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_ring *tx_ring,
 604                                       struct ixgbe_tx_buffer *tx_buffer_info)
 605 {
 606         if (tx_buffer_info->dma) {
 607                 if (tx_buffer_info->mapped_as_page)
 608                         dma_unmap_page(tx_ring->dev,
 609                                        tx_buffer_info->dma,
 610                                        tx_buffer_info->length,
 611                                        DMA_TO_DEVICE);
 612                 else
 613                         dma_unmap_single(tx_ring->dev,
 614                                          tx_buffer_info->dma,
 615                                          tx_buffer_info->length,
 616                                          DMA_TO_DEVICE);
 617                 tx_buffer_info->dma = 0;
 618         }
 619         if (tx_buffer_info->skb) {
 620                 dev_kfree_skb_any(tx_buffer_info->skb);
 621                 tx_buffer_info->skb = NULL;
 622         }
 623         tx_buffer_info->time_stamp = 0;
 624         /* tx_buffer_info must be completely set up in the transmit path */
 625 }
 626
 627 /**
 628  * ixgbe_tx_xon_state - check the tx ring xon state
 629  * @adapter: the ixgbe adapter
 630  * @tx_ring: the corresponding tx_ring
 631  *
 632  * If not in DCB mode, checks TFCS.TXOFF, otherwise, find out the
 633  * corresponding TC of this tx_ring when checking TFCS.
 634  *
 635  * Returns : true if in xon state (currently not paused)
 636  */
 637 static inline bool ixgbe_tx_xon_state(struct ixgbe_adapter *adapter,
 638                                       struct ixgbe_ring *tx_ring)
 639 {
 640         u32 txoff = IXGBE_TFCS_TXOFF;
 641
 642 #ifdef CONFIG_IXGBE_DCB
 643         if (adapter->dcb_cfg.pfc_mode_enable) {
 644                 int tc;
 645                 int reg_idx = tx_ring->reg_idx;
 646                 int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
 647
 648                 switch (adapter->hw.mac.type) {
 649                 case ixgbe_mac_82598EB:
 650                         tc = reg_idx >> 2;
 651                         txoff = IXGBE_TFCS_TXOFF0;
 652                         break;
 653                 case ixgbe_mac_82599EB:
 654                         tc = 0;
 655                         txoff = IXGBE_TFCS_TXOFF;
 656                         if (dcb_i == 8) {
 657                                 /* TC0, TC1 */
 658                                 tc = reg_idx >> 5;
 659                                 if (tc == 2) /* TC2, TC3 */
 660                                         tc += (reg_idx - 64) >> 4;
 661                                 else if (tc == 3) /* TC4, TC5, TC6, TC7 */
 662                                         tc += 1 + ((reg_idx - 96) >> 3);
 663                         } else if (dcb_i == 4) {
 664                                 /* TC0, TC1 */
 665                                 tc = reg_idx >> 6;
 666                                 if (tc == 1) {
 667                                         tc += (reg_idx - 64) >> 5;
 668                                         if (tc == 2) /* TC2, TC3 */
 669                                                 tc += (reg_idx - 96) >> 4;
 670                                 }
 671                         }
 672                         break;
 673                 default:
 674                         tc = 0;
 675                 }
 676                 txoff <<= tc;
 677         }
 678 #endif
 679         return IXGBE_READ_REG(&adapter->hw, IXGBE_TFCS) & txoff;
 680 }
 681
 682 static inline bool ixgbe_check_tx_hang(struct ixgbe_adapter *adapter,
 683                                        struct ixgbe_ring *tx_ring,
 684                                        unsigned int eop)
 685 {
 686         struct ixgbe_hw *hw = &adapter->hw;
 687
 688         /* Detect a transmit hang in hardware, this serializes the
 689          * check with the clearing of time_stamp and movement of eop */
 690         adapter->detect_tx_hung = false;
 691         if (tx_ring->tx_buffer_info[eop].time_stamp &&
 692             time_after(jiffies, tx_ring->tx_buffer_info[eop].time_stamp + HZ) &&
 693             ixgbe_tx_xon_state(adapter, tx_ring)) {
 694                 /* detected Tx unit hang */
 695                 union ixgbe_adv_tx_desc *tx_desc;
 696                 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
 697                 e_err(drv, "Detected Tx Unit Hang\n"
 698                       "  Tx Queue             <%d>\n"
 699                       "  TDH, TDT             <%x>, <%x>\n"
 700                       "  next_to_use          <%x>\n"
 701                       "  next_to_clean        <%x>\n"
 702                       "tx_buffer_info[next_to_clean]\n"
 703                       "  time_stamp           <%lx>\n"
 704                       "  jiffies              <%lx>\n",
 705                       tx_ring->queue_index,
 706                       IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
 707                       IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
 708                       tx_ring->next_to_use, eop,
 709                       tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
 710                 return true;
 711         }
 712
 713         return false;
 714 }
 715
 716 #define IXGBE_MAX_TXD_PWR       14
 717 #define IXGBE_MAX_DATA_PER_TXD  (1 << IXGBE_MAX_TXD_PWR)
 718
 719 /* Tx Descriptors needed, worst case */
 720 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
 721                          (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
 722 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
 723         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
 724
 725 static void ixgbe_tx_timeout(struct net_device *netdev);
 726
 727 /**
 728  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
 729  * @q_vector: structure containing interrupt and ring information
 730  * @tx_ring: tx ring to clean
 731  **/
 732 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
 733                                struct ixgbe_ring *tx_ring)
 734 {
 735         struct ixgbe_adapter *adapter = q_vector->adapter;
 736         union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
 737         struct ixgbe_tx_buffer *tx_buffer_info;
 738         unsigned int i, eop, count = 0;
 739         unsigned int total_bytes = 0, total_packets = 0;
 740
 741         i = tx_ring->next_to_clean;
 742         eop = tx_ring->tx_buffer_info[i].next_to_watch;
 743         eop_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
 744
 745         while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
 746                (count < tx_ring->work_limit)) {
 747                 bool cleaned = false;
 748                 rmb(); /* read buffer_info after eop_desc */
 749                 for ( ; !cleaned; count++) {
 750                         tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
 751                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
 752
 753                         tx_desc->wb.status = 0;
 754                         cleaned = (i == eop);
 755
 756                         i++;
 757                         if (i == tx_ring->count)
 758                                 i = 0;
 759
 760                         if (cleaned && tx_buffer_info->skb) {
 761                                 total_bytes += tx_buffer_info->bytecount;
 762                                 total_packets += tx_buffer_info->gso_segs;
 763                         }
 764
 765                         ixgbe_unmap_and_free_tx_resource(tx_ring,
 766                                                          tx_buffer_info);
 767                 }
 768
 769                 eop = tx_ring->tx_buffer_info[i].next_to_watch;
 770                 eop_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
 771         }
 772
 773         tx_ring->next_to_clean = i;
 774
 775 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
 776         if (unlikely(count && netif_carrier_ok(tx_ring->netdev) &&
 777                      (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
 778                 /* Make sure that anybody stopping the queue after this
 779                  * sees the new next_to_clean.
 780                  */
 781                 smp_mb();
 782                 if (__netif_subqueue_stopped(tx_ring->netdev, tx_ring->queue_index) &&
 783                     !test_bit(__IXGBE_DOWN, &adapter->state)) {
 784                         netif_wake_subqueue(tx_ring->netdev, tx_ring->queue_index);
 785                         ++tx_ring->tx_stats.restart_queue;
 786                 }
 787         }
 788
 789         if (adapter->detect_tx_hung) {
 790                 if (ixgbe_check_tx_hang(adapter, tx_ring, i)) {
 791                         /* schedule immediate reset if we believe we hung */
 792                         e_info(probe, "tx hang %d detected, resetting "
 793                                "adapter\n", adapter->tx_timeout_count + 1);
 794                         ixgbe_tx_timeout(adapter->netdev);
 795                 }
 796         }
 797
 798         /* re-arm the interrupt */
 799         if (count >= tx_ring->work_limit)
 800                 ixgbe_irq_rearm_queues(adapter, ((u64)1 << q_vector->v_idx));
 801
 802         tx_ring->total_bytes += total_bytes;
 803         tx_ring->total_packets += total_packets;
 804         u64_stats_update_begin(&tx_ring->syncp);
 805         tx_ring->stats.packets += total_packets;
 806         tx_ring->stats.bytes += total_bytes;
 807         u64_stats_update_end(&tx_ring->syncp);
 808         return count < tx_ring->work_limit;
 809 }
 810
 811 #ifdef CONFIG_IXGBE_DCA
 812 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
 813                                 struct ixgbe_ring *rx_ring)
 814 {
 815         u32 rxctrl;
 816         int cpu = get_cpu();
 817         int q = rx_ring->reg_idx;
 818
 819         if (rx_ring->cpu != cpu) {
 820                 rxctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q));
 821                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
 822                         rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
 823                         rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
 824                 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
 825                         rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
 826                         rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
 827                                    IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
 828                 }
 829                 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
 830                 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
 831                 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
 832                 rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN |
 833                             IXGBE_DCA_RXCTRL_DESC_HSRO_EN);
 834                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q), rxctrl);
 835                 rx_ring->cpu = cpu;
 836         }
 837         put_cpu();
 838 }
 839
 840 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
 841                                 struct ixgbe_ring *tx_ring)
 842 {
 843         u32 txctrl;
 844         int cpu = get_cpu();
 845         int q = tx_ring->reg_idx;
 846         struct ixgbe_hw *hw = &adapter->hw;
 847
 848         if (tx_ring->cpu != cpu) {
 849                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
 850                         txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(q));
 851                         txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
 852                         txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
 853                         txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
 854                         IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(q), txctrl);
 855                 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
 856                         txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(q));
 857                         txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
 858                         txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
 859                                   IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
 860                         txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
 861                         IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(q), txctrl);
 862                 }
 863                 tx_ring->cpu = cpu;
 864         }
 865         put_cpu();
 866 }
 867
 868 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
 869 {
 870         int i;
 871
 872         if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
 873                 return;
 874
 875         /* always use CB2 mode, difference is masked in the CB driver */
 876         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
 877
 878         for (i = 0; i < adapter->num_tx_queues; i++) {
 879                 adapter->tx_ring[i]->cpu = -1;
 880                 ixgbe_update_tx_dca(adapter, adapter->tx_ring[i]);
 881         }
 882         for (i = 0; i < adapter->num_rx_queues; i++) {
 883                 adapter->rx_ring[i]->cpu = -1;
 884                 ixgbe_update_rx_dca(adapter, adapter->rx_ring[i]);
 885         }
 886 }
 887
 888 static int __ixgbe_notify_dca(struct device *dev, void *data)
 889 {
 890         struct net_device *netdev = dev_get_drvdata(dev);
 891         struct ixgbe_adapter *adapter = netdev_priv(netdev);
 892         unsigned long event = *(unsigned long *)data;
 893
 894         switch (event) {
 895         case DCA_PROVIDER_ADD:
 896                 /* if we're already enabled, don't do it again */
 897                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
 898                         break;
 899                 if (dca_add_requester(dev) == 0) {
 900                         adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
 901                         ixgbe_setup_dca(adapter);
 902                         break;
 903                 }
 904                 /* Fall Through since DCA is disabled. */
 905         case DCA_PROVIDER_REMOVE:
 906                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
 907                         dca_remove_requester(dev);
 908                         adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
 909                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
 910                 }
 911                 break;
 912         }
 913
 914         return 0;
 915 }
 916
 917 #endif /* CONFIG_IXGBE_DCA */
 918 /**
 919  * ixgbe_receive_skb - Send a completed packet up the stack
 920  * @adapter: board private structure
 921  * @skb: packet to send up
 922  * @status: hardware indication of status of receive
 923  * @rx_ring: rx descriptor ring (for a specific queue) to setup
 924  * @rx_desc: rx descriptor
 925  **/
 926 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
 927                               struct sk_buff *skb, u8 status,
 928                               struct ixgbe_ring *ring,
 929                               union ixgbe_adv_rx_desc *rx_desc)
 930 {
 931         struct ixgbe_adapter *adapter = q_vector->adapter;
 932         struct napi_struct *napi = &q_vector->napi;
 933         bool is_vlan = (status & IXGBE_RXD_STAT_VP);
 934         u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
 935
 936         if (is_vlan && (tag & VLAN_VID_MASK))
 937                 __vlan_hwaccel_put_tag(skb, tag);
 938
 939         if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
 940                 napi_gro_receive(napi, skb);
 941         else
 942                 netif_rx(skb);
 943 }
 944
 945 /**
 946  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
 947  * @adapter: address of board private structure
 948  * @status_err: hardware indication of status of receive
 949  * @skb: skb currently being received and modified
 950  **/
 951 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
 952                                      union ixgbe_adv_rx_desc *rx_desc,
 953                                      struct sk_buff *skb)
 954 {
 955         u32 status_err = le32_to_cpu(rx_desc->wb.upper.status_error);
 956
 957         skb_checksum_none_assert(skb);
 958
 959         /* Rx csum disabled */
 960         if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
 961                 return;
 962
 963         /* if IP and error */
 964         if ((status_err & IXGBE_RXD_STAT_IPCS) &&
 965             (status_err & IXGBE_RXDADV_ERR_IPE)) {
 966                 adapter->hw_csum_rx_error++;
 967                 return;
 968         }
 969
 970         if (!(status_err & IXGBE_RXD_STAT_L4CS))
 971                 return;
 972
 973         if (status_err & IXGBE_RXDADV_ERR_TCPE) {
 974                 u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
 975
 976                 /*
 977                  * 82599 errata, UDP frames with a 0 checksum can be marked as
 978                  * checksum errors.
 979                  */
 980                 if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
 981                     (adapter->hw.mac.type == ixgbe_mac_82599EB))
 982                         return;
 983
 984                 adapter->hw_csum_rx_error++;
 985                 return;
 986         }
 987
 988         /* It must be a TCP or UDP packet with a valid checksum */
 989         skb->ip_summed = CHECKSUM_UNNECESSARY;
 990 }
 991
 992 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
 993 {
 994         /*
 995          * Force memory writes to complete before letting h/w
 996          * know there are new descriptors to fetch.  (Only
 997          * applicable for weak-ordered memory model archs,
 998          * such as IA-64).
 999          */
1000         wmb();
1001         writel(val, rx_ring->tail);
1002 }
1003
1004 /**
1005  * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
1006  * @rx_ring: ring to place buffers on
1007  * @cleaned_count: number of buffers to replace
1008  **/
1009 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1010 {
1011         union ixgbe_adv_rx_desc *rx_desc;
1012         struct ixgbe_rx_buffer *bi;
1013         struct sk_buff *skb;
1014         u16 i = rx_ring->next_to_use;
1015
1016         /* do nothing if no valid netdev defined */
1017         if (!rx_ring->netdev)
1018                 return;
1019
1020         while (cleaned_count--) {
1021                 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1022                 bi = &rx_ring->rx_buffer_info[i];
1023                 skb = bi->skb;
1024
1025                 if (!skb) {
1026                         skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1027                                                         rx_ring->rx_buf_len);
1028                         if (!skb) {
1029                                 rx_ring->rx_stats.alloc_rx_buff_failed++;
1030                                 goto no_buffers;
1031                         }
1032                         /* initialize queue mapping */
1033                         skb_record_rx_queue(skb, rx_ring->queue_index);
1034                         bi->skb = skb;
1035                 }
1036
1037                 if (!bi->dma) {
1038                         bi->dma = dma_map_single(rx_ring->dev,
1039                                                  skb->data,
1040                                                  rx_ring->rx_buf_len,
1041                                                  DMA_FROM_DEVICE);
1042                         if (dma_mapping_error(rx_ring->dev, bi->dma)) {
1043                                 rx_ring->rx_stats.alloc_rx_buff_failed++;
1044                                 bi->dma = 0;
1045                                 goto no_buffers;
1046                         }
1047                 }
1048
1049                 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
1050                         if (!bi->page) {
1051                                 bi->page = netdev_alloc_page(rx_ring->netdev);
1052                                 if (!bi->page) {
1053                                         rx_ring->rx_stats.alloc_rx_page_failed++;
1054                                         goto no_buffers;
1055                                 }
1056                         }
1057
1058                         if (!bi->page_dma) {
1059                                 /* use a half page if we're re-using */
1060                                 bi->page_offset ^= PAGE_SIZE / 2;
1061                                 bi->page_dma = dma_map_page(rx_ring->dev,
1062                                                             bi->page,
1063                                                             bi->page_offset,
1064                                                             PAGE_SIZE / 2,
1065                                                             DMA_FROM_DEVICE);
1066                                 if (dma_mapping_error(rx_ring->dev,
1067                                                       bi->page_dma)) {
1068                                         rx_ring->rx_stats.alloc_rx_page_failed++;
1069                                         bi->page_dma = 0;
1070                                         goto no_buffers;
1071                                 }
1072                         }
1073
1074                         /* Refresh the desc even if buffer_addrs didn't change
1075                          * because each write-back erases this info. */
1076                         rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
1077                         rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
1078                 } else {
1079                         rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
1080                         rx_desc->read.hdr_addr = 0;
1081                 }
1082
1083                 i++;
1084                 if (i == rx_ring->count)
1085                         i = 0;
1086         }
1087
1088 no_buffers:
1089         if (rx_ring->next_to_use != i) {
1090                 rx_ring->next_to_use = i;
1091                 ixgbe_release_rx_desc(rx_ring, i);
1092         }
1093 }
1094
1095 static inline u16 ixgbe_get_hdr_info(union ixgbe_adv_rx_desc *rx_desc)
1096 {
1097         return rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
1098 }
1099
1100 static inline u16 ixgbe_get_pkt_info(union ixgbe_adv_rx_desc *rx_desc)
1101 {
1102         return rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1103 }
1104
1105 static inline u32 ixgbe_get_rsc_count(union ixgbe_adv_rx_desc *rx_desc)
1106 {
1107         return (le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
1108                 IXGBE_RXDADV_RSCCNT_MASK) >>
1109                 IXGBE_RXDADV_RSCCNT_SHIFT;
1110 }
1111
1112 /**
1113  * ixgbe_transform_rsc_queue - change rsc queue into a full packet
1114  * @skb: pointer to the last skb in the rsc queue
1115  * @count: pointer to number of packets coalesced in this context
1116  *
1117  * This function changes a queue full of hw rsc buffers into a completed
1118  * packet.  It uses the ->prev pointers to find the first packet and then
1119  * turns it into the frag list owner.
1120  **/
1121 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb,
1122                                                         u64 *count)
1123 {
1124         unsigned int frag_list_size = 0;
1125
1126         while (skb->prev) {
1127                 struct sk_buff *prev = skb->prev;
1128                 frag_list_size += skb->len;
1129                 skb->prev = NULL;
1130                 skb = prev;
1131                 *count += 1;
1132         }
1133
1134         skb_shinfo(skb)->frag_list = skb->next;
1135         skb->next = NULL;
1136         skb->len += frag_list_size;
1137         skb->data_len += frag_list_size;
1138         skb->truesize += frag_list_size;
1139         return skb;
1140 }
1141
1142 struct ixgbe_rsc_cb {
1143         dma_addr_t dma;
1144         bool delay_unmap;
1145 };
1146
1147 #define IXGBE_RSC_CB(skb) ((struct ixgbe_rsc_cb *)(skb)->cb)
1148
1149 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1150                                struct ixgbe_ring *rx_ring,
1151                                int *work_done, int work_to_do)
1152 {
1153         struct ixgbe_adapter *adapter = q_vector->adapter;
1154         union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
1155         struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
1156         struct sk_buff *skb;
1157         unsigned int i, rsc_count = 0;
1158         u32 len, staterr;
1159         u16 hdr_info;
1160         bool cleaned = false;
1161         int cleaned_count = 0;
1162         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1163 #ifdef IXGBE_FCOE
1164         int ddp_bytes = 0;
1165 #endif /* IXGBE_FCOE */
1166
1167         i = rx_ring->next_to_clean;
1168         rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1169         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1170         rx_buffer_info = &rx_ring->rx_buffer_info[i];
1171
1172         while (staterr & IXGBE_RXD_STAT_DD) {
1173                 u32 upper_len = 0;
1174                 if (*work_done >= work_to_do)
1175                         break;
1176                 (*work_done)++;
1177
1178                 rmb(); /* read descriptor and rx_buffer_info after status DD */
1179                 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
1180                         hdr_info = le16_to_cpu(ixgbe_get_hdr_info(rx_desc));
1181                         len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
1182                                IXGBE_RXDADV_HDRBUFLEN_SHIFT;
1183                         upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1184                         if ((len > IXGBE_RX_HDR_SIZE) ||
1185                             (upper_len && !(hdr_info & IXGBE_RXDADV_SPH)))
1186                                 len = IXGBE_RX_HDR_SIZE;
1187                 } else {
1188                         len = le16_to_cpu(rx_desc->wb.upper.length);
1189                 }
1190
1191                 cleaned = true;
1192                 skb = rx_buffer_info->skb;
1193                 prefetch(skb->data);
1194                 rx_buffer_info->skb = NULL;
1195
1196                 if (rx_buffer_info->dma) {
1197                         if ((adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
1198                             (!(staterr & IXGBE_RXD_STAT_EOP)) &&
1199                                  (!(skb->prev))) {
1200                                 /*
1201                                  * When HWRSC is enabled, delay unmapping
1202                                  * of the first packet. It carries the
1203                                  * header information, HW may still
1204                                  * access the header after the writeback.
