_timeout(adapter->netdev);
 818                 }
 819         }
 820
 821         /* re-arm the interrupt */
 822         if (count >= tx_ring->work_limit)
 823                 ixgbe_irq_rearm_queues(adapter, ((u64)1 << q_vector->v_idx));
 824
 825         tx_ring->total_bytes += total_bytes;
 826         tx_ring->total_packets += total_packets;
 827         tx_ring->stats.packets += total_packets;
 828         tx_ring->stats.bytes += total_bytes;
 829         return count < tx_ring->work_limit;
 830 }
 831
 832 #ifdef CONFIG_IXGBE_DCA
 833 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
 834                                 struct ixgbe_ring *rx_ring)
 835 {
 836         u32 rxctrl;
 837         int cpu = get_cpu();
 838         int q = rx_ring->reg_idx;
 839
 840         if (rx_ring->cpu != cpu) {
 841                 rxctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q));
 842                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
 843                         rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
 844                         rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
 845                 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
 846                         rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
 847                         rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
 848                                    IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
 849                 }
 850                 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
 851                 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
 852                 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
 853                 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN |
 854                             IXGBE_DCA_RXCTRL_DESC_HSRO_EN);
 855                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q), rxctrl);
 856                 rx_ring->cpu = cpu;
 857         }
 858         put_cpu();
 859 }
 860
 861 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
 862                                 struct ixgbe_ring *tx_ring)
 863 {
 864         u32 txctrl;
 865         int cpu = get_cpu();
 866         int q = tx_ring->reg_idx;
 867         struct ixgbe_hw *hw = &adapter->hw;
 868
 869         if (tx_ring->cpu != cpu) {
 870                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
 871                         txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(q));
 872                         txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
 873                         txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
 874                         txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
 875                         IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(q), txctrl);
 876                 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
 877                         txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(q));
 878                         txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
 879                         txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
 880                                   IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
 881                         txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
 882                         IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(q), txctrl);
 883                 }
 884                 tx_ring->cpu = cpu;
 885         }
 886         put_cpu();
 887 }
 888
 889 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
 890 {
 891         int i;
 892
 893         if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
 894                 return;
 895
 896         /* always use CB2 mode, difference is masked in the CB driver */
 897         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
 898
 899         for (i = 0; i < adapter->num_tx_queues; i++) {
 900                 adapter->tx_ring[i]->cpu = -1;
 901                 ixgbe_update_tx_dca(adapter, adapter->tx_ring[i]);
 902         }
 903         for (i = 0; i < adapter->num_rx_queues; i++) {
 904                 adapter->rx_ring[i]->cpu = -1;
 905                 ixgbe_update_rx_dca(adapter, adapter->rx_ring[i]);
 906         }
 907 }
 908
 909 static int __ixgbe_notify_dca(struct device *dev, void *data)
 910 {
 911         struct net_device *netdev = dev_get_drvdata(dev);
 912         struct ixgbe_adapter *adapter = netdev_priv(netdev);
 913         unsigned long event = *(unsigned long *)data;
 914
 915         switch (event) {
 916         case DCA_PROVIDER_ADD:
 917                 /* if we're already enabled, don't do it again */
 918                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
 919                         break;
 920                 if (dca_add_requester(dev) == 0) {
 921                         adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
 922                         ixgbe_setup_dca(adapter);
 923                         break;
 924                 }
 925                 /* Fall Through since DCA is disabled. */
 926         case DCA_PROVIDER_REMOVE:
 927                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
 928                         dca_remove_requester(dev);
 929                         adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
 930                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
 931                 }
 932                 break;
 933         }
 934
 935         return 0;
 936 }
 937
 938 #endif /* CONFIG_IXGBE_DCA */
 939 /**
 940  * ixgbe_receive_skb - Send a completed packet up the stack
 941  * @adapter: board private structure
 942  * @skb: packet to send up
 943  * @status: hardware indication of status of receive
 944  * @rx_ring: rx descriptor ring (for a specific queue) to setup
 945  * @rx_desc: rx descriptor
 946  **/
 947 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
 948                               struct sk_buff *skb, u8 status,
 949                               struct ixgbe_ring *ring,
 950                               union ixgbe_adv_rx_desc *rx_desc)
 951 {
 952         struct ixgbe_adapter *adapter = q_vector->adapter;
 953         struct napi_struct *napi = &q_vector->napi;
 954         bool is_vlan = (status & IXGBE_RXD_STAT_VP);
 955         u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
 956
 957         if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL)) {
 958                 if (adapter->vlgrp && is_vlan && (tag & VLAN_VID_MASK))
 959                         vlan_gro_receive(napi, adapter->vlgrp, tag, skb);
 960                 else
 961                         napi_gro_receive(napi, skb);
 962         } else {
 963                 if (adapter->vlgrp && is_vlan && (tag & VLAN_VID_MASK))
 964                         vlan_hwaccel_rx(skb, adapter->vlgrp, tag);
 965                 else
 966                         netif_rx(skb);
 967         }
 968 }
 969
 970 /**
 971  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
 972  * @adapter: address of board private structure
 973  * @status_err: hardware indication of status of receive
 974  * @skb: skb currently being received and modified
 975  **/
 976 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
 977                                      union ixgbe_adv_rx_desc *rx_desc,
 978                                      struct sk_buff *skb)
 979 {
 980         u32 status_err = le32_to_cpu(rx_desc->wb.upper.status_error);
 981
 982         skb_checksum_none_assert(skb);
 983
 984         /* Rx csum disabled */
 985         if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
 986                 return;
 987
 988         /* if IP and error */
 989         if ((status_err & IXGBE_RXD_STAT_IPCS) &&
 990             (status_err & IXGBE_RXDADV_ERR_IPE)) {
 991                 adapter->hw_csum_rx_error++;
 992                 return;
 993         }
 994
 995         if (!(status_err & IXGBE_RXD_STAT_L4CS))
 996                 return;
 997
 998         if (status_err & IXGBE_RXDADV_ERR_TCPE) {
 999                 u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1000
1001                 /*
1002                  * 82599 errata, UDP frames with a 0 checksum can be marked as
1003                  * checksum errors.
1004                  */
1005                 if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
1006                     (adapter->hw.mac.type == ixgbe_mac_82599EB))
1007                         return;
1008
1009                 adapter->hw_csum_rx_error++;
1010                 return;
1011         }
1012
1013         /* It must be a TCP or UDP packet with a valid checksum */
1014         skb->ip_summed = CHECKSUM_UNNECESSARY;
1015 }
1016
1017 static inline void ixgbe_release_rx_desc(struct ixgbe_hw *hw,
1018                                          struct ixgbe_ring *rx_ring, u32 val)
1019 {
1020         /*
1021          * Force memory writes to complete before letting h/w
1022          * know there are new descriptors to fetch.  (Only
1023          * applicable for weak-ordered memory model archs,
1024          * such as IA-64).
1025          */
1026         wmb();
1027         IXGBE_WRITE_REG(hw, IXGBE_RDT(rx_ring->reg_idx), val);
1028 }
1029
1030 /**
1031  * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
1032  * @adapter: address of board private structure
1033  **/
1034 void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
1035                             struct ixgbe_ring *rx_ring,
1036                             int cleaned_count)
1037 {
1038         struct net_device *netdev = adapter->netdev;
1039         struct pci_dev *pdev = adapter->pdev;
1040         union ixgbe_adv_rx_desc *rx_desc;
1041         struct ixgbe_rx_buffer *bi;
1042         unsigned int i;
1043         unsigned int bufsz = rx_ring->rx_buf_len;
1044
1045         i = rx_ring->next_to_use;
1046         bi = &rx_ring->rx_buffer_info[i];
1047
1048         while (cleaned_count--) {
1049                 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1050
1051                 if (!bi->page_dma &&
1052                     (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED)) {
1053                         if (!bi->page) {
1054                                 bi->page = netdev_alloc_page(netdev);
1055                                 if (!bi->page) {
1056                                         adapter->alloc_rx_page_failed++;
1057                                         goto no_buffers;
1058                                 }
1059                                 bi->page_offset = 0;
1060                         } else {
1061                                 /* use a half page if we're re-using */
1062                                 bi->page_offset ^= (PAGE_SIZE / 2);
1063                         }
1064
1065                         bi->page_dma = dma_map_page(&pdev->dev, bi->page,
1066                                                     bi->page_offset,
1067                                                     (PAGE_SIZE / 2),
1068                                                     DMA_FROM_DEVICE);
1069                 }
1070
1071                 if (!bi->skb) {
1072                         struct sk_buff *skb = netdev_alloc_skb_ip_align(netdev,
1073                                                                         bufsz);
1074                         bi->skb = skb;
1075
1076                         if (!skb) {
1077                                 adapter->alloc_rx_buff_failed++;
1078                                 goto no_buffers;
1079                         }
1080                         /* initialize queue mapping */
1081                         skb_record_rx_queue(skb, rx_ring->queue_index);
1082                 }
1083
1084                 if (!bi->dma) {
1085                         bi->dma = dma_map_single(&pdev->dev,
1086                                                  bi->skb->data,
1087                                                  rx_ring->rx_buf_len,
1088                                                  DMA_FROM_DEVICE);
1089                 }
1090                 /* Refresh the desc even if buffer_addrs didn't change because
1091                  * each write-back erases this info. */
1092                 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
1093                         rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
1094                         rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
1095                 } else {
1096                         rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
1097                         rx_desc->read.hdr_addr = 0;
1098                 }
1099
1100                 i++;
1101                 if (i == rx_ring->count)
1102                         i = 0;
1103                 bi = &rx_ring->rx_buffer_info[i];
1104         }
1105
1106 no_buffers:
1107         if (rx_ring->next_to_use != i) {
1108                 rx_ring->next_to_use = i;
1109                 if (i-- == 0)
1110                         i = (rx_ring->count - 1);
1111
1112                 ixgbe_release_rx_desc(&adapter->hw, rx_ring, i);
1113         }
1114 }
1115
1116 static inline u16 ixgbe_get_hdr_info(union ixgbe_adv_rx_desc *rx_desc)
1117 {
1118         return rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
1119 }
1120
1121 static inline u16 ixgbe_get_pkt_info(union ixgbe_adv_rx_desc *rx_desc)
1122 {
1123         return rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1124 }
1125
1126 static inline u32 ixgbe_get_rsc_count(union ixgbe_adv_rx_desc *rx_desc)
1127 {
1128         return (le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
1129                 IXGBE_RXDADV_RSCCNT_MASK) >>
1130                 IXGBE_RXDADV_RSCCNT_SHIFT;
1131 }
1132
1133 /**
1134  * ixgbe_transform_rsc_queue - change rsc queue into a full packet
1135  * @skb: pointer to the last skb in the rsc queue
1136  * @count: pointer to number of packets coalesced in this context
1137  *
1138  * This function changes a queue full of hw rsc buffers into a completed
1139  * packet.  It uses the ->prev pointers to find the first packet and then
1140  * turns it into the frag list owner.
1141  **/
1142 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb,
1143                                                         u64 *count)
1144 {
1145         unsigned int frag_list_size = 0;
1146
1147         while (skb->prev) {
1148                 struct sk_buff *prev = skb->prev;
1149                 frag_list_size += skb->len;
1150                 skb->prev = NULL;
1151                 skb = prev;
1152                 *count += 1;
1153         }
1154
1155         skb_shinfo(skb)->frag_list = skb->next;
1156         skb->next = NULL;
1157         skb->len += frag_list_size;
1158         skb->data_len += frag_list_size;
1159         skb->truesize += frag_list_size;
1160         return skb;
1161 }
1162
1163 struct ixgbe_rsc_cb {
1164         dma_addr_t dma;
1165         bool delay_unmap;
1166 };
1167
1168 #define IXGBE_RSC_CB(skb) ((struct ixgbe_rsc_cb *)(skb)->cb)
1169
1170 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1171                                struct ixgbe_ring *rx_ring,
1172                                int *work_done, int work_to_do)
1173 {
1174         struct ixgbe_adapter *adapter = q_vector->adapter;
1175         struct net_device *netdev = adapter->netdev;
1176         struct pci_dev *pdev = adapter->pdev;
1177         union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
1178         struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
1179         struct sk_buff *skb;
1180         unsigned int i, rsc_count = 0;
1181         u32 len, staterr;
1182         u16 hdr_info;
1183         bool cleaned = false;
1184         int cleaned_count = 0;
1185         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1186 #ifdef IXGBE_FCOE
1187         int ddp_bytes = 0;
1188 #endif /* IXGBE_FCOE */
1189
1190         i = rx_ring->next_to_clean;
1191         rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1192         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1193         rx_buffer_info = &rx_ring->rx_buffer_info[i];
1194
1195         while (staterr & IXGBE_RXD_STAT_DD) {
1196                 u32 upper_len = 0;
1197                 if (*work_done >= work_to_do)
1198                         break;
1199                 (*work_done)++;
1200
1201                 rmb(); /* read descriptor and rx_buffer_info after status DD */
1202                 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
1203                         hdr_info = le16_to_cpu(ixgbe_get_hdr_info(rx_desc));
1204                         len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
1205                                IXGBE_RXDADV_HDRBUFLEN_SHIFT;
1206                         upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1207                         if ((len > IXGBE_RX_HDR_SIZE) ||
1208                             (upper_len && !(hdr_info & IXGBE_RXDADV_SPH)))
1209                                 len = IXGBE_RX_HDR_SIZE;
1210                 } else {
1211                         len = le16_to_cpu(rx_desc->wb.upper.length);
1212                 }
1213
1214                 cleaned = true;
1215                 skb = rx_buffer_info->skb;
1216                 prefetch(skb->data);
1217                 rx_buffer_info->skb = NULL;
1218
1219                 if (rx_buffer_info->dma) {
1220                         if ((adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
1221                             (!(staterr & IXGBE_RXD_STAT_EOP)) &&
1222                                  (!(skb->prev))) {
1223                                 /*
1224                                  * When HWRSC is enabled, delay unmapping
1225                                  * of the first packet. It carries the
1226                                  * header information, HW may still
1227                                  * access the header after the writeback.
1228                                  * Only unmap it when EOP is reached
1229                                  */
1230                                 IXGBE_RSC_CB(skb)->delay_unmap = true;
1231                                 IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
1232                         } else {
1233                                 dma_unmap_single(&pdev->dev,
1234                                                  rx_buffer_info->dma,
1235                                                  rx_ring->rx_buf_len,
1236                                                  DMA_FROM_DEVICE);
1237                         }
1238                         rx_buffer_info->dma = 0;
1239                         skb_put(skb, len);
1240                 }
1241
1242                 if (upper_len) {
1243                         dma_unmap_page(&pdev->dev, rx_buffer_info->page_dma,
1244                                        PAGE_SIZE / 2, DMA_FROM_DEVICE);
1245                         rx_buffer_info->page_dma = 0;
1246                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1247                                            rx_buffer_info->page,
1248                                            rx_buffer_info->page_offset,
1249                                            upper_len);
1250
1251                         if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
1252                             (page_count(rx_buffer_info->page) != 1))
1253                                 rx_buffer_info->page = NULL;
1254                         else
1255                                 get_page(rx_buffer_info->page);
1256
1257                         skb->len += upper_len;
1258                         skb->data_len += upper_len;
1259                         skb->truesize += upper_len;
1260                 }
1261
1262                 i++;
1263                 if (i == rx_ring->count)
1264                         i = 0;
1265
1266                 next_rxd = IXGBE_RX_DESC_ADV(rx_ring, i);
1267                 prefetch(next_rxd);
1268                 cleaned_count++;
1269
1270                 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
1271                         rsc_count = ixgbe_get_rsc_count(rx_desc);
1272
1273                 if (rsc_count) {
1274                         u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
1275                                      IXGBE_RXDADV_NEXTP_SHIFT;
1276                         next_buffer = &rx_ring->rx_buffer_info[nextp];
1277                 } else {
1278                         next_buffer = &rx_ring->rx_buffer_info[i];
1279                 }
1280
1281                 if (staterr & IXGBE_RXD_STAT_EOP) {
1282                         if (skb->prev)
1283                                 skb = ixgbe_transform_rsc_queue(skb,
1284                                                                 &(rx_ring->rsc_count));
1285                         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
1286                                 if (IXGBE_RSC_CB(skb)->delay_unmap) {
1287                                         dma_unmap_single(&pdev->dev,
1288                                                          IXGBE_RSC_CB(skb)->dma,
1289                                                          rx_ring->rx_buf_len,
1290                                                          DMA_FROM_DEVICE);
1291                                         IXGBE_RSC_CB(skb)->dma = 0;
1292                                         IXGBE_RSC_CB(skb)->delay_unmap = false;
1293                                 }
1294                                 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED)
1295                                         rx_ring->rsc_count +=
1296                                                 skb_shinfo(skb)->nr_frags;
1297                                 else
1298                                         rx_ring->rsc_count++;
1299                                 rx_ring->rsc_flush++;
1300                         }
1301                         rx_ring->stats.packets++;
1302                         rx_ring->stats.bytes += skb->len;
1303                 } else {
1304                         if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
1305                                 rx_buffer_info->skb = next_buffer->skb;
1306                                 rx_buffer_info->dma = next_buffer->dma;
1307                                 next_buffer->skb = skb;
1308                                 next_buffer->dma = 0;
1309                         } else {
1310                                 skb->next = next_buffer->skb;
1311                                 skb->next->prev = skb;
1312                         }
1313                         rx_ring->non_eop_descs++;
1314                         goto next_desc;
1315                 }
1316
1317                 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
1318                         dev_kfree_skb_irq(skb);
1319                         goto next_desc;
1320                 }
1321
1322                 ixgbe_rx_checksum(adapter, rx_desc, skb);
1323
1324                 /* probably a little skewed due to removing CRC */
1325                 total_rx_bytes += skb->len;
1326                 total_rx_packets++;
1327
1328                 skb->protocol = eth_type_trans(skb, adapter->netdev);
1329 #ifdef IXGBE_FCOE
1330                 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1331                 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
1332                         ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1333                         if (!ddp_bytes)
1334                                 goto next_desc;
1335                 }
1336 #endif /* IXGBE_FCOE */
1337                 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
1338
1339 next_desc:
1340                 rx_desc->wb.upper.status_error = 0;
1341
1342                 /* return some buffers to hardware, one at a time is too slow */
1343                 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1344                         ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
1345                         cleaned_count = 0;
1346                 }
1347
1348                 /* use prefetched values */
1349                 rx_desc = next_rxd;
1350                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1351
1352                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1353         }
1354
1355         rx_ring->next_to_clean = i;
1356         cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
1357
1358         if (cleaned_count)
1359                 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
1360
1361 #ifdef IXGBE_FCOE
1362         /* include DDPed FCoE data */
1363         if (ddp_bytes > 0) {
1364                 unsigned int mss;
1365
1366                 mss = adapter->netdev->mtu - sizeof(struct fcoe_hdr) -
1367                         sizeof(struct fc_frame_header) -
1368                         sizeof(struct fcoe_crc_eof);
1369                 if (mss > 512)
1370                         mss &= ~511;
1371                 total_rx_bytes += ddp_bytes;
1372                 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1373         }
1374 #endif /* IXGBE_FCOE */
1375
1376         rx_ring->total_packets += total_rx_packets;
1377         rx_ring->total_bytes += total_rx_bytes;
1378         netdev->stats.rx_bytes += total_rx_bytes;
1379         netdev->stats.rx_packets += total_rx_packets;
1380
1381         return cleaned;
1382 }
1383
1384 static int ixgbe_clean_rxonly(struct napi_struct *, int);
1385 /**
1386  * ixgbe_configure_msix - Configure MSI-X hardware
1387  * @adapter: board private structure
1388  *
1389  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1390  * interrupts.
1391  **/
1392 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1393 {
1394         struct ixgbe_q_vector *q_vector;
1395         int i, j, q_vectors, v_idx, r_idx;
1396         u32 mask;
1397
1398         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1399
1400         /*
1401          * Populate the IVAR table and set the ITR values to the
1402          * corresponding register.
1403          */
1404         for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1405                 q_vector = adapter->q_vector[v_idx];
1406                 /* XXX for_each_set_bit(...) */
1407                 r_idx = find_first_bit(q_vector->rxr_idx,
1408                                        adapter->num_rx_queues);
1409
1410                 for (i = 0; i < q_vector->rxr_count; i++) {
1411                         j = adapter->rx_ring[r_idx]->reg_idx;
1412                         ixgbe_set_ivar(adapter, 0, j, v_idx);
1413                         r_idx = find_next_bit(q_vector->rxr_idx,
1414                                               adapter->num_rx_queues,
1415                                               r_idx + 1);
1416                 }
1417                 r_idx = find_first_bit(q_vector->txr_idx,
1418                                        adapter->num_tx_queues);
1419
1420                 for (i = 0; i < q_vector->txr_count; i++) {
1421                         j = adapter->tx_ring[r_idx]->reg_idx;
1422                         ixgbe_set_ivar(adapter, 1, j, v_idx);
1423                         r_idx = find_next_bit(q_vector->txr_idx,
1424                                               adapter->num_tx_queues,
1425                                               r_idx + 1);
1426                 }
1427
1428                 if (q_vector->txr_count && !q_vector->rxr_count)
1429                         /* tx only */
1430                         q_vector->eitr = adapter->tx_eitr_param;
1431                 else if (q_vector->rxr_count)
1432                         /* rx or mixed */
1433                         q_vector->eitr = adapter->rx_eitr_param;
1434
1435                 ixgbe_write_eitr(q_vector);
1436                 /* If Flow Director is enabled, set interrupt affinity */
1437                 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
1438                     (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
1439                         /*
1440                          * Allocate the affinity_hint cpumask, assign the mask
1441                          * for this vector, and set our affinity_hint for
1442                          * this irq.
1443                          */
1444                         if (!alloc_cpumask_var(&q_vector->affinity_mask,
1445                                                GFP_KERNEL))
1446                                 return;
1447                         cpumask_set_cpu(v_idx, q_vector->affinity_mask);
1448                         irq_set_affinity_hint(adapter->msix_entries[v_idx].vector,
1449                                               q_vector->affinity_mask);
1450                 }
1451         }
1452
1453         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
1454                 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1455                                v_idx);
1456         else if (adapter->hw.mac.type == ixgbe_mac_82599EB)
1457                 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1458         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1459
1460         /* set up to autoclear timer, and the vectors */
1461         mask = IXGBE_EIMS_ENABLE_MASK;
1462         if (adapter->num_vfs)
1463                 mask &= ~(IXGBE_EIMS_OTHER |
1464                           IXGBE_EIMS_MAILBOX |
1465                           IXGBE_EIMS_LSC);
1466         else
1467                 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
1468         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1469 }
1470
1471 enum latency_range {
1472         lowest_latency = 0,
1473         low_latency = 1,
1474         bulk_latency = 2,
1475         latency_invalid = 255
1476 };
1477
1478 /**
1479  * ixgbe_update_itr - update the dynamic ITR value based on statistics
1480  * @adapter: pointer to adapter
1481  * @eitr: eitr setting (ints per sec) to give last timeslice
1482  * @itr_setting: current throttle rate in ints/second
1483  * @packets: the number of packets during this measurement interval
1484  * @bytes: the number of bytes during this measurement interval
1485  *
1486  *      Stores a new ITR value based on packets and byte
1487  *      counts during the last interrupt.  The advantage of per interrupt
1488  *      computation is faster updates and more accurate ITR for the current
1489  *      traffic pattern.  Constants in this function were computed
1490  *      based on theoretical maximum wire speed and thresholds were set based
1491  *      on testing data as well as attempting to minimize response time
1492  *      while increasing bulk throughput.
1493  *      this functionality is controlled by the InterruptThrottleRate module
1494  *      parameter (see ixgbe_param.c)
1495  **/
1496 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
1497                            u32 eitr, u8 itr_setting,
1498                            int packets, int bytes)
1499 {
1500         unsigned int retval = itr_setting;
1501         u32 timepassed_us;
1502         u64 bytes_perint;
1503
1504         if (packets == 0)
1505                 goto update_itr_done;
1506
1507
1508         /* simple throttlerate management
1509          *    0-20MB/s lowest (100000 ints/s)
1510          *   20-100MB/s low   (20000 ints/s)
1511          *  100-1249MB/s bulk (8000 ints/s)
1512          */
1513         /* what was last interrupt timeslice? */
1514         timepassed_us = 1000000/eitr;
1515         bytes_perint = bytes / timepassed_us; /* bytes/usec */
1516
1517         switch (itr_setting) {
1518         case lowest_latency:
1519                 if (bytes_perint > adapter->eitr_low)
1520                         retval = low_latency;
1521                 break;
1522         case low_latency:
1523                 if (bytes_perint > adapter->eitr_high)
1524                         retval = bulk_latency;
1525                 else if (bytes_perint <= adapter->eitr_low)
1526                         retval = lowest_latency;
1527                 break;
1528         case bulk_latency:
1529                 if (bytes_perint <= adapter->eitr_high)
1530                         retval = low_latency;
1531                 break;
1532         }
1533
1534 update_itr_done:
1535         return retval;
1536 }
1537
1538 /**
1539  * ixgbe_write_eitr - write EITR register in hardware specific way
1540  * @q_vector: structure containing interrupt and ring information
1541  *
1542  * This function is made to be called by ethtool and by the driver
1543  * when it needs to update EITR registers at runtime.  Hardware
1544  * specific quirks/differences are taken care of here.