1205                                  * Only unmap it when EOP is reached
1206                                  */
1207                                 IXGBE_RSC_CB(skb)->delay_unmap = true;
1208                                 IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
1209                         } else {
1210                                 dma_unmap_single(rx_ring->dev,
1211                                                  rx_buffer_info->dma,
1212                                                  rx_ring->rx_buf_len,
1213                                                  DMA_FROM_DEVICE);
1214                         }
1215                         rx_buffer_info->dma = 0;
1216                         skb_put(skb, len);
1217                 }
1218
1219                 if (upper_len) {
1220                         dma_unmap_page(rx_ring->dev,
1221                                        rx_buffer_info->page_dma,
1222                                        PAGE_SIZE / 2,
1223                                        DMA_FROM_DEVICE);
1224                         rx_buffer_info->page_dma = 0;
1225                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1226                                            rx_buffer_info->page,
1227                                            rx_buffer_info->page_offset,
1228                                            upper_len);
1229
1230                         if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
1231                             (page_count(rx_buffer_info->page) != 1))
1232                                 rx_buffer_info->page = NULL;
1233                         else
1234                                 get_page(rx_buffer_info->page);
1235
1236                         skb->len += upper_len;
1237                         skb->data_len += upper_len;
1238                         skb->truesize += upper_len;
1239                 }
1240
1241                 i++;
1242                 if (i == rx_ring->count)
1243                         i = 0;
1244
1245                 next_rxd = IXGBE_RX_DESC_ADV(rx_ring, i);
1246                 prefetch(next_rxd);
1247                 cleaned_count++;
1248
1249                 if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
1250                         rsc_count = ixgbe_get_rsc_count(rx_desc);
1251
1252                 if (rsc_count) {
1253                         u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
1254                                      IXGBE_RXDADV_NEXTP_SHIFT;
1255                         next_buffer = &rx_ring->rx_buffer_info[nextp];
1256                 } else {
1257                         next_buffer = &rx_ring->rx_buffer_info[i];
1258                 }
1259
1260                 if (staterr & IXGBE_RXD_STAT_EOP) {
1261                         if (skb->prev)
1262                                 skb = ixgbe_transform_rsc_queue(skb,
1263                                                 &(rx_ring->rx_stats.rsc_count));
1264                         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
1265                                 if (IXGBE_RSC_CB(skb)->delay_unmap) {
1266                                         dma_unmap_single(rx_ring->dev,
1267                                                          IXGBE_RSC_CB(skb)->dma,
1268                                                          rx_ring->rx_buf_len,
1269                                                          DMA_FROM_DEVICE);
1270                                         IXGBE_RSC_CB(skb)->dma = 0;
1271                                         IXGBE_RSC_CB(skb)->delay_unmap = false;
1272                                 }
1273                                 if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED)
1274                                         rx_ring->rx_stats.rsc_count +=
1275                                                  skb_shinfo(skb)->nr_frags;
1276                                 else
1277                                         rx_ring->rx_stats.rsc_count++;
1278                                 rx_ring->rx_stats.rsc_flush++;
1279                         }
1280                         u64_stats_update_begin(&rx_ring->syncp);
1281                         rx_ring->stats.packets++;
1282                         rx_ring->stats.bytes += skb->len;
1283                         u64_stats_update_end(&rx_ring->syncp);
1284                 } else {
1285                         if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
1286                                 rx_buffer_info->skb = next_buffer->skb;
1287                                 rx_buffer_info->dma = next_buffer->dma;
1288                                 next_buffer->skb = skb;
1289                                 next_buffer->dma = 0;
1290                         } else {
1291                                 skb->next = next_buffer->skb;
1292                                 skb->next->prev = skb;
1293                         }
1294                         rx_ring->rx_stats.non_eop_descs++;
1295                         goto next_desc;
1296                 }
1297
1298                 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
1299                         dev_kfree_skb_irq(skb);
1300                         goto next_desc;
1301                 }
1302
1303                 ixgbe_rx_checksum(adapter, rx_desc, skb);
1304
1305                 /* probably a little skewed due to removing CRC */
1306                 total_rx_bytes += skb->len;
1307                 total_rx_packets++;
1308
1309                 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1310 #ifdef IXGBE_FCOE
1311                 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1312                 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
1313                         ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1314                         if (!ddp_bytes)
1315                                 goto next_desc;
1316                 }
1317 #endif /* IXGBE_FCOE */
1318                 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
1319
1320 next_desc:
1321                 rx_desc->wb.upper.status_error = 0;
1322
1323                 /* return some buffers to hardware, one at a time is too slow */
1324                 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1325                         ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1326                         cleaned_count = 0;
1327                 }
1328
1329                 /* use prefetched values */
1330                 rx_desc = next_rxd;
1331                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1332
1333                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1334         }
1335
1336         rx_ring->next_to_clean = i;
1337         cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
1338
1339         if (cleaned_count)
1340                 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1341
1342 #ifdef IXGBE_FCOE
1343         /* include DDPed FCoE data */
1344         if (ddp_bytes > 0) {
1345                 unsigned int mss;
1346
1347                 mss = rx_ring->netdev->mtu - sizeof(struct fcoe_hdr) -
1348                         sizeof(struct fc_frame_header) -
1349                         sizeof(struct fcoe_crc_eof);
1350                 if (mss > 512)
1351                         mss &= ~511;
1352                 total_rx_bytes += ddp_bytes;
1353                 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1354         }
1355 #endif /* IXGBE_FCOE */
1356
1357         rx_ring->total_packets += total_rx_packets;
1358         rx_ring->total_bytes += total_rx_bytes;
1359
1360         return cleaned;
1361 }
1362
1363 static int ixgbe_clean_rxonly(struct napi_struct *, int);
1364 /**
1365  * ixgbe_configure_msix - Configure MSI-X hardware
1366  * @adapter: board private structure
1367  *
1368  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1369  * interrupts.
1370  **/
1371 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1372 {
1373         struct ixgbe_q_vector *q_vector;
1374         int i, j, q_vectors, v_idx, r_idx;
1375         u32 mask;
1376
1377         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1378
1379         /*
1380          * Populate the IVAR table and set the ITR values to the
1381          * corresponding register.
1382          */
1383         for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1384                 q_vector = adapter->q_vector[v_idx];
1385                 /* XXX for_each_set_bit(...) */
1386                 r_idx = find_first_bit(q_vector->rxr_idx,
1387                                        adapter->num_rx_queues);
1388
1389                 for (i = 0; i < q_vector->rxr_count; i++) {
1390                         j = adapter->rx_ring[r_idx]->reg_idx;
1391                         ixgbe_set_ivar(adapter, 0, j, v_idx);
1392                         r_idx = find_next_bit(q_vector->rxr_idx,
1393                                               adapter->num_rx_queues,
1394                                               r_idx + 1);
1395                 }
1396                 r_idx = find_first_bit(q_vector->txr_idx,
1397                                        adapter->num_tx_queues);
1398
1399                 for (i = 0; i < q_vector->txr_count; i++) {
1400                         j = adapter->tx_ring[r_idx]->reg_idx;
1401                         ixgbe_set_ivar(adapter, 1, j, v_idx);
1402                         r_idx = find_next_bit(q_vector->txr_idx,
1403                                               adapter->num_tx_queues,
1404                                               r_idx + 1);
1405                 }
1406
1407                 if (q_vector->txr_count && !q_vector->rxr_count)
1408                         /* tx only */
1409                         q_vector->eitr = adapter->tx_eitr_param;
1410                 else if (q_vector->rxr_count)
1411                         /* rx or mixed */
1412                         q_vector->eitr = adapter->rx_eitr_param;
1413
1414                 ixgbe_write_eitr(q_vector);
1415                 /* If Flow Director is enabled, set interrupt affinity */
1416                 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
1417                     (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
1418                         /*
1419                          * Allocate the affinity_hint cpumask, assign the mask
1420                          * for this vector, and set our affinity_hint for
1421                          * this irq.
1422                          */
1423                         if (!alloc_cpumask_var(&q_vector->affinity_mask,
1424                                                GFP_KERNEL))
1425                                 return;
1426                         cpumask_set_cpu(v_idx, q_vector->affinity_mask);
1427                         irq_set_affinity_hint(adapter->msix_entries[v_idx].vector,
1428                                               q_vector->affinity_mask);
1429                 }
1430         }
1431
1432         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
1433                 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1434                                v_idx);
1435         else if (adapter->hw.mac.type == ixgbe_mac_82599EB)
1436                 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1437         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1438
1439         /* set up to autoclear timer, and the vectors */
1440         mask = IXGBE_EIMS_ENABLE_MASK;
1441         if (adapter->num_vfs)
1442                 mask &= ~(IXGBE_EIMS_OTHER |
1443                           IXGBE_EIMS_MAILBOX |
1444                           IXGBE_EIMS_LSC);
1445         else
1446                 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
1447         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1448 }
1449
1450 enum latency_range {
1451         lowest_latency = 0,
1452         low_latency = 1,
1453         bulk_latency = 2,
1454         latency_invalid = 255
1455 };
1456
1457 /**
1458  * ixgbe_update_itr - update the dynamic ITR value based on statistics
1459  * @adapter: pointer to adapter
1460  * @eitr: eitr setting (ints per sec) to give last timeslice
1461  * @itr_setting: current throttle rate in ints/second
1462  * @packets: the number of packets during this measurement interval
1463  * @bytes: the number of bytes during this measurement interval
1464  *
1465  *      Stores a new ITR value based on packets and byte
1466  *      counts during the last interrupt.  The advantage of per interrupt
1467  *      computation is faster updates and more accurate ITR for the current
1468  *      traffic pattern.  Constants in this function were computed
1469  *      based on theoretical maximum wire speed and thresholds were set based
1470  *      on testing data as well as attempting to minimize response time
1471  *      while increasing bulk throughput.
1472  *      this functionality is controlled by the InterruptThrottleRate module
1473  *      parameter (see ixgbe_param.c)
1474  **/
1475 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
1476                            u32 eitr, u8 itr_setting,
1477                            int packets, int bytes)
1478 {
1479         unsigned int retval = itr_setting;
1480         u32 timepassed_us;
1481         u64 bytes_perint;
1482
1483         if (packets == 0)
1484                 goto update_itr_done;
1485
1486
1487         /* simple throttlerate management
1488          *    0-20MB/s lowest (100000 ints/s)
1489          *   20-100MB/s low   (20000 ints/s)
1490          *  100-1249MB/s bulk (8000 ints/s)
1491          */
1492         /* what was last interrupt timeslice? */
1493         timepassed_us = 1000000/eitr;
1494         bytes_perint = bytes / timepassed_us; /* bytes/usec */
1495
1496         switch (itr_setting) {
1497         case lowest_latency:
1498                 if (bytes_perint > adapter->eitr_low)
1499                         retval = low_latency;
1500                 break;
1501         case low_latency:
1502                 if (bytes_perint > adapter->eitr_high)
1503                         retval = bulk_latency;
1504                 else if (bytes_perint <= adapter->eitr_low)
1505                         retval = lowest_latency;
1506                 break;
1507         case bulk_latency:
1508                 if (bytes_perint <= adapter->eitr_high)
1509                         retval = low_latency;
1510                 break;
1511         }
1512
1513 update_itr_done:
1514         return retval;
1515 }
1516
1517 /**
1518  * ixgbe_write_eitr - write EITR register in hardware specific way
1519  * @q_vector: structure containing interrupt and ring information
1520  *
1521  * This function is made to be called by ethtool and by the driver
1522  * when it needs to update EITR registers at runtime.  Hardware
1523  * specific quirks/differences are taken care of here.
1524  */
1525 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1526 {
1527         struct ixgbe_adapter *adapter = q_vector->adapter;
1528         struct ixgbe_hw *hw = &adapter->hw;
1529         int v_idx = q_vector->v_idx;
1530         u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1531
1532         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1533                 /* must write high and low 16 bits to reset counter */
1534                 itr_reg |= (itr_reg << 16);
1535         } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
1536                 /*
1537                  * 82599 can support a value of zero, so allow it for
1538                  * max interrupt rate, but there is an errata where it can
1539                  * not be zero with RSC
1540                  */
1541                 if (itr_reg == 8 &&
1542                     !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
1543                         itr_reg = 0;
1544
1545                 /*
1546                  * set the WDIS bit to not clear the timer bits and cause an
1547                  * immediate assertion of the interrupt
1548                  */
1549                 itr_reg |= IXGBE_EITR_CNT_WDIS;
1550         }
1551         IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1552 }
1553
1554 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1555 {
1556         struct ixgbe_adapter *adapter = q_vector->adapter;
1557         u32 new_itr;
1558         u8 current_itr, ret_itr;
1559         int i, r_idx;
1560         struct ixgbe_ring *rx_ring, *tx_ring;
1561
1562         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1563         for (i = 0; i < q_vector->txr_count; i++) {
1564                 tx_ring = adapter->tx_ring[r_idx];
1565                 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1566                                            q_vector->tx_itr,
1567                                            tx_ring->total_packets,
1568                                            tx_ring->total_bytes);
1569                 /* if the result for this queue would decrease interrupt
1570                  * rate for this vector then use that result */
1571                 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1572                                     q_vector->tx_itr - 1 : ret_itr);
1573                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1574                                       r_idx + 1);
1575         }
1576
1577         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1578         for (i = 0; i < q_vector->rxr_count; i++) {
1579                 rx_ring = adapter->rx_ring[r_idx];
1580                 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1581                                            q_vector->rx_itr,
1582                                            rx_ring->total_packets,
1583                                            rx_ring->total_bytes);
1584                 /* if the result for this queue would decrease interrupt
1585                  * rate for this vector then use that result */
1586                 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1587                                     q_vector->rx_itr - 1 : ret_itr);
1588                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1589                                       r_idx + 1);
1590         }
1591
1592         current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1593
1594         switch (current_itr) {
1595         /* counts and packets in update_itr are dependent on these numbers */
1596         case lowest_latency:
1597                 new_itr = 100000;
1598                 break;
1599         case low_latency:
1600                 new_itr = 20000; /* aka hwitr = ~200 */
1601                 break;
1602         case bulk_latency:
1603         default:
1604                 new_itr = 8000;
1605                 break;
1606         }
1607
1608         if (new_itr != q_vector->eitr) {
1609                 /* do an exponential smoothing */
1610                 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1611
1612                 /* save the algorithm value here, not the smoothed one */
1613                 q_vector->eitr = new_itr;
1614
1615                 ixgbe_write_eitr(q_vector);
1616         }
1617 }
1618
1619 /**
1620  * ixgbe_check_overtemp_task - worker thread to check over tempurature
1621  * @work: pointer to work_struct containing our data
1622  **/
1623 static void ixgbe_check_overtemp_task(struct work_struct *work)
1624 {
1625         struct ixgbe_adapter *adapter = container_of(work,
1626                                                      struct ixgbe_adapter,
1627                                                      check_overtemp_task);
1628         struct ixgbe_hw *hw = &adapter->hw;
1629         u32 eicr = adapter->interrupt_event;
1630
1631         if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
1632                 return;
1633
1634         switch (hw->device_id) {
1635         case IXGBE_DEV_ID_82599_T3_LOM: {
1636                 u32 autoneg;
1637                 bool link_up = false;
1638
1639                 if (hw->mac.ops.check_link)
1640                         hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
1641
1642                 if (((eicr & IXGBE_EICR_GPI_SDP0) && (!link_up)) ||
1643                     (eicr & IXGBE_EICR_LSC))
1644                         /* Check if this is due to overtemp */
1645                         if (hw->phy.ops.check_overtemp(hw) == IXGBE_ERR_OVERTEMP)
1646                                 break;
1647                 return;
1648         }
1649         default:
1650                 if (!(eicr & IXGBE_EICR_GPI_SDP0))
1651                         return;
1652                 break;
1653         }
1654         e_crit(drv,
1655                "Network adapter has been stopped because it has over heated. "
1656                "Restart the computer. If the problem persists, "
1657                "power off the system and replace the adapter\n");
1658         /* write to clear the interrupt */
1659         IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP0);
1660 }
1661
1662 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1663 {
1664         struct ixgbe_hw *hw = &adapter->hw;
1665
1666         if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1667             (eicr & IXGBE_EICR_GPI_SDP1)) {
1668                 e_crit(probe, "Fan has stopped, replace the adapter\n");
1669                 /* write to clear the interrupt */
1670                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1671         }
1672 }
1673
1674 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1675 {
1676         struct ixgbe_hw *hw = &adapter->hw;
1677
1678         if (eicr & IXGBE_EICR_GPI_SDP1) {
1679                 /* Clear the interrupt */
1680                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1681                 schedule_work(&adapter->multispeed_fiber_task);
1682         } else if (eicr & IXGBE_EICR_GPI_SDP2) {
1683                 /* Clear the interrupt */
1684                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1685                 schedule_work(&adapter->sfp_config_module_task);
1686         } else {
1687                 /* Interrupt isn't for us... */
1688                 return;
1689         }
1690 }
1691
1692 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1693 {
1694         struct ixgbe_hw *hw = &adapter->hw;
1695
1696         adapter->lsc_int++;
1697         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1698         adapter->link_check_timeout = jiffies;
1699         if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1700                 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1701                 IXGBE_WRITE_FLUSH(hw);
1702                 schedule_work(&adapter->watchdog_task);
1703         }
1704 }
1705
1706 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1707 {
1708         struct net_device *netdev = data;
1709         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1710         struct ixgbe_hw *hw = &adapter->hw;
1711         u32 eicr;
1712
1713         /*
1714          * Workaround for Silicon errata.  Use clear-by-write instead
1715          * of clear-by-read.  Reading with EICS will return the
1716          * interrupt causes without clearing, which later be done
1717          * with the write to EICR.
1718          */
1719         eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1720         IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1721
1722         if (eicr & IXGBE_EICR_LSC)
1723                 ixgbe_check_lsc(adapter);
1724
1725         if (eicr & IXGBE_EICR_MAILBOX)
1726                 ixgbe_msg_task(adapter);
1727
1728         if (hw->mac.type == ixgbe_mac_82598EB)
1729                 ixgbe_check_fan_failure(adapter, eicr);
1730
1731         if (hw->mac.type == ixgbe_mac_82599EB) {
1732                 ixgbe_check_sfp_event(adapter, eicr);
1733                 adapter->interrupt_event = eicr;
1734                 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
1735                     ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)))
1736                         schedule_work(&adapter->check_overtemp_task);
1737
1738                 /* Handle Flow Director Full threshold interrupt */
1739                 if (eicr & IXGBE_EICR_FLOW_DIR) {
1740                         int i;
1741                         IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1742                         /* Disable transmits before FDIR Re-initialization */
1743                         netif_tx_stop_all_queues(netdev);
1744                         for (i = 0; i < adapter->num_tx_queues; i++) {
1745                                 struct ixgbe_ring *tx_ring =
1746                                                             adapter->tx_ring[i];
1747                                 if (test_and_clear_bit(__IXGBE_FDIR_INIT_DONE,
1748                                                        &tx_ring->reinit_state))
1749                                         schedule_work(&adapter->fdir_reinit_task);
1750                         }
1751                 }
1752         }
1753         if (!test_bit(__IXGBE_DOWN, &adapter->state))
1754                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1755
1756         return IRQ_HANDLED;
1757 }
1758
1759 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1760                                            u64 qmask)
1761 {
1762         u32 mask;
1763
1764         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1765                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1766                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1767         } else {
1768                 mask = (qmask & 0xFFFFFFFF);
1769                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(0), mask);
1770                 mask = (qmask >> 32);
1771                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS_EX(1), mask);
1772         }
1773         /* skip the flush */
1774 }
1775
1776 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1777                                             u64 qmask)
1778 {
1779         u32 mask;
1780
1781         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
1782                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1783                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, mask);
1784         } else {
1785                 mask = (qmask & 0xFFFFFFFF);
1786                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), mask);
1787                 mask = (qmask >> 32);
1788                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), mask);
1789         }
1790         /* skip the flush */
1791 }
1792
1793 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1794 {
1795         struct ixgbe_q_vector *q_vector = data;
1796         struct ixgbe_adapter  *adapter = q_vector->adapter;
1797         struct ixgbe_ring     *tx_ring;
1798         int i, r_idx;
1799
1800         if (!q_vector->txr_count)
1801                 return IRQ_HANDLED;
1802
1803         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1804         for (i = 0; i < q_vector->txr_count; i++) {
1805                 tx_ring = adapter->tx_ring[r_idx];
1806                 tx_ring->total_bytes = 0;
1807                 tx_ring->total_packets = 0;
1808                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1809                                       r_idx + 1);
1810         }
1811
1812         /* EIAM disabled interrupts (on this vector) for us */
1813         napi_schedule(&q_vector->napi);
1814
1815         return IRQ_HANDLED;
1816 }
1817
1818 /**
1819  * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
1820  * @irq: unused
1821  * @data: pointer to our q_vector struct for this interrupt vector
1822  **/
1823 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
1824 {
1825         struct ixgbe_q_vector *q_vector = data;
1826         struct ixgbe_adapter  *adapter = q_vector->adapter;
1827         struct ixgbe_ring  *rx_ring;
1828         int r_idx;
1829         int i;
1830
1831         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1832         for (i = 0;  i < q_vector->rxr_count; i++) {
1833                 rx_ring = adapter->rx_ring[r_idx];
1834                 rx_ring->total_bytes = 0;
1835                 rx_ring->total_packets = 0;
1836                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1837                                       r_idx + 1);
1838         }
1839
1840         if (!q_vector->rxr_count)
1841                 return IRQ_HANDLED;
1842
1843         /* disable interrupts on this vector only */
1844         /* EIAM disabled interrupts (on this vector) for us */
1845         napi_schedule(&q_vector->napi);
1846
1847         return IRQ_HANDLED;
1848 }
1849
1850 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1851 {
1852         struct ixgbe_q_vector *q_vector = data;
1853         struct ixgbe_adapter  *adapter = q_vector->adapter;
1854         struct ixgbe_ring  *ring;
1855         int r_idx;
1856         int i;
1857
1858         if (!q_vector->txr_count && !q_vector->rxr_count)
1859                 return IRQ_HANDLED;
1860
1861         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1862         for (i = 0; i < q_vector->txr_count; i++) {
1863                 ring = adapter->tx_ring[r_idx];
1864                 ring->total_bytes = 0;
1865                 ring->total_packets = 0;
1866                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1867                                       r_idx + 1);
1868         }
1869
1870         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1871         for (i = 0; i < q_vector->rxr_count; i++) {
1872                 ring = adapter->rx_ring[r_idx];
1873                 ring->total_bytes = 0;
1874                 ring->total_packets = 0;
1875                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1876                                       r_idx + 1);
1877         }
1878
1879         /* EIAM disabled interrupts (on this vector) for us */
1880         napi_schedule(&q_vector->napi);
1881
1882         return IRQ_HANDLED;
1883 }
1884
1885 /**
1886  * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1887  * @napi: napi struct with our devices info in it
1888  * @budget: amount of work driver is allowed to do this pass, in packets
1889  *
1890  * This function is optimized for cleaning one queue only on a single
1891  * q_vector!!!
1892  **/
1893 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1894 {
1895         struct ixgbe_q_vector *q_vector =
1896                                container_of(napi, struct ixgbe_q_vector, napi);
1897         struct ixgbe_adapter *adapter = q_vector->adapter;
1898         struct ixgbe_ring *rx_ring = NULL;
1899         int work_done = 0;
1900         long r_idx;
1901
1902         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1903         rx_ring = adapter->rx_ring[r_idx];
1904 #ifdef CONFIG_IXGBE_DCA
1905         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1906                 ixgbe_update_rx_dca(adapter, rx_ring);
1907 #endif
1908
1909         ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
1910
1911         /* If all Rx work done, exit the polling mode */
1912         if (work_done < budget) {
1913                 napi_complete(napi);
1914                 if (adapter->rx_itr_setting & 1)
1915                         ixgbe_set_itr_msix(q_vector);
1916                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1917                         ixgbe_irq_enable_queues(adapter,
1918                                                 ((u64)1 << q_vector->v_idx));
1919         }
1920
1921         return work_done;
1922 }
1923
1924 /**
1925  * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
1926  * @napi: napi struct with our devices info in it
1927  * @budget: amount of work driver is allowed to do this pass, in packets
1928  *
1929  * This function will clean more than one rx queue associated with a
1930  * q_vector.
1931  **/
1932 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
1933 {
1934         struct ixgbe_q_vector *q_vector =
1935                                container_of(napi, struct ixgbe_q_vector, napi);
1936         struct ixgbe_adapter *adapter = q_vector->adapter;
1937         struct ixgbe_ring *ring = NULL;
1938         int work_done = 0, i;
1939         long r_idx;
1940         bool tx_clean_complete = true;
1941
1942         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1943         for (i = 0; i < q_vector->txr_count; i++) {
1944                 ring = adapter->tx_ring[r_idx];
1945 #ifdef CONFIG_IXGBE_DCA
1946                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1947                         ixgbe_update_tx_dca(adapter, ring);
1948 #endif
1949                 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
1950                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1951                                       r_idx + 1);
1952         }
1953
1954         /* attempt to distribute budget to each queue fairly, but don't allow
1955          * the budget to go below 1 because we'll exit polling */
1956         budget /= (q_vector->rxr_count ?: 1);
1957         budget = max(budget, 1);
1958         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1959         for (i = 0; i < q_vector->rxr_count; i++) {
1960                 ring = adapter->rx_ring[r_idx];
1961 #ifdef CONFIG_IXGBE_DCA
1962                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1963                         ixgbe_update_rx_dca(adapter, ring);
1964 #endif
1965                 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
1966                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1967                                       r_idx + 1);
1968         }
1969
1970         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1971         ring = adapter->rx_ring[r_idx];
1972         /* If all Rx work done, exit the polling mode */
1973         if (work_done < budget) {
1974                 napi_complete(napi);
1975                 if (adapter->rx_itr_setting & 1)
1976                         ixgbe_set_itr_msix(q_vector);
1977                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1978                         ixgbe_irq_enable_queues(adapter,
1979                                                 ((u64)1 << q_vector->v_idx));
1980                 return 0;
1981         }
1982
1983         return work_done;
1984 }
1985
1986 /**
1987  * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
1988  * @napi: napi struct with our devices info in it
1989  * @budget: amount of work driver is allowed to do this pass, in packets
1990  *
1991  * This function is optimized for cleaning one queue only on a single
1992  * q_vector!!!
1993  **/
1994 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
1995 {
1996         struct ixgbe_q_vector *q_vector =
1997                                container_of(napi, struct ixgbe_q_vector, napi);
1998         struct ixgbe_adapter *adapter = q_vector->adapter;
1999         struct ixgbe_ring *tx_ring = NULL;
2000         int work_done = 0;
2001         long r_idx;
2002
2003         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2004         tx_ring = adapter->tx_ring[r_idx];
2005 #ifdef CONFIG_IXGBE_DCA
2006         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2007                 ixgbe_update_tx_dca(adapter, tx_ring);
2008 #endif
2009
2010         if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
2011                 work_done = budget;
2012
2013         /* If all Tx work done, exit the polling mode */
2014         if (work_done < budget) {
2015                 napi_complete(napi);
2016                 if (adapter->tx_itr_setting & 1)
2017                         ixgbe_set_itr_msix(q_vector);
2018                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2019                         ixgbe_irq_enable_queues(adapter,
2020                                                 ((u64)1 << q_vector->v_idx));
2021         }
2022
2023         return work_done;
2024 }
2025
2026 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
2027                                      int r_idx)
2028 {
2029         struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2030
2031         set_bit(r_idx, q_vector->rxr_idx);
2032         q_vector->rxr_count++;
2033 }
2034
2035 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
2036                                      int t_idx)
2037 {
2038         struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2039
2040         set_bit(t_idx, q_vector->txr_idx);
2041         q_vector->txr_count++;
2042 }
2043
2044 /**
2045  * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
2046  * @adapter: board private structure to initialize
2047  * @vectors: allotted vector count for descriptor rings
2048  *
2049  * This function maps descriptor rings to the queue-specific vectors
2050  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
2051  * one vector per ring/queue, but on a constrained vector budget, we
2052  * group the rings as "efficiently" as possible.  You would add new
2053  * mapping configurations in here.
2054  **/
2055 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
2056                                       int vectors)
2057 {
2058         int v_start = 0;
2059         int rxr_idx = 0, txr_idx = 0;
2060         int rxr_remaining = adapter->num_rx_queues;
2061         int txr_remaining = adapter->num_tx_queues;
2062         int i, j;
2063         int rqpv, tqpv;
2064         int err = 0;
2065
2066         /* No mapping required if MSI-X is disabled. */
2067         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2068                 goto out;
2069
2070         /*
2071          * The ideal configuration...
2072          * We have enough vectors to map one per queue.
2073          */
2074         if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
2075                 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
2076                         map_vector_to_rxq(adapter, v_start, rxr_idx);
2077
2078                 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
2079                         map_vector_to_txq(adapter, v_start, txr_idx);
2080
2081                 goto out;
2082         }
2083
2084         /*
2085          * If we don't have enough vectors for a 1-to-1
2086          * mapping, we'll have to group them so there are
2087          * multiple queues per vector.
2088          */
2089         /* Re-adjusting *qpv takes care of the remainder. */
2090         for (i = v_start; i < vectors; i++) {
2091                 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
2092                 for (j = 0; j < rqpv; j++) {
2093                         map_vector_to_rxq(adapter, i, rxr_idx);
2094                         rxr_idx++;
2095                         rxr_remaining--;
2096                 }
2097         }
2098         for (i = v_start; i < vectors; i++) {
2099                 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
2100                 for (j = 0; j < tqpv; j++) {
2101                         map_vector_to_txq(adapter, i, txr_idx);
2102                         txr_idx++;
2103                         txr_remaining--;
2104                 }
2105         }
2106
2107 out:
2108         return err;
2109 }
2110
2111 /**
2112  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2113  * @adapter: board private structure
2114  *
2115  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2116  * interrupts from the kernel.
2117  **/
2118 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2119 {
2120         struct net_device *netdev = adapter->netdev;
2121         irqreturn_t (*handler)(int, void *);
2122         int i, vector, q_vectors, err;
2123         int ri = 0, ti = 0;
2124
2125         /* Decrement for Other and TCP Timer vectors */
2126         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2127
2128         /* Map the Tx/Rx rings to the vectors we were allotted. */
2129         err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
2130         if (err)
2131                 goto out;
2132
2133 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
2134                          (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
2135                          &ixgbe_msix_clean_many)
2136         for (vector = 0; vector < q_vectors; vector++) {
2137                 handler = SET_HANDLER(adapter->q_vector[vector]);
2138
2139                 if (handler == &ixgbe_msix_clean_rx) {
2140                         sprintf(adapter->name[vector], "%s-%s-%d",
2141                                 netdev->name, "rx", ri++);
2142                 } else if (handler == &ixgbe_msix_clean_tx) {
2143                         sprintf(adapter->name[vector], "%s-%s-%d",
2144                                 netdev->name, "tx", ti++);
2145                 } else
2146                         sprintf(adapter->name[vector], "%s-%s-%d",
2147                                 netdev->name, "TxRx", vector);
2148
2149                 err = request_irq(adapter->msix_entries[vector].vector,
2150                                   handler, 0, adapter->name[vector],
2151                                   adapter->q_vector[vector]);
2152                 if (err) {
2153                         e_err(probe, "request_irq failed for MSIX interrupt "
2154                               "Error: %d\n", err);
2155                         goto free_queue_irqs;
2156                 }
2157         }
2158
2159         sprintf(adapter->name[vector], "%s:lsc", netdev->name);
2160         err = request_irq(adapter->msix_entries[vector].vector,
2161                           ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
2162         if (err) {
2163                 e_err(probe, "request_irq for msix_lsc failed: %d\n", err);
2164                 goto free_queue_irqs;
2165         }
2166
2167         return 0;
2168
2169 free_queue_irqs:
2170         for (i = vector - 1; i >= 0; i--)
2171                 free_irq(adapter->msix_entries[--vector].vector,
2172                          adapter->q_vector[i]);
2173         adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2174         pci_disable_msix(adapter->pdev);
2175         kfree(adapter->msix_entries);
2176         adapter->msix_entries = NULL;
2177 out:
2178         return err;
2179 }
2180
2181 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
2182 {
2183         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2184         u8 current_itr;
2185         u32 new_itr = q_vector->eitr;
2186         struct ixgbe_ring *rx_ring = adapter->rx_ring[0];
2187         struct ixgbe_ring *tx_ring = adapter->tx_ring[0];
2188
2189         q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
2190                                             q_vector->tx_itr,
2191                                             tx_ring->total_packets,
2192                                             tx_ring->total_bytes);
2193         q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
2194                                             q_vector->rx_itr,
2195                                             rx_ring->total_packets,
2196                                             rx_ring->total_bytes);
2197
2198         current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
2199
2200         switch (current_itr) {
2201         /* counts and packets in update_itr are dependent on these numbers */
2202         case lowest_latency:
2203                 new_itr = 100000;
2204                 break;
2205         case low_latency:
2206                 new_itr = 20000; /* aka hwitr = ~200 */
2207                 break;
2208         case bulk_latency:
2209                 new_itr = 8000;
2210                 break;
2211         default:
2212                 break;
2213         }
2214
2215         if (new_itr != q_vector->eitr) {
2216                 /* do an exponential smoothing */
2217                 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
2218
2219                 /* save the algorithm value here, not the smoothed one */
2220                 q_vector->eitr = new_itr;
2221
2222                 ixgbe_write_eitr(q_vector);
2223         }
2224 }
2225
2226 /**
2227  * ixgbe_irq_enable - Enable default interrupt generation settings
2228  * @adapter: board private structure
2229  **/
2230 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2231                                     bool flush)
2232 {
2233         u32 mask;
2234
2235         mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2236         if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2237                 mask |= IXGBE_EIMS_GPI_SDP0;
2238         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2239                 mask |= IXGBE_EIMS_GPI_SDP1;
2240         if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
2241                 mask |= IXGBE_EIMS_ECC;
2242                 mask |= IXGBE_EIMS_GPI_SDP1;
2243                 mask |= IXGBE_EIMS_GPI_SDP2;
2244                 if (adapter->num_vfs)
2245                         mask |= IXGBE_EIMS_MAILBOX;
2246         }
2247         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
2248             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
2249                 mask |= IXGBE_EIMS_FLOW_DIR;
2250
2251         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2252         if (queues)
2253                 ixgbe_irq_enable_queues(adapter, ~0);
2254         if (flush)
2255                 IXGBE_WRITE_FLUSH(&adapter->hw);
2256
2257         if (adapter->num_vfs > 32) {
2258                 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2259                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2260         }
2261 }
2262
2263 /**
2264  * ixgbe_intr - legacy mode Interrupt Handler
2265  * @irq: interrupt number
2266  * @data: pointer to a network interface device structure
2267  **/
2268 static irqreturn_t ixgbe_intr(int irq, void *data)
2269 {
2270         struct net_device *netdev = data;
2271         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2272         struct ixgbe_hw *hw = &adapter->hw;
2273         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2274         u32 eicr;
2275
2276         /*
2277          * Workaround for silicon errata on 82598.  Mask the interrupts
2278          * before the read of EICR.
2279          */
2280         IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2281
2282         /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2283          * therefore no explict interrupt disable is necessary */
2284         eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2285         if (!eicr) {
2286                 /*
2287                  * shared interrupt alert!
2288                  * make sure interrupts are enabled because the read will
2289                  * have disabled interrupts due to EIAM
2290                  * finish the workaround of silicon errata on 82598.  Unmask
2291                  * the interrupt that we masked before the EICR read.
2292                  */
2293                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2294                         ixgbe_irq_enable(adapter, true, true);
2295                 return IRQ_NONE;        /* Not our interrupt */
2296         }
2297
2298         if (eicr & IXGBE_EICR_LSC)
2299                 ixgbe_check_lsc(adapter);
2300
2301         if (hw->mac.type == ixgbe_mac_82599EB)
2302                 ixgbe_check_sfp_event(adapter, eicr);
2303
2304         ixgbe_check_fan_failure(adapter, eicr);
2305         if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2306             ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC)))
2307                 schedule_work(&adapter->check_overtemp_task);
2308
2309         if (napi_schedule_prep(&(q_vector->napi))) {
2310                 adapter->tx_ring[0]->total_packets = 0;
2311                 adapter->tx_ring[0]->total_bytes = 0;
2312                 adapter->rx_ring[0]->total_packets = 0;
2313                 adapter->rx_ring[0]->total_bytes = 0;
2314                 /* would disable interrupts here but EIAM disabled it */
2315                 __napi_schedule(&(q_vector->napi));
2316         }
2317
2318         /*
2319          * re-enable link(maybe) and non-queue interrupts, no flush.
2320          * ixgbe_poll will re-enable the queue interrupts
2321          */
2322
2323         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2324                 ixgbe_irq_enable(adapter, false, false);
2325
2326         return IRQ_HANDLED;
2327 }
2328
2329 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
2330 {
2331         int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2332
2333         for (i = 0; i < q_vectors; i++) {
2334                 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
2335                 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
2336                 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
2337                 q_vector->rxr_count = 0;
2338                 q_vector->txr_count = 0;
2339         }
2340 }
2341
2342 /**
2343  * ixgbe_request_irq - initialize interrupts
2344  * @adapter: board private structure
2345  *
2346  * Attempts to configure interrupts using the best available
2347  * capabilities of the hardware and kernel.
2348  **/
2349 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2350 {
2351         struct net_device *netdev = adapter->netdev;
2352         int err;
2353
2354         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2355                 err = ixgbe_request_msix_irqs(adapter);
2356         } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
2357                 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2358                                   netdev->name, netdev);
2359         } else {
2360                 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2361                                   netdev->name, netdev);
2362         }
2363
2364         if (err)
2365                 e_err(probe, "request_irq failed, Error %d\n", err);
2366
2367         return err;
2368 }
2369
2370 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2371 {
2372         struct net_device *netdev = adapter->netdev;
2373
2374         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2375                 int i, q_vectors;
2376
2377                 q_vectors = adapter->num_msix_vectors;
2378
2379                 i = q_vectors - 1;
2380                 free_irq(adapter->msix_entries[i].vector, netdev);
2381
2382                 i--;
2383                 for (; i >= 0; i--) {
2384                         free_irq(adapter->msix_entries[i].vector,
2385                                  adapter->q_vector[i]);
2386                 }
2387
2388                 ixgbe_reset_q_vectors(adapter);
2389         } else {
2390                 free_irq(adapter->pdev->irq, netdev);
2391         }
2392 }
2393
2394 /**
2395  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2396  * @adapter: board private structure
2397  **/
2398 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2399 {
2400         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2401                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2402         } else {
2403                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2404                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2405                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2406                 if (adapter->num_vfs > 32)
2407                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
2408         }
2409         IXGBE_WRITE_FLUSH(&adapter->hw);
2410         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2411                 int i;
2412                 for (i = 0; i < adapter->num_msix_vectors; i++)
2413                         synchronize_irq(adapter->msix_entries[i].vector);
2414         } else {
2415                 synchronize_irq(adapter->pdev->irq);
2416         }
2417 }
2418
2419 /**
2420  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2421  *
2422  **/
2423 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2424 {
2425         struct ixgbe_hw *hw = &adapter->hw;
2426
2427         IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
2428                         EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
2429
2430         ixgbe_set_ivar(adapter, 0, 0, 0);
2431         ixgbe_set_ivar(adapter, 1, 0, 0);
2432
2433         map_vector_to_rxq(adapter, 0, 0);
2434         map_vector_to_txq(adapter, 0, 0);
2435
2436         e_info(hw, "Legacy interrupt IVAR setup done\n");
2437 }
2438
2439 /**
2440  * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2441  * @adapter: board private structure
2442  * @ring: structure containing ring specific data
2443  *
2444  * Configure the Tx descriptor ring after a reset.
2445  **/
2446 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2447                              struct ixgbe_ring *ring)
2448 {
2449         struct ixgbe_hw *hw = &adapter->hw;
2450         u64 tdba = ring->dma;
2451         int wait_loop = 10;
2452         u32 txdctl;
2453         u16 reg_idx = ring->reg_idx;
2454
2455         /* disable queue to avoid issues while updating state */
2456         txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2457         IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx),
2458                         txdctl & ~IXGBE_TXDCTL_ENABLE);
2459         IXGBE_WRITE_FLUSH(hw);
2460
2461         IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2462                         (tdba & DMA_BIT_MASK(32)));
2463         IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2464         IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2465                         ring->count * sizeof(union ixgbe_adv_tx_desc));
2466         IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2467         IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2468         ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2469
2470         /* configure fetching thresholds */
2471         if (adapter->rx_itr_setting == 0) {
2472                 /* cannot set wthresh when itr==0 */
2473                 txdctl &= ~0x007F0000;
2474         } else {
2475                 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2476                 txdctl |= (8 << 16);
2477         }
2478         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2479                 /* PThresh workaround for Tx hang with DFP enabled. */
2480                 txdctl |= 32;
2481         }
2482
2483         /* reinitialize flowdirector state */
2484         set_bit(__IXGBE_FDIR_INIT_DONE, &ring->reinit_state);
2485
2486         /* enable queue */
2487         txdctl |= IXGBE_TXDCTL_ENABLE;
2488         IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2489
2490         /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2491         if (hw->mac.type == ixgbe_mac_82598EB &&
2492             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2493                 return;
2494
2495         /* poll to verify queue is enabled */
2496         do {
2497                 msleep(1);
2498                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2499         } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2500         if (!wait_loop)
2501                 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2502 }
2503
2504 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2505 {
2506         struct ixgbe_hw *hw = &adapter->hw;
2507         u32 rttdcs;
2508         u32 mask;
2509
2510         if (hw->mac.type == ixgbe_mac_82598EB)
2511                 return;
2512
2513         /* disable the arbiter while setting MTQC */
2514         rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2515         rttdcs |= IXGBE_RTTDCS_ARBDIS;
2516         IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2517
2518         /* set transmit pool layout */
2519         mask = (IXGBE_FLAG_SRIOV_ENABLED | IXGBE_FLAG_DCB_ENABLED);
2520         switch (adapter->flags & mask) {
2521
2522         case (IXGBE_FLAG_SRIOV_ENABLED):
2523                 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2524                                 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2525                 break;
2526
2527         case (IXGBE_FLAG_DCB_ENABLED):
2528                 /* We enable 8 traffic classes, DCB only */
2529                 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2530                               (IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ));
2531                 break;
2532
2533         default:
2534                 IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
2535                 break;
2536         }
2537
2538         /* re-enable the arbiter */
2539         rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2540         IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2541 }
2542
2543 /**
2544  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2545  * @adapter: board private structure
2546  *
2547  * Configure the Tx unit of the MAC after a reset.
2548  **/
2549 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2550 {
2551         struct ixgbe_hw *hw = &adapter->hw;
2552         u32 dmatxctl;
2553         u32 i;
2554
2555         ixgbe_setup_mtqc(adapter);
2556
2557         if (hw->mac.type != ixgbe_mac_82598EB) {
2558                 /* DMATXCTL.EN must be before Tx queues are enabled */
2559                 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2560                 dmatxctl |= IXGBE_DMATXCTL_TE;
2561                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2562         }
2563
2564         /* Setup the HW Tx Head and Tail descriptor pointers */
2565         for (i = 0; i < adapter->num_tx_queues; i++)
2566                 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2567 }
2568
2569 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2570
2571 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2572                                    struct ixgbe_ring *rx_ring)
2573 {
2574         u32 srrctl;
2575         int index;
2576         struct ixgbe_ring_feature *feature = adapter->ring_feature;
2577
2578         index = rx_ring->reg_idx;
2579         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
2580                 unsigned long mask;
2581                 mask = (unsigned long) feature[RING_F_RSS].mask;
2582                 index = index & mask;
2583         }
2584         srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
2585
2586         srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2587         srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2588         if (adapter->num_vfs)
2589                 srrctl |= IXGBE_SRRCTL_DROP_EN;
2590
2591         srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2592                   IXGBE_SRRCTL_BSIZEHDR_MASK;
2593
2594         if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
2595 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2596                 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2597 #else
2598                 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2599 #endif
2600                 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2601         } else {
2602                 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2603                           IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2604                 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2605         }
2606
2607         IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
2608 }
2609
2610 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2611 {
2612         struct ixgbe_hw *hw = &adapter->hw;
2613         static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2614                           0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2615                           0x6A3E67EA, 0x14364D17, 0x3BED200D};
2616         u32 mrqc = 0, reta = 0;
2617         u32 rxcsum;
2618         int i, j;
2619         int mask;
2620
2621         /* Fill out hash function seeds */
2622         for (i = 0; i < 10; i++)
2623                 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2624
2625         /* Fill out redirection table */
2626         for (i = 0, j = 0; i < 128; i++, j++) {
2627                 if (j == adapter->ring_feature[RING_F_RSS].indices)
2628                         j = 0;
2629                 /* reta = 4-byte sliding window of
2630                  * 0x00..(indices-1)(indices-1)00..etc. */
2631                 reta = (reta << 8) | (j * 0x11);
2632                 if ((i & 3) == 3)
2633                         IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2634         }
2635
2636         /* Disable indicating checksum in descriptor, enables RSS hash */
2637         rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2638         rxcsum |= IXGBE_RXCSUM_PCSD;
2639         IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2640
2641         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
2642                 mask = adapter->flags & IXGBE_FLAG_RSS_ENABLED;
2643         else
2644                 mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2645 #ifdef CONFIG_IXGBE_DCB
2646                                          | IXGBE_FLAG_DCB_ENABLED
2647 #endif
2648                                          | IXGBE_FLAG_SRIOV_ENABLED
2649                                         );
2650
2651         switch (mask) {
2652         case (IXGBE_FLAG_RSS_ENABLED):
2653                 mrqc = IXGBE_MRQC_RSSEN;
2654                 break;
2655         case (IXGBE_FLAG_SRIOV_ENABLED):
2656                 mrqc = IXGBE_MRQC_VMDQEN;
2657                 break;
2658 #ifdef CONFIG_IXGBE_DCB
2659         case (IXGBE_FLAG_DCB_ENABLED):
2660                 mrqc = IXGBE_MRQC_RT8TCEN;
2661                 break;
2662 #endif /* CONFIG_IXGBE_DCB */
2663         default:
2664                 break;
2665         }
2666
2667         /* Perform hash on these packet types */
2668         mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2669               | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2670               | IXGBE_MRQC_RSS_FIELD_IPV6
2671               | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2672
2673         IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2674 }
2675
2676 /**
2677  * ixgbe_configure_rscctl - enable RSC for the indicated ring
2678  * @adapter:    address of board private structure
2679  * @index:      index of ring to set
2680  **/
2681 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2682                                    struct ixgbe_ring *ring)
2683 {
2684         struct ixgbe_hw *hw = &adapter->hw;
2685         u32 rscctrl;
2686         int rx_buf_len;
2687         u16 reg_idx = ring->reg_idx;
2688
2689         if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
2690                 return;
2691
2692         rx_buf_len = ring->rx_buf_len;
2693         rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2694         rscctrl |= IXGBE_RSCCTL_RSCEN;
2695         /*
2696          * we must limit the number of descriptors so that the
2697          * total size of max desc * buf_len is not greater
2698          * than 65535
2699          */
2700         if (ring->flags & IXGBE_RING_RX_PS_ENABLED) {
2701 #if (MAX_SKB_FRAGS > 16)
2702                 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2703 #elif (MAX_SKB_FRAGS > 8)
2704                 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2705 #elif (MAX_SKB_FRAGS > 4)
2706                 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2707 #else
2708                 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2709 #endif
2710         } else {
2711                 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2712                         rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2713                 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2714                         rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2715                 else
2716                         rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2717         }
2718         IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2719 }
2720
2721 /**
2722  *  ixgbe_set_uta - Set unicast filter table address
2723  *  @adapter: board private structure
2724  *
2725  *  The unicast table address is a register array of 32-bit registers.