1545  */
1546 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1547 {
1548         struct ixgbe_adapter *adapter = q_vector->adapter;
1549         struct ixgbe_hw *hw = &adapter->hw;
1550         int v_idx = q_vector->v_idx;
1551         u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1552
1553         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1554                 /* must write high and low 16 bits to reset counter */
1555                 itr_reg |= (itr_reg << 16);
1556         } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1557                 /*
1558                  * 82599 can support a value of zero, so allow it for
1559                  * max interrupt rate, but there is an errata where it can
1560                  * not be zero with RSC
1561                  */
1562                 if (itr_reg == 8 &&
1563                     !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
1564                         itr_reg = 0;
1565
1566                 /*
1567                  * set the WDIS bit to not clear the timer bits and cause an
1568                  * immediate assertion of the interrupt
1569                  */
1570                 itr_reg |= IXGBE_EITR_CNT_WDIS;
1571         }
1572         IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1573 }
1574
1575 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1576 {
1577         struct ixgbe_adapter *adapter = q_vector->adapter;
1578         u32 new_itr;
1579         u8 current_itr, ret_itr;
1580         int i, r_idx;
1581         struct ixgbe_ring *rx_ring, *tx_ring;
1582
1583         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1584         for (i = 0; i < q_vector->txr_count; i++) {
1585                 tx_ring = adapter->tx_ring[r_idx];
1586                 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1587                                            q_vector->tx_itr,
1588                                            tx_ring->total_packets,
1589                                            tx_ring->total_bytes);
1590                 /* if the result for this queue would decrease interrupt
1591                  * rate for this vector then use that result */
1592                 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1593                                     q_vector->tx_itr - 1 : ret_itr);
1594                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1595                                       r_idx + 1);
1596         }
1597
1598         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1599         for (i = 0; i < q_vector->rxr_count; i++) {
1600                 rx_ring = adapter->rx_ring[r_idx];
1601                 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1602                                            q_vector->rx_itr,
1603                                            rx_ring->total_packets,
1604                                            rx_ring->total_bytes);
1605                 /* if the result for this queue would decrease interrupt
1606                  * rate for this vector then use that result */
1607                 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1608                                     q_vector->rx_itr - 1 : ret_itr);
1609                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1610                                       r_idx + 1);
1611         }
1612
1613         current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1614
1615         switch (current_itr) {
1616         /* counts and packets in update_itr are dependent on these numbers */
1617         case lowest_latency:
1618                 new_itr = 100000;
1619                 break;
1620         case low_latency:
1621                 new_itr = 20000; /* aka hwitr = ~200 */
1622                 break;
1623         case bulk_latency:
1624         default:
1625                 new_itr = 8000;
1626                 break;
1627         }
1628
1629         if (new_itr != q_vector->eitr) {
1630                 /* do an exponential smoothing */
1631                 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1632
1633                 /* save the algorithm value here, not the smoothed one */
1634                 q_vector->eitr = new_itr;
1635
1636                 ixgbe_write_eitr(q_vector);
1637         }
1638 }
1639
1640 /**
1641  * ixgbe_check_overtemp_task - worker thread to check over tempurature
1642  * @work: pointer to work_struct containing our data
1643  **/
1644 static void ixgbe_check_overtemp_task(struct work_struct *work)
1645 {
1646         struct ixgbe_adapter *adapter = container_of(work,
1647                                                      struct ixgbe_adapter,
1648                                                      check_overtemp_task);
1649         struct ixgbe_hw *hw = &adapter->hw;
1650         u32 eicr = adapter->interrupt_event;
1651
1652         if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
1653                 return;
1654
1655         switch (hw->device_id) {
1656         case IXGBE_DEV_ID_82599_T3_LOM: {
1657                 u32 autoneg;
1658                 bool link_up = false;
1659
1660                 if (hw->mac.ops.check_link)
1661                         hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
1662
1663                 if (((eicr & IXGBE_EICR_GPI_SDP0) && (!link_up)) ||
1664                     (eicr & IXGBE_EICR_LSC))
1665                         /* Check if this is due to overtemp */
1666                         if (hw->phy.ops.check_overtemp(hw) == IXGBE_ERR_OVERTEMP)
1667                                 break;
1668                 return;
1669         }
1670         default:
1671                 if (!(eicr & IXGBE_EICR_GPI_SDP0))
1672                         return;
1673                 break;
1674         }
1675         e_crit(drv,
1676                "Network adapter has been stopped because it has over heated. "
1677                "Restart the computer. If the problem persists, "
1678                "power off the system and replace the adapter\n");
1679         /* write to clear the interrupt */
1680         IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP0);
1681 }
1682
1683 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1684 {
1685         struct ixgbe_hw *hw = &adapter->hw;
1686
1687         if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1688             (eicr & IXGBE_EICR_GPI_SDP1)) {
1689                 e_crit(probe, "Fan has stopped, replace the adapter\n");
1690                 /* write to clear the interrupt */
1691                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1692         }
1693 }
1694
1695 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1696 {
1697         struct ixgbe_hw *hw = &adapter->hw;
1698
1699         if (eicr & IXGBE_EICR_GPI_SDP1) {
1700                 /* Clear the interrupt */
1701                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1702                 schedule_work(&adapter->multispeed_fiber_task);
1703         } else if (eicr & IXGBE_EICR_GPI_SDP2) {
1704                 /* Clear the interrupt */
1705                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1706                 schedule_work(&adapter->sfp_config_module_task);
1707         } else {
1708                 /* Interrupt isn't for us... */
1709                 return;
1710         }
1711 }
1712
1713 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1714 {
1715         struct ixgbe_hw *hw = &adapter->hw;
1716
1717         adapter->lsc_int++;
1718         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1719         adapter->link_check_timeout = jiffies;
1720         if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1721                 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1722                 IXGBE_WRITE_FLUSH(hw);
1723                 schedule_work(&adapter->watchdog_task);
1724         }
1725 }
1726
1727 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1728 {
1729         struct net_device *netdev = data;
1730         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1731         struct ixgbe_hw *hw = &adapter->hw;
1732         u32 eicr;
1733
1734         /*
1735          * Workaround for Silicon errata.  Use clear-by-write instead
1736          * of clear-by-read.  Reading with EICS will return the
1737          * interrupt causes without clearing, which later be done
1738          * with the write to EICR.
1739          */
1740         eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1741         IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1742
1743         if (eicr & IXGBE_EICR_LSC)
1744                 ixgbe_check_lsc(adapter);
1745
1746         if (eicr & IXGBE_EICR_MAILBOX)
1747                 ixgbe_msg_task(adapter);
1748
1749         if (hw->mac.type == ixgbe_mac_82598EB)
1750                 ixgbe_check_fan_failure(adapter, eicr);
1751
1752         if (hw->mac.type == ixgbe_mac_82599EB) {
1753                 ixgbe_check_sfp_event(adapter, eicr);
1754                 adapter->interrupt_event = eicr;
1755                 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
1756                     ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)))
1757                         schedule_work(&adapter->check_overtemp_task);
1758
1759                 /* Handle Flow Director Full threshold interrupt */
1760                 if (eicr & IXGBE_EICR_FLOW_DIR) {
1761                         int i;
1762                         IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1763                         /* Disable transmits before FDIR Re-initialization */
1764                         netif_tx_stop_all_queues(netdev);
1765                         for (i = 0; i < adapter->num_tx_queues; i++) {
1766                                 struct ixgbe_ring *tx_ring =
1767                                                             adapter->tx_ring[i];
1768                                 if (test_and_clear_bit(__IXGBE_FDIR_INIT_DONE,
1769                                                        &tx_ring->reinit_state))
1770                                         schedule_work(&adapter->fdir_reinit_task);
1771                         }
1772                 }
1773         }
1774         if (!test_bit(__IXGBE_DOWN, &adapter->state))
1775                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1776
1777         return IRQ_HANDLED;
1778 }
1779
1780 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1781                                            u64 qmask)
1782 {
1783         u32 mask;
1784
1785         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1786                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1787                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1788         } else {
1789                 mask = (qmask & 0xFFFFFFFF);
1790                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(0), mask);
1791                 mask = (qmask >> 32);
1792                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(1), mask);
1793         }
1794         /* skip the flush */
1795 }
1796
1797 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1798                                             u64 qmask)
1799 {
1800         u32 mask;
1801
1802         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1803                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1804                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, mask);
1805         } else {
1806                 mask = (qmask & 0xFFFFFFFF);
1807                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), mask);
1808                 mask = (qmask >> 32);
1809                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), mask);
1810         }
1811         /* skip the flush */
1812 }
1813
1814 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1815 {
1816         struct ixgbe_q_vector *q_vector = data;
1817         struct ixgbe_adapter  *adapter = q_vector->adapter;
1818         struct ixgbe_ring     *tx_ring;
1819         int i, r_idx;
1820
1821         if (!q_vector->txr_count)
1822                 return IRQ_HANDLED;
1823
1824         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1825         for (i = 0; i < q_vector->txr_count; i++) {
1826                 tx_ring = adapter->tx_ring[r_idx];
1827                 tx_ring->total_bytes = 0;
1828                 tx_ring->total_packets = 0;
1829                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1830                                       r_idx + 1);
1831         }
1832
1833         /* EIAM disabled interrupts (on this vector) for us */
1834         napi_schedule(&q_vector->napi);
1835
1836         return IRQ_HANDLED;
1837 }
1838
1839 /**
1840  * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
1841  * @irq: unused
1842  * @data: pointer to our q_vector struct for this interrupt vector
1843  **/
1844 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
1845 {
1846         struct ixgbe_q_vector *q_vector = data;
1847         struct ixgbe_adapter  *adapter = q_vector->adapter;
1848         struct ixgbe_ring  *rx_ring;
1849         int r_idx;
1850         int i;
1851
1852         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1853         for (i = 0;  i < q_vector->rxr_count; i++) {
1854                 rx_ring = adapter->rx_ring[r_idx];
1855                 rx_ring->total_bytes = 0;
1856                 rx_ring->total_packets = 0;
1857                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1858                                       r_idx + 1);
1859         }
1860
1861         if (!q_vector->rxr_count)
1862                 return IRQ_HANDLED;
1863
1864         /* disable interrupts on this vector only */
1865         /* EIAM disabled interrupts (on this vector) for us */
1866         napi_schedule(&q_vector->napi);
1867
1868         return IRQ_HANDLED;
1869 }
1870
1871 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1872 {
1873         struct ixgbe_q_vector *q_vector = data;
1874         struct ixgbe_adapter  *adapter = q_vector->adapter;
1875         struct ixgbe_ring  *ring;
1876         int r_idx;
1877         int i;
1878
1879         if (!q_vector->txr_count && !q_vector->rxr_count)
1880                 return IRQ_HANDLED;
1881
1882         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1883         for (i = 0; i < q_vector->txr_count; i++) {
1884                 ring = adapter->tx_ring[r_idx];
1885                 ring->total_bytes = 0;
1886                 ring->total_packets = 0;
1887                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1888                                       r_idx + 1);
1889         }
1890
1891         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1892         for (i = 0; i < q_vector->rxr_count; i++) {
1893                 ring = adapter->rx_ring[r_idx];
1894                 ring->total_bytes = 0;
1895                 ring->total_packets = 0;
1896                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1897                                       r_idx + 1);
1898         }
1899
1900         /* EIAM disabled interrupts (on this vector) for us */
1901         napi_schedule(&q_vector->napi);
1902
1903         return IRQ_HANDLED;
1904 }
1905
1906 /**
1907  * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1908  * @napi: napi struct with our devices info in it
1909  * @budget: amount of work driver is allowed to do this pass, in packets
1910  *
1911  * This function is optimized for cleaning one queue only on a single
1912  * q_vector!!!
1913  **/
1914 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1915 {
1916         struct ixgbe_q_vector *q_vector =
1917                                container_of(napi, struct ixgbe_q_vector, napi);
1918         struct ixgbe_adapter *adapter = q_vector->adapter;
1919         struct ixgbe_ring *rx_ring = NULL;
1920         int work_done = 0;
1921         long r_idx;
1922
1923         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1924         rx_ring = adapter->rx_ring[r_idx];
1925 #ifdef CONFIG_IXGBE_DCA
1926         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1927                 ixgbe_update_rx_dca(adapter, rx_ring);
1928 #endif
1929
1930         ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1931
1932         /* If all Rx work done, exit the polling mode */
1933         if (work_done < budget) {
1934                 napi_complete(napi);
1935                 if (adapter->rx_itr_setting & 1)
1936                         ixgbe_set_itr_msix(q_vector);
1937                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1938                         ixgbe_irq_enable_queues(adapter,
1939                                                 ((u64)1 << q_vector->v_idx));
1940         }
1941
1942         return work_done;
1943 }
1944
1945 /**
1946  * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
1947  * @napi: napi struct with our devices info in it
1948  * @budget: amount of work driver is allowed to do this pass, in packets
1949  *
1950  * This function will clean more than one rx queue associated with a
1951  * q_vector.
1952  **/
1953 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
1954 {
1955         struct ixgbe_q_vector *q_vector =
1956                                container_of(napi, struct ixgbe_q_vector, napi);
1957         struct ixgbe_adapter *adapter = q_vector->adapter;
1958         struct ixgbe_ring *ring = NULL;
1959         int work_done = 0, i;
1960         long r_idx;
1961         bool tx_clean_complete = true;
1962
1963         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1964         for (i = 0; i < q_vector->txr_count; i++) {
1965                 ring = adapter->tx_ring[r_idx];
1966 #ifdef CONFIG_IXGBE_DCA
1967                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1968                         ixgbe_update_tx_dca(adapter, ring);
1969 #endif
1970                 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
1971                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1972                                       r_idx + 1);
1973         }
1974
1975         /* attempt to distribute budget to each queue fairly, but don't allow
1976          * the budget to go below 1 because we'll exit polling */
1977         budget /= (q_vector->rxr_count ?: 1);
1978         budget = max(budget, 1);
1979         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1980         for (i = 0; i < q_vector->rxr_count; i++) {
1981                 ring = adapter->rx_ring[r_idx];
1982 #ifdef CONFIG_IXGBE_DCA
1983                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1984                         ixgbe_update_rx_dca(adapter, ring);
1985 #endif
1986                 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
1987                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1988                                       r_idx + 1);
1989         }
1990
1991         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1992         ring = adapter->rx_ring[r_idx];
1993         /* If all Rx work done, exit the polling mode */
1994         if (work_done < budget) {
1995                 napi_complete(napi);
1996                 if (adapter->rx_itr_setting & 1)
1997                         ixgbe_set_itr_msix(q_vector);
1998                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1999                         ixgbe_irq_enable_queues(adapter,
2000                                                 ((u64)1 << q_vector->v_idx));
2001                 return 0;
2002         }
2003
2004         return work_done;
2005 }
2006
2007 /**
2008  * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
2009  * @napi: napi struct with our devices info in it
2010  * @budget: amount of work driver is allowed to do this pass, in packets
2011  *
2012  * This function is optimized for cleaning one queue only on a single
2013  * q_vector!!!
2014  **/
2015 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
2016 {
2017         struct ixgbe_q_vector *q_vector =
2018                                container_of(napi, struct ixgbe_q_vector, napi);
2019         struct ixgbe_adapter *adapter = q_vector->adapter;
2020         struct ixgbe_ring *tx_ring = NULL;
2021         int work_done = 0;
2022         long r_idx;
2023
2024         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2025         tx_ring = adapter->tx_ring[r_idx];
2026 #ifdef CONFIG_IXGBE_DCA
2027         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2028                 ixgbe_update_tx_dca(adapter, tx_ring);
2029 #endif
2030
2031         if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
2032                 work_done = budget;
2033
2034         /* If all Tx work done, exit the polling mode */
2035         if (work_done < budget) {
2036                 napi_complete(napi);
2037                 if (adapter->tx_itr_setting & 1)
2038                         ixgbe_set_itr_msix(q_vector);
2039                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2040                         ixgbe_irq_enable_queues(adapter,
2041                                                 ((u64)1 << q_vector->v_idx));
2042         }
2043
2044         return work_done;
2045 }
2046
2047 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
2048                                      int r_idx)
2049 {
2050         struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2051
2052         set_bit(r_idx, q_vector->rxr_idx);
2053         q_vector->rxr_count++;
2054 }
2055
2056 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
2057                                      int t_idx)
2058 {
2059         struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2060
2061         set_bit(t_idx, q_vector->txr_idx);
2062         q_vector->txr_count++;
2063 }
2064
2065 /**
2066  * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
2067  * @adapter: board private structure to initialize
2068  * @vectors: allotted vector count for descriptor rings
2069  *
2070  * This function maps descriptor rings to the queue-specific vectors
2071  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
2072  * one vector per ring/queue, but on a constrained vector budget, we
2073  * group the rings as "efficiently" as possible.  You would add new
2074  * mapping configurations in here.
2075  **/
2076 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
2077                                       int vectors)
2078 {
2079         int v_start = 0;
2080         int rxr_idx = 0, txr_idx = 0;
2081         int rxr_remaining = adapter->num_rx_queues;
2082         int txr_remaining = adapter->num_tx_queues;
2083         int i, j;
2084         int rqpv, tqpv;
2085         int err = 0;
2086
2087         /* No mapping required if MSI-X is disabled. */
2088         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2089                 goto out;
2090
2091         /*
2092          * The ideal configuration...
2093          * We have enough vectors to map one per queue.
2094          */
2095         if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
2096                 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
2097                         map_vector_to_rxq(adapter, v_start, rxr_idx);
2098
2099                 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
2100                         map_vector_to_txq(adapter, v_start, txr_idx);
2101
2102                 goto out;
2103         }
2104
2105         /*
2106          * If we don't have enough vectors for a 1-to-1
2107          * mapping, we'll have to group them so there are
2108          * multiple queues per vector.
2109          */
2110         /* Re-adjusting *qpv takes care of the remainder. */
2111         for (i = v_start; i < vectors; i++) {
2112                 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
2113                 for (j = 0; j < rqpv; j++) {
2114                         map_vector_to_rxq(adapter, i, rxr_idx);
2115                         rxr_idx++;
2116                         rxr_remaining--;
2117                 }
2118         }
2119         for (i = v_start; i < vectors; i++) {
2120                 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
2121                 for (j = 0; j < tqpv; j++) {
2122                         map_vector_to_txq(adapter, i, txr_idx);
2123                         txr_idx++;
2124                         txr_remaining--;
2125                 }
2126         }
2127
2128 out:
2129         return err;
2130 }
2131
2132 /**
2133  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2134  * @adapter: board private structure
2135  *
2136  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2137  * interrupts from the kernel.
2138  **/
2139 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2140 {
2141         struct net_device *netdev = adapter->netdev;
2142         irqreturn_t (*handler)(int, void *);
2143         int i, vector, q_vectors, err;
2144         int ri = 0, ti = 0;
2145
2146         /* Decrement for Other and TCP Timer vectors */
2147         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2148
2149         /* Map the Tx/Rx rings to the vectors we were allotted. */
2150         err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
2151         if (err)
2152                 goto out;
2153
2154 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
2155                          (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
2156                          &ixgbe_msix_clean_many)
2157         for (vector = 0; vector < q_vectors; vector++) {
2158                 handler = SET_HANDLER(adapter->q_vector[vector]);
2159
2160                 if (handler == &ixgbe_msix_clean_rx) {
2161                         sprintf(adapter->name[vector], "%s-%s-%d",
2162                                 netdev->name, "rx", ri++);
2163                 } else if (handler == &ixgbe_msix_clean_tx) {
2164                         sprintf(adapter->name[vector], "%s-%s-%d",
2165                                 netdev->name, "tx", ti++);
2166                 } else
2167                         sprintf(adapter->name[vector], "%s-%s-%d",
2168                                 netdev->name, "TxRx", vector);
2169
2170                 err = request_irq(adapter->msix_entries[vector].vector,
2171                                   handler, 0, adapter->name[vector],
2172                                   adapter->q_vector[vector]);
2173                 if (err) {
2174                         e_err(probe, "request_irq failed for MSIX interrupt "
2175                               "Error: %d\n", err);
2176                         goto free_queue_irqs;
2177                 }
2178         }
2179
2180         sprintf(adapter->name[vector], "%s:lsc", netdev->name);
2181         err = request_irq(adapter->msix_entries[vector].vector,
2182                           ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
2183         if (err) {
2184                 e_err(probe, "request_irq for msix_lsc failed: %d\n", err);
2185                 goto free_queue_irqs;
2186         }
2187
2188         return 0;
2189
2190 free_queue_irqs:
2191         for (i = vector - 1; i >= 0; i--)
2192                 free_irq(adapter->msix_entries[--vector].vector,
2193                          adapter->q_vector[i]);
2194         adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2195         pci_disable_msix(adapter->pdev);
2196         kfree(adapter->msix_entries);
2197         adapter->msix_entries = NULL;
2198 out:
2199         return err;
2200 }
2201
2202 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
2203 {
2204         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2205         u8 current_itr;
2206         u32 new_itr = q_vector->eitr;
2207         struct ixgbe_ring *rx_ring = adapter->rx_ring[0];
2208         struct ixgbe_ring *tx_ring = adapter->tx_ring[0];
2209
2210         q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
2211                                             q_vector->tx_itr,
2212                                             tx_ring->total_packets,
2213                                             tx_ring->total_bytes);
2214         q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
2215                                             q_vector->rx_itr,
2216                                             rx_ring->total_packets,
2217                                             rx_ring->total_bytes);
2218
2219         current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
2220
2221         switch (current_itr) {
2222         /* counts and packets in update_itr are dependent on these numbers */
2223         case lowest_latency:
2224                 new_itr = 100000;
2225                 break;
2226         case low_latency:
2227                 new_itr = 20000; /* aka hwitr = ~200 */
2228                 break;
2229         case bulk_latency:
2230                 new_itr = 8000;
2231                 break;
2232         default:
2233                 break;
2234         }
2235
2236         if (new_itr != q_vector->eitr) {
2237                 /* do an exponential smoothing */
2238                 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
2239
2240                 /* save the algorithm value here, not the smoothed one */
2241                 q_vector->eitr = new_itr;
2242
2243                 ixgbe_write_eitr(q_vector);
2244         }
2245 }
2246
2247 /**
2248  * ixgbe_irq_enable - Enable default interrupt generation settings
2249  * @adapter: board private structure
2250  **/
2251 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2252                                     bool flush)
2253 {
2254         u32 mask;
2255
2256         mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2257         if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2258                 mask |= IXGBE_EIMS_GPI_SDP0;
2259         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2260                 mask |= IXGBE_EIMS_GPI_SDP1;
2261         if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
2262                 mask |= IXGBE_EIMS_ECC;
2263                 mask |= IXGBE_EIMS_GPI_SDP1;
2264                 mask |= IXGBE_EIMS_GPI_SDP2;
2265                 if (adapter->num_vfs)
2266                         mask |= IXGBE_EIMS_MAILBOX;
2267         }
2268         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
2269             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
2270                 mask |= IXGBE_EIMS_FLOW_DIR;
2271
2272         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2273         if (queues)
2274                 ixgbe_irq_enable_queues(adapter, ~0);
2275         if (flush)
2276                 IXGBE_WRITE_FLUSH(&adapter->hw);
2277
2278         if (adapter->num_vfs > 32) {
2279                 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2280                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2281         }
2282 }
2283
2284 /**
2285  * ixgbe_intr - legacy mode Interrupt Handler
2286  * @irq: interrupt number
2287  * @data: pointer to a network interface device structure
2288  **/
2289 static irqreturn_t ixgbe_intr(int irq, void *data)
2290 {
2291         struct net_device *netdev = data;
2292         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2293         struct ixgbe_hw *hw = &adapter->hw;
2294         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2295         u32 eicr;
2296
2297         /*
2298          * Workaround for silicon errata on 82598.  Mask the interrupts
2299          * before the read of EICR.
2300          */
2301         IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2302
2303         /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2304          * therefore no explict interrupt disable is necessary */
2305         eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2306         if (!eicr) {
2307                 /*
2308                  * shared interrupt alert!
2309                  * make sure interrupts are enabled because the read will
2310                  * have disabled interrupts due to EIAM
2311                  * finish the workaround of silicon errata on 82598.  Unmask
2312                  * the interrupt that we masked before the EICR read.
2313                  */
2314                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2315                         ixgbe_irq_enable(adapter, true, true);
2316                 return IRQ_NONE;        /* Not our interrupt */
2317         }
2318
2319         if (eicr & IXGBE_EICR_LSC)
2320                 ixgbe_check_lsc(adapter);
2321
2322         if (hw->mac.type == ixgbe_mac_82599EB)
2323                 ixgbe_check_sfp_event(adapter, eicr);
2324
2325         ixgbe_check_fan_failure(adapter, eicr);
2326         if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2327             ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)))
2328                 schedule_work(&adapter->check_overtemp_task);
2329
2330         if (napi_schedule_prep(&(q_vector->napi))) {
2331                 adapter->tx_ring[0]->total_packets = 0;
2332                 adapter->tx_ring[0]->total_bytes = 0;
2333                 adapter->rx_ring[0]->total_packets = 0;
2334                 adapter->rx_ring[0]->total_bytes = 0;
2335                 /* would disable interrupts here but EIAM disabled it */
2336                 __napi_schedule(&(q_vector->napi));
2337         }
2338
2339         /*
2340          * re-enable link(maybe) and non-queue interrupts, no flush.
2341          * ixgbe_poll will re-enable the queue interrupts
2342          */
2343
2344         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2345                 ixgbe_irq_enable(adapter, false, false);
2346
2347         return IRQ_HANDLED;
2348 }
2349
2350 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
2351 {
2352         int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2353
2354         for (i = 0; i < q_vectors; i++) {
2355                 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
2356                 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
2357                 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
2358                 q_vector->rxr_count = 0;
2359                 q_vector->txr_count = 0;
2360         }
2361 }
2362
2363 /**
2364  * ixgbe_request_irq - initialize interrupts
2365  * @adapter: board private structure
2366  *
2367  * Attempts to configure interrupts using the best available
2368  * capabilities of the hardware and kernel.