2726  *  The table is meant to be used in a way similar to how the MTA is used
2727  *  however due to certain limitations in the hardware it is necessary to
2728  *  set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
2729  *  enable bit to allow vlan tag stripping when promiscuous mode is enabled
2730  **/
2731 static void ixgbe_set_uta(struct ixgbe_adapter *adapter)
2732 {
2733         struct ixgbe_hw *hw = &adapter->hw;
2734         int i;
2735
2736         /* The UTA table only exists on 82599 hardware and newer */
2737         if (hw->mac.type < ixgbe_mac_82599EB)
2738                 return;
2739
2740         /* we only need to do this if VMDq is enabled */
2741         if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2742                 return;
2743
2744         for (i = 0; i < 128; i++)
2745                 IXGBE_WRITE_REG(hw, IXGBE_UTA(i), ~0);
2746 }
2747
2748 #define IXGBE_MAX_RX_DESC_POLL 10
2749 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
2750                                        struct ixgbe_ring *ring)
2751 {
2752         struct ixgbe_hw *hw = &adapter->hw;
2753         int reg_idx = ring->reg_idx;
2754         int wait_loop = IXGBE_MAX_RX_DESC_POLL;
2755         u32 rxdctl;
2756
2757         /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2758         if (hw->mac.type == ixgbe_mac_82598EB &&
2759             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2760                 return;
2761
2762         do {
2763                 msleep(1);
2764                 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2765         } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
2766
2767         if (!wait_loop) {
2768                 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
2769                       "the polling period\n", reg_idx);
2770         }
2771 }
2772
2773 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
2774                              struct ixgbe_ring *ring)
2775 {
2776         struct ixgbe_hw *hw = &adapter->hw;
2777         u64 rdba = ring->dma;
2778         u32 rxdctl;
2779         u16 reg_idx = ring->reg_idx;
2780
2781         /* disable queue to avoid issues while updating state */
2782         rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
2783         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx),
2784                         rxdctl & ~IXGBE_RXDCTL_ENABLE);
2785         IXGBE_WRITE_FLUSH(hw);
2786
2787         IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
2788         IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
2789         IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
2790                         ring->count * sizeof(union ixgbe_adv_rx_desc));
2791         IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
2792         IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
2793         ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
2794
2795         ixgbe_configure_srrctl(adapter, ring);
2796         ixgbe_configure_rscctl(adapter, ring);
2797
2798         if (hw->mac.type == ixgbe_mac_82598EB) {
2799                 /*
2800                  * enable cache line friendly hardware writes:
2801                  * PTHRESH=32 descriptors (half the internal cache),
2802                  * this also removes ugly rx_no_buffer_count increment
2803                  * HTHRESH=4 descriptors (to minimize latency on fetch)
2804                  * WTHRESH=8 burst writeback up to two cache lines
2805                  */
2806                 rxdctl &= ~0x3FFFFF;
2807                 rxdctl |=  0x080420;
2808         }
2809
2810         /* enable receive descriptor ring */
2811         rxdctl |= IXGBE_RXDCTL_ENABLE;
2812         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
2813
2814         ixgbe_rx_desc_queue_enable(adapter, ring);
2815         ixgbe_alloc_rx_buffers(ring, IXGBE_DESC_UNUSED(ring));
2816 }
2817
2818 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
2819 {
2820         struct ixgbe_hw *hw = &adapter->hw;
2821         int p;
2822
2823         /* PSRTYPE must be initialized in non 82598 adapters */
2824         u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
2825                       IXGBE_PSRTYPE_UDPHDR |
2826                       IXGBE_PSRTYPE_IPV4HDR |
2827                       IXGBE_PSRTYPE_L2HDR |
2828                       IXGBE_PSRTYPE_IPV6HDR;
2829
2830         if (hw->mac.type == ixgbe_mac_82598EB)
2831                 return;
2832
2833         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
2834                 psrtype |= (adapter->num_rx_queues_per_pool << 29);
2835
2836         for (p = 0; p < adapter->num_rx_pools; p++)
2837                 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
2838                                 psrtype);
2839 }
2840
2841 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
2842 {
2843         struct ixgbe_hw *hw = &adapter->hw;
2844         u32 gcr_ext;
2845         u32 vt_reg_bits;
2846         u32 reg_offset, vf_shift;
2847         u32 vmdctl;
2848
2849         if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2850                 return;
2851
2852         vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
2853         vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN | IXGBE_VT_CTL_REPLEN;
2854         vt_reg_bits |= (adapter->num_vfs << IXGBE_VT_CTL_POOL_SHIFT);
2855         IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
2856
2857         vf_shift = adapter->num_vfs % 32;
2858         reg_offset = (adapter->num_vfs > 32) ? 1 : 0;
2859
2860         /* Enable only the PF's pool for Tx/Rx */
2861         IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
2862         IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), 0);
2863         IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
2864         IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), 0);
2865         IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2866
2867         /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
2868         hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
2869
2870         /*
2871          * Set up VF register offsets for selected VT Mode,
2872          * i.e. 32 or 64 VFs for SR-IOV
2873          */
2874         gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
2875         gcr_ext |= IXGBE_GCR_EXT_MSIX_EN;
2876         gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
2877         IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
2878
2879         /* enable Tx loopback for VF/PF communication */
2880         IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
2881 }
2882
2883 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
2884 {
2885         struct ixgbe_hw *hw = &adapter->hw;
2886         struct net_device *netdev = adapter->netdev;
2887         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
2888         int rx_buf_len;
2889         struct ixgbe_ring *rx_ring;
2890         int i;
2891         u32 mhadd, hlreg0;
2892
2893         /* Decide whether to use packet split mode or not */
2894         /* Do not use packet split if we're in SR-IOV Mode */
2895         if (!adapter->num_vfs)
2896                 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
2897
2898         /* Set the RX buffer length according to the mode */
2899         if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
2900                 rx_buf_len = IXGBE_RX_HDR_SIZE;
2901         } else {
2902                 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
2903                     (netdev->mtu <= ETH_DATA_LEN))
2904                         rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2905                 else
2906                         rx_buf_len = ALIGN(max_frame + VLAN_HLEN, 1024);
2907         }
2908
2909 #ifdef IXGBE_FCOE
2910         /* adjust max frame to be able to do baby jumbo for FCoE */
2911         if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
2912             (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
2913                 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
2914
2915 #endif /* IXGBE_FCOE */
2916         mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
2917         if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
2918                 mhadd &= ~IXGBE_MHADD_MFS_MASK;
2919                 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
2920
2921                 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
2922         }
2923
2924         hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
2925         /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
2926         hlreg0 |= IXGBE_HLREG0_JUMBOEN;
2927         IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
2928
2929         /*
2930          * Setup the HW Rx Head and Tail Descriptor Pointers and
2931          * the Base and Length of the Rx Descriptor Ring
2932          */
2933         for (i = 0; i < adapter->num_rx_queues; i++) {
2934                 rx_ring = adapter->rx_ring[i];
2935                 rx_ring->rx_buf_len = rx_buf_len;
2936
2937                 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
2938                         rx_ring->flags |= IXGBE_RING_RX_PS_ENABLED;
2939                 else
2940                         rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2941
2942 #ifdef IXGBE_FCOE
2943                 if (netdev->features & NETIF_F_FCOE_MTU) {
2944                         struct ixgbe_ring_feature *f;
2945                         f = &adapter->ring_feature[RING_F_FCOE];
2946                         if ((i >= f->mask) && (i < f->mask + f->indices)) {
2947                                 rx_ring->flags &= ~IXGBE_RING_RX_PS_ENABLED;
2948                                 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
2949                                         rx_ring->rx_buf_len =
2950                                                 IXGBE_FCOE_JUMBO_FRAME_SIZE;
2951                         }
2952                 }
2953 #endif /* IXGBE_FCOE */
2954         }
2955
2956 }
2957
2958 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
2959 {
2960         struct ixgbe_hw *hw = &adapter->hw;
2961         u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
2962
2963         switch (hw->mac.type) {
2964         case ixgbe_mac_82598EB:
2965                 /*
2966                  * For VMDq support of different descriptor types or
2967                  * buffer sizes through the use of multiple SRRCTL
2968                  * registers, RDRXCTL.MVMEN must be set to 1
2969                  *
2970                  * also, the manual doesn't mention it clearly but DCA hints
2971                  * will only use queue 0's tags unless this bit is set.  Side
2972                  * effects of setting this bit are only that SRRCTL must be
2973                  * fully programmed [0..15]
2974                  */
2975                 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
2976                 break;
2977         case ixgbe_mac_82599EB:
2978                 /* Disable RSC for ACK packets */
2979                 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
2980                    (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
2981                 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
2982                 /* hardware requires some bits to be set by default */
2983                 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
2984                 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
2985                 break;
2986         default:
2987                 /* We should do nothing since we don't know this hardware */
2988                 return;
2989         }
2990
2991         IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
2992 }
2993
2994 /**
2995  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
2996  * @adapter: board private structure
2997  *
2998  * Configure the Rx unit of the MAC after a reset.
2999  **/
3000 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3001 {
3002         struct ixgbe_hw *hw = &adapter->hw;
3003         int i;
3004         u32 rxctrl;
3005
3006         /* disable receives while setting up the descriptors */
3007         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3008         IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3009
3010         ixgbe_setup_psrtype(adapter);
3011         ixgbe_setup_rdrxctl(adapter);
3012
3013         /* Program registers for the distribution of queues */
3014         ixgbe_setup_mrqc(adapter);
3015
3016         ixgbe_set_uta(adapter);
3017
3018         /* set_rx_buffer_len must be called before ring initialization */
3019         ixgbe_set_rx_buffer_len(adapter);
3020
3021         /*
3022          * Setup the HW Rx Head and Tail Descriptor Pointers and
3023          * the Base and Length of the Rx Descriptor Ring
3024          */
3025         for (i = 0; i < adapter->num_rx_queues; i++)
3026                 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3027
3028         /* disable drop enable for 82598 parts */
3029         if (hw->mac.type == ixgbe_mac_82598EB)
3030                 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3031
3032         /* enable all receives */
3033         rxctrl |= IXGBE_RXCTRL_RXEN;
3034         hw->mac.ops.enable_rx_dma(hw, rxctrl);
3035 }
3036
3037 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3038 {
3039         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3040         struct ixgbe_hw *hw = &adapter->hw;
3041         int pool_ndx = adapter->num_vfs;
3042
3043         /* add VID to filter table */
3044         hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3045         set_bit(vid, adapter->active_vlans);
3046 }
3047
3048 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3049 {
3050         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3051         struct ixgbe_hw *hw = &adapter->hw;
3052         int pool_ndx = adapter->num_vfs;
3053
3054         /* remove VID from filter table */
3055         hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3056         clear_bit(vid, adapter->active_vlans);
3057 }
3058
3059 /**
3060  * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3061  * @adapter: driver data
3062  */
3063 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3064 {
3065         struct ixgbe_hw *hw = &adapter->hw;
3066         u32 vlnctrl;
3067
3068         vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3069         vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3070         IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3071 }
3072
3073 /**
3074  * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3075  * @adapter: driver data
3076  */
3077 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3078 {
3079         struct ixgbe_hw *hw = &adapter->hw;
3080         u32 vlnctrl;
3081
3082         vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3083         vlnctrl |= IXGBE_VLNCTRL_VFE;
3084         vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3085         IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3086 }
3087
3088 /**
3089  * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3090  * @adapter: driver data
3091  */
3092 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3093 {
3094         struct ixgbe_hw *hw = &adapter->hw;
3095         u32 vlnctrl;
3096         int i, j;
3097
3098         switch (hw->mac.type) {
3099         case ixgbe_mac_82598EB:
3100                 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3101                 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3102                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3103                 break;
3104         case ixgbe_mac_82599EB:
3105                 for (i = 0; i < adapter->num_rx_queues; i++) {
3106                         j = adapter->rx_ring[i]->reg_idx;
3107                         vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3108                         vlnctrl &= ~IXGBE_RXDCTL_VME;
3109                         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3110                 }
3111                 break;
3112         default:
3113                 break;
3114         }
3115 }
3116
3117 /**
3118  * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3119  * @adapter: driver data
3120  */
3121 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3122 {
3123         struct ixgbe_hw *hw = &adapter->hw;
3124         u32 vlnctrl;
3125         int i, j;
3126
3127         switch (hw->mac.type) {
3128         case ixgbe_mac_82598EB:
3129                 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3130                 vlnctrl |= IXGBE_VLNCTRL_VME;
3131                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3132                 break;
3133         case ixgbe_mac_82599EB:
3134                 for (i = 0; i < adapter->num_rx_queues; i++) {
3135                         j = adapter->rx_ring[i]->reg_idx;
3136                         vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3137                         vlnctrl |= IXGBE_RXDCTL_VME;
3138                         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3139                 }
3140                 break;
3141         default:
3142                 break;
3143         }
3144 }
3145
3146 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3147 {
3148         u16 vid;
3149
3150         ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3151
3152         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3153                 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3154 }
3155
3156 /**
3157  * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3158  * @netdev: network interface device structure
3159  *
3160  * Writes unicast address list to the RAR table.
3161  * Returns: -ENOMEM on failure/insufficient address space
3162  *                0 on no addresses written
3163  *                X on writing X addresses to the RAR table
3164  **/
3165 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3166 {
3167         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3168         struct ixgbe_hw *hw = &adapter->hw;
3169         unsigned int vfn = adapter->num_vfs;
3170         unsigned int rar_entries = hw->mac.num_rar_entries - (vfn + 1);
3171         int count = 0;
3172
3173         /* return ENOMEM indicating insufficient memory for addresses */
3174         if (netdev_uc_count(netdev) > rar_entries)
3175                 return -ENOMEM;
3176
3177         if (!netdev_uc_empty(netdev) && rar_entries) {
3178                 struct netdev_hw_addr *ha;
3179                 /* return error if we do not support writing to RAR table */
3180                 if (!hw->mac.ops.set_rar)
3181                         return -ENOMEM;
3182
3183                 netdev_for_each_uc_addr(ha, netdev) {
3184                         if (!rar_entries)
3185                                 break;
3186                         hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3187                                             vfn, IXGBE_RAH_AV);
3188                         count++;
3189                 }
3190         }
3191         /* write the addresses in reverse order to avoid write combining */
3192         for (; rar_entries > 0 ; rar_entries--)
3193                 hw->mac.ops.clear_rar(hw, rar_entries);
3194
3195         return count;
3196 }
3197
3198 /**
3199  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3200  * @netdev: network interface device structure
3201  *
3202  * The set_rx_method entry point is called whenever the unicast/multicast
3203  * address list or the network interface flags are updated.  This routine is
3204  * responsible for configuring the hardware for proper unicast, multicast and
3205  * promiscuous mode.
3206  **/
3207 void ixgbe_set_rx_mode(struct net_device *netdev)
3208 {
3209         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3210         struct ixgbe_hw *hw = &adapter->hw;
3211         u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3212         int count;
3213
3214         /* Check for Promiscuous and All Multicast modes */
3215
3216         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3217
3218         /* set all bits that we expect to always be set */
3219         fctrl |= IXGBE_FCTRL_BAM;
3220         fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3221         fctrl |= IXGBE_FCTRL_PMCF;
3222
3223         /* clear the bits we are changing the status of */
3224         fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3225
3226         if (netdev->flags & IFF_PROMISC) {
3227                 hw->addr_ctrl.user_set_promisc = true;
3228                 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3229                 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3230                 /* don't hardware filter vlans in promisc mode */
3231                 ixgbe_vlan_filter_disable(adapter);
3232         } else {
3233                 if (netdev->flags & IFF_ALLMULTI) {
3234                         fctrl |= IXGBE_FCTRL_MPE;
3235                         vmolr |= IXGBE_VMOLR_MPE;
3236                 } else {
3237                         /*
3238                          * Write addresses to the MTA, if the attempt fails
3239                          * then we should just turn on promiscous mode so
3240                          * that we can at least receive multicast traffic
3241                          */
3242                         hw->mac.ops.update_mc_addr_list(hw, netdev);
3243                         vmolr |= IXGBE_VMOLR_ROMPE;
3244                 }
3245                 ixgbe_vlan_filter_enable(adapter);
3246                 hw->addr_ctrl.user_set_promisc = false;
3247                 /*
3248                  * Write addresses to available RAR registers, if there is not
3249                  * sufficient space to store all the addresses then enable
3250                  * unicast promiscous mode
3251                  */
3252                 count = ixgbe_write_uc_addr_list(netdev);
3253                 if (count < 0) {
3254                         fctrl |= IXGBE_FCTRL_UPE;
3255                         vmolr |= IXGBE_VMOLR_ROPE;
3256                 }
3257         }
3258
3259         if (adapter->num_vfs) {
3260                 ixgbe_restore_vf_multicasts(adapter);
3261                 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3262                          ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3263                            IXGBE_VMOLR_ROPE);
3264                 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3265         }
3266
3267         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3268
3269         if (netdev->features & NETIF_F_HW_VLAN_RX)
3270                 ixgbe_vlan_strip_enable(adapter);
3271         else
3272                 ixgbe_vlan_strip_disable(adapter);
3273 }
3274
3275 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3276 {
3277         int q_idx;
3278         struct ixgbe_q_vector *q_vector;
3279         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3280
3281         /* legacy and MSI only use one vector */
3282         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3283                 q_vectors = 1;
3284
3285         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3286                 struct napi_struct *napi;
3287                 q_vector = adapter->q_vector[q_idx];
3288                 napi = &q_vector->napi;
3289                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3290                         if (!q_vector->rxr_count || !q_vector->txr_count) {
3291                                 if (q_vector->txr_count == 1)
3292                                         napi->poll = &ixgbe_clean_txonly;
3293                                 else if (q_vector->rxr_count == 1)
3294                                         napi->poll = &ixgbe_clean_rxonly;
3295                         }
3296                 }
3297
3298                 napi_enable(napi);
3299         }
3300 }
3301
3302 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3303 {
3304         int q_idx;
3305         struct ixgbe_q_vector *q_vector;
3306         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3307
3308         /* legacy and MSI only use one vector */
3309         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3310                 q_vectors = 1;
3311
3312         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3313                 q_vector = adapter->q_vector[q_idx];
3314                 napi_disable(&q_vector->napi);
3315         }
3316 }
3317
3318 #ifdef CONFIG_IXGBE_DCB
3319 /*
3320  * ixgbe_configure_dcb - Configure DCB hardware
3321  * @adapter: ixgbe adapter struct
3322  *
3323  * This is called by the driver on open to configure the DCB hardware.
3324  * This is also called by the gennetlink interface when reconfiguring
3325  * the DCB state.
3326  */
3327 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3328 {
3329         struct ixgbe_hw *hw = &adapter->hw;
3330         int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3331         u32 txdctl;
3332         int i, j;
3333
3334         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3335                 if (hw->mac.type == ixgbe_mac_82598EB)
3336                         netif_set_gso_max_size(adapter->netdev, 65536);
3337                 return;
3338         }
3339
3340         if (hw->mac.type == ixgbe_mac_82598EB)
3341                 netif_set_gso_max_size(adapter->netdev, 32768);
3342
3343 #ifdef CONFIG_FCOE
3344         if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3345                 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3346 #endif
3347
3348         ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3349                                         DCB_TX_CONFIG);
3350         ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3351                                         DCB_RX_CONFIG);
3352
3353         /* reconfigure the hardware */
3354         ixgbe_dcb_hw_config(&adapter->hw, &adapter->dcb_cfg);
3355
3356         for (i = 0; i < adapter->num_tx_queues; i++) {
3357                 j = adapter->tx_ring[i]->reg_idx;
3358                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3359                 /* PThresh workaround for Tx hang with DFP enabled. */
3360                 txdctl |= 32;
3361                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
3362         }
3363         /* Enable VLAN tag insert/strip */
3364         adapter->netdev->features |= NETIF_F_HW_VLAN_RX;
3365
3366         hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3367 }
3368
3369 #endif
3370 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3371 {
3372         struct net_device *netdev = adapter->netdev;
3373         struct ixgbe_hw *hw = &adapter->hw;
3374         int i;
3375
3376 #ifdef CONFIG_IXGBE_DCB
3377         ixgbe_configure_dcb(adapter);
3378 #endif
3379
3380         ixgbe_set_rx_mode(netdev);
3381         ixgbe_restore_vlan(adapter);
3382
3383 #ifdef IXGBE_FCOE
3384         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3385                 ixgbe_configure_fcoe(adapter);
3386
3387 #endif /* IXGBE_FCOE */
3388         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3389                 for (i = 0; i < adapter->num_tx_queues; i++)
3390                         adapter->tx_ring[i]->atr_sample_rate =
3391                                                        adapter->atr_sample_rate;
3392                 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
3393         } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3394                 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
3395         }
3396         ixgbe_configure_virtualization(adapter);
3397
3398         ixgbe_configure_tx(adapter);
3399         ixgbe_configure_rx(adapter);
3400 }
3401
3402 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3403 {
3404         switch (hw->phy.type) {
3405         case ixgbe_phy_sfp_avago:
3406         case ixgbe_phy_sfp_ftl:
3407         case ixgbe_phy_sfp_intel:
3408         case ixgbe_phy_sfp_unknown:
3409         case ixgbe_phy_sfp_passive_tyco:
3410         case ixgbe_phy_sfp_passive_unknown:
3411         case ixgbe_phy_sfp_active_unknown:
3412         case ixgbe_phy_sfp_ftl_active:
3413                 return true;
3414         default:
3415                 return false;
3416         }
3417 }
3418
3419 /**
3420  * ixgbe_sfp_link_config - set up SFP+ link
3421  * @adapter: pointer to private adapter struct
3422  **/
3423 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3424 {
3425         struct ixgbe_hw *hw = &adapter->hw;
3426
3427                 if (hw->phy.multispeed_fiber) {
3428                         /*
3429                          * In multispeed fiber setups, the device may not have
3430                          * had a physical connection when the driver loaded.
3431                          * If that's the case, the initial link configuration
3432                          * couldn't get the MAC into 10G or 1G mode, so we'll
3433                          * never have a link status change interrupt fire.
3434                          * We need to try and force an autonegotiation
3435                          * session, then bring up link.