2369  **/
2370 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2371 {
2372         struct net_device *netdev = adapter->netdev;
2373         int err;
2374
2375         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2376                 err = ixgbe_request_msix_irqs(adapter);
2377         } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
2378                 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2379                                   netdev->name, netdev);
2380         } else {
2381                 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2382                                   netdev->name, netdev);
2383         }
2384
2385         if (err)
2386                 e_err(probe, "request_irq failed, Error %d\n", err);
2387
2388         return err;
2389 }
2390
2391 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2392 {
2393         struct net_device *netdev = adapter->netdev;
2394
2395         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2396                 int i, q_vectors;
2397
2398                 q_vectors = adapter->num_msix_vectors;
2399
2400                 i = q_vectors - 1;
2401                 free_irq(adapter->msix_entries[i].vector, netdev);
2402
2403                 i--;
2404                 for (; i >= 0; i--) {
2405                         free_irq(adapter->msix_entries[i].vector,
2406                                  adapter->q_vector[i]);
2407                 }
2408
2409                 ixgbe_reset_q_vectors(adapter);
2410         } else {
2411                 free_irq(adapter->pdev->irq, netdev);
2412         }
2413 }
2414
2415 /**
2416  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2417  * @adapter: board private structure
2418  **/
2419 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2420 {
2421         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2422                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2423         } else {
2424                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2425                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2426                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2427                 if (adapter->num_vfs > 32)
2428                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
2429         }
2430         IXGBE_WRITE_FLUSH(&adapter->hw);
2431         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2432                 int i;
2433                 for (i = 0; i < adapter->num_msix_vectors; i++)
2434                         synchronize_irq(adapter->msix_entries[i].vector);
2435         } else {
2436                 synchronize_irq(adapter->pdev->irq);
2437         }
2438 }
2439
2440 /**
2441  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2442  *
2443  **/
2444 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2445 {
2446         struct ixgbe_hw *hw = &adapter->hw;
2447
2448         IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
2449                         EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
2450
2451         ixgbe_set_ivar(adapter, 0, 0, 0);
2452         ixgbe_set_ivar(adapter, 1, 0, 0);
2453
2454         map_vector_to_rxq(adapter, 0, 0);
2455         map_vector_to_txq(adapter, 0, 0);
2456
2457         e_info(hw, "Legacy interrupt IVAR setup done\n");
2458 }
2459
2460 /**
2461  * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2462  * @adapter: board private structure
2463  * @ring: structure containing ring specific data
2464  *
2465  * Configure the Tx descriptor ring after a reset.
2466  **/
2467 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2468                              struct ixgbe_ring *ring)
2469 {
2470         struct ixgbe_hw *hw = &adapter->hw;
2471         u64 tdba = ring->dma;
2472         int wait_loop = 10;
2473         u32 txdctl;
2474         u16 reg_idx = ring->reg_idx;
2475
2476         /* disable queue to avoid issues while updating state */
2477         txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2478         IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx),
2479                         txdctl & ~IXGBE_TXDCTL_ENABLE);
2480         IXGBE_WRITE_FLUSH(hw);
2481
2482         IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2483                         (tdba & DMA_BIT_MASK(32)));
2484         IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2485         IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2486                         ring->count * sizeof(union ixgbe_adv_tx_desc));
2487         IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2488         IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2489         ring->head = IXGBE_TDH(reg_idx);
2490         ring->tail = IXGBE_TDT(reg_idx);
2491
2492         /* configure fetching thresholds */
2493         if (adapter->rx_itr_setting == 0) {
2494                 /* cannot set wthresh when itr==0 */
2495                 txdctl &= ~0x007F0000;
2496         } else {
2497                 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2498                 txdctl |= (8 << 16);
2499         }
2500         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2501                 /* PThresh workaround for Tx hang with DFP enabled. */
2502                 txdctl |= 32;
2503         }
2504
2505         /* reinitialize flowdirector state */
2506         set_bit(__IXGBE_FDIR_INIT_DONE, &ring->reinit_state);
2507
2508         /* enable queue */
2509         txdctl |= IXGBE_TXDCTL_ENABLE;
2510         IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2511
2512         /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2513         if (hw->mac.type == ixgbe_mac_82598EB &&
2514             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2515                 return;
2516
2517         /* poll to verify queue is enabled */
2518         do {
2519                 msleep(1);
2520                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2521         } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2522         if (!wait_loop)
2523                 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2524 }
2525
2526 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2527 {
2528         struct ixgbe_hw *hw = &adapter->hw;
2529         u32 rttdcs;
2530         u32 mask;
2531
2532         if (hw->mac.type == ixgbe_mac_82598EB)
2533                 return;
2534
2535         /* disable the arbiter while setting MTQC */
2536         rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2537         rttdcs |= IXGBE_RTTDCS_ARBDIS;
2538         IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2539
2540         /* set transmit pool layout */
2541         mask = (IXGBE_FLAG_SRIOV_ENABLED | IXGBE_FLAG_DCB_ENABLED);
2542         switch (adapter->flags & mask) {
2543
2544         case (IXGBE_FLAG_SRIOV_ENABLED):
2545                 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2546                                 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2547                 break;
2548
2549         case (IXGBE_FLAG_DCB_ENABLED):
2550                 /* We enable 8 traffic classes, DCB only */
2551                 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2552                               (IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ));
2553                 break;
2554
2555         default:
2556                 IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
2557                 break;
2558         }
2559
2560         /* re-enable the arbiter */
2561         rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2562         IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2563 }
2564
2565 /**
2566  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2567  * @adapter: board private structure
2568  *
2569  * Configure the Tx unit of the MAC after a reset.
2570  **/
2571 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2572 {
2573         struct ixgbe_hw *hw = &adapter->hw;
2574         u32 dmatxctl;
2575         u32 i;
2576
2577         ixgbe_setup_mtqc(adapter);
2578
2579         if (hw->mac.type != ixgbe_mac_82598EB) {
2580                 /* DMATXCTL.EN must be before Tx queues are enabled */
2581                 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2582                 dmatxctl |= IXGBE_DMATXCTL_TE;
2583                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2584         }
2585
2586         /* Setup the HW Tx Head and Tail descriptor pointers */
2587         for (i = 0; i < adapter->num_tx_queues; i++)
2588                 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2589 }
2590
2591 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2592
2593 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2594                                    struct ixgbe_ring *rx_ring)
2595 {
2596         u32 srrctl;
2597         int index;
2598         struct ixgbe_ring_feature *feature = adapter->ring_feature;
2599
2600         index = rx_ring->reg_idx;
2601         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2602                 unsigned long mask;
2603                 mask = (unsigned long) feature[RING_F_RSS].mask;
2604                 index = index & mask;
2605         }
2606         srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
2607
2608         srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2609         srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2610         if (adapter->num_vfs)
2611                 srrctl |= IXGBE_SRRCTL_DROP_EN;
2612
2613         srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2614                   IXGBE_SRRCTL_BSIZEHDR_MASK;
2615
2616         if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
2617 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2618                 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2619 #else
2620                 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2621 #endif
2622                 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2623         } else {
2624                 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2625                           IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2626                 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2627         }
2628
2629         IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
2630 }
2631
2632 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2633 {
2634         struct ixgbe_hw *hw = &adapter->hw;
2635         static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2636                           0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2637                           0x6A3E67EA, 0x14364D17, 0x3BED200D};
2638         u32 mrqc = 0, reta = 0;
2639         u32 rxcsum;
2640         int i, j;
2641         int mask;
2642
2643         /* Fill out hash function seeds */
2644         for (i = 0; i < 10; i++)
2645                 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2646
2647         /* Fill out redirection table */
2648         for (i = 0, j = 0; i < 128; i++, j++) {
2649                 if (j == adapter->ring_feature[RING_F_RSS].indices)
2650                         j = 0;
2651                 /* reta = 4-byte sliding window of
2652                  * 0x00..(indices-1)(indices-1)00..etc. */
2653                 reta = (reta << 8) | (j * 0x11);
2654                 if ((i & 3) == 3)
2655                         IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2656         }
2657
2658         /* Disable indicating checksum in descriptor, enables RSS hash */
2659         rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2660         rxcsum |= IXGBE_RXCSUM_PCSD;
2661         IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2662
2663         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
2664                 mask = adapter->flags & IXGBE_FLAG_RSS_ENABLED;
2665         else
2666                 mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2667 #ifdef CONFIG_IXGBE_DCB
2668                                          | IXGBE_FLAG_DCB_ENABLED
2669 #endif
2670                                          | IXGBE_FLAG_SRIOV_ENABLED
2671                                         );
2672
2673         switch (mask) {
2674         case (IXGBE_FLAG_RSS_ENABLED):
2675                 mrqc = IXGBE_MRQC_RSSEN;
2676                 break;
2677         case (IXGBE_FLAG_SRIOV_ENABLED):
2678                 mrqc = IXGBE_MRQC_VMDQEN;
2679                 break;
2680 #ifdef CONFIG_IXGBE_DCB
2681         case (IXGBE_FLAG_DCB_ENABLED):
2682                 mrqc = IXGBE_MRQC_RT8TCEN;
2683                 break;
2684 #endif /* CONFIG_IXGBE_DCB */
2685         default:
2686                 break;
2687         }
2688
2689         /* Perform hash on these packet types */
2690         mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2691               | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2692               | IXGBE_MRQC_RSS_FIELD_IPV6
2693               | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2694
2695         IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2696 }
2697
2698 /**
2699  * ixgbe_configure_rscctl - enable RSC for the indicated ring
2700  * @adapter:    address of board private structure
2701  * @index:      index of ring to set
2702  **/
2703 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2704                                    struct ixgbe_ring *ring)
2705 {
2706         struct ixgbe_hw *hw = &adapter->hw;
2707         u32 rscctrl;
2708         int rx_buf_len;
2709         u16 reg_idx = ring->reg_idx;
2710
2711         if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
2712                 return;
2713
2714         rx_buf_len = ring->rx_buf_len;
2715         rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2716         rscctrl |= IXGBE_RSCCTL_RSCEN;
2717         /*
2718          * we must limit the number of descriptors so that the
2719          * total size of max desc * buf_len is not greater
2720          * than 65535
2721          */
2722         if (ring->flags & IXGBE_RING_RX_PS_ENABLED) {
2723 #if (MAX_SKB_FRAGS > 16)
2724                 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2725 #elif (MAX_SKB_FRAGS > 8)
2726                 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2727 #elif (MAX_SKB_FRAGS > 4)
2728                 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2729 #else
2730                 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2731 #endif
2732         } else {
2733                 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2734                         rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2735                 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2736                         rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2737                 else
2738                         rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2739         }
2740         IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2741 }
2742
2743 /**
2744  *  ixgbe_set_uta - Set unicast filter table address
2745  *  @adapter: board private structure
2746  *
2747  *  The unicast table address is a register array of 32-bit registers.
2748  *  The table is meant to be used in a way similar to how the MTA is used
2749  *  however due to certain limitations in the hardware it is necessary to
2750  *  set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
2751  *  enable bit to allow vlan tag stripping when promiscuous mode is enabled
2752  **/
2753 static void ixgbe_set_uta(struct ixgbe_adapter *adapter)
2754 {
2755         struct ixgbe_hw *hw = &adapter->hw;
2756         int i;
2757
2758         /* The UTA table only exists on 82599 hardware and newer */
2759         if (hw->mac.type < ixgbe_mac_82599EB)
2760                 return;
2761
2762         /* we only need to do this if VMDq is enabled */
2763         if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2764                 return;
2765
2766         for (i = 0; i < 128; i++)
2767                 IXGBE_WRITE_REG(hw, IXGBE_UTA(i), ~0);
2768 }
2769
2770 #define IXGBE_MAX_RX_DESC_POLL 10
2771 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2772                                        struct ixgbe_ring *ring)
2773 {
2774         struct ixgbe_hw *hw = &adapter->hw;
2775         int reg_idx = ring->reg_idx;
2776         int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2777         u32 rxdctl;
2778
2779         /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2780         if (hw->mac.type == ixgbe_mac_82598EB &&
2781             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2782                 return;
2783
2784         do {
2785                 msleep(1);
2786                 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2787         } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
2788
2789         if (!wait_loop) {
2790                 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
2791                       "the polling period\n", reg_idx);
2792         }
2793 }
2794
2795 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
2796                              struct ixgbe_ring *ring)
2797 {
2798         struct ixgbe_hw *hw = &adapter->hw;
2799         u64 rdba = ring->dma;
2800         u32 rxdctl;
2801         u16 reg_idx = ring->reg_idx;
2802
2803         /* disable queue to avoid issues while updating state */
2804         rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2805         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx),
2806                         rxdctl & ~IXGBE_RXDCTL_ENABLE);
2807         IXGBE_WRITE_FLUSH(hw);
2808
2809         IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
2810         IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
2811         IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
2812                         ring->count * sizeof(union ixgbe_adv_rx_desc));
2813         IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
2814         IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
2815         ring->head = IXGBE_RDH(reg_idx);
2816         ring->tail = IXGBE_RDT(reg_idx);
2817
2818         ixgbe_configure_srrctl(adapter, ring);
2819         ixgbe_configure_rscctl(adapter, ring);
2820
2821         if (hw->mac.type == ixgbe_mac_82598EB) {
2822                 /*
2823                  * enable cache line friendly hardware writes:
2824                  * PTHRESH=32 descriptors (half the internal cache),
2825                  * this also removes ugly rx_no_buffer_count increment
2826                  * HTHRESH=4 descriptors (to minimize latency on fetch)
2827                  * WTHRESH=8 burst writeback up to two cache lines
2828                  */
2829                 rxdctl &= ~0x3FFFFF;
2830                 rxdctl |=  0x080420;
2831         }
2832
2833         /* enable receive descriptor ring */
2834         rxdctl |= IXGBE_RXDCTL_ENABLE;
2835         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
2836
2837         ixgbe_rx_desc_queue_enable(adapter, ring);
2838         ixgbe_alloc_rx_buffers(adapter, ring, IXGBE_DESC_UNUSED(ring));
2839 }
2840
2841 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
2842 {
2843         struct ixgbe_hw *hw = &adapter->hw;
2844         int p;
2845
2846         /* PSRTYPE must be initialized in non 82598 adapters */
2847         u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2848                       IXGBE_PSRTYPE_UDPHDR |
2849                       IXGBE_PSRTYPE_IPV4HDR |
2850                       IXGBE_PSRTYPE_L2HDR |
2851                       IXGBE_PSRTYPE_IPV6HDR;
2852
2853         if (hw->mac.type == ixgbe_mac_82598EB)
2854                 return;
2855
2856         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
2857                 psrtype |= (adapter->num_rx_queues_per_pool << 29);
2858
2859         for (p = 0; p < adapter->num_rx_pools; p++)
2860                 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
2861                                 psrtype);
2862 }
2863
2864 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
2865 {
2866         struct ixgbe_hw *hw = &adapter->hw;
2867         u32 gcr_ext;
2868         u32 vt_reg_bits;
2869         u32 reg_offset, vf_shift;
2870         u32 vmdctl;
2871
2872         if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2873                 return;
2874
2875         vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
2876         vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN | IXGBE_VT_CTL_REPLEN;
2877         vt_reg_bits |= (adapter->num_vfs << IXGBE_VT_CTL_POOL_SHIFT);
2878         IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
2879
2880         vf_shift = adapter->num_vfs % 32;
2881         reg_offset = (adapter->num_vfs > 32) ? 1 : 0;
2882
2883         /* Enable only the PF's pool for Tx/Rx */
2884         IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
2885         IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), 0);
2886         IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
2887         IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), 0);
2888         IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2889
2890         /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
2891         hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
2892
2893         /*
2894          * Set up VF register offsets for selected VT Mode,
2895          * i.e. 32 or 64 VFs for SR-IOV
2896          */
2897         gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
2898         gcr_ext |= IXGBE_GCR_EXT_MSIX_EN;
2899         gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
2900         IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
2901
2902         /* enable Tx loopback for VF/PF communication */
2903         IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2904 }
2905
2906 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
2907 {
2908         struct ixgbe_hw *hw = &adapter->hw;
2909         struct net_device *netdev = adapter->netdev;
2910         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2911         int rx_buf_len;
2912         struct ixgbe_ring *rx_ring;
2913         int i;
2914         u32 mhadd, hlreg0;
2915
2916         /* Decide whether to use packet split mode or not */
2917         /* Do not use packet split if we're in SR-IOV Mode */
2918         if (!adapter->num_vfs)
2919                 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
2920
2921         /* Set the RX buffer length according to the mode */
2922         if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2923                 rx_buf_len = IXGBE_RX_HDR_SIZE;
2924         } else {
2925                 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
2926                     (netdev->mtu <= ETH_DATA_LEN))
2927                         rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2928                 else
2929                         rx_buf_len = ALIGN(max_frame + VLAN_HLEN, 1024);
2930         }
2931
2932 #ifdef IXGBE_FCOE
2933         /* adjust max frame to be able to do baby jumbo for FCoE */
2934         if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
2935             (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
2936                 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
2937
2938 #endif /* IXGBE_FCOE */
2939         mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
2940         if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
2941                 mhadd &= ~IXGBE_MHADD_MFS_MASK;
2942                 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
2943
2944                 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
2945         }
2946
2947         hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2948         /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
2949         hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2950         IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
2951
2952         /*
2953          * Setup the HW Rx Head and Tail Descriptor Pointers and
2954          * the Base and Length of the Rx Descriptor Ring
2955          */
2956         for (i = 0; i < adapter->num_rx_queues; i++) {
2957                 rx_ring = adapter->rx_ring[i];
2958                 rx_ring->rx_buf_len = rx_buf_len;
2959
2960                 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
2961                         rx_ring->flags |= IXGBE_RING_RX_PS_ENABLED;
2962                 else
2963                         rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2964
2965 #ifdef IXGBE_FCOE
2966                 if (netdev->features & NETIF_F_FCOE_MTU) {
2967                         struct ixgbe_ring_feature *f;
2968                         f = &adapter->ring_feature[RING_F_FCOE];
2969                         if ((i >= f->mask) && (i < f->mask + f->indices)) {
2970                                 rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2971                                 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
2972                                         rx_ring->rx_buf_len =
2973                                                 IXGBE_FCOE_JUMBO_FRAME_SIZE;
2974                         }
2975                 }
2976 #endif /* IXGBE_FCOE */
2977         }
2978
2979 }
2980
2981 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
2982 {
2983         struct ixgbe_hw *hw = &adapter->hw;
2984         u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2985
2986         switch (hw->mac.type) {
2987         case ixgbe_mac_82598EB:
2988                 /*
2989                  * For VMDq support of different descriptor types or
2990                  * buffer sizes through the use of multiple SRRCTL
2991                  * registers, RDRXCTL.MVMEN must be set to 1
2992                  *
2993                  * also, the manual doesn't mention it clearly but DCA hints
2994                  * will only use queue 0's tags unless this bit is set.  Side
2995                  * effects of setting this bit are only that SRRCTL must be
2996                  * fully programmed [0..15]
2997                  */
2998                 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
2999                 break;
3000         case ixgbe_mac_82599EB:
3001                 /* Disable RSC for ACK packets */
3002                 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3003                    (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3004                 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3005                 /* hardware requires some bits to be set by default */
3006                 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3007                 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3008                 break;
3009         default:
3010                 /* We should do nothing since we don't know this hardware */
3011                 return;
3012         }
3013
3014         IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3015 }
3016
3017 /**
3018  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3019  * @adapter: board private structure
3020  *
3021  * Configure the Rx unit of the MAC after a reset.
3022  **/
3023 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3024 {
3025         struct ixgbe_hw *hw = &adapter->hw;
3026         int i;
3027         u32 rxctrl;
3028
3029         /* disable receives while setting up the descriptors */
3030         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3031         IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3032
3033         ixgbe_setup_psrtype(adapter);
3034         ixgbe_setup_rdrxctl(adapter);
3035
3036         /* Program registers for the distribution of queues */
3037         ixgbe_setup_mrqc(adapter);
3038
3039         ixgbe_set_uta(adapter);
3040
3041         /* set_rx_buffer_len must be called before ring initialization */
3042         ixgbe_set_rx_buffer_len(adapter);
3043
3044         /*
3045          * Setup the HW Rx Head and Tail Descriptor Pointers and
3046          * the Base and Length of the Rx Descriptor Ring
3047          */
3048         for (i = 0; i < adapter->num_rx_queues; i++)
3049                 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3050
3051         /* disable drop enable for 82598 parts */
3052         if (hw->mac.type == ixgbe_mac_82598EB)
3053                 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3054
3055         /* enable all receives */
3056         rxctrl |= IXGBE_RXCTRL_RXEN;
3057         hw->mac.ops.enable_rx_dma(hw, rxctrl);
3058 }
3059
3060 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3061 {
3062         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3063         struct ixgbe_hw *hw = &adapter->hw;
3064         int pool_ndx = adapter->num_vfs;
3065
3066         /* add VID to filter table */
3067         hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3068 }
3069
3070 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3071 {
3072         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3073         struct ixgbe_hw *hw = &adapter->hw;
3074         int pool_ndx = adapter->num_vfs;
3075
3076         if (!test_bit(__IXGBE_DOWN, &adapter->state))
3077                 ixgbe_irq_disable(adapter);
3078
3079         vlan_group_set_device(adapter->vlgrp, vid, NULL);
3080
3081         if (!test_bit(__IXGBE_DOWN, &adapter->state))
3082                 ixgbe_irq_enable(adapter, true, true);
3083
3084         /* remove VID from filter table */
3085         hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3086 }
3087
3088 /**
3089  * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3090  * @adapter: driver data
3091  */
3092 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3093 {
3094         struct ixgbe_hw *hw = &adapter->hw;
3095         u32 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3096         int i, j;
3097
3098         switch (hw->mac.type) {
3099         case ixgbe_mac_82598EB:
3100                 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
3101 #ifdef CONFIG_IXGBE_DCB
3102                 if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
3103                         vlnctrl &= ~IXGBE_VLNCTRL_VME;
3104 #endif
3105                 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3106                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3107                 break;
3108         case ixgbe_mac_82599EB:
3109                 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
3110                 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3111                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3112 #ifdef CONFIG_IXGBE_DCB
3113                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
3114                         break;
3115 #endif
3116                 for (i = 0; i < adapter->num_rx_queues; i++) {
3117                         j = adapter->rx_ring[i]->reg_idx;
3118                         vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3119                         vlnctrl &= ~IXGBE_RXDCTL_VME;
3120                         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3121                 }
3122                 break;
3123         default:
3124                 break;
3125         }
3126 }
3127
3128 /**
3129  * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3130  * @adapter: driver data
3131  */
3132 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3133 {
3134         struct ixgbe_hw *hw = &adapter->hw;
3135         u32 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3136         int i, j;
3137
3138         switch (hw->mac.type) {
3139         case ixgbe_mac_82598EB:
3140                 vlnctrl |= IXGBE_VLNCTRL_VME | IXGBE_VLNCTRL_VFE;
3141                 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3142                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3143                 break;
3144         case ixgbe_mac_82599EB:
3145                 vlnctrl |= IXGBE_VLNCTRL_VFE;
3146                 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3147                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3148                 for (i = 0; i < adapter->num_rx_queues; i++) {
3149                         j = adapter->rx_ring[i]->reg_idx;
3150                         vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3151                         vlnctrl |= IXGBE_RXDCTL_VME;
3152                         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3153                 }
3154                 break;
3155         default:
3156                 break;
3157         }
3158 }
3159
3160 static void ixgbe_vlan_rx_register(struct net_device *netdev,
3161                                    struct vlan_group *grp)
3162 {
3163         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3164
3165         if (!test_bit(__IXGBE_DOWN, &adapter->state))
3166                 ixgbe_irq_disable(adapter);
3167         adapter->vlgrp = grp;
3168
3169         /*
3170          * For a DCB driver, always enable VLAN tag stripping so we can
3171          * still receive traffic from a DCB-enabled host even if we're
3172          * not in DCB mode.
3173          */
3174         ixgbe_vlan_filter_enable(adapter);
3175
3176         ixgbe_vlan_rx_add_vid(netdev, 0);
3177
3178         if (!test_bit(__IXGBE_DOWN, &adapter->state))
3179                 ixgbe_irq_enable(adapter, true, true);
3180 }
3181
3182 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3183 {
3184         ixgbe_vlan_rx_register(adapter->netdev, adapter->vlgrp);
3185
3186         if (adapter->vlgrp) {
3187                 u16 vid;
3188                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
3189                         if (!vlan_group_get_device(adapter->vlgrp, vid))
3190                                 continue;
3191                         ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3192                 }
3193         }
3194 }
3195
3196 /**
3197  * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3198  * @netdev: network interface device structure
3199  *
3200  * Writes unicast address list to the RAR table.
3201  * Returns: -ENOMEM on failure/insufficient address space
3202  *                0 on no addresses written
3203  *                X on writing X addresses to the RAR table
3204  **/
3205 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3206 {
3207         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3208         struct ixgbe_hw *hw = &adapter->hw;
3209         unsigned int vfn = adapter->num_vfs;
3210         unsigned int rar_entries = hw->mac.num_rar_entries - (vfn + 1);
3211         int count = 0;
3212
3213         /* return ENOMEM indicating insufficient memory for addresses */
3214         if (netdev_uc_count(netdev) > rar_entries)
3215                 return -ENOMEM;
3216
3217         if (!netdev_uc_empty(netdev) && rar_entries) {
3218                 struct netdev_hw_addr *ha;
3219                 /* return error if we do not support writing to RAR table */
3220                 if (!hw->mac.ops.set_rar)
3221                         return -ENOMEM;
3222
3223                 netdev_for_each_uc_addr(ha, netdev) {
3224                         if (!rar_entries)
3225                                 break;
3226                         hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3227                                             vfn, IXGBE_RAH_AV);
3228                         count++;
3229                 }
3230         }
3231         /* write the addresses in reverse order to avoid write combining */
3232         for (; rar_entries > 0 ; rar_entries--)
3233                 hw->mac.ops.clear_rar(hw, rar_entries);
3234
3235         return count;
3236 }
3237
3238 /**
3239  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3240  * @netdev: network interface device structure
3241  *
3242  * The set_rx_method entry point is called whenever the unicast/multicast
3243  * address list or the network interface flags are updated.  This routine is
3244  * responsible for configuring the hardware for proper unicast, multicast and
3245  * promiscuous mode.
3246  **/
3247 void ixgbe_set_rx_mode(struct net_device *netdev)
3248 {
3249         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3250         struct ixgbe_hw *hw = &adapter->hw;
3251         u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3252         int count;
3253
3254         /* Check for Promiscuous and All Multicast modes */
3255
3256         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3257
3258         /* set all bits that we expect to always be set */
3259         fctrl |= IXGBE_FCTRL_BAM;
3260         fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3261         fctrl |= IXGBE_FCTRL_PMCF;
3262
3263         /* clear the bits we are changing the status of */
3264         fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3265
3266         if (netdev->flags & IFF_PROMISC) {
3267                 hw->addr_ctrl.user_set_promisc = true;
3268                 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3269                 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3270                 /* don't hardware filter vlans in promisc mode */
3271                 ixgbe_vlan_filter_disable(adapter);
3272         } else {
3273                 if (netdev->flags & IFF_ALLMULTI) {
3274                         fctrl |= IXGBE_FCTRL_MPE;
3275                         vmolr |= IXGBE_VMOLR_MPE;
3276                 } else {
3277                         /*
3278                          * Write addresses to the MTA, if the attempt fails
3279                          * then we should just turn on promiscous mode so
3280                          * that we can at least receive multicast traffic
3281                          */
3282                         hw->mac.ops.update_mc_addr_list(hw, netdev);
3283                         vmolr |= IXGBE_VMOLR_ROMPE;
3284                 }
3285                 ixgbe_vlan_filter_enable(adapter);
3286                 hw->addr_ctrl.user_set_promisc = false;
3287                 /*
3288                  * Write addresses to available RAR registers, if there is not
3289                  * sufficient space to store all the addresses then enable
3290                  * unicast promiscous mode
3291                  */
3292                 count = ixgbe_write_uc_addr_list(netdev);
3293                 if (count < 0) {
3294                         fctrl |= IXGBE_FCTRL_UPE;
3295                         vmolr |= IXGBE_VMOLR_ROPE;
3296                 }
3297         }
3298
3299         if (adapter->num_vfs) {
3300                 ixgbe_restore_vf_multicasts(adapter);
3301                 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3302                          ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3303                            IXGBE_VMOLR_ROPE);
3304                 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3305         }
3306
3307         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3308 }
3309
3310 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3311 {
3312         int q_idx;
3313         struct ixgbe_q_vector *q_vector;
3314         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3315
3316         /* legacy and MSI only use one vector */
3317         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3318                 q_vectors = 1;
3319
3320         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3321                 struct napi_struct *napi;
3322                 q_vector = adapter->q_vector[q_idx];
3323                 napi = &q_vector->napi;
3324                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3325                         if (!q_vector->rxr_count || !q_vector->txr_count) {
3326                                 if (q_vector->txr_count == 1)
3327                                         napi->poll = &ixgbe_clean_txonly;
3328                                 else if (q_vector->rxr_count == 1)
3329                                         napi->poll = &ixgbe_clean_rxonly;
3330                         }
3331                 }
3332
3333                 napi_enable(napi);
3334         }
3335 }
3336
3337 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3338 {
3339         int q_idx;
3340         struct ixgbe_q_vector *q_vector;
3341         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3342
3343         /* legacy and MSI only use one vector */
3344         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3345                 q_vectors = 1;
3346
3347         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3348                 q_vector = adapter->q_vector[q_idx];
3349                 napi_disable(&q_vector->napi);
3350         }
3351 }
3352
3353 #ifdef CONFIG_IXGBE_DCB
3354 /*
3355  * ixgbe_configure_dcb - Configure DCB hardware
3356  * @adapter: ixgbe adapter struct
3357  *
3358  * This is called by the driver on open to configure the DCB hardware.
3359  * This is also called by the gennetlink interface when reconfiguring
3360  * the DCB state.
3361  */
3362 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3363 {
3364         struct ixgbe_hw *hw = &adapter->hw;
3365         u32 txdctl;
3366         int i, j;
3367
3368         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3369                 if (hw->mac.type == ixgbe_mac_82598EB)
3370                         netif_set_gso_max_size(adapter->netdev, 65536);
3371                 return;
3372         }
3373
3374         if (hw->mac.type == ixgbe_mac_82598EB)
3375                 netif_set_gso_max_size(adapter->netdev, 32768);
3376
3377         ixgbe_dcb_check_config(&adapter->dcb_cfg);
3378         ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_TX_CONFIG);
3379         ixgbe_dcb_calculate_tc_credits(&adapter->dcb_cfg, DCB_RX_CONFIG);
3380
3381         /* reconfigure the hardware */
3382         ixgbe_dcb_hw_config(&adapter->hw, &adapter->dcb_cfg);
3383
3384         for (i = 0; i < adapter->num_tx_queues; i++) {
3385                 j = adapter->tx_ring[i]->reg_idx;
3386                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3387                 /* PThresh workaround for Tx hang with DFP enabled. */
3388                 txdctl |= 32;
3389                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
3390         }
3391         /* Enable VLAN tag insert/strip */
3392         ixgbe_vlan_filter_enable(adapter);
3393
3394         hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3395 }
3396
3397 #endif
3398 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3399 {
3400         struct net_device *netdev = adapter->netdev;
3401         struct ixgbe_hw *hw = &adapter->hw;
3402         int i;
3403
3404         ixgbe_set_rx_mode(netdev);
3405
3406         ixgbe_restore_vlan(adapter);
3407 #ifdef CONFIG_IXGBE_DCB
3408         ixgbe_configure_dcb(adapter);
3409 #endif
3410
3411 #ifdef IXGBE_FCOE
3412         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3413                 ixgbe_configure_fcoe(adapter);
3414
3415 #endif /* IXGBE_FCOE */
3416         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3417                 for (i = 0; i < adapter->num_tx_queues; i++)
3418                         adapter->tx_ring[i]->atr_sample_rate =
3419                                                        adapter->atr_sample_rate;
3420                 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
3421         } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3422                 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
3423         }
3424         ixgbe_configure_virtualization(adapter);
3425
3426         ixgbe_configure_tx(adapter);
3427         ixgbe_configure_rx(adapter);
3428 }
3429
3430 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3431 {
3432         switch (hw->phy.type) {
3433         case ixgbe_phy_sfp_avago:
3434         case ixgbe_phy_sfp_ftl:
3435         case ixgbe_phy_sfp_intel:
3436         case ixgbe_phy_sfp_unknown:
3437         case ixgbe_phy_sfp_passive_tyco:
3438         case ixgbe_phy_sfp_passive_unknown:
3439         case ixgbe_phy_sfp_active_unknown:
3440         case ixgbe_phy_sfp_ftl_active:
3441                 return true;
3442         default:
3443                 return false;
3444         }
3445 }
3446
3447 /**
3448  * ixgbe_sfp_link_config - set up SFP+ link
3449  * @adapter: pointer to private adapter struct
3450  **/
3451 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3452 {
3453         struct ixgbe_hw *hw = &adapter->hw;
3454
3455                 if (hw->phy.multispeed_fiber) {
3456                         /*
3457                          * In multispeed fiber setups, the device may not have
3458                          * had a physical connection when the driver loaded.
3459                          * If that's the case, the initial link configuration
3460                          * couldn't get the MAC into 10G or 1G mode, so we'll
3461                          * never have a link status change interrupt fire.
3462                          * We need to try and force an autonegotiation
3463                          * session, then bring up link.
3464                          */
3465                         hw->mac.ops.setup_sfp(hw);
3466                         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
3467                                 schedule_work(&adapter->multispeed_fiber_task);
3468                 } else {
3469                         /*
3470                          * Direct Attach Cu and non-multispeed fiber modules
3471                          * still need to be configured properly prior to
3472                          * attempting link.
3473                          */
3474                         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
3475                                 schedule_work(&adapter->sfp_config_module_task);
3476                 }
3477 }
3478
3479 /**
3480  * ixgbe_non_sfp_link_config - set up non-SFP+ link
3481  * @hw: pointer to private hardware struct
3482  *
3483  * Returns 0 on success, negative on failure
3484  **/
3485 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3486 {
3487         u32 autoneg;
3488         bool negotiation, link_up = false;
3489         u32 ret = IXGBE_ERR_LINK_SETUP;
3490
3491         if (hw->mac.ops.check_link)
3492                 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3493
3494         if (ret)
3495                 goto link_cfg_out;
3496
3497         if (hw->mac.ops.get_link_capabilities)
3498                 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3499                                                         &negotiation);
3500         if (ret)
3501                 goto link_cfg_out;
3502
3503         if (hw->mac.ops.setup_link)
3504                 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3505 link_cfg_out:
3506         return ret;
3507 }
3508
3509 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3510 {
3511         struct ixgbe_hw *hw = &adapter->hw;
3512         u32 gpie = 0;
3513
3514         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3515                 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3516                        IXGBE_GPIE_OCD;
3517                 gpie |= IXGBE_GPIE_EIAME;
3518                 /*
3519                  * use EIAM to auto-mask when MSI-X interrupt is asserted
3520                  * this saves a register write for every interrupt
3521                  */
3522                 switch (hw->mac.type) {
3523                 case ixgbe_mac_82598EB:
3524                         IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3525                         break;
3526                 default:
3527                 case ixgbe_mac_82599EB:
3528                         IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3529                         IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3530                         break;
3531                 }
3532         } else {
3533                 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3534                  * specifically only auto mask tx and rx interrupts */
3535                 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3536         }
3537
3538         /* XXX: to interrupt immediately for EICS writes, enable this */
3539         /* gpie |= IXGBE_GPIE_EIMEN; */
3540
3541         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3542                 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3543                 gpie |= IXGBE_GPIE_VTMODE_64;
3544         }
3545
3546         /* Enable fan failure interrupt */
3547         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3548                 gpie |= IXGBE_SDP1_GPIEN;
3549
3550         if (hw->mac.type == ixgbe_mac_82599EB)
3551                 gpie |= IXGBE_SDP1_GPIEN;
3552                 gpie |= IXGBE_SDP2_GPIEN;
3553
3554         IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3555 }
3556
3557 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
3558 {
3559         struct ixgbe_hw *hw = &adapter->hw;
3560         int err;
3561         u32 ctrl_ext;
3562
3563         ixgbe_get_hw_control(adapter);
3564         ixgbe_setup_gpie(adapter);
3565
3566         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3567                 ixgbe_configure_msix(adapter);
3568         else
3569                 ixgbe_configure_msi_and_legacy(adapter);
3570
3571         /* enable the optics */
3572         if (hw->phy.multispeed_fiber)
3573                 hw->mac.ops.enable_tx_laser(hw);
3574
3575         clear_bit(__IXGBE_DOWN, &adapter->state);
3576         ixgbe_napi_enable_all(adapter);
3577
3578         /* clear any pending interrupts, may auto mask */
3579         IXGBE_READ_REG(hw, IXGBE_EICR);
3580         ixgbe_irq_enable(adapter, true, true);
3581
3582         /*
3583          * If this adapter has a fan, check to see if we had a failure
3584          * before we enabled the interrupt.
3585          */
3586         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3587                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3588                 if (esdp & IXGBE_ESDP_SDP1)
3589                         e_crit(drv, "Fan has stopped, replace the adapter\n");
3590         }
3591
3592         /*
3593          * For hot-pluggable SFP+ devices, a new SFP+ module may have
3594          * arrived before interrupts were enabled but after probe.  Such
3595          * devices wouldn't have their type identified yet. We need to
3596          * kick off the SFP+ module setup first, then try to bring up link.
3597          * If we're not hot-pluggable SFP+, we just need to configure link
3598          * and bring it up.
3599          */
3600         if (hw->phy.type == ixgbe_phy_unknown) {
3601                 err = hw->phy.ops.identify(hw);
3602                 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
3603                         /*
3604                          * Take the device down and schedule the sfp tasklet
3605                          * which will unregister_netdev and log it.
3606                          */
3607                         ixgbe_down(adapter);
3608                         schedule_work(&adapter->sfp_config_module_task);
3609                         return err;
3610                 }
3611         }
3612
3613         if (ixgbe_is_sfp(hw)) {
3614                 ixgbe_sfp_link_config(adapter);
3615         } else {
3616                 err = ixgbe_non_sfp_link_config(hw);
3617                 if (err)
3618                         e_err(probe, "link_config FAILED %d\n", err);
3619         }
3620
3621         /* enable transmits */
3622         netif_tx_start_all_queues(adapter->netdev);
3623
3624         /* bring the link up in the watchdog, this could race with our first
3625          * link up interrupt but shouldn't be a problem */
3626         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3627         adapter->link_check_timeout = jiffies;
3628         mod_timer(&adapter->watchdog_timer, jiffies);
3629
3630         /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3631         ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3632         ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3633         IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3634
3635         return 0;
3636 }
3637
3638 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3639 {
3640         WARN_ON(in_interrupt());
3641         while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3642                 msleep(1);
3643         ixgbe_down(adapter);
3644         /*
3645          * If SR-IOV enabled then wait a bit before bringing the adapter
3646          * back up to give the VFs time to respond to the reset.  The
3647          * two second wait is based upon the watchdog timer cycle in
3648          * the VF driver.
3649          */
3650         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3651                 msleep(2000);
3652         ixgbe_up(adapter);
3653         clear_bit(__IXGBE_RESETTING, &adapter->state);
3654 }
3655
3656 int ixgbe_up(struct ixgbe_adapter *adapter)
3657 {
3658         /* hardware has been reset, we need to reload some things */
3659         ixgbe_configure(adapter);
3660
3661         return ixgbe_up_complete(adapter);
3662 }
3663
3664 void ixgbe_reset(struct ixgbe_adapter *adapter)
3665 {
3666         struct ixgbe_hw *hw = &adapter->hw;
3667         int err;
3668
3669         err = hw->mac.ops.init_hw(hw);
3670         switch (err) {
3671         case 0:
3672         case IXGBE_ERR_SFP_NOT_PRESENT:
3673                 break;
3674         case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3675                 e_dev_err("master disable timed out\n");
3676                 break;
3677         case IXGBE_ERR_EEPROM_VERSION:
3678                 /* We are running on a pre-production device, log a warning */
3679                 e_dev_warn("This device is a pre-production adapter/LOM. "
3680                            "Please be aware there may be issuesassociated with "
3681                            "your hardware.  If you are experiencing problems "
3682                            "please contact your Intel or hardware "
3683                            "representative who provided you with this "
3684                            "hardware.\n");
3685                 break;
3686         default:
3687                 e_dev_err("Hardware Error: %d\n", err);
3688         }
3689
3690         /* reprogram the RAR[0] in case user changed it. */
3691         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3692                             IXGBE_RAH_AV);
3693 }
3694
3695 /**
3696  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3697  * @adapter: board private structure
3698  * @rx_ring: ring to free buffers from
3699  **/
3700 static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
3701                                 struct ixgbe_ring *rx_ring)
3702 {
3703         struct pci_dev *pdev = adapter->pdev;
3704         unsigned long size;
3705         unsigned int i;
3706
3707         /* ring already cleared, nothing to do */
3708         if (!rx_ring->rx_buffer_info)
3709                 return;
3710
3711         /* Free all the Rx ring sk_buffs */
3712         for (i = 0; i < rx_ring->count; i++) {
3713                 struct ixgbe_rx_buffer *rx_buffer_info;
3714
3715                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
3716                 if (rx_buffer_info->dma) {
3717                         dma_unmap_single(&pdev->dev, rx_buffer_info->dma,
3718                                          rx_ring->rx_buf_len,
3719                                          DMA_FROM_DEVICE);
3720                         rx_buffer_info->dma = 0;
3721                 }
3722                 if (rx_buffer_info->skb) {
3723                         struct sk_buff *skb = rx_buffer_info->skb;
3724                         rx_buffer_info->skb = NULL;
3725                         do {
3726                                 struct sk_buff *this = skb;
3727                                 if (IXGBE_RSC_CB(this)->delay_unmap) {
3728                                         dma_unmap_single(&pdev->dev,
3729                                                          IXGBE_RSC_CB(this)->dma,
3730                                                          rx_ring->rx_buf_len,
3731                                                          DMA_FROM_DEVICE);
3732                                         IXGBE_RSC_CB(this)->dma = 0;
3733                                         IXGBE_RSC_CB(skb)->delay_unmap = false;
3734                                 }
3735                                 skb = skb->prev;
3736                                 dev_kfree_skb(this);
3737                         } while (skb);
3738                 }
3739                 if (!rx_buffer_info->page)
3740                         continue;
3741                 if (rx_buffer_info->page_dma) {
3742                         dma_unmap_page(&pdev->dev, rx_buffer_info->page_dma,
3743                                        PAGE_SIZE / 2, DMA_FROM_DEVICE);
3744                         rx_buffer_info->page_dma = 0;
3745                 }
3746                 put_page(rx_buffer_info->page);
3747                 rx_buffer_info->page = NULL;
3748                 rx_buffer_info->page_offset = 0;
3749         }
3750
3751         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3752         memset(rx_ring->rx_buffer_info, 0, size);
3753
3754         /* Zero out the descriptor ring */
3755         memset(rx_ring->desc, 0, rx_ring->size);
3756
3757         rx_ring->next_to_clean = 0;
3758         rx_ring->next_to_use = 0;
3759
3760         if (rx_ring->head)
3761                 writel(0, adapter->hw.hw_addr + rx_ring->head);
3762         if (rx_ring->tail)
3763                 writel(0, adapter->hw.hw_addr + rx_ring->tail);
3764 }
3765
3766 /**
3767  * ixgbe_clean_tx_ring - Free Tx Buffers
3768  * @adapter: board private structure
3769  * @tx_ring: ring to be cleaned
3770  **/
3771 static void ixgbe_clean_tx_ring(struct ixgbe_adapter *adapter,
3772                                 struct ixgbe_ring *tx_ring)
3773 {
3774         struct ixgbe_tx_buffer *tx_buffer_info;
3775         unsigned long size;
3776         unsigned int i;
3777
3778         /* ring already cleared, nothing to do */
3779         if (!tx_ring->tx_buffer_info)
3780                 return;
3781
3782         /* Free all the Tx ring sk_buffs */
3783         for (i = 0; i < tx_ring->count; i++) {
3784                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3785                 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
3786         }
3787
3788         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3789         memset(tx_ring->tx_buffer_info, 0, size);
3790
3791         /* Zero out the descriptor ring */
3792         memset(tx_ring->desc, 0, tx_ring->size);
3793
3794         tx_ring->next_to_use = 0;
3795         tx_ring->next_to_clean = 0;
3796
3797         if (tx_ring->head)
3798                 writel(0, adapter->hw.hw_addr + tx_ring->head);
3799         if (tx_ring->tail)
3800                 writel(0, adapter->hw.hw_addr + tx_ring->tail);
3801 }
3802
3803 /**
3804  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
3805  * @adapter: board private structure
3806  **/
3807 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
3808 {
3809         int i;
3810
3811         for (i = 0; i < adapter->num_rx_queues; i++)
3812                 ixgbe_clean_rx_ring(adapter, adapter->rx_ring[i]);
3813 }
3814
3815 /**
3816  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
3817  * @adapter: board private structure
3818  **/
3819 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
3820 {
3821         int i;
3822
3823         for (i = 0; i < adapter->num_tx_queues; i++)
3824                 ixgbe_clean_tx_ring(adapter, adapter->tx_ring[i]);
3825 }
3826
3827 void ixgbe_down(struct ixgbe_adapter *adapter)
3828 {
3829         struct net_device *netdev = adapter->netdev;
3830         struct ixgbe_hw *hw = &adapter->hw;
3831         u32 rxctrl;
3832         u32 txdctl;
3833         int i, j;
3834         int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3835
3836         /* signal that we are down to the interrupt handler */
3837         set_bit(__IXGBE_DOWN, &adapter->state);
3838
3839         /* disable receive for all VFs and wait one second */
3840         if (adapter->num_vfs) {
3841                 /* ping all the active vfs to let them know we are going down */
3842                 ixgbe_ping_all_vfs(adapter);
3843
3844                 /* Disable all VFTE/VFRE TX/RX */
3845                 ixgbe_disable_tx_rx(adapter);
3846
3847                 /* Mark all the VFs as inactive */
3848                 for (i = 0 ; i < adapter->num_vfs; i++)
3849                         adapter->vfinfo[i].clear_to_send = 0;
3850         }
3851
3852         /* disable receives */
3853         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3854         IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3855
3856         IXGBE_WRITE_FLUSH(hw);
3857         msleep(10);
3858
3859         netif_tx_stop_all_queues(netdev);
3860
3861         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3862         del_timer_sync(&adapter->sfp_timer);
3863         del_timer_sync(&adapter->watchdog_timer);
3864         cancel_work_sync(&adapter->watchdog_task);
3865
3866         netif_carrier_off(netdev);
3867         netif_tx_disable(netdev);
3868
3869         ixgbe_irq_disable(adapter);
3870
3871         ixgbe_napi_disable_all(adapter);
3872
3873         /* Cleanup the affinity_hint CPU mask memory and callback */
3874         for (i = 0; i < num_q_vectors; i++) {
3875                 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
3876                 /* clear the affinity_mask in the IRQ descriptor */
3877                 irq_set_affinity_hint(adapter->msix_entries[i]. vector, NULL);
3878                 /* release the CPU mask memory */
3879                 free_cpumask_var(q_vector->affinity_mask);
3880         }
3881
3882         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3883             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3884                 cancel_work_sync(&adapter->fdir_reinit_task);
3885
3886         if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
3887                 cancel_work_sync(&adapter->check_overtemp_task);
3888
3889         /* disable transmits in the hardware now that interrupts are off */
3890         for (i = 0; i < adapter->num_tx_queues; i++) {
3891                 j = adapter->tx_ring[i]->reg_idx;
3892                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3893                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
3894                                 (txdctl & ~IXGBE_TXDCTL_ENABLE));
3895         }
3896         /* Disable the Tx DMA engine on 82599 */
3897         if (hw->mac.type == ixgbe_mac_82599EB)
3898                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
3899                                 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
3900                                  ~IXGBE_DMATXCTL_TE));
3901
3902         /* power down the optics */
3903         if (hw->phy.multispeed_fiber)
3904                 hw->mac.ops.disable_tx_laser(hw);
3905
3906         /* clear n-tuple filters that are cached */
3907         ethtool_ntuple_flush(netdev);
3908
3909         if (!pci_channel_offline(adapter->pdev))
3910                 ixgbe_reset(adapter);
3911         ixgbe_clean_all_tx_rings(adapter);
3912         ixgbe_clean_all_rx_rings(adapter);
3913
3914 #ifdef CONFIG_IXGBE_DCA
3915         /* since we reset the hardware DCA settings were cleared */
3916         ixgbe_setup_dca(adapter);
3917 #endif
3918 }
3919
3920 /**
3921  * ixgbe_poll - NAPI Rx polling callback
3922  * @napi: structure for representing this polling device
3923  * @budget: how many packets driver is allowed to clean
3924  *
3925  * This function is used for legacy and MSI, NAPI mode
3926  **/
3927 static int ixgbe_poll(struct napi_struct *napi, int budget)
3928 {
3929         struct ixgbe_q_vector *q_vector =
3930                                 container_of(napi, struct ixgbe_q_vector, napi);
3931         struct ixgbe_adapter *adapter = q_vector->adapter;
3932         int tx_clean_complete, work_done = 0;
3933
3934 #ifdef CONFIG_IXGBE_DCA
3935         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
3936                 ixgbe_update_tx_dca(adapter, adapter->tx_ring[0]);
3937                 ixgbe_update_rx_dca(adapter, adapter->rx_ring[0]);
3938         }
3939 #endif
3940
3941         tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring[0]);
3942         ixgbe_clean_rx_irq(q_vector, adapter->rx_ring[0], &work_done, budget);
3943
3944         if (!tx_clean_complete)
3945                 work_done = budget;
3946
3947         /* If budget not fully consumed, exit the polling mode */
3948         if (work_done < budget) {
3949                 napi_complete(napi);
3950                 if (adapter->rx_itr_setting & 1)
3951                         ixgbe_set_itr(adapter);
3952                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3953                         ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
3954         }
3955         return work_done;
3956 }
3957
3958 /**
3959  * ixgbe_tx_timeout - Respond to a Tx Hang
3960  * @netdev: network interface device structure
3961  **/
3962 static void ixgbe_tx_timeout(struct net_device *netdev)
3963 {
3964         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3965
3966         /* Do the reset outside of interrupt context */
3967         schedule_work(&adapter->reset_task);
3968 }
3969
3970 static void ixgbe_reset_task(struct work_struct *work)
3971 {
3972         struct ixgbe_adapter *adapter;
3973         adapter = container_of(work, struct ixgbe_adapter, reset_task);
3974
3975         /* If we're already down or resetting, just bail */
3976         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
3977             test_bit(__IXGBE_RESETTING, &adapter->state))
3978                 return;
3979
3980         adapter->tx_timeout_count++;
3981
3982         ixgbe_dump(adapter);
3983         netdev_err(adapter->netdev, "Reset adapter\n");
3984         ixgbe_reinit_locked(adapter);
3985 }
3986
3987 #ifdef CONFIG_IXGBE_DCB
3988 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
3989 {
3990         bool ret = false;
3991         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
3992
3993         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
3994                 return ret;
3995
3996         f->mask = 0x7 << 3;
3997         adapter->num_rx_queues = f->indices;
3998         adapter->num_tx_queues = f->indices;
3999         ret = true;
4000
4001         return ret;
4002 }
4003 #endif
4004
4005 /**
4006  * ixgbe_set_rss_queues: Allocate queues for RSS
4007  * @adapter: board private structure to initialize
4008  *
4009  * This is our "base" multiqueue mode.  RSS (Receive Side Scaling) will try
4010  * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
4011  *
4012  **/
4013 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
4014 {
4015         bool ret = false;
4016         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
4017
4018         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4019                 f->mask = 0xF;
4020                 adapter->num_rx_queues = f->indices;
4021                 adapter->num_tx_queues = f->indices;
4022                 ret = true;
4023         } else {
4024                 ret = false;
4025         }
4026
4027         return ret;
4028 }
4029
4030 /**
4031  * ixgbe_set_fdir_queues: Allocate queues for Flow Director
4032  * @adapter: board private structure to initialize
4033  *
4034  * Flow Director is an advanced Rx filter, attempting to get Rx flows back
4035  * to the original CPU that initiated the Tx session.  This runs in addition
4036  * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
4037  * Rx load across CPUs using RSS.
4038  *
4039  **/
4040 static inline bool ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
4041 {
4042         bool ret = false;
4043         struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
4044
4045         f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
4046         f_fdir->mask = 0;
4047
4048         /* Flow Director must have RSS enabled */
4049         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4050             ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4051              (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
4052                 adapter->num_tx_queues = f_fdir->indices;
4053                 adapter->num_rx_queues = f_fdir->indices;
4054                 ret = true;
4055         } else {
4056                 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4057                 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4058         }
4059         return ret;
4060 }
4061
4062 #ifdef IXGBE_FCOE
4063 /**
4064  * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
4065  * @adapter: board private structure to initialize
4066  *
4067  * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
4068  * The ring feature mask is not used as a mask for FCoE, as it can take any 8
4069  * rx queues out of the max number of rx queues, instead, it is used as the
4070  * index of the first rx queue used by FCoE.