3436                          */
3437                         hw->mac.ops.setup_sfp(hw);
3438                         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
3439                                 schedule_work(&adapter->multispeed_fiber_task);
3440                 } else {
3441                         /*
3442                          * Direct Attach Cu and non-multispeed fiber modules
3443                          * still need to be configured properly prior to
3444                          * attempting link.
3445                          */
3446                         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
3447                                 schedule_work(&adapter->sfp_config_module_task);
3448                 }
3449 }
3450
3451 /**
3452  * ixgbe_non_sfp_link_config - set up non-SFP+ link
3453  * @hw: pointer to private hardware struct
3454  *
3455  * Returns 0 on success, negative on failure
3456  **/
3457 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3458 {
3459         u32 autoneg;
3460         bool negotiation, link_up = false;
3461         u32 ret = IXGBE_ERR_LINK_SETUP;
3462
3463         if (hw->mac.ops.check_link)
3464                 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3465
3466         if (ret)
3467                 goto link_cfg_out;
3468
3469         if (hw->mac.ops.get_link_capabilities)
3470                 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3471                                                         &negotiation);
3472         if (ret)
3473                 goto link_cfg_out;
3474
3475         if (hw->mac.ops.setup_link)
3476                 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3477 link_cfg_out:
3478         return ret;
3479 }
3480
3481 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3482 {
3483         struct ixgbe_hw *hw = &adapter->hw;
3484         u32 gpie = 0;
3485
3486         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3487                 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3488                        IXGBE_GPIE_OCD;
3489                 gpie |= IXGBE_GPIE_EIAME;
3490                 /*
3491                  * use EIAM to auto-mask when MSI-X interrupt is asserted
3492                  * this saves a register write for every interrupt
3493                  */
3494                 switch (hw->mac.type) {
3495                 case ixgbe_mac_82598EB:
3496                         IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3497                         break;
3498                 default:
3499                 case ixgbe_mac_82599EB:
3500                         IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3501                         IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3502                         break;
3503                 }
3504         } else {
3505                 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3506                  * specifically only auto mask tx and rx interrupts */
3507                 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3508         }
3509
3510         /* XXX: to interrupt immediately for EICS writes, enable this */
3511         /* gpie |= IXGBE_GPIE_EIMEN; */
3512
3513         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3514                 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3515                 gpie |= IXGBE_GPIE_VTMODE_64;
3516         }
3517
3518         /* Enable fan failure interrupt */
3519         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3520                 gpie |= IXGBE_SDP1_GPIEN;
3521
3522         if (hw->mac.type == ixgbe_mac_82599EB)
3523                 gpie |= IXGBE_SDP1_GPIEN;
3524                 gpie |= IXGBE_SDP2_GPIEN;
3525
3526         IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3527 }
3528
3529 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
3530 {
3531         struct ixgbe_hw *hw = &adapter->hw;
3532         int err;
3533         u32 ctrl_ext;
3534
3535         ixgbe_get_hw_control(adapter);
3536         ixgbe_setup_gpie(adapter);
3537
3538         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3539                 ixgbe_configure_msix(adapter);
3540         else
3541                 ixgbe_configure_msi_and_legacy(adapter);
3542
3543         /* enable the optics */
3544         if (hw->phy.multispeed_fiber)
3545                 hw->mac.ops.enable_tx_laser(hw);
3546
3547         clear_bit(__IXGBE_DOWN, &adapter->state);
3548         ixgbe_napi_enable_all(adapter);
3549
3550         /* clear any pending interrupts, may auto mask */
3551         IXGBE_READ_REG(hw, IXGBE_EICR);
3552         ixgbe_irq_enable(adapter, true, true);
3553
3554         /*
3555          * If this adapter has a fan, check to see if we had a failure
3556          * before we enabled the interrupt.
3557          */
3558         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3559                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3560                 if (esdp & IXGBE_ESDP_SDP1)
3561                         e_crit(drv, "Fan has stopped, replace the adapter\n");
3562         }
3563
3564         /*
3565          * For hot-pluggable SFP+ devices, a new SFP+ module may have
3566          * arrived before interrupts were enabled but after probe.  Such
3567          * devices wouldn't have their type identified yet. We need to
3568          * kick off the SFP+ module setup first, then try to bring up link.
3569          * If we're not hot-pluggable SFP+, we just need to configure link
3570          * and bring it up.
3571          */
3572         if (hw->phy.type == ixgbe_phy_unknown) {
3573                 err = hw->phy.ops.identify(hw);
3574                 if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
3575                         /*
3576                          * Take the device down and schedule the sfp tasklet
3577                          * which will unregister_netdev and log it.
3578                          */
3579                         ixgbe_down(adapter);
3580                         schedule_work(&adapter->sfp_config_module_task);
3581                         return err;
3582                 }
3583         }
3584
3585         if (ixgbe_is_sfp(hw)) {
3586                 ixgbe_sfp_link_config(adapter);
3587         } else {
3588                 err = ixgbe_non_sfp_link_config(hw);
3589                 if (err)
3590                         e_err(probe, "link_config FAILED %d\n", err);
3591         }
3592
3593         /* enable transmits */
3594         netif_tx_start_all_queues(adapter->netdev);
3595
3596         /* bring the link up in the watchdog, this could race with our first
3597          * link up interrupt but shouldn't be a problem */
3598         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3599         adapter->link_check_timeout = jiffies;
3600         mod_timer(&adapter->watchdog_timer, jiffies);
3601
3602         /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3603         ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3604         ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3605         IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3606
3607         return 0;
3608 }
3609
3610 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3611 {
3612         WARN_ON(in_interrupt());
3613         while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3614                 msleep(1);
3615         ixgbe_down(adapter);
3616         /*
3617          * If SR-IOV enabled then wait a bit before bringing the adapter
3618          * back up to give the VFs time to respond to the reset.  The
3619          * two second wait is based upon the watchdog timer cycle in
3620          * the VF driver.
3621          */
3622         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3623                 msleep(2000);
3624         ixgbe_up(adapter);
3625         clear_bit(__IXGBE_RESETTING, &adapter->state);
3626 }
3627
3628 int ixgbe_up(struct ixgbe_adapter *adapter)
3629 {
3630         /* hardware has been reset, we need to reload some things */
3631         ixgbe_configure(adapter);
3632
3633         return ixgbe_up_complete(adapter);
3634 }
3635
3636 void ixgbe_reset(struct ixgbe_adapter *adapter)
3637 {
3638         struct ixgbe_hw *hw = &adapter->hw;
3639         int err;
3640
3641         err = hw->mac.ops.init_hw(hw);
3642         switch (err) {
3643         case 0:
3644         case IXGBE_ERR_SFP_NOT_PRESENT:
3645                 break;
3646         case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3647                 e_dev_err("master disable timed out\n");
3648                 break;
3649         case IXGBE_ERR_EEPROM_VERSION:
3650                 /* We are running on a pre-production device, log a warning */
3651                 e_dev_warn("This device is a pre-production adapter/LOM. "
3652                            "Please be aware there may be issuesassociated with "
3653                            "your hardware.  If you are experiencing problems "
3654                            "please contact your Intel or hardware "
3655                            "representative who provided you with this "
3656                            "hardware.\n");
3657                 break;
3658         default:
3659                 e_dev_err("Hardware Error: %d\n", err);
3660         }
3661
3662         /* reprogram the RAR[0] in case user changed it. */
3663         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3664                             IXGBE_RAH_AV);
3665 }
3666
3667 /**
3668  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3669  * @rx_ring: ring to free buffers from
3670  **/
3671 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
3672 {
3673         struct device *dev = rx_ring->dev;
3674         unsigned long size;
3675         u16 i;
3676
3677         /* ring already cleared, nothing to do */
3678         if (!rx_ring->rx_buffer_info)
3679                 return;
3680
3681         /* Free all the Rx ring sk_buffs */
3682         for (i = 0; i < rx_ring->count; i++) {
3683                 struct ixgbe_rx_buffer *rx_buffer_info;
3684
3685                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
3686                 if (rx_buffer_info->dma) {
3687                         dma_unmap_single(rx_ring->dev, rx_buffer_info->dma,
3688                                          rx_ring->rx_buf_len,
3689                                          DMA_FROM_DEVICE);
3690                         rx_buffer_info->dma = 0;
3691                 }
3692                 if (rx_buffer_info->skb) {
3693                         struct sk_buff *skb = rx_buffer_info->skb;
3694                         rx_buffer_info->skb = NULL;
3695                         do {
3696                                 struct sk_buff *this = skb;
3697                                 if (IXGBE_RSC_CB(this)->delay_unmap) {
3698                                         dma_unmap_single(dev,
3699                                                          IXGBE_RSC_CB(this)->dma,
3700                                                          rx_ring->rx_buf_len,
3701                                                          DMA_FROM_DEVICE);
3702                                         IXGBE_RSC_CB(this)->dma = 0;
3703                                         IXGBE_RSC_CB(skb)->delay_unmap = false;
3704                                 }
3705                                 skb = skb->prev;
3706                                 dev_kfree_skb(this);
3707                         } while (skb);
3708                 }
3709                 if (!rx_buffer_info->page)
3710                         continue;
3711                 if (rx_buffer_info->page_dma) {
3712                         dma_unmap_page(dev, rx_buffer_info->page_dma,
3713                                        PAGE_SIZE / 2, DMA_FROM_DEVICE);
3714                         rx_buffer_info->page_dma = 0;
3715                 }
3716                 put_page(rx_buffer_info->page);
3717                 rx_buffer_info->page = NULL;
3718                 rx_buffer_info->page_offset = 0;
3719         }
3720
3721         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3722         memset(rx_ring->rx_buffer_info, 0, size);
3723
3724         /* Zero out the descriptor ring */
3725         memset(rx_ring->desc, 0, rx_ring->size);
3726
3727         rx_ring->next_to_clean = 0;
3728         rx_ring->next_to_use = 0;
3729 }
3730
3731 /**
3732  * ixgbe_clean_tx_ring - Free Tx Buffers
3733  * @tx_ring: ring to be cleaned
3734  **/
3735 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
3736 {
3737         struct ixgbe_tx_buffer *tx_buffer_info;
3738         unsigned long size;
3739         u16 i;
3740
3741         /* ring already cleared, nothing to do */
3742         if (!tx_ring->tx_buffer_info)
3743                 return;
3744
3745         /* Free all the Tx ring sk_buffs */
3746         for (i = 0; i < tx_ring->count; i++) {
3747                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3748                 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
3749         }
3750
3751         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3752         memset(tx_ring->tx_buffer_info, 0, size);
3753
3754         /* Zero out the descriptor ring */
3755         memset(tx_ring->desc, 0, tx_ring->size);
3756
3757         tx_ring->next_to_use = 0;
3758         tx_ring->next_to_clean = 0;
3759 }
3760
3761 /**
3762  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
3763  * @adapter: board private structure
3764  **/
3765 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
3766 {
3767         int i;
3768
3769         for (i = 0; i < adapter->num_rx_queues; i++)
3770                 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
3771 }
3772
3773 /**
3774  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
3775  * @adapter: board private structure
3776  **/
3777 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
3778 {
3779         int i;
3780
3781         for (i = 0; i < adapter->num_tx_queues; i++)
3782                 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
3783 }
3784
3785 void ixgbe_down(struct ixgbe_adapter *adapter)
3786 {
3787         struct net_device *netdev = adapter->netdev;
3788         struct ixgbe_hw *hw = &adapter->hw;
3789         u32 rxctrl;
3790         u32 txdctl;
3791         int i, j;
3792         int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3793
3794         /* signal that we are down to the interrupt handler */
3795         set_bit(__IXGBE_DOWN, &adapter->state);
3796
3797         /* disable receive for all VFs and wait one second */
3798         if (adapter->num_vfs) {
3799                 /* ping all the active vfs to let them know we are going down */
3800                 ixgbe_ping_all_vfs(adapter);
3801
3802                 /* Disable all VFTE/VFRE TX/RX */
3803                 ixgbe_disable_tx_rx(adapter);
3804
3805                 /* Mark all the VFs as inactive */
3806                 for (i = 0 ; i < adapter->num_vfs; i++)
3807                         adapter->vfinfo[i].clear_to_send = 0;
3808         }
3809
3810         /* disable receives */
3811         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3812         IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3813
3814         IXGBE_WRITE_FLUSH(hw);
3815         msleep(10);
3816
3817         netif_tx_stop_all_queues(netdev);
3818
3819         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
3820         del_timer_sync(&adapter->sfp_timer);
3821         del_timer_sync(&adapter->watchdog_timer);
3822         cancel_work_sync(&adapter->watchdog_task);
3823
3824         netif_carrier_off(netdev);
3825         netif_tx_disable(netdev);
3826
3827         ixgbe_irq_disable(adapter);
3828
3829         ixgbe_napi_disable_all(adapter);
3830
3831         /* Cleanup the affinity_hint CPU mask memory and callback */
3832         for (i = 0; i < num_q_vectors; i++) {
3833                 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
3834                 /* clear the affinity_mask in the IRQ descriptor */
3835                 irq_set_affinity_hint(adapter->msix_entries[i]. vector, NULL);
3836                 /* release the CPU mask memory */
3837                 free_cpumask_var(q_vector->affinity_mask);
3838         }
3839
3840         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
3841             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
3842                 cancel_work_sync(&adapter->fdir_reinit_task);
3843
3844         if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
3845                 cancel_work_sync(&adapter->check_overtemp_task);
3846
3847         /* disable transmits in the hardware now that interrupts are off */
3848         for (i = 0; i < adapter->num_tx_queues; i++) {
3849                 j = adapter->tx_ring[i]->reg_idx;
3850                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
3851                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
3852                                 (txdctl & ~IXGBE_TXDCTL_ENABLE));
3853         }
3854         /* Disable the Tx DMA engine on 82599 */
3855         if (hw->mac.type == ixgbe_mac_82599EB)
3856                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
3857                                 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
3858                                  ~IXGBE_DMATXCTL_TE));
3859
3860         /* power down the optics */
3861         if (hw->phy.multispeed_fiber)
3862                 hw->mac.ops.disable_tx_laser(hw);
3863
3864         /* clear n-tuple filters that are cached */
3865         ethtool_ntuple_flush(netdev);
3866
3867         if (!pci_channel_offline(adapter->pdev))
3868                 ixgbe_reset(adapter);
3869         ixgbe_clean_all_tx_rings(adapter);
3870         ixgbe_clean_all_rx_rings(adapter);
3871
3872 #ifdef CONFIG_IXGBE_DCA
3873         /* since we reset the hardware DCA settings were cleared */
3874         ixgbe_setup_dca(adapter);
3875 #endif
3876 }
3877
3878 /**
3879  * ixgbe_poll - NAPI Rx polling callback
3880  * @napi: structure for representing this polling device
3881  * @budget: how many packets driver is allowed to clean
3882  *
3883  * This function is used for legacy and MSI, NAPI mode
3884  **/
3885 static int ixgbe_poll(struct napi_struct *napi, int budget)
3886 {
3887         struct ixgbe_q_vector *q_vector =
3888                                 container_of(napi, struct ixgbe_q_vector, napi);
3889         struct ixgbe_adapter *adapter = q_vector->adapter;
3890         int tx_clean_complete, work_done = 0;
3891
3892 #ifdef CONFIG_IXGBE_DCA
3893         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
3894                 ixgbe_update_tx_dca(adapter, adapter->tx_ring[0]);
3895                 ixgbe_update_rx_dca(adapter, adapter->rx_ring[0]);
3896         }
3897 #endif
3898
3899         tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring[0]);
3900         ixgbe_clean_rx_irq(q_vector, adapter->rx_ring[0], &work_done, budget);
3901
3902         if (!tx_clean_complete)
3903                 work_done = budget;
3904
3905         /* If budget not fully consumed, exit the polling mode */
3906         if (work_done < budget) {
3907                 napi_complete(napi);
3908                 if (adapter->rx_itr_setting & 1)
3909                         ixgbe_set_itr(adapter);
3910                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3911                         ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
3912         }
3913         return work_done;
3914 }
3915
3916 /**
3917  * ixgbe_tx_timeout - Respond to a Tx Hang
3918  * @netdev: network interface device structure
3919  **/
3920 static void ixgbe_tx_timeout(struct net_device *netdev)
3921 {
3922         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3923
3924         /* Do the reset outside of interrupt context */
3925         schedule_work(&adapter->reset_task);
3926 }
3927
3928 static void ixgbe_reset_task(struct work_struct *work)
3929 {
3930         struct ixgbe_adapter *adapter;
3931         adapter = container_of(work, struct ixgbe_adapter, reset_task);
3932
3933         /* If we're already down or resetting, just bail */
3934         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
3935             test_bit(__IXGBE_RESETTING, &adapter->state))
3936                 return;
3937
3938         adapter->tx_timeout_count++;
3939
3940         ixgbe_dump(adapter);
3941         netdev_err(adapter->netdev, "Reset adapter\n");
3942         ixgbe_reinit_locked(adapter);
3943 }
3944
3945 #ifdef CONFIG_IXGBE_DCB
3946 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
3947 {
3948         bool ret = false;
3949         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
3950
3951         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
3952                 return ret;
3953
3954         f->mask = 0x7 << 3;
3955         adapter->num_rx_queues = f->indices;
3956         adapter->num_tx_queues = f->indices;
3957         ret = true;
3958
3959         return ret;
3960 }
3961 #endif
3962
3963 /**
3964  * ixgbe_set_rss_queues: Allocate queues for RSS
3965  * @adapter: board private structure to initialize
3966  *
3967  * This is our "base" multiqueue mode.  RSS (Receive Side Scaling) will try
3968  * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
3969  *
3970  **/
3971 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
3972 {
3973         bool ret = false;
3974         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
3975
3976         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
3977                 f->mask = 0xF;
3978                 adapter->num_rx_queues = f->indices;
3979                 adapter->num_tx_queues = f->indices;
3980                 ret = true;
3981         } else {
3982                 ret = false;
3983         }
3984
3985         return ret;
3986 }
3987
3988 /**
3989  * ixgbe_set_fdir_queues: Allocate queues for Flow Director
3990  * @adapter: board private structure to initialize
3991  *
3992  * Flow Director is an advanced Rx filter, attempting to get Rx flows back
3993  * to the original CPU that initiated the Tx session.  This runs in addition
3994  * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
3995  * Rx load across CPUs using RSS.
3996  *
3997  **/
3998 static inline bool ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
3999 {
4000         bool ret = false;
4001         struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
4002
4003         f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
4004         f_fdir->mask = 0;
4005
4006         /* Flow Director must have RSS enabled */
4007         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4008             ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4009              (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
4010                 adapter->num_tx_queues = f_fdir->indices;
4011                 adapter->num_rx_queues = f_fdir->indices;
4012                 ret = true;
4013         } else {
4014                 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4015                 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4016         }
4017         return ret;
4018 }
4019
4020 #ifdef IXGBE_FCOE
4021 /**
4022  * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
4023  * @adapter: board private structure to initialize
4024  *
4025  * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
4026  * The ring feature mask is not used as a mask for FCoE, as it can take any 8
4027  * rx queues out of the max number of rx queues, instead, it is used as the
4028  * index of the first rx queue used by FCoE.
4029  *
4030  **/
4031 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
4032 {
4033         bool ret = false;
4034         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4035
4036         f->indices = min((int)num_online_cpus(), f->indices);
4037         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4038                 adapter->num_rx_queues = 1;
4039                 adapter->num_tx_queues = 1;
4040 #ifdef CONFIG_IXGBE_DCB
4041                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4042                         e_info(probe, "FCoE enabled with DCB\n");
4043                         ixgbe_set_dcb_queues(adapter);
4044                 }
4045 #endif
4046                 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4047                         e_info(probe, "FCoE enabled with RSS\n");
4048                         if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4049                             (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4050                                 ixgbe_set_fdir_queues(adapter);
4051                         else
4052                                 ixgbe_set_rss_queues(adapter);
4053                 }
4054                 /* adding FCoE rx rings to the end */
4055                 f->mask = adapter->num_rx_queues;
4056                 adapter->num_rx_queues += f->indices;
4057                 adapter->num_tx_queues += f->indices;
4058
4059                 ret = true;
4060         }
4061
4062         return ret;
4063 }
4064
4065 #endif /* IXGBE_FCOE */
4066 /**
4067  * ixgbe_set_sriov_queues: Allocate queues for IOV use
4068  * @adapter: board private structure to initialize
4069  *
4070  * IOV doesn't actually use anything, so just NAK the
4071  * request for now and let the other queue routines
4072  * figure out what to do.
4073  */
4074 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
4075 {
4076         return false;
4077 }
4078
4079 /*
4080  * ixgbe_set_num_queues: Allocate queues for device, feature dependant
4081  * @adapter: board private structure to initialize
4082  *
4083  * This is the top level queue allocation routine.  The order here is very
4084  * important, starting with the "most" number of features turned on at once,
4085  * and ending with the smallest set of features.  This way large combinations
4086  * can be allocated if they're turned on, and smaller combinations are the
4087  * fallthrough conditions.
4088  *
4089  **/
4090 static int ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
4091 {
4092         /* Start with base case */
4093         adapter->num_rx_queues = 1;
4094         adapter->num_tx_queues = 1;
4095         adapter->num_rx_pools = adapter->num_rx_queues;
4096         adapter->num_rx_queues_per_pool = 1;
4097
4098         if (ixgbe_set_sriov_queues(adapter))
4099                 goto done;
4100
4101 #ifdef IXGBE_FCOE
4102         if (ixgbe_set_fcoe_queues(adapter))
4103                 goto done;
4104
4105 #endif /* IXGBE_FCOE */
4106 #ifdef CONFIG_IXGBE_DCB
4107         if (ixgbe_set_dcb_queues(adapter))
4108                 goto done;
4109
4110 #endif
4111         if (ixgbe_set_fdir_queues(adapter))
4112                 goto done;
4113
4114         if (ixgbe_set_rss_queues(adapter))
4115                 goto done;
4116
4117         /* fallback to base case */
4118         adapter->num_rx_queues = 1;
4119         adapter->num_tx_queues = 1;
4120
4121 done:
4122         /* Notify the stack of the (possibly) reduced queue counts. */
4123         netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
4124         return netif_set_real_num_rx_queues(adapter->netdev,
4125                                             adapter->num_rx_queues);
4126 }
4127
4128 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4129                                        int vectors)
4130 {
4131         int err, vector_threshold;
4132
4133         /* We'll want at least 3 (vector_threshold):
4134          * 1) TxQ[0] Cleanup
4135          * 2) RxQ[0] Cleanup
4136          * 3) Other (Link Status Change, etc.)
4137          * 4) TCP Timer (optional)
4138          */
4139         vector_threshold = MIN_MSIX_COUNT;
4140
4141         /* The more we get, the more we will assign to Tx/Rx Cleanup
4142          * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4143          * Right now, we simply care about how many we'll get; we'll
4144          * set them up later while requesting irq's.