4071  *
4072  **/
4073 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
4074 {
4075         bool ret = false;
4076         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4077
4078         f->indices = min((int)num_online_cpus(), f->indices);
4079         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4080                 adapter->num_rx_queues = 1;
4081                 adapter->num_tx_queues = 1;
4082 #ifdef CONFIG_IXGBE_DCB
4083                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4084                         e_info(probe, "FCoE enabled with DCB\n");
4085                         ixgbe_set_dcb_queues(adapter);
4086                 }
4087 #endif
4088                 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4089                         e_info(probe, "FCoE enabled with RSS\n");
4090                         if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4091                             (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4092                                 ixgbe_set_fdir_queues(adapter);
4093                         else
4094                                 ixgbe_set_rss_queues(adapter);
4095                 }
4096                 /* adding FCoE rx rings to the end */
4097                 f->mask = adapter->num_rx_queues;
4098                 adapter->num_rx_queues += f->indices;
4099                 adapter->num_tx_queues += f->indices;
4100
4101                 ret = true;
4102         }
4103
4104         return ret;
4105 }
4106
4107 #endif /* IXGBE_FCOE */
4108 /**
4109  * ixgbe_set_sriov_queues: Allocate queues for IOV use
4110  * @adapter: board private structure to initialize
4111  *
4112  * IOV doesn't actually use anything, so just NAK the
4113  * request for now and let the other queue routines
4114  * figure out what to do.
4115  */
4116 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
4117 {
4118         return false;
4119 }
4120
4121 /*
4122  * ixgbe_set_num_queues: Allocate queues for device, feature dependant
4123  * @adapter: board private structure to initialize
4124  *
4125  * This is the top level queue allocation routine.  The order here is very
4126  * important, starting with the "most" number of features turned on at once,
4127  * and ending with the smallest set of features.  This way large combinations
4128  * can be allocated if they're turned on, and smaller combinations are the
4129  * fallthrough conditions.
4130  *
4131  **/
4132 static int ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
4133 {
4134         /* Start with base case */
4135         adapter->num_rx_queues = 1;
4136         adapter->num_tx_queues = 1;
4137         adapter->num_rx_pools = adapter->num_rx_queues;
4138         adapter->num_rx_queues_per_pool = 1;
4139
4140         if (ixgbe_set_sriov_queues(adapter))
4141                 goto done;
4142
4143 #ifdef IXGBE_FCOE
4144         if (ixgbe_set_fcoe_queues(adapter))
4145                 goto done;
4146
4147 #endif /* IXGBE_FCOE */
4148 #ifdef CONFIG_IXGBE_DCB
4149         if (ixgbe_set_dcb_queues(adapter))
4150                 goto done;
4151
4152 #endif
4153         if (ixgbe_set_fdir_queues(adapter))
4154                 goto done;
4155
4156         if (ixgbe_set_rss_queues(adapter))
4157                 goto done;
4158
4159         /* fallback to base case */
4160         adapter->num_rx_queues = 1;
4161         adapter->num_tx_queues = 1;
4162
4163 done:
4164         /* Notify the stack of the (possibly) reduced queue counts. */
4165         netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
4166         return netif_set_real_num_rx_queues(adapter->netdev,
4167                                             adapter->num_rx_queues);
4168 }
4169
4170 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4171                                        int vectors)
4172 {
4173         int err, vector_threshold;
4174
4175         /* We'll want at least 3 (vector_threshold):
4176          * 1) TxQ[0] Cleanup
4177          * 2) RxQ[0] Cleanup
4178          * 3) Other (Link Status Change, etc.)
4179          * 4) TCP Timer (optional)
4180          */
4181         vector_threshold = MIN_MSIX_COUNT;
4182
4183         /* The more we get, the more we will assign to Tx/Rx Cleanup
4184          * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4185          * Right now, we simply care about how many we'll get; we'll
4186          * set them up later while requesting irq's.
4187          */
4188         while (vectors >= vector_threshold) {
4189                 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4190                                       vectors);
4191                 if (!err) /* Success in acquiring all requested vectors. */
4192                         break;
4193                 else if (err < 0)
4194                         vectors = 0; /* Nasty failure, quit now */
4195                 else /* err == number of vectors we should try again with */
4196                         vectors = err;
4197         }
4198
4199         if (vectors < vector_threshold) {
4200                 /* Can't allocate enough MSI-X interrupts?  Oh well.
4201                  * This just means we'll go with either a single MSI
4202                  * vector or fall back to legacy interrupts.
4203                  */
4204                 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4205                              "Unable to allocate MSI-X interrupts\n");
4206                 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4207                 kfree(adapter->msix_entries);
4208                 adapter->msix_entries = NULL;
4209         } else {
4210                 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4211                 /*
4212                  * Adjust for only the vectors we'll use, which is minimum
4213                  * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4214                  * vectors we were allocated.
4215                  */
4216                 adapter->num_msix_vectors = min(vectors,
4217                                    adapter->max_msix_q_vectors + NON_Q_VECTORS);
4218         }
4219 }
4220
4221 /**
4222  * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4223  * @adapter: board private structure to initialize
4224  *
4225  * Cache the descriptor ring offsets for RSS to the assigned rings.
4226  *
4227  **/
4228 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4229 {
4230         int i;
4231         bool ret = false;
4232
4233         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4234                 for (i = 0; i < adapter->num_rx_queues; i++)
4235                         adapter->rx_ring[i]->reg_idx = i;
4236                 for (i = 0; i < adapter->num_tx_queues; i++)
4237                         adapter->tx_ring[i]->reg_idx = i;
4238                 ret = true;
4239         } else {
4240                 ret = false;
4241         }
4242
4243         return ret;
4244 }
4245
4246 #ifdef CONFIG_IXGBE_DCB
4247 /**
4248  * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4249  * @adapter: board private structure to initialize
4250  *
4251  * Cache the descriptor ring offsets for DCB to the assigned rings.
4252  *
4253  **/
4254 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4255 {
4256         int i;
4257         bool ret = false;
4258         int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
4259
4260         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4261                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
4262                         /* the number of queues is assumed to be symmetric */
4263                         for (i = 0; i < dcb_i; i++) {
4264                                 adapter->rx_ring[i]->reg_idx = i << 3;
4265                                 adapter->tx_ring[i]->reg_idx = i << 2;
4266                         }
4267                         ret = true;
4268                 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
4269                         if (dcb_i == 8) {
4270                                 /*
4271                                  * Tx TC0 starts at: descriptor queue 0
4272                                  * Tx TC1 starts at: descriptor queue 32
4273                                  * Tx TC2 starts at: descriptor queue 64
4274                                  * Tx TC3 starts at: descriptor queue 80
4275                                  * Tx TC4 starts at: descriptor queue 96
4276                                  * Tx TC5 starts at: descriptor queue 104
4277                                  * Tx TC6 starts at: descriptor queue 112
4278                                  * Tx TC7 starts at: descriptor queue 120
4279                                  *
4280                                  * Rx TC0-TC7 are offset by 16 queues each
4281                                  */
4282                                 for (i = 0; i < 3; i++) {
4283                                         adapter->tx_ring[i]->reg_idx = i << 5;
4284                                         adapter->rx_ring[i]->reg_idx = i << 4;
4285                                 }
4286                                 for ( ; i < 5; i++) {
4287                                         adapter->tx_ring[i]->reg_idx =
4288                                                                  ((i + 2) << 4);
4289                                         adapter->rx_ring[i]->reg_idx = i << 4;
4290                                 }
4291                                 for ( ; i < dcb_i; i++) {
4292                                         adapter->tx_ring[i]->reg_idx =
4293                                                                  ((i + 8) << 3);
4294                                         adapter->rx_ring[i]->reg_idx = i << 4;
4295                                 }
4296
4297                                 ret = true;
4298                         } else if (dcb_i == 4) {
4299                                 /*
4300                                  * Tx TC0 starts at: descriptor queue 0
4301                                  * Tx TC1 starts at: descriptor queue 64
4302                                  * Tx TC2 starts at: descriptor queue 96
4303                                  * Tx TC3 starts at: descriptor queue 112
4304                                  *
4305                                  * Rx TC0-TC3 are offset by 32 queues each
4306                                  */
4307                                 adapter->tx_ring[0]->reg_idx = 0;
4308                                 adapter->tx_ring[1]->reg_idx = 64;
4309                                 adapter->tx_ring[2]->reg_idx = 96;
4310                                 adapter->tx_ring[3]->reg_idx = 112;
4311                                 for (i = 0 ; i < dcb_i; i++)
4312                                         adapter->rx_ring[i]->reg_idx = i << 5;
4313
4314                                 ret = true;
4315                         } else {
4316                                 ret = false;
4317                         }
4318                 } else {
4319                         ret = false;
4320                 }
4321         } else {
4322                 ret = false;
4323         }
4324
4325         return ret;
4326 }
4327 #endif
4328
4329 /**
4330  * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4331  * @adapter: board private structure to initialize
4332  *
4333  * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4334  *
4335  **/
4336 static inline bool ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4337 {
4338         int i;
4339         bool ret = false;
4340
4341         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4342             ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4343              (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
4344                 for (i = 0; i < adapter->num_rx_queues; i++)
4345                         adapter->rx_ring[i]->reg_idx = i;
4346                 for (i = 0; i < adapter->num_tx_queues; i++)
4347                         adapter->tx_ring[i]->reg_idx = i;
4348                 ret = true;
4349         }
4350
4351         return ret;
4352 }
4353
4354 #ifdef IXGBE_FCOE
4355 /**
4356  * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4357  * @adapter: board private structure to initialize
4358  *
4359  * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4360  *
4361  */
4362 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4363 {
4364         int i, fcoe_rx_i = 0, fcoe_tx_i = 0;
4365         bool ret = false;
4366         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4367
4368         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4369 #ifdef CONFIG_IXGBE_DCB
4370                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4371                         struct ixgbe_fcoe *fcoe = &adapter->fcoe;
4372
4373                         ixgbe_cache_ring_dcb(adapter);
4374                         /* find out queues in TC for FCoE */
4375                         fcoe_rx_i = adapter->rx_ring[fcoe->tc]->reg_idx + 1;
4376                         fcoe_tx_i = adapter->tx_ring[fcoe->tc]->reg_idx + 1;
4377                         /*
4378                          * In 82599, the number of Tx queues for each traffic
4379                          * class for both 8-TC and 4-TC modes are:
4380                          * TCs  : TC0 TC1 TC2 TC3 TC4 TC5 TC6 TC7
4381                          * 8 TCs:  32  32  16  16   8   8   8   8
4382                          * 4 TCs:  64  64  32  32
4383                          * We have max 8 queues for FCoE, where 8 the is
4384                          * FCoE redirection table size. If TC for FCoE is
4385                          * less than or equal to TC3, we have enough queues
4386                          * to add max of 8 queues for FCoE, so we start FCoE
4387                          * tx descriptor from the next one, i.e., reg_idx + 1.
4388                          * If TC for FCoE is above TC3, implying 8 TC mode,
4389                          * and we need 8 for FCoE, we have to take all queues
4390                          * in that traffic class for FCoE.
4391                          */
4392                         if ((f->indices == IXGBE_FCRETA_SIZE) && (fcoe->tc > 3))
4393                                 fcoe_tx_i--;
4394                 }
4395 #endif /* CONFIG_IXGBE_DCB */
4396                 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4397                         if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4398                             (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4399                                 ixgbe_cache_ring_fdir(adapter);
4400                         else
4401                                 ixgbe_cache_ring_rss(adapter);
4402
4403                         fcoe_rx_i = f->mask;
4404                         fcoe_tx_i = f->mask;
4405                 }
4406                 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4407                         adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4408                         adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4409                 }
4410                 ret = true;
4411         }
4412         return ret;
4413 }
4414
4415 #endif /* IXGBE_FCOE */
4416 /**
4417  * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4418  * @adapter: board private structure to initialize
4419  *
4420  * SR-IOV doesn't use any descriptor rings but changes the default if
4421  * no other mapping is used.
4422  *
4423  */
4424 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4425 {
4426         adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4427         adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4428         if (adapter->num_vfs)
4429                 return true;
4430         else
4431                 return false;
4432 }
4433
4434 /**
4435  * ixgbe_cache_ring_register - Descriptor ring to register mapping
4436  * @adapter: board private structure to initialize
4437  *
4438  * Once we know the feature-set enabled for the device, we'll cache
4439  * the register offset the descriptor ring is assigned to.
4440  *
4441  * Note, the order the various feature calls is important.  It must start with
4442  * the "most" features enabled at the same time, then trickle down to the
4443  * least amount of features turned on at once.
4444  **/
4445 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4446 {
4447         /* start with default case */
4448         adapter->rx_ring[0]->reg_idx = 0;
4449         adapter->tx_ring[0]->reg_idx = 0;
4450
4451         if (ixgbe_cache_ring_sriov(adapter))
4452                 return;
4453
4454 #ifdef IXGBE_FCOE
4455         if (ixgbe_cache_ring_fcoe(adapter))
4456                 return;
4457
4458 #endif /* IXGBE_FCOE */
4459 #ifdef CONFIG_IXGBE_DCB
4460         if (ixgbe_cache_ring_dcb(adapter))
4461                 return;
4462
4463 #endif
4464         if (ixgbe_cache_ring_fdir(adapter))
4465                 return;
4466
4467         if (ixgbe_cache_ring_rss(adapter))
4468                 return;
4469 }
4470
4471 /**
4472  * ixgbe_alloc_queues - Allocate memory for all rings
4473  * @adapter: board private structure to initialize
4474  *
4475  * We allocate one ring per queue at run-time since we don't know the
4476  * number of queues at compile-time.  The polling_netdev array is
4477  * intended for Multiqueue, but should work fine with a single queue.
4478  **/
4479 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
4480 {
4481         int i;
4482         int orig_node = adapter->node;
4483
4484         for (i = 0; i < adapter->num_tx_queues; i++) {
4485                 struct ixgbe_ring *ring = adapter->tx_ring[i];
4486                 if (orig_node == -1) {
4487                         int cur_node = next_online_node(adapter->node);
4488                         if (cur_node == MAX_NUMNODES)
4489                                 cur_node = first_online_node;
4490                         adapter->node = cur_node;
4491                 }
4492                 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4493                                     adapter->node);
4494                 if (!ring)
4495                         ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4496                 if (!ring)
4497                         goto err_tx_ring_allocation;
4498                 ring->count = adapter->tx_ring_count;
4499                 ring->queue_index = i;
4500                 ring->numa_node = adapter->node;
4501
4502                 adapter->tx_ring[i] = ring;
4503         }
4504
4505         /* Restore the adapter's original node */
4506         adapter->node = orig_node;
4507
4508         for (i = 0; i < adapter->num_rx_queues; i++) {
4509                 struct ixgbe_ring *ring = adapter->rx_ring[i];
4510                 if (orig_node == -1) {
4511                         int cur_node = next_online_node(adapter->node);
4512                         if (cur_node == MAX_NUMNODES)
4513                                 cur_node = first_online_node;
4514                         adapter->node = cur_node;
4515                 }
4516                 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4517                                     adapter->node);
4518                 if (!ring)
4519                         ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4520                 if (!ring)
4521                         goto err_rx_ring_allocation;
4522                 ring->count = adapter->rx_ring_count;
4523                 ring->queue_index = i;
4524                 ring->numa_node = adapter->node;
4525
4526                 adapter->rx_ring[i] = ring;
4527         }
4528
4529         /* Restore the adapter's original node */
4530         adapter->node = orig_node;
4531
4532         ixgbe_cache_ring_register(adapter);
4533
4534         return 0;
4535
4536 err_rx_ring_allocation:
4537         for (i = 0; i < adapter->num_tx_queues; i++)
4538                 kfree(adapter->tx_ring[i]);
4539 err_tx_ring_allocation:
4540         return -ENOMEM;
4541 }
4542
4543 /**
4544  * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4545  * @adapter: board private structure to initialize
4546  *
4547  * Attempt to configure the interrupts using the best available
4548  * capabilities of the hardware and the kernel.
4549  **/
4550 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4551 {
4552         struct ixgbe_hw *hw = &adapter->hw;
4553         int err = 0;
4554         int vector, v_budget;
4555
4556         /*
4557          * It's easy to be greedy for MSI-X vectors, but it really
4558          * doesn't do us much good if we have a lot more vectors
4559          * than CPU's.  So let's be conservative and only ask for
4560          * (roughly) the same number of vectors as there are CPU's.
4561          */
4562         v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
4563                        (int)num_online_cpus()) + NON_Q_VECTORS;
4564
4565         /*
4566          * At the same time, hardware can only support a maximum of
4567          * hw.mac->max_msix_vectors vectors.  With features
4568          * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4569          * descriptor queues supported by our device.  Thus, we cap it off in
4570          * those rare cases where the cpu count also exceeds our vector limit.
4571          */
4572         v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
4573
4574         /* A failure in MSI-X entry allocation isn't fatal, but it does
4575          * mean we disable MSI-X capabilities of the adapter. */
4576         adapter->msix_entries = kcalloc(v_budget,
4577                                         sizeof(struct msix_entry), GFP_KERNEL);
4578         if (adapter->msix_entries) {
4579                 for (vector = 0; vector < v_budget; vector++)
4580                         adapter->msix_entries[vector].entry = vector;
4581
4582                 ixgbe_acquire_msix_vectors(adapter, v_budget);
4583
4584                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4585                         goto out;
4586         }
4587
4588         adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4589         adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4590         adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4591         adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4592         adapter->atr_sample_rate = 0;
4593         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4594                 ixgbe_disable_sriov(adapter);
4595
4596         err = ixgbe_set_num_queues(adapter);
4597         if (err)
4598                 return err;
4599
4600         err = pci_enable_msi(adapter->pdev);
4601         if (!err) {
4602                 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4603         } else {
4604                 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4605                              "Unable to allocate MSI interrupt, "
4606                              "falling back to legacy.  Error: %d\n", err);
4607                 /* reset err */
4608                 err = 0;
4609         }
4610
4611 out:
4612         return err;
4613 }
4614
4615 /**
4616  * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4617  * @adapter: board private structure to initialize
4618  *
4619  * We allocate one q_vector per queue interrupt.  If allocation fails we
4620  * return -ENOMEM.
4621  **/
4622 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4623 {
4624         int q_idx, num_q_vectors;
4625         struct ixgbe_q_vector *q_vector;
4626         int napi_vectors;
4627         int (*poll)(struct napi_struct *, int);
4628
4629         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4630                 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4631                 napi_vectors = adapter->num_rx_queues;
4632                 poll = &ixgbe_clean_rxtx_many;
4633         } else {
4634                 num_q_vectors = 1;
4635                 napi_vectors = 1;
4636                 poll = &ixgbe_poll;
4637         }
4638
4639         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4640                 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4641                                         GFP_KERNEL, adapter->node);
4642                 if (!q_vector)
4643                         q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4644                                            GFP_KERNEL);
4645                 if (!q_vector)
4646                         goto err_out;
4647                 q_vector->adapter = adapter;
4648                 if (q_vector->txr_count && !q_vector->rxr_count)
4649                         q_vector->eitr = adapter->tx_eitr_param;
4650                 else
4651                         q_vector->eitr = adapter->rx_eitr_param;
4652                 q_vector->v_idx = q_idx;
4653                 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
4654                 adapter->q_vector[q_idx] = q_vector;
4655         }
4656
4657         return 0;
4658
4659 err_out:
4660         while (q_idx) {
4661                 q_idx--;
4662                 q_vector = adapter->q_vector[q_idx];
4663                 netif_napi_del(&q_vector->napi);
4664                 kfree(q_vector);
4665                 adapter->q_vector[q_idx] = NULL;
4666         }
4667         return -ENOMEM;
4668 }
4669
4670 /**
4671  * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4672  * @adapter: board private structure to initialize
4673  *
4674  * This function frees the memory allocated to the q_vectors.  In addition if
4675  * NAPI is enabled it will delete any references to the NAPI struct prior
4676  * to freeing the q_vector.
4677  **/
4678 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4679 {
4680         int q_idx, num_q_vectors;
4681
4682         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4683                 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4684         else
4685                 num_q_vectors = 1;
4686
4687         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4688                 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
4689                 adapter->q_vector[q_idx] = NULL;
4690                 netif_napi_del(&q_vector->napi);
4691                 kfree(q_vector);
4692         }
4693 }
4694
4695 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4696 {
4697         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4698                 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4699                 pci_disable_msix(adapter->pdev);
4700                 kfree(adapter->msix_entries);
4701                 adapter->msix_entries = NULL;
4702         } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4703                 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4704                 pci_disable_msi(adapter->pdev);
4705         }
4706 }
4707
4708 /**
4709  * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4710  * @adapter: board private structure to initialize
4711  *
4712  * We determine which interrupt scheme to use based on...
4713  * - Kernel support (MSI, MSI-X)
4714  *   - which can be user-defined (via MODULE_PARAM)
4715  * - Hardware queue count (num_*_queues)
4716  *   - defined by miscellaneous hardware support/features (RSS, etc.)
4717  **/
4718 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4719 {
4720         int err;
4721
4722         /* Number of supported queues */
4723         err = ixgbe_set_num_queues(adapter);
4724         if (err)
4725                 return err;
4726
4727         err = ixgbe_set_interrupt_capability(adapter);
4728         if (err) {
4729                 e_dev_err("Unable to setup interrupt capabilities\n");
4730                 goto err_set_interrupt;
4731         }
4732
4733         err = ixgbe_alloc_q_vectors(adapter);
4734         if (err) {
4735                 e_dev_err("Unable to allocate memory for queue vectors\n");
4736                 goto err_alloc_q_vectors;
4737         }
4738
4739         err = ixgbe_alloc_queues(adapter);
4740         if (err) {
4741                 e_dev_err("Unable to allocate memory for queues\n");
4742                 goto err_alloc_queues;
4743         }
4744
4745         e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
4746                    (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
4747                    adapter->num_rx_queues, adapter->num_tx_queues);
4748
4749         set_bit(__IXGBE_DOWN, &adapter->state);
4750
4751         return 0;
4752
4753 err_alloc_queues:
4754         ixgbe_free_q_vectors(adapter);
4755 err_alloc_q_vectors:
4756         ixgbe_reset_interrupt_capability(adapter);
4757 err_set_interrupt:
4758         return err;
4759 }
4760
4761 /**
4762  * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
4763  * @adapter: board private structure to clear interrupt scheme on
4764  *
4765  * We go through and clear interrupt specific resources and reset the structure
4766  * to pre-load conditions
4767  **/
4768 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
4769 {
4770         int i;
4771
4772         for (i = 0; i < adapter->num_tx_queues; i++) {
4773                 kfree(adapter->tx_ring[i]);
4774                 adapter->tx_ring[i] = NULL;
4775         }
4776         for (i = 0; i < adapter->num_rx_queues; i++) {
4777                 kfree(adapter->rx_ring[i]);
4778                 adapter->rx_ring[i] = NULL;
4779         }
4780
4781         ixgbe_free_q_vectors(adapter);
4782         ixgbe_reset_interrupt_capability(adapter);
4783 }
4784
4785 /**
4786  * ixgbe_sfp_timer - worker thread to find a missing module
4787  * @data: pointer to our adapter struct
4788  **/
4789 static void ixgbe_sfp_timer(unsigned long data)
4790 {
4791         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4792
4793         /*
4794          * Do the sfp_timer outside of interrupt context due to the
4795          * delays that sfp+ detection requires
4796          */
4797         schedule_work(&adapter->sfp_task);
4798 }
4799
4800 /**
4801  * ixgbe_sfp_task - worker thread to find a missing module
4802  * @work: pointer to work_struct containing our data
4803  **/
4804 static void ixgbe_sfp_task(struct work_struct *work)
4805 {
4806         struct ixgbe_adapter *adapter = container_of(work,
4807                                                      struct ixgbe_adapter,
4808                                                      sfp_task);
4809         struct ixgbe_hw *hw = &adapter->hw;
4810
4811         if ((hw->phy.type == ixgbe_phy_nl) &&
4812             (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
4813                 s32 ret = hw->phy.ops.identify_sfp(hw);
4814                 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
4815                         goto reschedule;
4816                 ret = hw->phy.ops.reset(hw);
4817                 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4818                         e_dev_err("failed to initialize because an unsupported "
4819                                   "SFP+ module type was detected.\n");
4820                         e_dev_err("Reload the driver after installing a "
4821                                   "supported module.\n");
4822                         unregister_netdev(adapter->netdev);
4823                 } else {
4824                         e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
4825                 }
4826                 /* don't need this routine any more */
4827                 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4828         }
4829         return;
4830 reschedule:
4831         if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
4832                 mod_timer(&adapter->sfp_timer,
4833                           round_jiffies(jiffies + (2 * HZ)));
4834 }
4835
4836 /**
4837  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4838  * @adapter: board private structure to initialize
4839  *
4840  * ixgbe_sw_init initializes the Adapter private data structure.
4841  * Fields are initialized based on PCI device information and
4842  * OS network device settings (MTU size).