4145          */
4146         while (vectors >= vector_threshold) {
4147                 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4148                                       vectors);
4149                 if (!err) /* Success in acquiring all requested vectors. */
4150                         break;
4151                 else if (err < 0)
4152                         vectors = 0; /* Nasty failure, quit now */
4153                 else /* err == number of vectors we should try again with */
4154                         vectors = err;
4155         }
4156
4157         if (vectors < vector_threshold) {
4158                 /* Can't allocate enough MSI-X interrupts?  Oh well.
4159                  * This just means we'll go with either a single MSI
4160                  * vector or fall back to legacy interrupts.
4161                  */
4162                 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4163                              "Unable to allocate MSI-X interrupts\n");
4164                 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4165                 kfree(adapter->msix_entries);
4166                 adapter->msix_entries = NULL;
4167         } else {
4168                 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4169                 /*
4170                  * Adjust for only the vectors we'll use, which is minimum
4171                  * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4172                  * vectors we were allocated.
4173                  */
4174                 adapter->num_msix_vectors = min(vectors,
4175                                    adapter->max_msix_q_vectors + NON_Q_VECTORS);
4176         }
4177 }
4178
4179 /**
4180  * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4181  * @adapter: board private structure to initialize
4182  *
4183  * Cache the descriptor ring offsets for RSS to the assigned rings.
4184  *
4185  **/
4186 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4187 {
4188         int i;
4189         bool ret = false;
4190
4191         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4192                 for (i = 0; i < adapter->num_rx_queues; i++)
4193                         adapter->rx_ring[i]->reg_idx = i;
4194                 for (i = 0; i < adapter->num_tx_queues; i++)
4195                         adapter->tx_ring[i]->reg_idx = i;
4196                 ret = true;
4197         } else {
4198                 ret = false;
4199         }
4200
4201         return ret;
4202 }
4203
4204 #ifdef CONFIG_IXGBE_DCB
4205 /**
4206  * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4207  * @adapter: board private structure to initialize
4208  *
4209  * Cache the descriptor ring offsets for DCB to the assigned rings.
4210  *
4211  **/
4212 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4213 {
4214         int i;
4215         bool ret = false;
4216         int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
4217
4218         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4219                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
4220                         /* the number of queues is assumed to be symmetric */
4221                         for (i = 0; i < dcb_i; i++) {
4222                                 adapter->rx_ring[i]->reg_idx = i << 3;
4223                                 adapter->tx_ring[i]->reg_idx = i << 2;
4224                         }
4225                         ret = true;
4226                 } else if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
4227                         if (dcb_i == 8) {
4228                                 /*
4229                                  * Tx TC0 starts at: descriptor queue 0
4230                                  * Tx TC1 starts at: descriptor queue 32
4231                                  * Tx TC2 starts at: descriptor queue 64
4232                                  * Tx TC3 starts at: descriptor queue 80
4233                                  * Tx TC4 starts at: descriptor queue 96
4234                                  * Tx TC5 starts at: descriptor queue 104
4235                                  * Tx TC6 starts at: descriptor queue 112
4236                                  * Tx TC7 starts at: descriptor queue 120
4237                                  *
4238                                  * Rx TC0-TC7 are offset by 16 queues each
4239                                  */
4240                                 for (i = 0; i < 3; i++) {
4241                                         adapter->tx_ring[i]->reg_idx = i << 5;
4242                                         adapter->rx_ring[i]->reg_idx = i << 4;
4243                                 }
4244                                 for ( ; i < 5; i++) {
4245                                         adapter->tx_ring[i]->reg_idx =
4246                                                                  ((i + 2) << 4);
4247                                         adapter->rx_ring[i]->reg_idx = i << 4;
4248                                 }
4249                                 for ( ; i < dcb_i; i++) {
4250                                         adapter->tx_ring[i]->reg_idx =
4251                                                                  ((i + 8) << 3);
4252                                         adapter->rx_ring[i]->reg_idx = i << 4;
4253                                 }
4254
4255                                 ret = true;
4256                         } else if (dcb_i == 4) {
4257                                 /*
4258                                  * Tx TC0 starts at: descriptor queue 0
4259                                  * Tx TC1 starts at: descriptor queue 64
4260                                  * Tx TC2 starts at: descriptor queue 96
4261                                  * Tx TC3 starts at: descriptor queue 112
4262                                  *
4263                                  * Rx TC0-TC3 are offset by 32 queues each
4264                                  */
4265                                 adapter->tx_ring[0]->reg_idx = 0;
4266                                 adapter->tx_ring[1]->reg_idx = 64;
4267                                 adapter->tx_ring[2]->reg_idx = 96;
4268                                 adapter->tx_ring[3]->reg_idx = 112;
4269                                 for (i = 0 ; i < dcb_i; i++)
4270                                         adapter->rx_ring[i]->reg_idx = i << 5;
4271
4272                                 ret = true;
4273                         } else {
4274                                 ret = false;
4275                         }
4276                 } else {
4277                         ret = false;
4278                 }
4279         } else {
4280                 ret = false;
4281         }
4282
4283         return ret;
4284 }
4285 #endif
4286
4287 /**
4288  * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4289  * @adapter: board private structure to initialize
4290  *
4291  * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4292  *
4293  **/
4294 static inline bool ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4295 {
4296         int i;
4297         bool ret = false;
4298
4299         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4300             ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4301              (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
4302                 for (i = 0; i < adapter->num_rx_queues; i++)
4303                         adapter->rx_ring[i]->reg_idx = i;
4304                 for (i = 0; i < adapter->num_tx_queues; i++)
4305                         adapter->tx_ring[i]->reg_idx = i;
4306                 ret = true;
4307         }
4308
4309         return ret;
4310 }
4311
4312 #ifdef IXGBE_FCOE
4313 /**
4314  * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4315  * @adapter: board private structure to initialize
4316  *
4317  * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4318  *
4319  */
4320 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4321 {
4322         int i, fcoe_rx_i = 0, fcoe_tx_i = 0;
4323         bool ret = false;
4324         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4325
4326         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4327 #ifdef CONFIG_IXGBE_DCB
4328                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4329                         struct ixgbe_fcoe *fcoe = &adapter->fcoe;
4330
4331                         ixgbe_cache_ring_dcb(adapter);
4332                         /* find out queues in TC for FCoE */
4333                         fcoe_rx_i = adapter->rx_ring[fcoe->tc]->reg_idx + 1;
4334                         fcoe_tx_i = adapter->tx_ring[fcoe->tc]->reg_idx + 1;
4335                         /*
4336                          * In 82599, the number of Tx queues for each traffic
4337                          * class for both 8-TC and 4-TC modes are:
4338                          * TCs  : TC0 TC1 TC2 TC3 TC4 TC5 TC6 TC7
4339                          * 8 TCs:  32  32  16  16   8   8   8   8
4340                          * 4 TCs:  64  64  32  32
4341                          * We have max 8 queues for FCoE, where 8 the is
4342                          * FCoE redirection table size. If TC for FCoE is
4343                          * less than or equal to TC3, we have enough queues
4344                          * to add max of 8 queues for FCoE, so we start FCoE
4345                          * tx descriptor from the next one, i.e., reg_idx + 1.
4346                          * If TC for FCoE is above TC3, implying 8 TC mode,
4347                          * and we need 8 for FCoE, we have to take all queues
4348                          * in that traffic class for FCoE.
4349                          */
4350                         if ((f->indices == IXGBE_FCRETA_SIZE) && (fcoe->tc > 3))
4351                                 fcoe_tx_i--;
4352                 }
4353 #endif /* CONFIG_IXGBE_DCB */
4354                 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4355                         if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4356                             (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4357                                 ixgbe_cache_ring_fdir(adapter);
4358                         else
4359                                 ixgbe_cache_ring_rss(adapter);
4360
4361                         fcoe_rx_i = f->mask;
4362                         fcoe_tx_i = f->mask;
4363                 }
4364                 for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4365                         adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4366                         adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4367                 }
4368                 ret = true;
4369         }
4370         return ret;
4371 }
4372
4373 #endif /* IXGBE_FCOE */
4374 /**
4375  * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4376  * @adapter: board private structure to initialize
4377  *
4378  * SR-IOV doesn't use any descriptor rings but changes the default if
4379  * no other mapping is used.
4380  *
4381  */
4382 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4383 {
4384         adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4385         adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4386         if (adapter->num_vfs)
4387                 return true;
4388         else
4389                 return false;
4390 }
4391
4392 /**
4393  * ixgbe_cache_ring_register - Descriptor ring to register mapping
4394  * @adapter: board private structure to initialize
4395  *
4396  * Once we know the feature-set enabled for the device, we'll cache
4397  * the register offset the descriptor ring is assigned to.
4398  *
4399  * Note, the order the various feature calls is important.  It must start with
4400  * the "most" features enabled at the same time, then trickle down to the
4401  * least amount of features turned on at once.
4402  **/
4403 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4404 {
4405         /* start with default case */
4406         adapter->rx_ring[0]->reg_idx = 0;
4407         adapter->tx_ring[0]->reg_idx = 0;
4408
4409         if (ixgbe_cache_ring_sriov(adapter))
4410                 return;
4411
4412 #ifdef IXGBE_FCOE
4413         if (ixgbe_cache_ring_fcoe(adapter))
4414                 return;
4415
4416 #endif /* IXGBE_FCOE */
4417 #ifdef CONFIG_IXGBE_DCB
4418         if (ixgbe_cache_ring_dcb(adapter))
4419                 return;
4420
4421 #endif
4422         if (ixgbe_cache_ring_fdir(adapter))
4423                 return;
4424
4425         if (ixgbe_cache_ring_rss(adapter))
4426                 return;
4427 }
4428
4429 /**
4430  * ixgbe_alloc_queues - Allocate memory for all rings
4431  * @adapter: board private structure to initialize
4432  *
4433  * We allocate one ring per queue at run-time since we don't know the
4434  * number of queues at compile-time.  The polling_netdev array is
4435  * intended for Multiqueue, but should work fine with a single queue.
4436  **/
4437 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
4438 {
4439         int i;
4440         int rx_count;
4441         int orig_node = adapter->node;
4442
4443         for (i = 0; i < adapter->num_tx_queues; i++) {
4444                 struct ixgbe_ring *ring = adapter->tx_ring[i];
4445                 if (orig_node == -1) {
4446                         int cur_node = next_online_node(adapter->node);
4447                         if (cur_node == MAX_NUMNODES)
4448                                 cur_node = first_online_node;
4449                         adapter->node = cur_node;
4450                 }
4451                 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4452                                     adapter->node);
4453                 if (!ring)
4454                         ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4455                 if (!ring)
4456                         goto err_tx_ring_allocation;
4457                 ring->count = adapter->tx_ring_count;
4458                 ring->queue_index = i;
4459                 ring->dev = &adapter->pdev->dev;
4460                 ring->netdev = adapter->netdev;
4461                 ring->numa_node = adapter->node;
4462
4463                 adapter->tx_ring[i] = ring;
4464         }
4465
4466         /* Restore the adapter's original node */
4467         adapter->node = orig_node;
4468
4469         rx_count = adapter->rx_ring_count;
4470         for (i = 0; i < adapter->num_rx_queues; i++) {
4471                 struct ixgbe_ring *ring = adapter->rx_ring[i];
4472                 if (orig_node == -1) {
4473                         int cur_node = next_online_node(adapter->node);
4474                         if (cur_node == MAX_NUMNODES)
4475                                 cur_node = first_online_node;
4476                         adapter->node = cur_node;
4477                 }
4478                 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4479                                     adapter->node);
4480                 if (!ring)
4481                         ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4482                 if (!ring)
4483                         goto err_rx_ring_allocation;
4484                 ring->count = rx_count;
4485                 ring->queue_index = i;
4486                 ring->dev = &adapter->pdev->dev;
4487                 ring->netdev = adapter->netdev;
4488                 ring->numa_node = adapter->node;
4489
4490                 adapter->rx_ring[i] = ring;
4491         }
4492
4493         /* Restore the adapter's original node */
4494         adapter->node = orig_node;
4495
4496         ixgbe_cache_ring_register(adapter);
4497
4498         return 0;
4499
4500 err_rx_ring_allocation:
4501         for (i = 0; i < adapter->num_tx_queues; i++)
4502                 kfree(adapter->tx_ring[i]);
4503 err_tx_ring_allocation:
4504         return -ENOMEM;
4505 }
4506
4507 /**
4508  * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4509  * @adapter: board private structure to initialize
4510  *
4511  * Attempt to configure the interrupts using the best available
4512  * capabilities of the hardware and the kernel.
4513  **/
4514 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4515 {
4516         struct ixgbe_hw *hw = &adapter->hw;
4517         int err = 0;
4518         int vector, v_budget;
4519
4520         /*
4521          * It's easy to be greedy for MSI-X vectors, but it really
4522          * doesn't do us much good if we have a lot more vectors
4523          * than CPU's.  So let's be conservative and only ask for
4524          * (roughly) the same number of vectors as there are CPU's.
4525          */
4526         v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
4527                        (int)num_online_cpus()) + NON_Q_VECTORS;
4528
4529         /*
4530          * At the same time, hardware can only support a maximum of
4531          * hw.mac->max_msix_vectors vectors.  With features
4532          * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4533          * descriptor queues supported by our device.  Thus, we cap it off in
4534          * those rare cases where the cpu count also exceeds our vector limit.
4535          */
4536         v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
4537
4538         /* A failure in MSI-X entry allocation isn't fatal, but it does
4539          * mean we disable MSI-X capabilities of the adapter. */
4540         adapter->msix_entries = kcalloc(v_budget,
4541                                         sizeof(struct msix_entry), GFP_KERNEL);
4542         if (adapter->msix_entries) {
4543                 for (vector = 0; vector < v_budget; vector++)
4544                         adapter->msix_entries[vector].entry = vector;
4545
4546                 ixgbe_acquire_msix_vectors(adapter, v_budget);
4547
4548                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4549                         goto out;
4550         }
4551
4552         adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4553         adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4554         adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4555         adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4556         adapter->atr_sample_rate = 0;
4557         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4558                 ixgbe_disable_sriov(adapter);
4559
4560         err = ixgbe_set_num_queues(adapter);
4561         if (err)
4562                 return err;
4563
4564         err = pci_enable_msi(adapter->pdev);
4565         if (!err) {
4566                 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4567         } else {
4568                 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4569                              "Unable to allocate MSI interrupt, "
4570                              "falling back to legacy.  Error: %d\n", err);
4571                 /* reset err */
4572                 err = 0;
4573         }
4574
4575 out:
4576         return err;
4577 }
4578
4579 /**
4580  * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4581  * @adapter: board private structure to initialize
4582  *
4583  * We allocate one q_vector per queue interrupt.  If allocation fails we
4584  * return -ENOMEM.
4585  **/
4586 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4587 {
4588         int q_idx, num_q_vectors;
4589         struct ixgbe_q_vector *q_vector;
4590         int napi_vectors;
4591         int (*poll)(struct napi_struct *, int);
4592
4593         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4594                 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4595                 napi_vectors = adapter->num_rx_queues;
4596                 poll = &ixgbe_clean_rxtx_many;
4597         } else {
4598                 num_q_vectors = 1;
4599                 napi_vectors = 1;
4600                 poll = &ixgbe_poll;
4601         }
4602
4603         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4604                 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4605                                         GFP_KERNEL, adapter->node);
4606                 if (!q_vector)
4607                         q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4608                                            GFP_KERNEL);
4609                 if (!q_vector)
4610                         goto err_out;
4611                 q_vector->adapter = adapter;
4612                 if (q_vector->txr_count && !q_vector->rxr_count)
4613                         q_vector->eitr = adapter->tx_eitr_param;
4614                 else
4615                         q_vector->eitr = adapter->rx_eitr_param;
4616                 q_vector->v_idx = q_idx;
4617                 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
4618                 adapter->q_vector[q_idx] = q_vector;
4619         }
4620
4621         return 0;
4622
4623 err_out:
4624         while (q_idx) {
4625                 q_idx--;
4626                 q_vector = adapter->q_vector[q_idx];
4627                 netif_napi_del(&q_vector->napi);
4628                 kfree(q_vector);
4629                 adapter->q_vector[q_idx] = NULL;
4630         }
4631         return -ENOMEM;
4632 }
4633
4634 /**
4635  * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4636  * @adapter: board private structure to initialize
4637  *
4638  * This function frees the memory allocated to the q_vectors.  In addition if
4639  * NAPI is enabled it will delete any references to the NAPI struct prior
4640  * to freeing the q_vector.
4641  **/
4642 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4643 {
4644         int q_idx, num_q_vectors;
4645
4646         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4647                 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4648         else
4649                 num_q_vectors = 1;
4650
4651         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4652                 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
4653                 adapter->q_vector[q_idx] = NULL;
4654                 netif_napi_del(&q_vector->napi);
4655                 kfree(q_vector);
4656         }
4657 }
4658
4659 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4660 {
4661         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4662                 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4663                 pci_disable_msix(adapter->pdev);
4664                 kfree(adapter->msix_entries);
4665                 adapter->msix_entries = NULL;
4666         } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4667                 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4668                 pci_disable_msi(adapter->pdev);
4669         }
4670 }
4671
4672 /**
4673  * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4674  * @adapter: board private structure to initialize
4675  *
4676  * We determine which interrupt scheme to use based on...
4677  * - Kernel support (MSI, MSI-X)
4678  *   - which can be user-defined (via MODULE_PARAM)
4679  * - Hardware queue count (num_*_queues)
4680  *   - defined by miscellaneous hardware support/features (RSS, etc.)
4681  **/
4682 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4683 {
4684         int err;
4685
4686         /* Number of supported queues */
4687         err = ixgbe_set_num_queues(adapter);
4688         if (err)
4689                 return err;
4690
4691         err = ixgbe_set_interrupt_capability(adapter);
4692         if (err) {
4693                 e_dev_err("Unable to setup interrupt capabilities\n");
4694                 goto err_set_interrupt;
4695         }
4696
4697         err = ixgbe_alloc_q_vectors(adapter);
4698         if (err) {
4699                 e_dev_err("Unable to allocate memory for queue vectors\n");
4700                 goto err_alloc_q_vectors;
4701         }
4702
4703         err = ixgbe_alloc_queues(adapter);
4704         if (err) {
4705                 e_dev_err("Unable to allocate memory for queues\n");
4706                 goto err_alloc_queues;
4707         }
4708
4709         e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
4710                    (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
4711                    adapter->num_rx_queues, adapter->num_tx_queues);
4712
4713         set_bit(__IXGBE_DOWN, &adapter->state);
4714
4715         return 0;
4716
4717 err_alloc_queues:
4718         ixgbe_free_q_vectors(adapter);
4719 err_alloc_q_vectors:
4720         ixgbe_reset_interrupt_capability(adapter);
4721 err_set_interrupt:
4722         return err;
4723 }
4724
4725 static void ring_free_rcu(struct rcu_head *head)
4726 {
4727         kfree(container_of(head, struct ixgbe_ring, rcu));
4728 }
4729
4730 /**
4731  * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
4732  * @adapter: board private structure to clear interrupt scheme on
4733  *
4734  * We go through and clear interrupt specific resources and reset the structure
4735  * to pre-load conditions
4736  **/
4737 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
4738 {
4739         int i;
4740
4741         for (i = 0; i < adapter->num_tx_queues; i++) {
4742                 kfree(adapter->tx_ring[i]);
4743                 adapter->tx_ring[i] = NULL;
4744         }
4745         for (i = 0; i < adapter->num_rx_queues; i++) {
4746                 struct ixgbe_ring *ring = adapter->rx_ring[i];
4747
4748                 /* ixgbe_get_stats64() might access this ring, we must wait
4749                  * a grace period before freeing it.
4750                  */
4751                 call_rcu(&ring->rcu, ring_free_rcu);
4752                 adapter->rx_ring[i] = NULL;
4753         }
4754
4755         ixgbe_free_q_vectors(adapter);
4756         ixgbe_reset_interrupt_capability(adapter);
4757 }
4758
4759 /**
4760  * ixgbe_sfp_timer - worker thread to find a missing module
4761  * @data: pointer to our adapter struct
4762  **/
4763 static void ixgbe_sfp_timer(unsigned long data)
4764 {
4765         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
4766
4767         /*
4768          * Do the sfp_timer outside of interrupt context due to the
4769          * delays that sfp+ detection requires
4770          */
4771         schedule_work(&adapter->sfp_task);
4772 }
4773
4774 /**
4775  * ixgbe_sfp_task - worker thread to find a missing module
4776  * @work: pointer to work_struct containing our data
4777  **/
4778 static void ixgbe_sfp_task(struct work_struct *work)
4779 {
4780         struct ixgbe_adapter *adapter = container_of(work,
4781                                                      struct ixgbe_adapter,
4782                                                      sfp_task);
4783         struct ixgbe_hw *hw = &adapter->hw;
4784
4785         if ((hw->phy.type == ixgbe_phy_nl) &&
4786             (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
4787                 s32 ret = hw->phy.ops.identify_sfp(hw);
4788                 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
4789                         goto reschedule;
4790                 ret = hw->phy.ops.reset(hw);
4791                 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
4792                         e_dev_err("failed to initialize because an unsupported "
4793                                   "SFP+ module type was detected.\n");
4794                         e_dev_err("Reload the driver after installing a "
4795                                   "supported module.\n");
4796                         unregister_netdev(adapter->netdev);
4797                 } else {
4798                         e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
4799                 }
4800                 /* don't need this routine any more */
4801                 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4802         }
4803         return;
4804 reschedule:
4805         if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
4806                 mod_timer(&adapter->sfp_timer,
4807                           round_jiffies(jiffies + (2 * HZ)));
4808 }
4809
4810 /**
4811  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
4812  * @adapter: board private structure to initialize
4813  *
4814  * ixgbe_sw_init initializes the Adapter private data structure.
4815  * Fields are initialized based on PCI device information and
4816  * OS network device settings (MTU size).