4843  **/
4844 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4845 {
4846         struct ixgbe_hw *hw = &adapter->hw;
4847         struct pci_dev *pdev = adapter->pdev;
4848         struct net_device *dev = adapter->netdev;
4849         unsigned int rss;
4850 #ifdef CONFIG_IXGBE_DCB
4851         int j;
4852         struct tc_configuration *tc;
4853 #endif
4854
4855         /* PCI config space info */
4856
4857         hw->vendor_id = pdev->vendor;
4858         hw->device_id = pdev->device;
4859         hw->revision_id = pdev->revision;
4860         hw->subsystem_vendor_id = pdev->subsystem_vendor;
4861         hw->subsystem_device_id = pdev->subsystem_device;
4862
4863         /* Set capability flags */
4864         rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
4865         adapter->ring_feature[RING_F_RSS].indices = rss;
4866         adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4867         adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
4868         if (hw->mac.type == ixgbe_mac_82598EB) {
4869                 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4870                         adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4871                 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
4872         } else if (hw->mac.type == ixgbe_mac_82599EB) {
4873                 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
4874                 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4875                 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4876                 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4877                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4878                 if (dev->features & NETIF_F_NTUPLE) {
4879                         /* Flow Director perfect filter enabled */
4880                         adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4881                         adapter->atr_sample_rate = 0;
4882                         spin_lock_init(&adapter->fdir_perfect_lock);
4883                 } else {
4884                         /* Flow Director hash filters enabled */
4885                         adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4886                         adapter->atr_sample_rate = 20;
4887                 }
4888                 adapter->ring_feature[RING_F_FDIR].indices =
4889                                                          IXGBE_MAX_FDIR_INDICES;
4890                 adapter->fdir_pballoc = 0;
4891 #ifdef IXGBE_FCOE
4892                 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4893                 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4894                 adapter->ring_feature[RING_F_FCOE].indices = 0;
4895 #ifdef CONFIG_IXGBE_DCB
4896                 /* Default traffic class to use for FCoE */
4897                 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
4898                 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4899 #endif
4900 #endif /* IXGBE_FCOE */
4901         }
4902
4903 #ifdef CONFIG_IXGBE_DCB
4904         /* Configure DCB traffic classes */
4905         for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4906                 tc = &adapter->dcb_cfg.tc_config[j];
4907                 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4908                 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4909                 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4910                 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4911                 tc->dcb_pfc = pfc_disabled;
4912         }
4913         adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4914         adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4915         adapter->dcb_cfg.rx_pba_cfg = pba_equal;
4916         adapter->dcb_cfg.pfc_mode_enable = false;
4917         adapter->dcb_cfg.round_robin_enable = false;
4918         adapter->dcb_set_bitmap = 0x00;
4919         ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
4920                            adapter->ring_feature[RING_F_DCB].indices);
4921
4922 #endif
4923
4924         /* default flow control settings */
4925         hw->fc.requested_mode = ixgbe_fc_full;
4926         hw->fc.current_mode = ixgbe_fc_full;    /* init for ethtool output */
4927 #ifdef CONFIG_DCB
4928         adapter->last_lfc_mode = hw->fc.current_mode;
4929 #endif
4930         hw->fc.high_water = IXGBE_DEFAULT_FCRTH;
4931         hw->fc.low_water = IXGBE_DEFAULT_FCRTL;
4932         hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4933         hw->fc.send_xon = true;
4934         hw->fc.disable_fc_autoneg = false;
4935
4936         /* enable itr by default in dynamic mode */
4937         adapter->rx_itr_setting = 1;
4938         adapter->rx_eitr_param = 20000;
4939         adapter->tx_itr_setting = 1;
4940         adapter->tx_eitr_param = 10000;
4941
4942         /* set defaults for eitr in MegaBytes */
4943         adapter->eitr_low = 10;
4944         adapter->eitr_high = 20;
4945
4946         /* set default ring sizes */
4947         adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4948         adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4949
4950         /* initialize eeprom parameters */
4951         if (ixgbe_init_eeprom_params_generic(hw)) {
4952                 e_dev_err("EEPROM initialization failed\n");
4953                 return -EIO;
4954         }
4955
4956         /* enable rx csum by default */
4957         adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
4958
4959         /* get assigned NUMA node */
4960         adapter->node = dev_to_node(&pdev->dev);
4961
4962         set_bit(__IXGBE_DOWN, &adapter->state);
4963
4964         return 0;
4965 }
4966
4967 /**
4968  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4969  * @adapter: board private structure
4970  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
4971  *
4972  * Return 0 on success, negative on failure
4973  **/
4974 int ixgbe_setup_tx_resources(struct ixgbe_adapter *adapter,
4975                              struct ixgbe_ring *tx_ring)
4976 {
4977         struct pci_dev *pdev = adapter->pdev;
4978         int size;
4979
4980         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4981         tx_ring->tx_buffer_info = vmalloc_node(size, tx_ring->numa_node);
4982         if (!tx_ring->tx_buffer_info)
4983                 tx_ring->tx_buffer_info = vmalloc(size);
4984         if (!tx_ring->tx_buffer_info)
4985                 goto err;
4986         memset(tx_ring->tx_buffer_info, 0, size);
4987
4988         /* round up to nearest 4K */
4989         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4990         tx_ring->size = ALIGN(tx_ring->size, 4096);
4991
4992         tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
4993                                            &tx_ring->dma, GFP_KERNEL);
4994         if (!tx_ring->desc)
4995                 goto err;
4996
4997         tx_ring->next_to_use = 0;
4998         tx_ring->next_to_clean = 0;
4999         tx_ring->work_limit = tx_ring->count;
5000         return 0;
5001
5002 err:
5003         vfree(tx_ring->tx_buffer_info);
5004         tx_ring->tx_buffer_info = NULL;
5005         e_err(probe, "Unable to allocate memory for the Tx descriptor ring\n");
5006         return -ENOMEM;
5007 }
5008
5009 /**
5010  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
5011  * @adapter: board private structure
5012  *
5013  * If this function returns with an error, then it's possible one or
5014  * more of the rings is populated (while the rest are not).  It is the
5015  * callers duty to clean those orphaned rings.
5016  *
5017  * Return 0 on success, negative on failure
5018  **/
5019 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
5020 {
5021         int i, err = 0;
5022
5023         for (i = 0; i < adapter->num_tx_queues; i++) {
5024                 err = ixgbe_setup_tx_resources(adapter, adapter->tx_ring[i]);
5025                 if (!err)
5026                         continue;
5027                 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5028                 break;
5029         }
5030
5031         return err;
5032 }
5033
5034 /**
5035  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5036  * @adapter: board private structure
5037  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
5038  *
5039  * Returns 0 on success, negative on failure
5040  **/
5041 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
5042                              struct ixgbe_ring *rx_ring)
5043 {
5044         struct pci_dev *pdev = adapter->pdev;
5045         int size;
5046
5047         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5048         rx_ring->rx_buffer_info = vmalloc_node(size, adapter->node);
5049         if (!rx_ring->rx_buffer_info)
5050                 rx_ring->rx_buffer_info = vmalloc(size);
5051         if (!rx_ring->rx_buffer_info) {
5052                 e_err(probe, "vmalloc allocation failed for the Rx "
5053                       "descriptor ring\n");
5054                 goto alloc_failed;
5055         }
5056         memset(rx_ring->rx_buffer_info, 0, size);
5057
5058         /* Round up to nearest 4K */
5059         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5060         rx_ring->size = ALIGN(rx_ring->size, 4096);
5061
5062         rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
5063                                            &rx_ring->dma, GFP_KERNEL);
5064
5065         if (!rx_ring->desc) {
5066                 e_err(probe, "Memory allocation failed for the Rx "
5067                       "descriptor ring\n");
5068                 vfree(rx_ring->rx_buffer_info);
5069                 goto alloc_failed;
5070         }
5071
5072         rx_ring->next_to_clean = 0;
5073         rx_ring->next_to_use = 0;
5074
5075         return 0;
5076
5077 alloc_failed:
5078         return -ENOMEM;
5079 }
5080
5081 /**
5082  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5083  * @adapter: board private structure
5084  *
5085  * If this function returns with an error, then it's possible one or
5086  * more of the rings is populated (while the rest are not).  It is the
5087  * callers duty to clean those orphaned rings.
5088  *
5089  * Return 0 on success, negative on failure
5090  **/
5091
5092 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5093 {
5094         int i, err = 0;
5095
5096         for (i = 0; i < adapter->num_rx_queues; i++) {
5097                 err = ixgbe_setup_rx_resources(adapter, adapter->rx_ring[i]);
5098                 if (!err)
5099                         continue;
5100                 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5101                 break;
5102         }
5103
5104         return err;
5105 }
5106
5107 /**
5108  * ixgbe_free_tx_resources - Free Tx Resources per Queue
5109  * @adapter: board private structure
5110  * @tx_ring: Tx descriptor ring for a specific queue
5111  *
5112  * Free all transmit software resources
5113  **/
5114 void ixgbe_free_tx_resources(struct ixgbe_adapter *adapter,
5115                              struct ixgbe_ring *tx_ring)
5116 {
5117         struct pci_dev *pdev = adapter->pdev;
5118
5119         ixgbe_clean_tx_ring(adapter, tx_ring);
5120
5121         vfree(tx_ring->tx_buffer_info);
5122         tx_ring->tx_buffer_info = NULL;
5123
5124         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
5125                           tx_ring->dma);
5126
5127         tx_ring->desc = NULL;
5128 }
5129
5130 /**
5131  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5132  * @adapter: board private structure
5133  *
5134  * Free all transmit software resources
5135  **/
5136 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5137 {
5138         int i;
5139
5140         for (i = 0; i < adapter->num_tx_queues; i++)
5141                 if (adapter->tx_ring[i]->desc)
5142                         ixgbe_free_tx_resources(adapter, adapter->tx_ring[i]);
5143 }
5144
5145 /**
5146  * ixgbe_free_rx_resources - Free Rx Resources
5147  * @adapter: board private structure
5148  * @rx_ring: ring to clean the resources from
5149  *
5150  * Free all receive software resources
5151  **/
5152 void ixgbe_free_rx_resources(struct ixgbe_adapter *adapter,
5153                              struct ixgbe_ring *rx_ring)
5154 {
5155         struct pci_dev *pdev = adapter->pdev;
5156
5157         ixgbe_clean_rx_ring(adapter, rx_ring);
5158
5159         vfree(rx_ring->rx_buffer_info);
5160         rx_ring->rx_buffer_info = NULL;
5161
5162         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
5163                           rx_ring->dma);
5164
5165         rx_ring->desc = NULL;
5166 }
5167
5168 /**
5169  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5170  * @adapter: board private structure
5171  *
5172  * Free all receive software resources
5173  **/
5174 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5175 {
5176         int i;
5177
5178         for (i = 0; i < adapter->num_rx_queues; i++)
5179                 if (adapter->rx_ring[i]->desc)
5180                         ixgbe_free_rx_resources(adapter, adapter->rx_ring[i]);
5181 }
5182
5183 /**
5184  * ixgbe_change_mtu - Change the Maximum Transfer Unit
5185  * @netdev: network interface device structure
5186  * @new_mtu: new value for maximum frame size
5187  *
5188  * Returns 0 on success, negative on failure
5189  **/
5190 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5191 {
5192         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5193         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5194
5195         /* MTU < 68 is an error and causes problems on some kernels */
5196         if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5197                 return -EINVAL;
5198
5199         e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5200         /* must set new MTU before calling down or up */
5201         netdev->mtu = new_mtu;
5202
5203         if (netif_running(netdev))
5204                 ixgbe_reinit_locked(adapter);
5205
5206         return 0;
5207 }
5208
5209 /**
5210  * ixgbe_open - Called when a network interface is made active
5211  * @netdev: network interface device structure
5212  *
5213  * Returns 0 on success, negative value on failure
5214  *
5215  * The open entry point is called when a network interface is made
5216  * active by the system (IFF_UP).  At this point all resources needed
5217  * for transmit and receive operations are allocated, the interrupt
5218  * handler is registered with the OS, the watchdog timer is started,
5219  * and the stack is notified that the interface is ready.
5220  **/
5221 static int ixgbe_open(struct net_device *netdev)
5222 {
5223         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5224         int err;
5225
5226         /* disallow open during test */
5227         if (test_bit(__IXGBE_TESTING, &adapter->state))
5228                 return -EBUSY;
5229
5230         netif_carrier_off(netdev);
5231
5232         /* allocate transmit descriptors */
5233         err = ixgbe_setup_all_tx_resources(adapter);
5234         if (err)
5235                 goto err_setup_tx;
5236
5237         /* allocate receive descriptors */
5238         err = ixgbe_setup_all_rx_resources(adapter);
5239         if (err)
5240                 goto err_setup_rx;
5241
5242         ixgbe_configure(adapter);
5243
5244         err = ixgbe_request_irq(adapter);
5245         if (err)
5246                 goto err_req_irq;
5247
5248         err = ixgbe_up_complete(adapter);
5249         if (err)
5250                 goto err_up;
5251
5252         netif_tx_start_all_queues(netdev);
5253
5254         return 0;
5255
5256 err_up:
5257         ixgbe_release_hw_control(adapter);
5258         ixgbe_free_irq(adapter);
5259 err_req_irq:
5260 err_setup_rx:
5261         ixgbe_free_all_rx_resources(adapter);
5262 err_setup_tx:
5263         ixgbe_free_all_tx_resources(adapter);
5264         ixgbe_reset(adapter);
5265
5266         return err;
5267 }
5268
5269 /**
5270  * ixgbe_close - Disables a network interface
5271  * @netdev: network interface device structure
5272  *
5273  * Returns 0, this is not allowed to fail
5274  *
5275  * The close entry point is called when an interface is de-activated
5276  * by the OS.  The hardware is still under the drivers control, but
5277  * needs to be disabled.  A global MAC reset is issued to stop the
5278  * hardware, and all transmit and receive resources are freed.
5279  **/
5280 static int ixgbe_close(struct net_device *netdev)
5281 {
5282         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5283
5284         ixgbe_down(adapter);
5285         ixgbe_free_irq(adapter);
5286
5287         ixgbe_free_all_tx_resources(adapter);
5288         ixgbe_free_all_rx_resources(adapter);
5289
5290         ixgbe_release_hw_control(adapter);
5291
5292         return 0;
5293 }
5294
5295 #ifdef CONFIG_PM
5296 static int ixgbe_resume(struct pci_dev *pdev)
5297 {
5298         struct net_device *netdev = pci_get_drvdata(pdev);
5299         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5300         u32 err;
5301
5302         pci_set_power_state(pdev, PCI_D0);
5303         pci_restore_state(pdev);
5304         /*
5305          * pci_restore_state clears dev->state_saved so call
5306          * pci_save_state to restore it.
5307          */
5308         pci_save_state(pdev);
5309
5310         err = pci_enable_device_mem(pdev);
5311         if (err) {
5312                 e_dev_err("Cannot enable PCI device from suspend\n");
5313                 return err;
5314         }
5315         pci_set_master(pdev);
5316
5317         pci_wake_from_d3(pdev, false);
5318
5319         err = ixgbe_init_interrupt_scheme(adapter);
5320         if (err) {
5321                 e_dev_err("Cannot initialize interrupts for device\n");
5322                 return err;
5323         }
5324
5325         ixgbe_reset(adapter);
5326
5327         IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5328
5329         if (netif_running(netdev)) {
5330                 err = ixgbe_open(adapter->netdev);
5331                 if (err)
5332                         return err;
5333         }
5334
5335         netif_device_attach(netdev);
5336
5337         return 0;
5338 }
5339 #endif /* CONFIG_PM */
5340
5341 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5342 {
5343         struct net_device *netdev = pci_get_drvdata(pdev);
5344         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5345         struct ixgbe_hw *hw = &adapter->hw;
5346         u32 ctrl, fctrl;
5347         u32 wufc = adapter->wol;
5348 #ifdef CONFIG_PM
5349         int retval = 0;
5350 #endif
5351
5352         netif_device_detach(netdev);
5353
5354         if (netif_running(netdev)) {
5355                 ixgbe_down(adapter);
5356                 ixgbe_free_irq(adapter);
5357                 ixgbe_free_all_tx_resources(adapter);
5358                 ixgbe_free_all_rx_resources(adapter);
5359         }
5360
5361 #ifdef CONFIG_PM
5362         retval = pci_save_state(pdev);
5363         if (retval)
5364                 return retval;
5365
5366 #endif
5367         if (wufc) {
5368                 ixgbe_set_rx_mode(netdev);
5369
5370                 /* turn on all-multi mode if wake on multicast is enabled */
5371                 if (wufc & IXGBE_WUFC_MC) {
5372                         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5373                         fctrl |= IXGBE_FCTRL_MPE;
5374                         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5375                 }
5376
5377                 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5378                 ctrl |= IXGBE_CTRL_GIO_DIS;
5379                 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5380
5381                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5382         } else {
5383                 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5384                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5385         }
5386
5387         if (wufc && hw->mac.type == ixgbe_mac_82599EB)
5388                 pci_wake_from_d3(pdev, true);
5389         else
5390                 pci_wake_from_d3(pdev, false);
5391
5392         *enable_wake = !!wufc;
5393
5394         ixgbe_clear_interrupt_scheme(adapter);
5395
5396         ixgbe_release_hw_control(adapter);
5397
5398         pci_disable_device(pdev);
5399
5400         return 0;
5401 }
5402
5403 #ifdef CONFIG_PM
5404 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5405 {
5406         int retval;
5407         bool wake;
5408
5409         retval = __ixgbe_shutdown(pdev, &wake);
5410         if (retval)
5411                 return retval;
5412
5413         if (wake) {
5414                 pci_prepare_to_sleep(pdev);
5415         } else {
5416                 pci_wake_from_d3(pdev, false);
5417                 pci_set_power_state(pdev, PCI_D3hot);
5418         }
5419
5420         return 0;
5421 }
5422 #endif /* CONFIG_PM */
5423
5424 static void ixgbe_shutdown(struct pci_dev *pdev)
5425 {
5426         bool wake;
5427
5428         __ixgbe_shutdown(pdev, &wake);
5429
5430         if (system_state == SYSTEM_POWER_OFF) {
5431                 pci_wake_from_d3(pdev, wake);
5432                 pci_set_power_state(pdev, PCI_D3hot);
5433         }
5434 }
5435
5436 /**
5437  * ixgbe_update_stats - Update the board statistics counters.
5438  * @adapter: board private structure
5439  **/
5440 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5441 {
5442         struct net_device *netdev = adapter->netdev;
5443         struct ixgbe_hw *hw = &adapter->hw;
5444         u64 total_mpc = 0;
5445         u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5446         u64 non_eop_descs = 0, restart_queue = 0;
5447         struct ixgbe_hw_stats *hwstats = &adapter->stats;
5448
5449         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5450             test_bit(__IXGBE_RESETTING, &adapter->state))
5451                 return;
5452
5453         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5454                 u64 rsc_count = 0;
5455                 u64 rsc_flush = 0;
5456                 for (i = 0; i < 16; i++)
5457                         adapter->hw_rx_no_dma_resources +=
5458                                 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5459                 for (i = 0; i < adapter->num_rx_queues; i++) {
5460                         rsc_count += adapter->rx_ring[i]->rsc_count;
5461                         rsc_flush += adapter->rx_ring[i]->rsc_flush;
5462                 }
5463                 adapter->rsc_total_count = rsc_count;
5464                 adapter->rsc_total_flush = rsc_flush;
5465         }
5466
5467         /* gather some stats to the adapter struct that are per queue */
5468         for (i = 0; i < adapter->num_tx_queues; i++)
5469                 restart_queue += adapter->tx_ring[i]->restart_queue;
5470         adapter->restart_queue = restart_queue;
5471
5472         for (i = 0; i < adapter->num_rx_queues; i++)
5473                 non_eop_descs += adapter->rx_ring[i]->non_eop_descs;
5474         adapter->non_eop_descs = non_eop_descs;
5475
5476         hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5477         for (i = 0; i < 8; i++) {
5478                 /* for packet buffers not used, the register should read 0 */
5479                 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5480                 missed_rx += mpc;
5481                 hwstats->mpc[i] += mpc;
5482                 total_mpc += hwstats->mpc[i];
5483                 if (hw->mac.type == ixgbe_mac_82598EB)
5484                         hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5485                 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5486                 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5487                 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5488                 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5489                 if (hw->mac.type == ixgbe_mac_82599EB) {
5490                         hwstats->pxonrxc[i] +=
5491                                 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5492                         hwstats->pxoffrxc[i] +=
5493                                 IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
5494                         hwstats->qprdc[i] += IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5495                 } else {
5496                         hwstats->pxonrxc[i] +=
5497                                 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5498                         hwstats->pxoffrxc[i] +=
5499                                 IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
5500                 }
5501                 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5502                 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5503         }
5504         hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5505         /* work around hardware counting issue */
5506         hwstats->gprc -= missed_rx;
5507
5508         /* 82598 hardware only has a 32 bit counter in the high register */
5509         if (hw->mac.type == ixgbe_mac_82599EB) {
5510                 u64 tmp;
5511                 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5512                 tmp = IXGBE_READ_REG(hw, IXGBE_GORCH) & 0xF;
5513                                                 /* 4 high bits of GORC */
5514                 hwstats->gorc += (tmp << 32);
5515                 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5516                 tmp = IXGBE_READ_REG(hw, IXGBE_GOTCH) & 0xF;
5517                                                 /* 4 high bits of GOTC */
5518                 hwstats->gotc += (tmp << 32);
5519                 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5520                 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5521                 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5522                 hwstats->lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
5523                 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5524                 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5525 #ifdef IXGBE_FCOE
5526                 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5527                 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5528                 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5529                 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5530                 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5531                 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5532 #endif /* IXGBE_FCOE */
5533         } else {
5534                 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5535                 hwstats->lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
5536                 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5537                 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5538                 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5539         }
5540         bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5541         hwstats->bprc += bprc;
5542         hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5543         if (hw->mac.type == ixgbe_mac_82598EB)
5544                 hwstats->mprc -= bprc;
5545         hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5546         hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5547         hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5548         hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5549         hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5550         hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5551         hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5552         hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5553         lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5554         hwstats->lxontxc += lxon;
5555         lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5556         hwstats->lxofftxc += lxoff;
5557         hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5558         hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5559         hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5560         /*
5561          * 82598 errata - tx of flow control packets is included in tx counters
5562          */
5563         xon_off_tot = lxon + lxoff;
5564         hwstats->gptc -= xon_off_tot;
5565         hwstats->mptc -= xon_off_tot;
5566         hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5567         hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5568         hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5569         hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5570         hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5571         hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5572         hwstats->ptc64 -= xon_off_tot;
5573         hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5574         hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5575         hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5576         hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5577         hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5578         hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5579
5580         /* Fill out the OS statistics structure */
5581         netdev->stats.multicast = hwstats->mprc;
5582
5583         /* Rx Errors */
5584         netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5585         netdev->stats.rx_dropped = 0;
5586         netdev->stats.rx_length_errors = hwstats->rlec;
5587         netdev->stats.rx_crc_errors = hwstats->crcerrs;
5588         netdev->stats.rx_missed_errors = total_mpc;
5589 }
5590
5591 /**
5592  * ixgbe_watchdog - Timer Call-back
5593  * @data: pointer to adapter cast into an unsigned long
5594  **/
5595 static void ixgbe_watchdog(unsigned long data)
5596 {
5597         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5598         struct ixgbe_hw *hw = &adapter->hw;
5599         u64 eics = 0;
5600         int i;
5601
5602         /*
5603          *  Do the watchdog outside of interrupt context due to the lovely
5604          * delays that some of the newer hardware requires
5605          */
5606
5607         if (test_bit(__IXGBE_DOWN, &adapter->state))
5608                 goto watchdog_short_circuit;
5609
5610         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5611                 /*
5612                  * for legacy and MSI interrupts don't set any bits
5613                  * that are enabled for EIAM, because this operation
5614                  * would set *both* EIMS and EICS for any bit in EIAM
5615                  */
5616                 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5617                         (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5618                 goto watchdog_reschedule;
5619         }
5620
5621         /* get one bit for every active tx/rx interrupt vector */
5622         for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5623                 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5624                 if (qv->rxr_count || qv->txr_count)
5625                         eics |= ((u64)1 << i);
5626         }
5627
5628         /* Cause software interrupt to ensure rx rings are cleaned */
5629         ixgbe_irq_rearm_queues(adapter, eics);
5630
5631 watchdog_reschedule:
5632         /* Reset the timer */
5633         mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
5634
5635 watchdog_short_circuit:
5636         schedule_work(&adapter->watchdog_task);
5637 }
5638
5639 /**
5640  * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
5641  * @work: pointer to work_struct containing our data
5642  **/
5643 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
5644 {
5645         struct ixgbe_adapter *adapter = container_of(work,
5646                                                      struct ixgbe_adapter,
5647                                                      multispeed_fiber_task);
5648         struct ixgbe_hw *hw = &adapter->hw;
5649         u32 autoneg;
5650         bool negotiation;
5651
5652         adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
5653         autoneg = hw->phy.autoneg_advertised;
5654         if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5655                 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5656         hw->mac.autotry_restart = false;
5657         if (hw->mac.ops.setup_link)
5658                 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5659         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5660         adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
5661 }
5662
5663 /**
5664  * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
5665  * @work: pointer to work_struct containing our data
5666  **/
5667 static void ixgbe_sfp_config_module_task(struct work_struct *work)
5668 {
5669         struct ixgbe_adapter *adapter = container_of(work,
5670                                                      struct ixgbe_adapter,
5671                                                      sfp_config_module_task);
5672         struct ixgbe_hw *hw = &adapter->hw;
5673         u32 err;
5674
5675         adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
5676
5677         /* Time for electrical oscillations to settle down */
5678         msleep(100);
5679         err = hw->phy.ops.identify_sfp(hw);
5680
5681         if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5682                 e_dev_err("failed to initialize because an unsupported SFP+ "
5683                           "module type was detected.\n");
5684                 e_dev_err("Reload the driver after installing a supported "
5685                           "module.\n");
5686                 unregister_netdev(adapter->netdev);
5687                 return;
5688         }
5689         hw->mac.ops.setup_sfp(hw);
5690
5691         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
5692                 /* This will also work for DA Twinax connections */
5693                 schedule_work(&adapter->multispeed_fiber_task);
5694         adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
5695 }
5696
5697 /**
5698  * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
5699  * @work: pointer to work_struct containing our data
5700  **/
5701 static void ixgbe_fdir_reinit_task(struct work_struct *work)
5702 {
5703         struct ixgbe_adapter *adapter = container_of(work,
5704                                                      struct ixgbe_adapter,
5705                                                      fdir_reinit_task);
5706         struct ixgbe_hw *hw = &adapter->hw;
5707         int i;
5708
5709         if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5710                 for (i = 0; i < adapter->num_tx_queues; i++)
5711                         set_bit(__IXGBE_FDIR_INIT_DONE,
5712                                 &(adapter->tx_ring[i]->reinit_state));
5713         } else {
5714                 e_err(probe, "failed to finish FDIR re-initialization, "
5715                       "ignored adding FDIR ATR filters\n");
5716         }
5717         /* Done FDIR Re-initialization, enable transmits */
5718         netif_tx_start_all_queues(adapter->netdev);
5719 }
5720
5721 static DEFINE_MUTEX(ixgbe_watchdog_lock);
5722
5723 /**
5724  * ixgbe_watchdog_task - worker thread to bring link up
5725  * @work: pointer to work_struct containing our data
5726  **/
5727 static void ixgbe_watchdog_task(struct work_struct *work)
5728 {
5729         struct ixgbe_adapter *adapter = container_of(work,
5730                                                      struct ixgbe_adapter,
5731                                                      watchdog_task);
5732         struct net_device *netdev = adapter->netdev;
5733         struct ixgbe_hw *hw = &adapter->hw;
5734         u32 link_speed;
5735         bool link_up;
5736         int i;
5737         struct ixgbe_ring *tx_ring;
5738         int some_tx_pending = 0;
5739
5740         mutex_lock(&ixgbe_watchdog_lock);
5741
5742         link_up = adapter->link_up;
5743         link_speed = adapter->link_speed;
5744
5745         if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
5746                 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5747                 if (link_up) {
5748 #ifdef CONFIG_DCB
5749                         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5750                                 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5751                                         hw->mac.ops.fc_enable(hw, i);
5752                         } else {
5753                                 hw->mac.ops.fc_enable(hw, 0);
5754                         }
5755 #else
5756                         hw->mac.ops.fc_enable(hw, 0);
5757 #endif
5758                 }
5759
5760                 if (link_up ||
5761                     time_after(jiffies, (adapter->link_check_timeout +
5762                                          IXGBE_TRY_LINK_TIMEOUT))) {
5763                         adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5764                         IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5765                 }
5766                 adapter->link_up = link_up;
5767                 adapter->link_speed = link_speed;
5768         }
5769
5770         if (link_up) {
5771                 if (!netif_carrier_ok(netdev)) {
5772                         bool flow_rx, flow_tx;
5773
5774                         if (hw->mac.type == ixgbe_mac_82599EB) {
5775                                 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5776                                 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5777                                 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5778                                 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5779                         } else {
5780                                 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5781                                 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5782                                 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5783                                 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5784                         }
5785
5786                         e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5787                                (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5788                                "10 Gbps" :
5789                                (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5790                                "1 Gbps" : "unknown speed")),
5791                                ((flow_rx && flow_tx) ? "RX/TX" :
5792                                (flow_rx ? "RX" :
5793                                (flow_tx ? "TX" : "None"))));
5794
5795                         netif_carrier_on(netdev);
5796                 } else {
5797                         /* Force detection of hung controller */
5798                         adapter->detect_tx_hung = true;
5799                 }
5800         } else {
5801                 adapter->link_up = false;
5802                 adapter->link_speed = 0;
5803                 if (netif_carrier_ok(netdev)) {
5804                         e_info(drv, "NIC Link is Down\n");
5805                         netif_carrier_off(netdev);
5806                 }
5807         }
5808
5809         if (!netif_carrier_ok(netdev)) {
5810                 for (i = 0; i < adapter->num_tx_queues; i++) {
5811                         tx_ring = adapter->tx_ring[i];
5812                         if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5813                                 some_tx_pending = 1;
5814                                 break;
5815                         }
5816                 }
5817
5818                 if (some_tx_pending) {
5819                         /* We've lost link, so the controller stops DMA,
5820                          * but we've got queued Tx work that's never going
5821                          * to get done, so reset controller to flush Tx.