4817  **/
4818 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
4819 {
4820         struct ixgbe_hw *hw = &adapter->hw;
4821         struct pci_dev *pdev = adapter->pdev;
4822         struct net_device *dev = adapter->netdev;
4823         unsigned int rss;
4824 #ifdef CONFIG_IXGBE_DCB
4825         int j;
4826         struct tc_configuration *tc;
4827 #endif
4828         int max_frame = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
4829
4830         /* PCI config space info */
4831
4832         hw->vendor_id = pdev->vendor;
4833         hw->device_id = pdev->device;
4834         hw->revision_id = pdev->revision;
4835         hw->subsystem_vendor_id = pdev->subsystem_vendor;
4836         hw->subsystem_device_id = pdev->subsystem_device;
4837
4838         /* Set capability flags */
4839         rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
4840         adapter->ring_feature[RING_F_RSS].indices = rss;
4841         adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
4842         adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
4843         if (hw->mac.type == ixgbe_mac_82598EB) {
4844                 if (hw->device_id == IXGBE_DEV_ID_82598AT)
4845                         adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
4846                 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
4847         } else if (hw->mac.type == ixgbe_mac_82599EB) {
4848                 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
4849                 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
4850                 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
4851                 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
4852                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
4853                 if (dev->features & NETIF_F_NTUPLE) {
4854                         /* Flow Director perfect filter enabled */
4855                         adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4856                         adapter->atr_sample_rate = 0;
4857                         spin_lock_init(&adapter->fdir_perfect_lock);
4858                 } else {
4859                         /* Flow Director hash filters enabled */
4860                         adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
4861                         adapter->atr_sample_rate = 20;
4862                 }
4863                 adapter->ring_feature[RING_F_FDIR].indices =
4864                                                          IXGBE_MAX_FDIR_INDICES;
4865                 adapter->fdir_pballoc = 0;
4866 #ifdef IXGBE_FCOE
4867                 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
4868                 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
4869                 adapter->ring_feature[RING_F_FCOE].indices = 0;
4870 #ifdef CONFIG_IXGBE_DCB
4871                 /* Default traffic class to use for FCoE */
4872                 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
4873                 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
4874 #endif
4875 #endif /* IXGBE_FCOE */
4876         }
4877
4878 #ifdef CONFIG_IXGBE_DCB
4879         /* Configure DCB traffic classes */
4880         for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
4881                 tc = &adapter->dcb_cfg.tc_config[j];
4882                 tc->path[DCB_TX_CONFIG].bwg_id = 0;
4883                 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
4884                 tc->path[DCB_RX_CONFIG].bwg_id = 0;
4885                 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
4886                 tc->dcb_pfc = pfc_disabled;
4887         }
4888         adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
4889         adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
4890         adapter->dcb_cfg.rx_pba_cfg = pba_equal;
4891         adapter->dcb_cfg.pfc_mode_enable = false;
4892         adapter->dcb_cfg.round_robin_enable = false;
4893         adapter->dcb_set_bitmap = 0x00;
4894         ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
4895                            adapter->ring_feature[RING_F_DCB].indices);
4896
4897 #endif
4898
4899         /* default flow control settings */
4900         hw->fc.requested_mode = ixgbe_fc_full;
4901         hw->fc.current_mode = ixgbe_fc_full;    /* init for ethtool output */
4902 #ifdef CONFIG_DCB
4903         adapter->last_lfc_mode = hw->fc.current_mode;
4904 #endif
4905         hw->fc.high_water = FC_HIGH_WATER(max_frame);
4906         hw->fc.low_water = FC_LOW_WATER(max_frame);
4907         hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
4908         hw->fc.send_xon = true;
4909         hw->fc.disable_fc_autoneg = false;
4910
4911         /* enable itr by default in dynamic mode */
4912         adapter->rx_itr_setting = 1;
4913         adapter->rx_eitr_param = 20000;
4914         adapter->tx_itr_setting = 1;
4915         adapter->tx_eitr_param = 10000;
4916
4917         /* set defaults for eitr in MegaBytes */
4918         adapter->eitr_low = 10;
4919         adapter->eitr_high = 20;
4920
4921         /* set default ring sizes */
4922         adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
4923         adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
4924
4925         /* initialize eeprom parameters */
4926         if (ixgbe_init_eeprom_params_generic(hw)) {
4927                 e_dev_err("EEPROM initialization failed\n");
4928                 return -EIO;
4929         }
4930
4931         /* enable rx csum by default */
4932         adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
4933
4934         /* get assigned NUMA node */
4935         adapter->node = dev_to_node(&pdev->dev);
4936
4937         set_bit(__IXGBE_DOWN, &adapter->state);
4938
4939         return 0;
4940 }
4941
4942 /**
4943  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
4944  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
4945  *
4946  * Return 0 on success, negative on failure
4947  **/
4948 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
4949 {
4950         struct device *dev = tx_ring->dev;
4951         int size;
4952
4953         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
4954         tx_ring->tx_buffer_info = vmalloc_node(size, tx_ring->numa_node);
4955         if (!tx_ring->tx_buffer_info)
4956                 tx_ring->tx_buffer_info = vmalloc(size);
4957         if (!tx_ring->tx_buffer_info)
4958                 goto err;
4959         memset(tx_ring->tx_buffer_info, 0, size);
4960
4961         /* round up to nearest 4K */
4962         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
4963         tx_ring->size = ALIGN(tx_ring->size, 4096);
4964
4965         tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
4966                                            &tx_ring->dma, GFP_KERNEL);
4967         if (!tx_ring->desc)
4968                 goto err;
4969
4970         tx_ring->next_to_use = 0;
4971         tx_ring->next_to_clean = 0;
4972         tx_ring->work_limit = tx_ring->count;
4973         return 0;
4974
4975 err:
4976         vfree(tx_ring->tx_buffer_info);
4977         tx_ring->tx_buffer_info = NULL;
4978         dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
4979         return -ENOMEM;
4980 }
4981
4982 /**
4983  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
4984  * @adapter: board private structure
4985  *
4986  * If this function returns with an error, then it's possible one or
4987  * more of the rings is populated (while the rest are not).  It is the
4988  * callers duty to clean those orphaned rings.
4989  *
4990  * Return 0 on success, negative on failure
4991  **/
4992 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
4993 {
4994         int i, err = 0;
4995
4996         for (i = 0; i < adapter->num_tx_queues; i++) {
4997                 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
4998                 if (!err)
4999                         continue;
5000                 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5001                 break;
5002         }
5003
5004         return err;
5005 }
5006
5007 /**
5008  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5009  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
5010  *
5011  * Returns 0 on success, negative on failure
5012  **/
5013 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5014 {
5015         struct device *dev = rx_ring->dev;
5016         int size;
5017
5018         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5019         rx_ring->rx_buffer_info = vmalloc_node(size, rx_ring->numa_node);
5020         if (!rx_ring->rx_buffer_info)
5021                 rx_ring->rx_buffer_info = vmalloc(size);
5022         if (!rx_ring->rx_buffer_info)
5023                 goto err;
5024         memset(rx_ring->rx_buffer_info, 0, size);
5025
5026         /* Round up to nearest 4K */
5027         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5028         rx_ring->size = ALIGN(rx_ring->size, 4096);
5029
5030         rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5031                                            &rx_ring->dma, GFP_KERNEL);
5032
5033         if (!rx_ring->desc)
5034                 goto err;
5035
5036         rx_ring->next_to_clean = 0;
5037         rx_ring->next_to_use = 0;
5038
5039         return 0;
5040 err:
5041         vfree(rx_ring->rx_buffer_info);
5042         rx_ring->rx_buffer_info = NULL;
5043         dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5044         return -ENOMEM;
5045 }
5046
5047 /**
5048  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5049  * @adapter: board private structure
5050  *
5051  * If this function returns with an error, then it's possible one or
5052  * more of the rings is populated (while the rest are not).  It is the
5053  * callers duty to clean those orphaned rings.
5054  *
5055  * Return 0 on success, negative on failure
5056  **/
5057 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5058 {
5059         int i, err = 0;
5060
5061         for (i = 0; i < adapter->num_rx_queues; i++) {
5062                 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5063                 if (!err)
5064                         continue;
5065                 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5066                 break;
5067         }
5068
5069         return err;
5070 }
5071
5072 /**
5073  * ixgbe_free_tx_resources - Free Tx Resources per Queue
5074  * @tx_ring: Tx descriptor ring for a specific queue
5075  *
5076  * Free all transmit software resources
5077  **/
5078 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5079 {
5080         ixgbe_clean_tx_ring(tx_ring);
5081
5082         vfree(tx_ring->tx_buffer_info);
5083         tx_ring->tx_buffer_info = NULL;
5084
5085         /* if not set, then don't free */
5086         if (!tx_ring->desc)
5087                 return;
5088
5089         dma_free_coherent(tx_ring->dev, tx_ring->size,
5090                           tx_ring->desc, tx_ring->dma);
5091
5092         tx_ring->desc = NULL;
5093 }
5094
5095 /**
5096  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5097  * @adapter: board private structure
5098  *
5099  * Free all transmit software resources
5100  **/
5101 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5102 {
5103         int i;
5104
5105         for (i = 0; i < adapter->num_tx_queues; i++)
5106                 if (adapter->tx_ring[i]->desc)
5107                         ixgbe_free_tx_resources(adapter->tx_ring[i]);
5108 }
5109
5110 /**
5111  * ixgbe_free_rx_resources - Free Rx Resources
5112  * @rx_ring: ring to clean the resources from
5113  *
5114  * Free all receive software resources
5115  **/
5116 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5117 {
5118         ixgbe_clean_rx_ring(rx_ring);
5119
5120         vfree(rx_ring->rx_buffer_info);
5121         rx_ring->rx_buffer_info = NULL;
5122
5123         /* if not set, then don't free */
5124         if (!rx_ring->desc)
5125                 return;
5126
5127         dma_free_coherent(rx_ring->dev, rx_ring->size,
5128                           rx_ring->desc, rx_ring->dma);
5129
5130         rx_ring->desc = NULL;
5131 }
5132
5133 /**
5134  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5135  * @adapter: board private structure
5136  *
5137  * Free all receive software resources
5138  **/
5139 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5140 {
5141         int i;
5142
5143         for (i = 0; i < adapter->num_rx_queues; i++)
5144                 if (adapter->rx_ring[i]->desc)
5145                         ixgbe_free_rx_resources(adapter->rx_ring[i]);
5146 }
5147
5148 /**
5149  * ixgbe_change_mtu - Change the Maximum Transfer Unit
5150  * @netdev: network interface device structure
5151  * @new_mtu: new value for maximum frame size
5152  *
5153  * Returns 0 on success, negative on failure
5154  **/
5155 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5156 {
5157         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5158         struct ixgbe_hw *hw = &adapter->hw;
5159         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5160
5161         /* MTU < 68 is an error and causes problems on some kernels */
5162         if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5163                 return -EINVAL;
5164
5165         e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5166         /* must set new MTU before calling down or up */
5167         netdev->mtu = new_mtu;
5168
5169         hw->fc.high_water = FC_HIGH_WATER(max_frame);
5170         hw->fc.low_water = FC_LOW_WATER(max_frame);
5171
5172         if (netif_running(netdev))
5173                 ixgbe_reinit_locked(adapter);
5174
5175         return 0;
5176 }
5177
5178 /**
5179  * ixgbe_open - Called when a network interface is made active
5180  * @netdev: network interface device structure
5181  *
5182  * Returns 0 on success, negative value on failure
5183  *
5184  * The open entry point is called when a network interface is made
5185  * active by the system (IFF_UP).  At this point all resources needed
5186  * for transmit and receive operations are allocated, the interrupt
5187  * handler is registered with the OS, the watchdog timer is started,
5188  * and the stack is notified that the interface is ready.
5189  **/
5190 static int ixgbe_open(struct net_device *netdev)
5191 {
5192         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5193         int err;
5194
5195         /* disallow open during test */
5196         if (test_bit(__IXGBE_TESTING, &adapter->state))
5197                 return -EBUSY;
5198
5199         netif_carrier_off(netdev);
5200
5201         /* allocate transmit descriptors */
5202         err = ixgbe_setup_all_tx_resources(adapter);
5203         if (err)
5204                 goto err_setup_tx;
5205
5206         /* allocate receive descriptors */
5207         err = ixgbe_setup_all_rx_resources(adapter);
5208         if (err)
5209                 goto err_setup_rx;
5210
5211         ixgbe_configure(adapter);
5212
5213         err = ixgbe_request_irq(adapter);
5214         if (err)
5215                 goto err_req_irq;
5216
5217         err = ixgbe_up_complete(adapter);
5218         if (err)
5219                 goto err_up;
5220
5221         netif_tx_start_all_queues(netdev);
5222
5223         return 0;
5224
5225 err_up:
5226         ixgbe_release_hw_control(adapter);
5227         ixgbe_free_irq(adapter);
5228 err_req_irq:
5229 err_setup_rx:
5230         ixgbe_free_all_rx_resources(adapter);
5231 err_setup_tx:
5232         ixgbe_free_all_tx_resources(adapter);
5233         ixgbe_reset(adapter);
5234
5235         return err;
5236 }
5237
5238 /**
5239  * ixgbe_close - Disables a network interface
5240  * @netdev: network interface device structure
5241  *
5242  * Returns 0, this is not allowed to fail
5243  *
5244  * The close entry point is called when an interface is de-activated
5245  * by the OS.  The hardware is still under the drivers control, but
5246  * needs to be disabled.  A global MAC reset is issued to stop the
5247  * hardware, and all transmit and receive resources are freed.
5248  **/
5249 static int ixgbe_close(struct net_device *netdev)
5250 {
5251         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5252
5253         ixgbe_down(adapter);
5254         ixgbe_free_irq(adapter);
5255
5256         ixgbe_free_all_tx_resources(adapter);
5257         ixgbe_free_all_rx_resources(adapter);
5258
5259         ixgbe_release_hw_control(adapter);
5260
5261         return 0;
5262 }
5263
5264 #ifdef CONFIG_PM
5265 static int ixgbe_resume(struct pci_dev *pdev)
5266 {
5267         struct net_device *netdev = pci_get_drvdata(pdev);
5268         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5269         u32 err;
5270
5271         pci_set_power_state(pdev, PCI_D0);
5272         pci_restore_state(pdev);
5273         /*
5274          * pci_restore_state clears dev->state_saved so call
5275          * pci_save_state to restore it.
5276          */
5277         pci_save_state(pdev);
5278
5279         err = pci_enable_device_mem(pdev);
5280         if (err) {
5281                 e_dev_err("Cannot enable PCI device from suspend\n");
5282                 return err;
5283         }
5284         pci_set_master(pdev);
5285
5286         pci_wake_from_d3(pdev, false);
5287
5288         err = ixgbe_init_interrupt_scheme(adapter);
5289         if (err) {
5290                 e_dev_err("Cannot initialize interrupts for device\n");
5291                 return err;
5292         }
5293
5294         ixgbe_reset(adapter);
5295
5296         IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5297
5298         if (netif_running(netdev)) {
5299                 err = ixgbe_open(adapter->netdev);
5300                 if (err)
5301                         return err;
5302         }
5303
5304         netif_device_attach(netdev);
5305
5306         return 0;
5307 }
5308 #endif /* CONFIG_PM */
5309
5310 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5311 {
5312         struct net_device *netdev = pci_get_drvdata(pdev);
5313         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5314         struct ixgbe_hw *hw = &adapter->hw;
5315         u32 ctrl, fctrl;
5316         u32 wufc = adapter->wol;
5317 #ifdef CONFIG_PM
5318         int retval = 0;
5319 #endif
5320
5321         netif_device_detach(netdev);
5322
5323         if (netif_running(netdev)) {
5324                 ixgbe_down(adapter);
5325                 ixgbe_free_irq(adapter);
5326                 ixgbe_free_all_tx_resources(adapter);
5327                 ixgbe_free_all_rx_resources(adapter);
5328         }
5329
5330         ixgbe_clear_interrupt_scheme(adapter);
5331
5332 #ifdef CONFIG_PM
5333         retval = pci_save_state(pdev);
5334         if (retval)
5335                 return retval;
5336
5337 #endif
5338         if (wufc) {
5339                 ixgbe_set_rx_mode(netdev);
5340
5341                 /* turn on all-multi mode if wake on multicast is enabled */
5342                 if (wufc & IXGBE_WUFC_MC) {
5343                         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5344                         fctrl |= IXGBE_FCTRL_MPE;
5345                         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5346                 }
5347
5348                 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5349                 ctrl |= IXGBE_CTRL_GIO_DIS;
5350                 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5351
5352                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5353         } else {
5354                 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5355                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5356         }
5357
5358         if (wufc && hw->mac.type == ixgbe_mac_82599EB)
5359                 pci_wake_from_d3(pdev, true);
5360         else
5361                 pci_wake_from_d3(pdev, false);
5362
5363         *enable_wake = !!wufc;
5364
5365         ixgbe_release_hw_control(adapter);
5366
5367         pci_disable_device(pdev);
5368
5369         return 0;
5370 }
5371
5372 #ifdef CONFIG_PM
5373 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5374 {
5375         int retval;
5376         bool wake;
5377
5378         retval = __ixgbe_shutdown(pdev, &wake);
5379         if (retval)
5380                 return retval;
5381
5382         if (wake) {
5383                 pci_prepare_to_sleep(pdev);
5384         } else {
5385                 pci_wake_from_d3(pdev, false);
5386                 pci_set_power_state(pdev, PCI_D3hot);
5387         }
5388
5389         return 0;
5390 }
5391 #endif /* CONFIG_PM */
5392
5393 static void ixgbe_shutdown(struct pci_dev *pdev)
5394 {
5395         bool wake;
5396
5397         __ixgbe_shutdown(pdev, &wake);
5398
5399         if (system_state == SYSTEM_POWER_OFF) {
5400                 pci_wake_from_d3(pdev, wake);
5401                 pci_set_power_state(pdev, PCI_D3hot);
5402         }
5403 }
5404
5405 /**
5406  * ixgbe_update_stats - Update the board statistics counters.