5822                          * (Do the reset outside of interrupt context).
5823                          */
5824                          schedule_work(&adapter->reset_task);
5825                 }
5826         }
5827
5828         ixgbe_update_stats(adapter);
5829         mutex_unlock(&ixgbe_watchdog_lock);
5830 }
5831
5832 static int ixgbe_tso(struct ixgbe_adapter *adapter,
5833                      struct ixgbe_ring *tx_ring, struct sk_buff *skb,
5834                      u32 tx_flags, u8 *hdr_len)
5835 {
5836         struct ixgbe_adv_tx_context_desc *context_desc;
5837         unsigned int i;
5838         int err;
5839         struct ixgbe_tx_buffer *tx_buffer_info;
5840         u32 vlan_macip_lens = 0, type_tucmd_mlhl;
5841         u32 mss_l4len_idx, l4len;
5842
5843         if (skb_is_gso(skb)) {
5844                 if (skb_header_cloned(skb)) {
5845                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5846                         if (err)
5847                                 return err;
5848                 }
5849                 l4len = tcp_hdrlen(skb);
5850                 *hdr_len += l4len;
5851
5852                 if (skb->protocol == htons(ETH_P_IP)) {
5853                         struct iphdr *iph = ip_hdr(skb);
5854                         iph->tot_len = 0;
5855                         iph->check = 0;
5856                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5857                                                                  iph->daddr, 0,
5858                                                                  IPPROTO_TCP,
5859                                                                  0);
5860                 } else if (skb_is_gso_v6(skb)) {
5861                         ipv6_hdr(skb)->payload_len = 0;
5862                         tcp_hdr(skb)->check =
5863                             ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5864                                              &ipv6_hdr(skb)->daddr,
5865                                              0, IPPROTO_TCP, 0);
5866                 }
5867
5868                 i = tx_ring->next_to_use;
5869
5870                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5871                 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
5872
5873                 /* VLAN MACLEN IPLEN */
5874                 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5875                         vlan_macip_lens |=
5876                             (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5877                 vlan_macip_lens |= ((skb_network_offset(skb)) <<
5878                                     IXGBE_ADVTXD_MACLEN_SHIFT);
5879                 *hdr_len += skb_network_offset(skb);
5880                 vlan_macip_lens |=
5881                     (skb_transport_header(skb) - skb_network_header(skb));
5882                 *hdr_len +=
5883                     (skb_transport_header(skb) - skb_network_header(skb));
5884                 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5885                 context_desc->seqnum_seed = 0;
5886
5887                 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5888                 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
5889                                    IXGBE_ADVTXD_DTYP_CTXT);
5890
5891                 if (skb->protocol == htons(ETH_P_IP))
5892                         type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5893                 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5894                 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5895
5896                 /* MSS L4LEN IDX */
5897                 mss_l4len_idx =
5898                     (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
5899                 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
5900                 /* use index 1 for TSO */
5901                 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5902                 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
5903
5904                 tx_buffer_info->time_stamp = jiffies;
5905                 tx_buffer_info->next_to_watch = i;
5906
5907                 i++;
5908                 if (i == tx_ring->count)
5909                         i = 0;
5910                 tx_ring->next_to_use = i;
5911
5912                 return true;
5913         }
5914         return false;
5915 }
5916
5917 static u32 ixgbe_psum(struct ixgbe_adapter *adapter, struct sk_buff *skb)
5918 {
5919         u32 rtn = 0;
5920         __be16 protocol;
5921
5922         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
5923                 protocol = ((const struct vlan_ethhdr *)skb->data)->
5924                                         h_vlan_encapsulated_proto;
5925         else
5926                 protocol = skb->protocol;
5927
5928         switch (protocol) {
5929         case cpu_to_be16(ETH_P_IP):
5930                 rtn |= IXGBE_ADVTXD_TUCMD_IPV4;
5931                 switch (ip_hdr(skb)->protocol) {
5932                 case IPPROTO_TCP:
5933                         rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5934                         break;
5935                 case IPPROTO_SCTP:
5936                         rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5937                         break;
5938                 }
5939                 break;
5940         case cpu_to_be16(ETH_P_IPV6):
5941                 /* XXX what about other V6 headers?? */
5942                 switch (ipv6_hdr(skb)->nexthdr) {
5943                 case IPPROTO_TCP:
5944                         rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5945                         break;
5946                 case IPPROTO_SCTP:
5947                         rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5948                         break;
5949                 }
5950                 break;
5951         default:
5952                 if (unlikely(net_ratelimit()))
5953                         e_warn(probe, "partial checksum but proto=%x!\n",
5954                                skb->protocol);
5955                 break;
5956         }
5957
5958         return rtn;
5959 }
5960
5961 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
5962                           struct ixgbe_ring *tx_ring,
5963                           struct sk_buff *skb, u32 tx_flags)
5964 {
5965         struct ixgbe_adv_tx_context_desc *context_desc;
5966         unsigned int i;
5967         struct ixgbe_tx_buffer *tx_buffer_info;
5968         u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
5969
5970         if (skb->ip_summed == CHECKSUM_PARTIAL ||
5971             (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
5972                 i = tx_ring->next_to_use;
5973                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5974                 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
5975
5976                 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5977                         vlan_macip_lens |=
5978                             (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5979                 vlan_macip_lens |= (skb_network_offset(skb) <<
5980                                     IXGBE_ADVTXD_MACLEN_SHIFT);
5981                 if (skb->ip_summed == CHECKSUM_PARTIAL)
5982                         vlan_macip_lens |= (skb_transport_header(skb) -
5983                                             skb_network_header(skb));
5984
5985                 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5986                 context_desc->seqnum_seed = 0;
5987
5988                 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
5989                                     IXGBE_ADVTXD_DTYP_CTXT);
5990
5991                 if (skb->ip_summed == CHECKSUM_PARTIAL)
5992                         type_tucmd_mlhl |= ixgbe_psum(adapter, skb);
5993
5994                 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5995                 /* use index zero for tx checksum offload */
5996                 context_desc->mss_l4len_idx = 0;
5997
5998                 tx_buffer_info->time_stamp = jiffies;
5999                 tx_buffer_info->next_to_watch = i;
6000
6001                 i++;
6002                 if (i == tx_ring->count)
6003                         i = 0;
6004                 tx_ring->next_to_use = i;
6005
6006                 return true;
6007         }
6008
6009         return false;
6010 }
6011
6012 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
6013                         struct ixgbe_ring *tx_ring,
6014                         struct sk_buff *skb, u32 tx_flags,
6015                         unsigned int first)
6016 {
6017         struct pci_dev *pdev = adapter->pdev;
6018         struct ixgbe_tx_buffer *tx_buffer_info;
6019         unsigned int len;
6020         unsigned int total = skb->len;
6021         unsigned int offset = 0, size, count = 0, i;
6022         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
6023         unsigned int f;
6024
6025         i = tx_ring->next_to_use;
6026
6027         if (tx_flags & IXGBE_TX_FLAGS_FCOE)
6028                 /* excluding fcoe_crc_eof for FCoE */
6029                 total -= sizeof(struct fcoe_crc_eof);
6030
6031         len = min(skb_headlen(skb), total);
6032         while (len) {
6033                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6034                 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6035
6036                 tx_buffer_info->length = size;
6037                 tx_buffer_info->mapped_as_page = false;
6038                 tx_buffer_info->dma = dma_map_single(&pdev->dev,
6039                                                      skb->data + offset,
6040                                                      size, DMA_TO_DEVICE);
6041                 if (dma_mapping_error(&pdev->dev, tx_buffer_info->dma))
6042                         goto dma_error;
6043                 tx_buffer_info->time_stamp = jiffies;
6044                 tx_buffer_info->next_to_watch = i;
6045
6046                 len -= size;
6047                 total -= size;
6048                 offset += size;
6049                 count++;
6050
6051                 if (len) {
6052                         i++;
6053                         if (i == tx_ring->count)
6054                                 i = 0;
6055                 }
6056         }
6057
6058         for (f = 0; f < nr_frags; f++) {
6059                 struct skb_frag_struct *frag;
6060
6061                 frag = &skb_shinfo(skb)->frags[f];
6062                 len = min((unsigned int)frag->size, total);
6063                 offset = frag->page_offset;
6064
6065                 while (len) {
6066                         i++;
6067                         if (i == tx_ring->count)
6068                                 i = 0;
6069
6070                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
6071                         size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6072
6073                         tx_buffer_info->length = size;
6074                         tx_buffer_info->dma = dma_map_page(&adapter->pdev->dev,
6075                                                            frag->page,
6076                                                            offset, size,
6077                                                            DMA_TO_DEVICE);
6078                         tx_buffer_info->mapped_as_page = true;
6079                         if (dma_mapping_error(&pdev->dev, tx_buffer_info->dma))
6080                                 goto dma_error;
6081                         tx_buffer_info->time_stamp = jiffies;
6082                         tx_buffer_info->next_to_watch = i;
6083
6084                         len -= size;
6085                         total -= size;
6086                         offset += size;
6087                         count++;
6088                 }
6089                 if (total == 0)
6090                         break;
6091         }
6092
6093         tx_ring->tx_buffer_info[i].skb = skb;
6094         tx_ring->tx_buffer_info[first].next_to_watch = i;
6095
6096         return count;
6097
6098 dma_error:
6099         e_dev_err("TX DMA map failed\n");
6100
6101         /* clear timestamp and dma mappings for failed tx_buffer_info map */
6102         tx_buffer_info->dma = 0;
6103         tx_buffer_info->time_stamp = 0;
6104         tx_buffer_info->next_to_watch = 0;
6105         if (count)
6106                 count--;
6107
6108         /* clear timestamp and dma mappings for remaining portion of packet */
6109         while (count--) {
6110                 if (i == 0)
6111                         i += tx_ring->count;
6112                 i--;
6113                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6114                 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
6115         }
6116
6117         return 0;
6118 }
6119
6120 static void ixgbe_tx_queue(struct ixgbe_adapter *adapter,
6121                            struct ixgbe_ring *tx_ring,
6122                            int tx_flags, int count, u32 paylen, u8 hdr_len)
6123 {
6124         union ixgbe_adv_tx_desc *tx_desc = NULL;
6125         struct ixgbe_tx_buffer *tx_buffer_info;
6126         u32 olinfo_status = 0, cmd_type_len = 0;
6127         unsigned int i;
6128         u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
6129
6130         cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
6131
6132         cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
6133
6134         if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6135                 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
6136
6137         if (tx_flags & IXGBE_TX_FLAGS_TSO) {
6138                 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6139
6140                 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6141                                  IXGBE_ADVTXD_POPTS_SHIFT;
6142
6143                 /* use index 1 context for tso */
6144                 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6145                 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6146                         olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
6147                                          IXGBE_ADVTXD_POPTS_SHIFT;
6148
6149         } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6150                 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6151                                  IXGBE_ADVTXD_POPTS_SHIFT;
6152
6153         if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6154                 olinfo_status |= IXGBE_ADVTXD_CC;
6155                 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6156                 if (tx_flags & IXGBE_TX_FLAGS_FSO)
6157                         cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6158         }
6159
6160         olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
6161
6162         i = tx_ring->next_to_use;
6163         while (count--) {
6164                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6165                 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
6166                 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
6167                 tx_desc->read.cmd_type_len =
6168                         cpu_to_le32(cmd_type_len | tx_buffer_info->length);
6169                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6170                 i++;
6171                 if (i == tx_ring->count)
6172                         i = 0;
6173         }
6174
6175         tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
6176
6177         /*
6178          * Force memory writes to complete before letting h/w
6179          * know there are new descriptors to fetch.  (Only
6180          * applicable for weak-ordered memory model archs,
6181          * such as IA-64).
6182          */
6183         wmb();
6184
6185         tx_ring->next_to_use = i;
6186         writel(i, adapter->hw.hw_addr + tx_ring->tail);
6187 }
6188
6189 static void ixgbe_atr(struct ixgbe_adapter *adapter, struct sk_buff *skb,
6190                       int queue, u32 tx_flags)
6191 {
6192         struct ixgbe_atr_input atr_input;
6193         struct tcphdr *th;
6194         struct iphdr *iph = ip_hdr(skb);
6195         struct ethhdr *eth = (struct ethhdr *)skb->data;
6196         u16 vlan_id, src_port, dst_port, flex_bytes;
6197         u32 src_ipv4_addr, dst_ipv4_addr;
6198         u8 l4type = 0;
6199
6200         /* Right now, we support IPv4 only */
6201         if (skb->protocol != htons(ETH_P_IP))
6202                 return;
6203         /* check if we're UDP or TCP */
6204         if (iph->protocol == IPPROTO_TCP) {
6205                 th = tcp_hdr(skb);
6206                 src_port = th->source;
6207                 dst_port = th->dest;
6208                 l4type |= IXGBE_ATR_L4TYPE_TCP;
6209                 /* l4type IPv4 type is 0, no need to assign */
6210         } else {
6211                 /* Unsupported L4 header, just bail here */
6212                 return;
6213         }
6214
6215         memset(&atr_input, 0, sizeof(struct ixgbe_atr_input));
6216
6217         vlan_id = (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK) >>
6218                    IXGBE_TX_FLAGS_VLAN_SHIFT;
6219         src_ipv4_addr = iph->saddr;
6220         dst_ipv4_addr = iph->daddr;
6221         flex_bytes = eth->h_proto;
6222
6223         ixgbe_atr_set_vlan_id_82599(&atr_input, vlan_id);
6224         ixgbe_atr_set_src_port_82599(&atr_input, dst_port);
6225         ixgbe_atr_set_dst_port_82599(&atr_input, src_port);
6226         ixgbe_atr_set_flex_byte_82599(&atr_input, flex_bytes);
6227         ixgbe_atr_set_l4type_82599(&atr_input, l4type);
6228         /* src and dst are inverted, think how the receiver sees them */
6229         ixgbe_atr_set_src_ipv4_82599(&atr_input, dst_ipv4_addr);
6230         ixgbe_atr_set_dst_ipv4_82599(&atr_input, src_ipv4_addr);
6231
6232         /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6233         ixgbe_fdir_add_signature_filter_82599(&adapter->hw, &atr_input, queue);
6234 }
6235
6236 static int __ixgbe_maybe_stop_tx(struct net_device *netdev,
6237                                  struct ixgbe_ring *tx_ring, int size)
6238 {
6239         netif_stop_subqueue(netdev, tx_ring->queue_index);
6240         /* Herbert's original patch had:
6241          *  smp_mb__after_netif_stop_queue();
6242          * but since that doesn't exist yet, just open code it. */
6243         smp_mb();
6244
6245         /* We need to check again in a case another CPU has just
6246          * made room available. */
6247         if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
6248                 return -EBUSY;
6249
6250         /* A reprieve! - use start_queue because it doesn't call schedule */
6251         netif_start_subqueue(netdev, tx_ring->queue_index);
6252         ++tx_ring->restart_queue;
6253         return 0;
6254 }
6255
6256 static int ixgbe_maybe_stop_tx(struct net_device *netdev,
6257                               struct ixgbe_ring *tx_ring, int size)
6258 {
6259         if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
6260                 return 0;
6261         return __ixgbe_maybe_stop_tx(netdev, tx_ring, size);
6262 }
6263
6264 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6265 {
6266         struct ixgbe_adapter *adapter = netdev_priv(dev);
6267         int txq = smp_processor_id();
6268
6269 #ifdef IXGBE_FCOE
6270         if ((skb->protocol == htons(ETH_P_FCOE)) ||
6271             (skb->protocol == htons(ETH_P_FIP))) {
6272                 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
6273                         txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6274                         txq += adapter->ring_feature[RING_F_FCOE].mask;
6275                         return txq;
6276 #ifdef CONFIG_IXGBE_DCB
6277                 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6278                         txq = adapter->fcoe.up;
6279                         return txq;
6280 #endif
6281                 }
6282         }
6283 #endif
6284
6285         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6286                 while (unlikely(txq >= dev->real_num_tx_queues))
6287                         txq -= dev->real_num_tx_queues;
6288                 return txq;
6289         }
6290
6291         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6292                 if (skb->priority == TC_PRIO_CONTROL)
6293                         txq = adapter->ring_feature[RING_F_DCB].indices-1;
6294                 else
6295                         txq = (skb->vlan_tci & IXGBE_TX_FLAGS_VLAN_PRIO_MASK)
6296                                >> 13;
6297                 return txq;
6298         }
6299
6300         return skb_tx_hash(dev, skb);
6301 }
6302
6303 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb, struct net_device *netdev,
6304                           struct ixgbe_adapter *adapter,
6305                           struct ixgbe_ring *tx_ring)
6306 {
6307         struct netdev_queue *txq;
6308         unsigned int first;
6309         unsigned int tx_flags = 0;
6310         u8 hdr_len = 0;
6311         int tso;
6312         int count = 0;
6313         unsigned int f;
6314
6315         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
6316                 tx_flags |= vlan_tx_tag_get(skb);
6317                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6318                         tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6319                         tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6320                 }
6321                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6322                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6323         } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED &&
6324                    skb->priority != TC_PRIO_CONTROL) {
6325                 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6326                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6327                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6328         }
6329
6330 #ifdef IXGBE_FCOE
6331         /* for FCoE with DCB, we force the priority to what
6332          * was specified by the switch */
6333         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED &&
6334             (skb->protocol == htons(ETH_P_FCOE) ||
6335              skb->protocol == htons(ETH_P_FIP))) {
6336 #ifdef CONFIG_IXGBE_DCB
6337                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6338                         tx_flags &= ~(IXGBE_TX_FLAGS_VLAN_PRIO_MASK
6339                                       << IXGBE_TX_FLAGS_VLAN_SHIFT);
6340                         tx_flags |= ((adapter->fcoe.up << 13)
6341                                       << IXGBE_TX_FLAGS_VLAN_SHIFT);
6342                 }
6343 #endif
6344                 /* flag for FCoE offloads */
6345                 if (skb->protocol == htons(ETH_P_FCOE))
6346                         tx_flags |= IXGBE_TX_FLAGS_FCOE;
6347         }
6348 #endif
6349
6350         /* four things can cause us to need a context descriptor */
6351         if (skb_is_gso(skb) ||
6352             (skb->ip_summed == CHECKSUM_PARTIAL) ||
6353             (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
6354             (tx_flags & IXGBE_TX_FLAGS_FCOE))
6355                 count++;
6356
6357         count += TXD_USE_COUNT(skb_headlen(skb));
6358         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6359                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6360
6361         if (ixgbe_maybe_stop_tx(netdev, tx_ring, count)) {
6362                 adapter->tx_busy++;
6363                 return NETDEV_TX_BUSY;
6364         }
6365
6366         first = tx_ring->next_to_use;
6367         if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6368 #ifdef IXGBE_FCOE
6369                 /* setup tx offload for FCoE */
6370                 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
6371                 if (tso < 0) {
6372                         dev_kfree_skb_any(skb);
6373                         return NETDEV_TX_OK;
6374                 }
6375                 if (tso)
6376                         tx_flags |= IXGBE_TX_FLAGS_FSO;
6377 #endif /* IXGBE_FCOE */
6378         } else {
6379                 if (skb->protocol == htons(ETH_P_IP))
6380                         tx_flags |= IXGBE_TX_FLAGS_IPV4;
6381                 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
6382                 if (tso < 0) {
6383                         dev_kfree_skb_any(skb);
6384                         return NETDEV_TX_OK;
6385                 }
6386
6387                 if (tso)
6388                         tx_flags |= IXGBE_TX_FLAGS_TSO;
6389                 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags) &&
6390                          (skb->ip_summed == CHECKSUM_PARTIAL))
6391                         tx_flags |= IXGBE_TX_FLAGS_CSUM;
6392         }
6393
6394         count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first);
6395         if (count) {
6396                 /* add the ATR filter if ATR is on */
6397                 if (tx_ring->atr_sample_rate) {
6398                         ++tx_ring->atr_count;
6399                         if ((tx_ring->atr_count >= tx_ring->atr_sample_rate) &&
6400                              test_bit(__IXGBE_FDIR_INIT_DONE,
6401                                       &tx_ring->reinit_state)) {
6402                                 ixgbe_atr(adapter, skb, tx_ring->queue_index,
6403                                           tx_flags);
6404                                 tx_ring->atr_count = 0;
6405                         }
6406                 }
6407                 txq = netdev_get_tx_queue(netdev, tx_ring->queue_index);
6408                 txq->tx_bytes += skb->len;
6409                 txq->tx_packets++;
6410                 ixgbe_tx_queue(adapter, tx_ring, tx_flags, count, skb->len,
6411                                hdr_len);
6412                 ixgbe_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
6413
6414         } else {
6415                 dev_kfree_skb_any(skb);
6416                 tx_ring->tx_buffer_info[first].time_stamp = 0;
6417                 tx_ring->next_to_use = first;
6418         }
6419
6420         return NETDEV_TX_OK;
6421 }
6422
6423 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
6424 {
6425         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6426         struct ixgbe_ring *tx_ring;
6427
6428         tx_ring = adapter->tx_ring[skb->queue_mapping];
6429         return ixgbe_xmit_frame_ring(skb, netdev, adapter, tx_ring);
6430 }
6431
6432 /**
6433  * ixgbe_set_mac - Change the Ethernet Address of the NIC
6434  * @netdev: network interface device structure
6435  * @p: pointer to an address structure
6436  *
6437  * Returns 0 on success, negative on failure
6438  **/
6439 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6440 {
6441         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6442         struct ixgbe_hw *hw = &adapter->hw;
6443         struct sockaddr *addr = p;
6444
6445         if (!is_valid_ether_addr(addr->sa_data))
6446                 return -EADDRNOTAVAIL;
6447
6448         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6449         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6450
6451         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6452                             IXGBE_RAH_AV);
6453
6454         return 0;
6455 }
6456
6457 static int
6458 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6459 {
6460         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6461         struct ixgbe_hw *hw = &adapter->hw;
6462         u16 value;
6463         int rc;
6464
6465         if (prtad != hw->phy.mdio.prtad)
6466                 return -EINVAL;
6467         rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6468         if (!rc)
6469                 rc = value;
6470         return rc;
6471 }
6472
6473 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6474                             u16 addr, u16 value)
6475 {
6476         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6477         struct ixgbe_hw *hw = &adapter->hw;
6478
6479         if (prtad != hw->phy.mdio.prtad)
6480                 return -EINVAL;
6481         return hw->phy.ops.write_reg(hw, addr, devad, value);
6482 }
6483
6484 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6485 {
6486         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6487
6488         return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6489 }
6490
6491 /**
6492  * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6493  * netdev->dev_addrs
6494  * @netdev: network interface device structure
6495  *
6496  * Returns non-zero on failure
6497  **/
6498 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6499 {
6500         int err = 0;
6501         struct ixgbe_adapter *adapter = netdev_priv(dev);
6502         struct ixgbe_mac_info *mac = &adapter->hw.mac;
6503
6504         if (is_valid_ether_addr(mac->san_addr)) {
6505                 rtnl_lock();
6506                 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6507                 rtnl_unlock();
6508         }
6509         return err;
6510 }
6511
6512 /**
6513  * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6514  * netdev->dev_addrs
6515  * @netdev: network interface device structure
6516  *
6517  * Returns non-zero on failure
6518  **/
6519 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6520 {
6521         int err = 0;
6522         struct ixgbe_adapter *adapter = netdev_priv(dev);
6523         struct ixgbe_mac_info *mac = &adapter->hw.mac;
6524
6525         if (is_valid_ether_addr(mac->san_addr)) {
6526                 rtnl_lock();
6527                 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6528                 rtnl_unlock();
6529         }
6530         return err;
6531 }
6532
6533 #ifdef CONFIG_NET_POLL_CONTROLLER
6534 /*
6535  * Polling 'interrupt' - used by things like netconsole to send skbs
6536  * without having to re-enable interrupts. It's not called while
6537  * the interrupt routine is executing.
6538  */
6539 static void ixgbe_netpoll(struct net_device *netdev)
6540 {
6541         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6542         int i;
6543
6544         /* if interface is down do nothing */
6545         if (test_bit(__IXGBE_DOWN, &adapter->state))
6546                 return;
6547
6548         adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6549         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6550                 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
6551                 for (i = 0; i < num_q_vectors; i++) {
6552                         struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
6553                         ixgbe_msix_clean_many(0, q_vector);
6554                 }
6555         } else {
6556                 ixgbe_intr(adapter->pdev->irq, netdev);
6557         }
6558         adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6559 }
6560 #endif
6561
6562 static const struct net_device_ops ixgbe_netdev_ops = {
6563         .ndo_open               = ixgbe_open,
6564         .ndo_stop               = ixgbe_close,
6565         .ndo_start_xmit         = ixgbe_xmit_frame,
6566         .ndo_select_queue       = ixgbe_select_queue,
6567         .ndo_set_rx_mode        = ixgbe_set_rx_mode,
6568         .ndo_set_multicast_list = ixgbe_set_rx_mode,
6569         .ndo_validate_addr      = eth_validate_addr,
6570         .ndo_set_mac_address    = ixgbe_set_mac,
6571         .ndo_change_mtu         = ixgbe_change_mtu,
6572         .ndo_tx_timeout         = ixgbe_tx_timeout,
6573         .ndo_vlan_rx_register   = ixgbe_vlan_rx_register,
6574         .ndo_vlan_rx_add_vid    = ixgbe_vlan_rx_add_vid,
6575         .ndo_vlan_rx_kill_vid   = ixgbe_vlan_rx_kill_vid,
6576         .ndo_do_ioctl           = ixgbe_ioctl,
6577         .ndo_set_vf_mac         = ixgbe_ndo_set_vf_mac,
6578         .ndo_set_vf_vlan        = ixgbe_ndo_set_vf_vlan,
6579         .ndo_set_vf_tx_rate     = ixgbe_ndo_set_vf_bw,
6580         .ndo_get_vf_config      = ixgbe_ndo_get_vf_config,
6581 #ifdef CONFIG_NET_POLL_CONTROLLER
6582         .ndo_poll_controller    = ixgbe_netpoll,
6583 #endif
6584 #ifdef IXGBE_FCOE
6585         .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
6586         .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
6587         .ndo_fcoe_enable = ixgbe_fcoe_enable,
6588         .ndo_fcoe_disable = ixgbe_fcoe_disable,
6589         .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
6590 #endif /* IXGBE_FCOE */
6591 };
6592
6593 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
6594                            const struct ixgbe_info *ii)
6595 {
6596 #ifdef CONFIG_PCI_IOV
6597         struct ixgbe_hw *hw = &adapter->hw;
6598         int err;
6599
6600         if (hw->mac.type != ixgbe_mac_82599EB || !max_vfs)
6601                 return;
6602
6603         /* The 82599 supports up to 64 VFs per physical function
6604          * but this implementation limits allocation to 63 so that
6605          * basic networking resources are still available to the
6606          * physical function
6607          */
6608         adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
6609         adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
6610         err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
6611         if (err) {
6612                 e_err(probe, "Failed to enable PCI sriov: %d\n", err);
6613                 goto err_novfs;
6614         }
6615         /* If call to enable VFs succeeded then allocate memory
6616          * for per VF control structures.
6617          */
6618         adapter->vfinfo =
6619                 kcalloc(adapter->num_vfs,
6620                         sizeof(struct vf_data_storage), GFP_KERNEL);
6621         if (adapter->vfinfo) {
6622                 /* Now that we're sure SR-IOV is enabled
6623                  * and memory allocated set up the mailbox parameters
6624                  */
6625                 ixgbe_init_mbx_params_pf(hw);
6626                 memcpy(&hw->mbx.ops, ii->mbx_ops,
6627                        sizeof(hw->mbx.ops));
6628
6629                 /* Disable RSC when in SR-IOV mode */
6630                 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
6631                                      IXGBE_FLAG2_RSC_ENABLED);
6632                 return;
6633         }
6634
6635         /* Oh oh */
6636         e_err(probe, "Unable to allocate memory for VF Data Storage - "
6637               "SRIOV disabled\n");
6638         pci_disable_sriov(adapter->pdev);
6639
6640 err_novfs:
6641         adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
6642         adapter->num_vfs = 0;
6643 #endif /* CONFIG_PCI_IOV */
6644 }
6645
6646 /**
6647  * ixgbe_probe - Device Initialization Routine
6648  * @pdev: PCI device information struct
6649  * @ent: entry in ixgbe_pci_tbl
6650  *
6651  * Returns 0 on success, negative on failure
6652  *
6653  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
6654  * The OS initialization, configuring of the adapter private structure,
6655  * and a hardware reset occur.
6656  **/
6657 static int __devinit ixgbe_probe(struct pci_dev *pdev,
6658                                  const struct pci_device_id *ent)
6659 {
6660         struct net_device *netdev;
6661         struct ixgbe_adapter *adapter = NULL;
6662         struct ixgbe_hw *hw;
6663         const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
6664         static int cards_found;
6665         int i, err, pci_using_dac;
6666         unsigned int indices = num_possible_cpus();
6667 #ifdef IXGBE_FCOE
6668         u16 device_caps;
6669 #endif
6670         u32 part_num, eec;
6671
6672         /* Catch broken hardware that put the wrong VF device ID in
6673          * the PCIe SR-IOV capability.
6674          */
6675         if (pdev->is_virtfn) {
6676                 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
6677                      pci_name(pdev), pdev->vendor, pdev->device);
6678                 return -EINVAL;
6679         }
6680
6681         err = pci_enable_device_mem(pdev);
6682         if (err)
6683                 return err;
6684
6685         if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
6686             !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
6687                 pci_using_dac = 1;
6688         } else {
6689                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6690                 if (err) {
6691                         err = dma_set_coherent_mask(&pdev->dev,
6692                                                     DMA_BIT_MASK(32));
6693                         if (err) {
6694                                 dev_err(&pdev->dev,
6695                                         "No usable DMA configuration, aborting\n");
6696                                 goto err_dma;
6697                         }
6698                 }
6699                 pci_using_dac = 0;
6700         }
6701
6702         err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
6703                                            IORESOURCE_MEM), ixgbe_driver_name);
6704         if (err) {
6705                 dev_err(&pdev->dev,
6706                         "pci_request_selected_regions failed 0x%x\n", err);
6707                 goto err_pci_reg;
6708         }
6709
6710         pci_enable_pcie_error_reporting(pdev);
6711
6712         pci_set_master(pdev);
6713         pci_save_state(pdev);
6714
6715         if (ii->mac == ixgbe_mac_82598EB)
6716                 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
6717         else
6718                 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
6719
6720         indices = max_t(unsigned int, indices, IXGBE_MAX_DCB_INDICES);
6721 #ifdef IXGBE_FCOE
6722         indices += min_t(unsigned int, num_possible_cpus(),
6723                          IXGBE_MAX_FCOE_INDICES);
6724 #endif
6725         netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
6726         if (!netdev) {
6727                 err = -ENOMEM;
6728                 goto err_alloc_etherdev;
6729         }
6730
6731         SET_NETDEV_DEV(netdev, &pdev->dev);
6732
6733         pci_set_drvdata(pdev, netdev);
6734         adapter = netdev_priv(netdev);
6735
6736         adapter->netdev = netdev;
6737         adapter->pdev = pdev;
6738         hw = &adapter->hw;
6739         hw->back = adapter;
6740         adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
6741
6742         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
6743                               pci_resource_len(pdev, 0));
6744         if (!hw->hw_addr) {
6745                 err = -EIO;
6746                 goto err_ioremap;
6747         }
6748
6749         for (i = 1; i <= 5; i++) {
6750                 if (pci_resource_len(pdev, i) == 0)
6751                         continue;
6752         }
6753
6754         netdev->netdev_ops = &ixgbe_netdev_ops;
6755         ixgbe_set_ethtool_ops(netdev);
6756         netdev->watchdog_timeo = 5 * HZ;
6757         strcpy(netdev->name, pci_name(pdev));
6758
6759         adapter->bd_number = cards_found;
6760
6761         /* Setup hw api */
6762         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
6763         hw->mac.type  = ii->mac;
6764
6765         /* EEPROM */
6766         memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
6767         eec = IXGBE_READ_REG(hw, IXGBE_EEC);
6768         /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
6769         if (!(eec & (1 << 8)))
6770                 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
6771
6772         /* PHY */
6773         memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
6774         hw->phy.sfp_type = ixgbe_sfp_type_unknown;
6775         /* ixgbe_identify_phy_generic will set prtad and mmds properly */
6776         hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
6777         hw->phy.mdio.mmds = 0;
6778         hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
6779         hw->phy.mdio.dev = netdev;
6780         hw->phy.mdio.mdio_read = ixgbe_mdio_read;
6781         hw->phy.mdio.mdio_write = ixgbe_mdio_write;
6782
6783         /* set up this timer and work struct before calling get_invariants
6784          * which might start the timer
6785          */
6786         init_timer(&adapter->sfp_timer);
6787         adapter->sfp_timer.function = ixgbe_sfp_timer;
6788         adapter->sfp_timer.data = (unsigned long) adapter;
6789
6790         INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
6791
6792         /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
6793         INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
6794
6795         /* a new SFP+ module arrival, called from GPI SDP2 context */
6796         INIT_WORK(&adapter->sfp_config_module_task,
6797                   ixgbe_sfp_config_module_task);
6798
6799         ii->get_invariants(hw);
6800
6801         /* setup the private structure */
6802         err = ixgbe_sw_init(adapter);
6803         if (err)
6804                 goto err_sw_init;
6805
6806         /* Make it possible the adapter to be woken up via WOL */
6807         if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6808                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6809
6810         /*
6811          * If there is a fan on this device and it has failed log the
6812          * failure.
6813          */
6814         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
6815                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
6816                 if (esdp & IXGBE_ESDP_SDP1)
6817                         e_crit(probe, "Fan has stopped, replace the adapter\n");
6818         }
6819
6820         /* reset_hw fills in the perm_addr as well */
6821         hw->phy.reset_if_overtemp = true;
6822         err = hw->mac.ops.reset_hw(hw);
6823         hw->phy.reset_if_overtemp = false;
6824         if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
6825             hw->mac.type == ixgbe_mac_82598EB) {
6826                 /*
6827                  * Start a kernel thread to watch for a module to arrive.
6828                  * Only do this for 82598, since 82599 will generate
6829                  * interrupts on module arrival.
6830                  */
6831                 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6832                 mod_timer(&adapter->sfp_timer,
6833                           round_jiffies(jiffies + (2 * HZ)));
6834                 err = 0;
6835         } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
6836                 e_dev_err("failed to initialize because an unsupported SFP+ "
6837                           "module type was detected.\n");
6838                 e_dev_err("Reload the driver after installing a supported "
6839                           "module.\n");
6840                 goto err_sw_init;
6841         } else if (err) {
6842                 e_dev_err("HW Init failed: %d\n", err);
6843                 goto err_sw_init;
6844         }
6845
6846         ixgbe_probe_vf(adapter, ii);
6847
6848         netdev->features = NETIF_F_SG |
6849                            NETIF_F_IP_CSUM |
6850                            NETIF_F_HW_VLAN_TX |
6851                            NETIF_F_HW_VLAN_RX |
6852                            NETIF_F_HW_VLAN_FILTER;
6853
6854         netdev->features |= NETIF_F_IPV6_CSUM;
6855         netdev->features |= NETIF_F_TSO;
6856         netdev->features |= NETIF_F_TSO6;
6857         netdev->features |= NETIF_F_GRO;
6858
6859         if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6860                 netdev->features |= NETIF_F_SCTP_CSUM;
6861
6862         netdev->vlan_features |= NETIF_F_TSO;
6863         netdev->vlan_features |= NETIF_F_TSO6;
6864         netdev->vlan_features |= NETIF_F_IP_CSUM;
6865         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
6866         netdev->vlan_features |= NETIF_F_SG;
6867
6868         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6869                 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
6870                                     IXGBE_FLAG_DCB_ENABLED);
6871         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
6872                 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
6873
6874 #ifdef CONFIG_IXGBE_DCB
6875         netdev->dcbnl_ops = &dcbnl_ops;
6876 #endif
6877
6878 #ifdef IXGBE_FCOE
6879         if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6880                 if (hw->mac.ops.get_device_caps) {
6881                         hw->mac.ops.get_device_caps(hw, &device_caps);
6882                         if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
6883                                 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6884                 }
6885         }
6886         if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6887                 netdev->vlan_features |= NETIF_F_FCOE_CRC;
6888                 netdev->vlan_features |= NETIF_F_FSO;
6889                 netdev->vlan_features |= NETIF_F_FCOE_MTU;
6890         }
6891 #endif /* IXGBE_FCOE */
6892         if (pci_using_dac) {
6893                 netdev->features |= NETIF_F_HIGHDMA;
6894                 netdev->vlan_features |= NETIF_F_HIGHDMA;
6895         }
6896
6897         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6898                 netdev->features |= NETIF_F_LRO;
6899
6900         /* make sure the EEPROM is good */
6901         if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
6902                 e_dev_err("The EEPROM Checksum Is Not Valid\n");
6903                 err = -EIO;
6904                 goto err_eeprom;
6905         }
6906
6907         memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
6908         memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
6909
6910         if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
6911                 e_dev_err("invalid MAC address\n");
6912                 err = -EIO;
6913                 goto err_eeprom;
6914         }
6915
6916         /* power down the optics */
6917         if (hw->phy.multispeed_fiber)
6918                 hw->mac.ops.disable_tx_laser(hw);
6919
6920         init_timer(&adapter->watchdog_timer);
6921         adapter->watchdog_timer.function = ixgbe_watchdog;
6922         adapter->watchdog_timer.data = (unsigned long)adapter;
6923
6924         INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
6925         INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
6926
6927         err = ixgbe_init_interrupt_scheme(adapter);
6928         if (err)
6929                 goto err_sw_init;
6930
6931         switch (pdev->device) {
6932         case IXGBE_DEV_ID_82599_KX4:
6933                 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
6934                                 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
6935                 break;
6936         default:
6937                 adapter->wol = 0;
6938                 break;
6939         }
6940         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6941
6942         /* pick up the PCI bus settings for reporting later */
6943         hw->mac.ops.get_bus_info(hw);
6944
6945         /* print bus type/speed/width info */
6946         e_dev_info("(PCI Express:%s:%s) %pM\n",
6947                    (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0Gb/s" :
6948                     hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5Gb/s" :
6949                     "Unknown"),
6950                    (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
6951                     hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
6952                     hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
6953                     "Unknown"),
6954                    netdev->dev_addr);
6955         ixgbe_read_pba_num_generic(hw, &part_num);
6956         if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
6957                 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, "
6958                            "PBA No: %06x-%03x\n",
6959                            hw->mac.type, hw->phy.type, hw->phy.sfp_type,
6960                            (part_num >> 8), (part_num & 0xff));
6961         else
6962                 e_dev_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
6963                            hw->mac.type, hw->phy.type,
6964                            (part_num >> 8), (part_num & 0xff));
6965
6966         if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
6967                 e_dev_warn("PCI-Express bandwidth available for this card is "
6968                            "not sufficient for optimal performance.\n");
6969                 e_dev_warn("For optimal performance a x8 PCI-Express slot "
6970                            "is required.\n");
6971         }
6972
6973         /* save off EEPROM version number */
6974         hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
6975
6976         /* reset the hardware with the new settings */
6977         err = hw->mac.ops.start_hw(hw);
6978
6979         if (err == IXGBE_ERR_EEPROM_VERSION) {
6980                 /* We are running on a pre-production device, log a warning */
6981                 e_dev_warn("This device is a pre-production adapter/LOM. "
6982                            "Please be aware there may be issues associated "
6983                            "with your hardware.  If you are experiencing "
6984                            "problems please contact your Intel or hardware "
6985                            "representative who provided you with this "
6986                            "hardware.\n");
6987         }
6988         strcpy(netdev->name, "eth%d");
6989         err = register_netdev(netdev);
6990         if (err)
6991                 goto err_register;
6992
6993         /* carrier off reporting is important to ethtool even BEFORE open */
6994         netif_carrier_off(netdev);
6995
6996         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
6997             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
6998                 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
6999
7000         if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
7001                 INIT_WORK(&adapter->check_overtemp_task,
7002                           ixgbe_check_overtemp_task);
7003 #ifdef CONFIG_IXGBE_DCA
7004         if (dca_add_requester(&pdev->dev) == 0) {
7005                 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7006                 ixgbe_setup_dca(adapter);
7007         }
7008 #endif
7009         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7010                 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7011                 for (i = 0; i < adapter->num_vfs; i++)
7012                         ixgbe_vf_configuration(pdev, (i | 0x10000000));
7013         }
7014
7015         /* add san mac addr to netdev */
7016         ixgbe_add_sanmac_netdev(netdev);
7017
7018         e_dev_info("Intel(R) 10 Gigabit Network Connection\n");
7019         cards_found++;
7020         return 0;
7021
7022 err_register:
7023         ixgbe_release_hw_control(adapter);
7024         ixgbe_clear_interrupt_scheme(adapter);
7025 err_sw_init:
7026 err_eeprom:
7027         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7028                 ixgbe_disable_sriov(adapter);
7029         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7030         del_timer_sync(&adapter->sfp_timer);
7031         cancel_work_sync(&adapter->sfp_task);
7032         cancel_work_sync(&adapter->multispeed_fiber_task);
7033         cancel_work_sync(&adapter->sfp_config_module_task);
7034         iounmap(hw->hw_addr);
7035 err_ioremap:
7036         free_netdev(netdev);
7037 err_alloc_etherdev:
7038         pci_release_selected_regions(pdev,
7039                                      pci_select_bars(pdev, IORESOURCE_MEM));
7040 err_pci_reg:
7041 err_dma:
7042         pci_disable_device(pdev);
7043         return err;
7044 }
7045
7046 /**
7047  * ixgbe_remove - Device Removal Routine
7048  * @pdev: PCI device information struct
7049  *
7050  * ixgbe_remove is called by the PCI subsystem to alert the driver
7051  * that it should release a PCI device.  The could be caused by a
7052  * Hot-Plug event, or because the driver is going to be removed from
7053  * memory.
7054  **/
7055 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7056 {
7057         struct net_device *netdev = pci_get_drvdata(pdev);
7058         struct ixgbe_adapter *adapter = netdev_priv(netdev);
7059
7060         set_bit(__IXGBE_DOWN, &adapter->state);
7061         /* clear the module not found bit to make sure the worker won't
7062          * reschedule
7063          */
7064         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7065         del_timer_sync(&adapter->watchdog_timer);
7066
7067         del_timer_sync(&adapter->sfp_timer);
7068         cancel_work_sync(&adapter->watchdog_task);
7069         cancel_work_sync(&adapter->sfp_task);
7070         cancel_work_sync(&adapter->multispeed_fiber_task);
7071         cancel_work_sync(&adapter->sfp_config_module_task);
7072         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7073             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7074                 cancel_work_sync(&adapter->fdir_reinit_task);
7075         flush_scheduled_work();
7076
7077 #ifdef CONFIG_IXGBE_DCA
7078         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7079                 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7080                 dca_remove_requester(&pdev->dev);
7081                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7082         }
7083
7084 #endif
7085 #ifdef IXGBE_FCOE
7086         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7087                 ixgbe_cleanup_fcoe(adapter);
7088
7089 #endif /* IXGBE_FCOE */
7090
7091         /* remove the added san mac */
7092         ixgbe_del_sanmac_netdev(netdev);
7093
7094         if (netdev->reg_state == NETREG_REGISTERED)
7095                 unregister_netdev(netdev);
7096
7097         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7098                 ixgbe_disable_sriov(adapter);
7099
7100         ixgbe_clear_interrupt_scheme(adapter);
7101
7102         ixgbe_release_hw_control(adapter);
7103
7104         iounmap(adapter->hw.hw_addr);
7105         pci_release_selected_regions(pdev, pci_select_bars(pdev,
7106                                      IORESOURCE_MEM));
7107
7108         e_dev_info("complete\n");
7109
7110         free_netdev(netdev);
7111
7112         pci_disable_pcie_error_reporting(pdev);
7113
7114         pci_disable_device(pdev);
7115 }
7116
7117 /**
7118  * ixgbe_io_error_detected - called when PCI error is detected
7119  * @pdev: Pointer to PCI device
7120  * @state: The current pci connection state
7121  *
7122  * This function is called after a PCI bus error affecting
7123  * this device has been detected.
7124  */
7125 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7126                                                 pci_channel_state_t state)
7127 {
7128         struct net_device *netdev = pci_get_drvdata(pdev);
7129         struct ixgbe_adapter *adapter = netdev_priv(netdev);
7130
7131         netif_device_detach(netdev);
7132
7133         if (state == pci_channel_io_perm_failure)
7134                 return PCI_ERS_RESULT_DISCONNECT;
7135
7136         if (netif_running(netdev))
7137                 ixgbe_down(adapter);
7138         pci_disable_device(pdev);
7139
7140         /* Request a slot reset. */
7141         return PCI_ERS_RESULT_NEED_RESET;
7142 }
7143
7144 /**
7145  * ixgbe_io_slot_reset - called after the pci bus has been reset.
7146  * @pdev: Pointer to PCI device
7147  *
7148  * Restart the card from scratch, as if from a cold-boot.
7149  */
7150 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7151 {
7152         struct net_device *netdev = pci_get_drvdata(pdev);
7153         struct ixgbe_adapter *adapter = netdev_priv(netdev);
7154         pci_ers_result_t result;
7155         int err;
7156
7157         if (pci_enable_device_mem(pdev)) {
7158                 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7159                 result = PCI_ERS_RESULT_DISCONNECT;
7160         } else {
7161                 pci_set_master(pdev);
7162                 pci_restore_state(pdev);
7163                 pci_save_state(pdev);
7164
7165                 pci_wake_from_d3(pdev, false);
7166
7167                 ixgbe_reset(adapter);
7168                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7169                 result = PCI_ERS_RESULT_RECOVERED;
7170         }
7171
7172         err = pci_cleanup_aer_uncorrect_error_status(pdev);
7173         if (err) {
7174                 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7175                           "failed 0x%0x\n", err);
7176                 /* non-fatal, continue */
7177         }
7178
7179         return result;
7180 }
7181
7182 /**
7183  * ixgbe_io_resume - called when traffic can start flowing again.
7184  * @pdev: Pointer to PCI device
7185  *
7186  * This callback is called when the error recovery driver tells us that
7187  * its OK to resume normal operation.
7188  */
7189 static void ixgbe_io_resume(struct pci_dev *pdev)
7190 {
7191         struct net_device *netdev = pci_get_drvdata(pdev);
7192         struct ixgbe_adapter *adapter = netdev_priv(netdev);
7193
7194         if (netif_running(netdev)) {
7195                 if (ixgbe_up(adapter)) {
7196                         e_info(probe, "ixgbe_up failed after reset\n");
7197                         return;
7198                 }
7199         }
7200
7201         netif_device_attach(netdev);
7202 }
7203
7204 static struct pci_error_handlers ixgbe_err_handler = {
7205         .error_detected = ixgbe_io_error_detected,
7206         .slot_reset = ixgbe_io_slot_reset,
7207         .resume = ixgbe_io_resume,
7208 };
7209
7210 static struct pci_driver ixgbe_driver = {
7211         .name     = ixgbe_driver_name,
7212         .id_table = ixgbe_pci_tbl,
7213         .probe    = ixgbe_probe,
7214         .remove   = __devexit_p(ixgbe_remove),
7215 #ifdef CONFIG_PM
7216         .suspend  = ixgbe_suspend,
7217         .resume   = ixgbe_resume,
7218 #endif
7219         .shutdown = ixgbe_shutdown,
7220         .err_handler = &ixgbe_err_handler
7221 };
7222
7223 /**
7224  * ixgbe_init_module - Driver Registration Routine
7225  *
7226  * ixgbe_init_module is the first routine called when the driver is
7227  * loaded. All it does is register with the PCI subsystem.
7228  **/
7229 static int __init ixgbe_init_module(void)
7230 {
7231         int ret;
7232         pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7233         pr_info("%s\n", ixgbe_copyright);
7234
7235 #ifdef CONFIG_IXGBE_DCA
7236         dca_register_notify(&dca_notifier);
7237 #endif
7238
7239         ret = pci_register_driver(&ixgbe_driver);
7240         return ret;
7241 }
7242
7243 module_init(ixgbe_init_module);
7244
7245 /**
7246  * ixgbe_exit_module - Driver Exit Cleanup Routine
7247  *
7248  * ixgbe_exit_module is called just before the driver is removed
7249  * from memory.
7250  **/
7251 static void __exit ixgbe_exit_module(void)
7252 {
7253 #ifdef CONFIG_IXGBE_DCA
7254         dca_unregister_notify(&dca_notifier);
7255 #endif
7256         pci_unregister_driver(&ixgbe_driver);
7257 }
7258
7259 #ifdef CONFIG_IXGBE_DCA
7260 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7261                             void *p)
7262 {
7263         int ret_val;
7264
7265         ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7266                                          __ixgbe_notify_dca);
7267
7268         return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7269 }
7270
7271 #endif /* CONFIG_IXGBE_DCA */
7272
7273 /**
7274  * ixgbe_get_hw_dev return device
7275  * used by hardware layer to print debugging information
7276  **/
7277 struct net_device *ixgbe_get_hw_dev(struct ixgbe_hw *hw)
7278 {
7279         struct ixgbe_adapter *adapter = hw->back;
7280         return adapter->netdev;
7281 }
7282
7283 module_exit(ixgbe_exit_module);
7284
7285 /* ixgbe_main.c */