5407  * @adapter: board private structure
5408  **/
5409 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5410 {
5411         struct net_device *netdev = adapter->netdev;
5412         struct ixgbe_hw *hw = &adapter->hw;
5413         struct ixgbe_hw_stats *hwstats = &adapter->stats;
5414         u64 total_mpc = 0;
5415         u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5416         u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5417         u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5418         u64 bytes = 0, packets = 0;
5419
5420         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5421             test_bit(__IXGBE_RESETTING, &adapter->state))
5422                 return;
5423
5424         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5425                 u64 rsc_count = 0;
5426                 u64 rsc_flush = 0;
5427                 for (i = 0; i < 16; i++)
5428                         adapter->hw_rx_no_dma_resources +=
5429                                 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5430                 for (i = 0; i < adapter->num_rx_queues; i++) {
5431                         rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5432                         rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5433                 }
5434                 adapter->rsc_total_count = rsc_count;
5435                 adapter->rsc_total_flush = rsc_flush;
5436         }
5437
5438         for (i = 0; i < adapter->num_rx_queues; i++) {
5439                 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5440                 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5441                 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5442                 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5443                 bytes += rx_ring->stats.bytes;
5444                 packets += rx_ring->stats.packets;
5445         }
5446         adapter->non_eop_descs = non_eop_descs;
5447         adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5448         adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5449         netdev->stats.rx_bytes = bytes;
5450         netdev->stats.rx_packets = packets;
5451
5452         bytes = 0;
5453         packets = 0;
5454         /* gather some stats to the adapter struct that are per queue */
5455         for (i = 0; i < adapter->num_tx_queues; i++) {
5456                 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5457                 restart_queue += tx_ring->tx_stats.restart_queue;
5458                 tx_busy += tx_ring->tx_stats.tx_busy;
5459                 bytes += tx_ring->stats.bytes;
5460                 packets += tx_ring->stats.packets;
5461         }
5462         adapter->restart_queue = restart_queue;
5463         adapter->tx_busy = tx_busy;
5464         netdev->stats.tx_bytes = bytes;
5465         netdev->stats.tx_packets = packets;
5466
5467         hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5468         for (i = 0; i < 8; i++) {
5469                 /* for packet buffers not used, the register should read 0 */
5470                 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5471                 missed_rx += mpc;
5472                 hwstats->mpc[i] += mpc;
5473                 total_mpc += hwstats->mpc[i];
5474                 if (hw->mac.type == ixgbe_mac_82598EB)
5475                         hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5476                 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5477                 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5478                 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5479                 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5480                 if (hw->mac.type == ixgbe_mac_82599EB) {
5481                         hwstats->pxonrxc[i] +=
5482                                 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5483                         hwstats->pxoffrxc[i] +=
5484                                 IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
5485                         hwstats->qprdc[i] += IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5486                 } else {
5487                         hwstats->pxonrxc[i] +=
5488                                 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5489                         hwstats->pxoffrxc[i] +=
5490                                 IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
5491                 }
5492                 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5493                 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5494         }
5495         hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5496         /* work around hardware counting issue */
5497         hwstats->gprc -= missed_rx;
5498
5499         /* 82598 hardware only has a 32 bit counter in the high register */
5500         if (hw->mac.type == ixgbe_mac_82599EB) {
5501                 u64 tmp;
5502                 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5503                 tmp = IXGBE_READ_REG(hw, IXGBE_GORCH) & 0xF;
5504                                                 /* 4 high bits of GORC */
5505                 hwstats->gorc += (tmp << 32);
5506                 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5507                 tmp = IXGBE_READ_REG(hw, IXGBE_GOTCH) & 0xF;
5508                                                 /* 4 high bits of GOTC */
5509                 hwstats->gotc += (tmp << 32);
5510                 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5511                 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5512                 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5513                 hwstats->lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
5514                 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5515                 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5516 #ifdef IXGBE_FCOE
5517                 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5518                 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5519                 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5520                 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5521                 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5522                 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5523 #endif /* IXGBE_FCOE */
5524         } else {
5525                 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5526                 hwstats->lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
5527                 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5528                 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5529                 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5530         }
5531         bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5532         hwstats->bprc += bprc;
5533         hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5534         if (hw->mac.type == ixgbe_mac_82598EB)
5535                 hwstats->mprc -= bprc;
5536         hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5537         hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5538         hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5539         hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5540         hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5541         hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5542         hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5543         hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5544         lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5545         hwstats->lxontxc += lxon;
5546         lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5547         hwstats->lxofftxc += lxoff;
5548         hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5549         hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5550         hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5551         /*
5552          * 82598 errata - tx of flow control packets is included in tx counters
5553          */
5554         xon_off_tot = lxon + lxoff;
5555         hwstats->gptc -= xon_off_tot;
5556         hwstats->mptc -= xon_off_tot;
5557         hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5558         hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5559         hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5560         hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5561         hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5562         hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5563         hwstats->ptc64 -= xon_off_tot;
5564         hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5565         hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5566         hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5567         hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5568         hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5569         hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5570
5571         /* Fill out the OS statistics structure */
5572         netdev->stats.multicast = hwstats->mprc;
5573
5574         /* Rx Errors */
5575         netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5576         netdev->stats.rx_dropped = 0;
5577         netdev->stats.rx_length_errors = hwstats->rlec;
5578         netdev->stats.rx_crc_errors = hwstats->crcerrs;
5579         netdev->stats.rx_missed_errors = total_mpc;
5580 }
5581
5582 /**
5583  * ixgbe_watchdog - Timer Call-back
5584  * @data: pointer to adapter cast into an unsigned long
5585  **/
5586 static void ixgbe_watchdog(unsigned long data)
5587 {
5588         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5589         struct ixgbe_hw *hw = &adapter->hw;
5590         u64 eics = 0;
5591         int i;
5592
5593         /*
5594          *  Do the watchdog outside of interrupt context due to the lovely
5595          * delays that some of the newer hardware requires
5596          */
5597
5598         if (test_bit(__IXGBE_DOWN, &adapter->state))
5599                 goto watchdog_short_circuit;
5600
5601         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5602                 /*
5603                  * for legacy and MSI interrupts don't set any bits
5604                  * that are enabled for EIAM, because this operation
5605                  * would set *both* EIMS and EICS for any bit in EIAM
5606                  */
5607                 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5608                         (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5609                 goto watchdog_reschedule;
5610         }
5611
5612         /* get one bit for every active tx/rx interrupt vector */
5613         for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5614                 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5615                 if (qv->rxr_count || qv->txr_count)
5616                         eics |= ((u64)1 << i);
5617         }
5618
5619         /* Cause software interrupt to ensure rx rings are cleaned */
5620         ixgbe_irq_rearm_queues(adapter, eics);
5621
5622 watchdog_reschedule:
5623         /* Reset the timer */
5624         mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
5625
5626 watchdog_short_circuit:
5627         schedule_work(&adapter->watchdog_task);
5628 }
5629
5630 /**
5631  * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
5632  * @work: pointer to work_struct containing our data
5633  **/
5634 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
5635 {
5636         struct ixgbe_adapter *adapter = container_of(work,
5637                                                      struct ixgbe_adapter,
5638                                                      multispeed_fiber_task);
5639         struct ixgbe_hw *hw = &adapter->hw;
5640         u32 autoneg;
5641         bool negotiation;
5642
5643         adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
5644         autoneg = hw->phy.autoneg_advertised;
5645         if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5646                 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5647         hw->mac.autotry_restart = false;
5648         if (hw->mac.ops.setup_link)
5649                 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5650         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5651         adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
5652 }
5653
5654 /**
5655  * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
5656  * @work: pointer to work_struct containing our data
5657  **/
5658 static void ixgbe_sfp_config_module_task(struct work_struct *work)
5659 {
5660         struct ixgbe_adapter *adapter = container_of(work,
5661                                                      struct ixgbe_adapter,
5662                                                      sfp_config_module_task);
5663         struct ixgbe_hw *hw = &adapter->hw;
5664         u32 err;
5665
5666         adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
5667
5668         /* Time for electrical oscillations to settle down */
5669         msleep(100);
5670         err = hw->phy.ops.identify_sfp(hw);
5671
5672         if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5673                 e_dev_err("failed to initialize because an unsupported SFP+ "
5674                           "module type was detected.\n");
5675                 e_dev_err("Reload the driver after installing a supported "
5676                           "module.\n");
5677                 unregister_netdev(adapter->netdev);
5678                 return;
5679         }
5680         hw->mac.ops.setup_sfp(hw);
5681
5682         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
5683                 /* This will also work for DA Twinax connections */
5684                 schedule_work(&adapter->multispeed_fiber_task);
5685         adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
5686 }
5687
5688 /**
5689  * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
5690  * @work: pointer to work_struct containing our data
5691  **/
5692 static void ixgbe_fdir_reinit_task(struct work_struct *work)
5693 {
5694         struct ixgbe_adapter *adapter = container_of(work,
5695                                                      struct ixgbe_adapter,
5696                                                      fdir_reinit_task);
5697         struct ixgbe_hw *hw = &adapter->hw;
5698         int i;
5699
5700         if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5701                 for (i = 0; i < adapter->num_tx_queues; i++)
5702                         set_bit(__IXGBE_FDIR_INIT_DONE,
5703                                 &(adapter->tx_ring[i]->reinit_state));
5704         } else {
5705                 e_err(probe, "failed to finish FDIR re-initialization, "
5706                       "ignored adding FDIR ATR filters\n");
5707         }
5708         /* Done FDIR Re-initialization, enable transmits */
5709         netif_tx_start_all_queues(adapter->netdev);
5710 }
5711
5712 static DEFINE_MUTEX(ixgbe_watchdog_lock);
5713
5714 /**
5715  * ixgbe_watchdog_task - worker thread to bring link up
5716  * @work: pointer to work_struct containing our data
5717  **/
5718 static void ixgbe_watchdog_task(struct work_struct *work)
5719 {
5720         struct ixgbe_adapter *adapter = container_of(work,
5721                                                      struct ixgbe_adapter,
5722                                                      watchdog_task);
5723         struct net_device *netdev = adapter->netdev;
5724         struct ixgbe_hw *hw = &adapter->hw;
5725         u32 link_speed;
5726         bool link_up;
5727         int i;
5728         struct ixgbe_ring *tx_ring;
5729         int some_tx_pending = 0;
5730
5731         mutex_lock(&ixgbe_watchdog_lock);
5732
5733         link_up = adapter->link_up;
5734         link_speed = adapter->link_speed;
5735
5736         if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
5737                 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5738                 if (link_up) {
5739 #ifdef CONFIG_DCB
5740                         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5741                                 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5742                                         hw->mac.ops.fc_enable(hw, i);
5743                         } else {
5744                                 hw->mac.ops.fc_enable(hw, 0);
5745                         }
5746 #else
5747                         hw->mac.ops.fc_enable(hw, 0);
5748 #endif
5749                 }
5750
5751                 if (link_up ||
5752                     time_after(jiffies, (adapter->link_check_timeout +
5753                                          IXGBE_TRY_LINK_TIMEOUT))) {
5754                         adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5755                         IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
5756                 }
5757                 adapter->link_up = link_up;
5758                 adapter->link_speed = link_speed;
5759         }
5760
5761         if (link_up) {
5762                 if (!netif_carrier_ok(netdev)) {
5763                         bool flow_rx, flow_tx;
5764
5765                         if (hw->mac.type == ixgbe_mac_82599EB) {
5766                                 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
5767                                 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
5768                                 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
5769                                 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
5770                         } else {
5771                                 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5772                                 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
5773                                 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
5774                                 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
5775                         }
5776
5777                         e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
5778                                (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
5779                                "10 Gbps" :
5780                                (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
5781                                "1 Gbps" : "unknown speed")),
5782                                ((flow_rx && flow_tx) ? "RX/TX" :
5783                                (flow_rx ? "RX" :
5784                                (flow_tx ? "TX" : "None"))));
5785
5786                         netif_carrier_on(netdev);
5787                 } else {
5788                         /* Force detection of hung controller */
5789                         adapter->detect_tx_hung = true;
5790                 }
5791         } else {
5792                 adapter->link_up = false;
5793                 adapter->link_speed = 0;
5794                 if (netif_carrier_ok(netdev)) {
5795                         e_info(drv, "NIC Link is Down\n");
5796                         netif_carrier_off(netdev);
5797                 }
5798         }
5799
5800         if (!netif_carrier_ok(netdev)) {
5801                 for (i = 0; i < adapter->num_tx_queues; i++) {
5802                         tx_ring = adapter->tx_ring[i];
5803                         if (tx_ring->next_to_use != tx_ring->next_to_clean) {
5804                                 some_tx_pending = 1;
5805                                 break;
5806                         }
5807                 }
5808
5809                 if (some_tx_pending) {
5810                         /* We've lost link, so the controller stops DMA,
5811                          * but we've got queued Tx work that's never going
5812                          * to get done, so reset controller to flush Tx.
5813                          * (Do the reset outside of interrupt context).
5814                          */
5815                          schedule_work(&adapter->reset_task);
5816                 }
5817         }
5818
5819         ixgbe_update_stats(adapter);
5820         mutex_unlock(&ixgbe_watchdog_lock);
5821 }
5822
5823 static int ixgbe_tso(struct ixgbe_adapter *adapter,
5824                      struct ixgbe_ring *tx_ring, struct sk_buff *skb,
5825                      u32 tx_flags, u8 *hdr_len, __be16 protocol)
5826 {
5827         struct ixgbe_adv_tx_context_desc *context_desc;
5828         unsigned int i;
5829         int err;
5830         struct ixgbe_tx_buffer *tx_buffer_info;
5831         u32 vlan_macip_lens = 0, type_tucmd_mlhl;
5832         u32 mss_l4len_idx, l4len;
5833
5834         if (skb_is_gso(skb)) {
5835                 if (skb_header_cloned(skb)) {
5836                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5837                         if (err)
5838                                 return err;
5839                 }
5840                 l4len = tcp_hdrlen(skb);
5841                 *hdr_len += l4len;
5842
5843                 if (protocol == htons(ETH_P_IP)) {
5844                         struct iphdr *iph = ip_hdr(skb);
5845                         iph->tot_len = 0;
5846                         iph->check = 0;
5847                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
5848                                                                  iph->daddr, 0,
5849                                                                  IPPROTO_TCP,
5850                                                                  0);
5851                 } else if (skb_is_gso_v6(skb)) {
5852                         ipv6_hdr(skb)->payload_len = 0;
5853                         tcp_hdr(skb)->check =
5854                             ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5855                                              &ipv6_hdr(skb)->daddr,
5856                                              0, IPPROTO_TCP, 0);
5857                 }
5858
5859                 i = tx_ring->next_to_use;
5860
5861                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5862                 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
5863
5864                 /* VLAN MACLEN IPLEN */
5865                 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5866                         vlan_macip_lens |=
5867                             (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5868                 vlan_macip_lens |= ((skb_network_offset(skb)) <<
5869                                     IXGBE_ADVTXD_MACLEN_SHIFT);
5870                 *hdr_len += skb_network_offset(skb);
5871                 vlan_macip_lens |=
5872                     (skb_transport_header(skb) - skb_network_header(skb));
5873                 *hdr_len +=
5874                     (skb_transport_header(skb) - skb_network_header(skb));
5875                 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5876                 context_desc->seqnum_seed = 0;
5877
5878                 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5879                 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
5880                                    IXGBE_ADVTXD_DTYP_CTXT);
5881
5882                 if (protocol == htons(ETH_P_IP))
5883                         type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
5884                 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5885                 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5886
5887                 /* MSS L4LEN IDX */
5888                 mss_l4len_idx =
5889                     (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
5890                 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
5891                 /* use index 1 for TSO */
5892                 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
5893                 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
5894
5895                 tx_buffer_info->time_stamp = jiffies;
5896                 tx_buffer_info->next_to_watch = i;
5897
5898                 i++;
5899                 if (i == tx_ring->count)
5900                         i = 0;
5901                 tx_ring->next_to_use = i;
5902
5903                 return true;
5904         }
5905         return false;
5906 }
5907
5908 static u32 ixgbe_psum(struct ixgbe_adapter *adapter, struct sk_buff *skb,
5909                       __be16 protocol)
5910 {
5911         u32 rtn = 0;
5912
5913         switch (protocol) {
5914         case cpu_to_be16(ETH_P_IP):
5915                 rtn |= IXGBE_ADVTXD_TUCMD_IPV4;
5916                 switch (ip_hdr(skb)->protocol) {
5917                 case IPPROTO_TCP:
5918                         rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5919                         break;
5920                 case IPPROTO_SCTP:
5921                         rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5922                         break;
5923                 }
5924                 break;
5925         case cpu_to_be16(ETH_P_IPV6):
5926                 /* XXX what about other V6 headers?? */
5927                 switch (ipv6_hdr(skb)->nexthdr) {
5928                 case IPPROTO_TCP:
5929                         rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
5930                         break;
5931                 case IPPROTO_SCTP:
5932                         rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
5933                         break;
5934                 }
5935                 break;
5936         default:
5937                 if (unlikely(net_ratelimit()))
5938                         e_warn(probe, "partial checksum but proto=%x!\n",
5939                                protocol);
5940                 break;
5941         }
5942
5943         return rtn;
5944 }
5945
5946 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
5947                           struct ixgbe_ring *tx_ring,
5948                           struct sk_buff *skb, u32 tx_flags,
5949                           __be16 protocol)
5950 {
5951         struct ixgbe_adv_tx_context_desc *context_desc;
5952         unsigned int i;
5953         struct ixgbe_tx_buffer *tx_buffer_info;
5954         u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
5955
5956         if (skb->ip_summed == CHECKSUM_PARTIAL ||
5957             (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
5958                 i = tx_ring->next_to_use;
5959                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5960                 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
5961
5962                 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
5963                         vlan_macip_lens |=
5964                             (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
5965                 vlan_macip_lens |= (skb_network_offset(skb) <<
5966                                     IXGBE_ADVTXD_MACLEN_SHIFT);
5967                 if (skb->ip_summed == CHECKSUM_PARTIAL)
5968                         vlan_macip_lens |= (skb_transport_header(skb) -
5969                                             skb_network_header(skb));
5970
5971                 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
5972                 context_desc->seqnum_seed = 0;
5973
5974                 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
5975                                     IXGBE_ADVTXD_DTYP_CTXT);
5976
5977                 if (skb->ip_summed == CHECKSUM_PARTIAL)
5978                         type_tucmd_mlhl |= ixgbe_psum(adapter, skb, protocol);
5979
5980                 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
5981                 /* use index zero for tx checksum offload */
5982                 context_desc->mss_l4len_idx = 0;
5983
5984                 tx_buffer_info->time_stamp = jiffies;
5985                 tx_buffer_info->next_to_watch = i;
5986
5987                 i++;
5988                 if (i == tx_ring->count)
5989                         i = 0;
5990                 tx_ring->next_to_use = i;
5991
5992                 return true;
5993         }
5994
5995         return false;
5996 }
5997
5998 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
5999                         struct ixgbe_ring *tx_ring,
6000                         struct sk_buff *skb, u32 tx_flags,
6001                         unsigned int first, const u8 hdr_len)
6002 {
6003         struct device *dev = tx_ring->dev;
6004         struct ixgbe_tx_buffer *tx_buffer_info;
6005         unsigned int len;
6006         unsigned int total = skb->len;
6007         unsigned int offset = 0, size, count = 0, i;
6008         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
6009         unsigned int f;
6010         unsigned int bytecount = skb->len;
6011         u16 gso_segs = 1;
6012
6013         i = tx_ring->next_to_use;
6014
6015         if (tx_flags & IXGBE_TX_FLAGS_FCOE)
6016                 /* excluding fcoe_crc_eof for FCoE */
6017                 total -= sizeof(struct fcoe_crc_eof);
6018
6019         len = min(skb_headlen(skb), total);
6020         while (len) {
6021                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6022                 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6023
6024                 tx_buffer_info->length = size;
6025                 tx_buffer_info->mapped_as_page = false;
6026                 tx_buffer_info->dma = dma_map_single(dev,
6027                                                      skb->data + offset,
6028                                                      size, DMA_TO_DEVICE);
6029                 if (dma_mapping_error(dev, tx_buffer_info->dma))
6030                         goto dma_error;
6031                 tx_buffer_info->time_stamp = jiffies;
6032                 tx_buffer_info->next_to_watch = i;
6033
6034                 len -= size;
6035                 total -= size;
6036                 offset += size;
6037                 count++;
6038
6039                 if (len) {
6040                         i++;
6041                         if (i == tx_ring->count)
6042                                 i = 0;
6043                 }
6044         }
6045
6046         for (f = 0; f < nr_frags; f++) {
6047                 struct skb_frag_struct *frag;
6048
6049                 frag = &skb_shinfo(skb)->frags[f];
6050                 len = min((unsigned int)frag->size, total);
6051                 offset = frag->page_offset;
6052
6053                 while (len) {
6054                         i++;
6055                         if (i == tx_ring->count)
6056                                 i = 0;
6057
6058                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
6059                         size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6060
6061                         tx_buffer_info->length = size;
6062                         tx_buffer_info->dma = dma_map_page(dev,
6063                                                            frag->page,
6064                                                            offset, size,
6065                                                            DMA_TO_DEVICE);
6066                         tx_buffer_info->mapped_as_page = true;
6067                         if (dma_mapping_error(dev, tx_buffer_info->dma))
6068                                 goto dma_error;
6069                         tx_buffer_info->time_stamp = jiffies;
6070                         tx_buffer_info->next_to_watch = i;
6071
6072                         len -= size;
6073                         total -= size;
6074                         offset += size;
6075                         count++;
6076                 }
6077                 if (total == 0)
6078                         break;
6079         }
6080
6081         if (tx_flags & IXGBE_TX_FLAGS_TSO)
6082                 gso_segs = skb_shinfo(skb)->gso_segs;
6083 #ifdef IXGBE_FCOE
6084         /* adjust for FCoE Sequence Offload */
6085         else if (tx_flags & IXGBE_TX_FLAGS_FSO)
6086                 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
6087                                         skb_shinfo(skb)->gso_size);
6088 #endif /* IXGBE_FCOE */
6089         bytecount += (gso_segs - 1) * hdr_len;
6090
6091         /* multiply data chunks by size of headers */
6092         tx_ring->tx_buffer_info[i].bytecount = bytecount;
6093         tx_ring->tx_buffer_info[i].gso_segs = gso_segs;
6094         tx_ring->tx_buffer_info[i].skb = skb;
6095         tx_ring->tx_buffer_info[first].next_to_watch = i;
6096
6097         return count;
6098
6099 dma_error:
6100         e_dev_err("TX DMA map failed\n");
6101
6102         /* clear timestamp and dma mappings for failed tx_buffer_info map */
6103         tx_buffer_info->dma = 0;
6104         tx_buffer_info->time_stamp = 0;
6105         tx_buffer_info->next_to_watch = 0;
6106         if (count)
6107                 count--;
6108
6109         /* clear timestamp and dma mappings for remaining portion of packet */
6110         while (count--) {
6111                 if (i == 0)
6112                         i += tx_ring->count;
6113                 i--;
6114                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6115                 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
6116         }
6117
6118         return 0;
6119 }
6120
6121 static void ixgbe_tx_queue(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, tx_ring->tail);
6187 }
6188
6189 static void ixgbe_atr(struct ixgbe_adapter *adapter, struct sk_buff *skb,
6190                       int queue, u32 tx_flags, __be16 protocol)
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 (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 ixgbe_ring *tx_ring, int size)
6237 {
6238         netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6239         /* Herbert's original patch had:
6240          *  smp_mb__after_netif_stop_queue();
6241          * but since that doesn't exist yet, just open code it. */
6242         smp_mb();
6243
6244         /* We need to check again in a case another CPU has just
6245          * made room available. */
6246         if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
6247                 return -EBUSY;
6248
6249         /* A reprieve! - use start_queue because it doesn't call schedule */
6250         netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6251         ++tx_ring->tx_stats.restart_queue;
6252         return 0;
6253 }
6254
6255 static int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, int size)
6256 {
6257         if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
6258                 return 0;
6259         return __ixgbe_maybe_stop_tx(tx_ring, size);
6260 }
6261
6262 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6263 {
6264         struct ixgbe_adapter *adapter = netdev_priv(dev);
6265         int txq = smp_processor_id();
6266 #ifdef IXGBE_FCOE
6267         __be16 protocol;
6268
6269         protocol = vlan_get_protocol(skb);
6270
6271         if ((protocol == htons(ETH_P_FCOE)) ||
6272             (protocol == htons(ETH_P_FIP))) {
6273                 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
6274                         txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6275                         txq += adapter->ring_feature[RING_F_FCOE].mask;
6276                         return txq;
6277 #ifdef CONFIG_IXGBE_DCB
6278                 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6279                         txq = adapter->fcoe.up;
6280                         return txq;
6281 #endif
6282                 }
6283         }
6284 #endif
6285
6286         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6287                 while (unlikely(txq >= dev->real_num_tx_queues))
6288                         txq -= dev->real_num_tx_queues;
6289                 return txq;
6290         }
6291
6292         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6293                 if (skb->priority == TC_PRIO_CONTROL)
6294                         txq = adapter->ring_feature[RING_F_DCB].indices-1;
6295                 else
6296                         txq = (skb->vlan_tci & IXGBE_TX_FLAGS_VLAN_PRIO_MASK)
6297                                >> 13;
6298                 return txq;
6299         }
6300
6301         return skb_tx_hash(dev, skb);
6302 }
6303
6304 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6305                           struct ixgbe_adapter *adapter,
6306                           struct ixgbe_ring *tx_ring)
6307 {
6308         struct net_device *netdev = tx_ring->netdev;
6309         struct netdev_queue *txq;
6310         unsigned int first;
6311         unsigned int tx_flags = 0;
6312         u8 hdr_len = 0;
6313         int tso;
6314         int count = 0;
6315         unsigned int f;
6316         __be16 protocol;
6317
6318         protocol = vlan_get_protocol(skb);
6319
6320         if (vlan_tx_tag_present(skb)) {
6321                 tx_flags |= vlan_tx_tag_get(skb);
6322                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6323                         tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6324                         tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6325                 }
6326                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6327                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6328         } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED &&
6329                    skb->priority != TC_PRIO_CONTROL) {
6330                 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6331                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6332                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6333         }
6334
6335 #ifdef IXGBE_FCOE
6336         /* for FCoE with DCB, we force the priority to what
6337          * was specified by the switch */
6338         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED &&
6339             (protocol == htons(ETH_P_FCOE) ||
6340              protocol == htons(ETH_P_FIP))) {
6341 #ifdef CONFIG_IXGBE_DCB
6342                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6343                         tx_flags &= ~(IXGBE_TX_FLAGS_VLAN_PRIO_MASK
6344                                       << IXGBE_TX_FLAGS_VLAN_SHIFT);
6345                         tx_flags |= ((adapter->fcoe.up << 13)
6346                                       << IXGBE_TX_FLAGS_VLAN_SHIFT);
6347                 }
6348 #endif
6349                 /* flag for FCoE offloads */
6350                 if (protocol == htons(ETH_P_FCOE))
6351                         tx_flags |= IXGBE_TX_FLAGS_FCOE;
6352         }
6353 #endif
6354
6355         /* four things can cause us to need a context descriptor */
6356         if (skb_is_gso(skb) ||
6357             (skb->ip_summed == CHECKSUM_PARTIAL) ||
6358             (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
6359             (tx_flags & IXGBE_TX_FLAGS_FCOE))
6360                 count++;
6361
6362         count += TXD_USE_COUNT(skb_headlen(skb));
6363         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6364                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6365
6366         if (ixgbe_maybe_stop_tx(tx_ring, count)) {
6367                 tx_ring->tx_stats.tx_busy++;
6368                 return NETDEV_TX_BUSY;
6369         }
6370
6371         first = tx_ring->next_to_use;
6372         if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6373 #ifdef IXGBE_FCOE
6374                 /* setup tx offload for FCoE */
6375                 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
6376                 if (tso < 0) {
6377                         dev_kfree_skb_any(skb);
6378                         return NETDEV_TX_OK;
6379                 }
6380                 if (tso)
6381                         tx_flags |= IXGBE_TX_FLAGS_FSO;
6382 #endif /* IXGBE_FCOE */
6383         } else {
6384                 if (protocol == htons(ETH_P_IP))
6385                         tx_flags |= IXGBE_TX_FLAGS_IPV4;
6386                 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len,
6387                                 protocol);
6388                 if (tso < 0) {
6389                         dev_kfree_skb_any(skb);
6390                         return NETDEV_TX_OK;
6391                 }
6392
6393                 if (tso)
6394                         tx_flags |= IXGBE_TX_FLAGS_TSO;
6395                 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags,
6396                                        protocol) &&
6397                          (skb->ip_summed == CHECKSUM_PARTIAL))
6398                         tx_flags |= IXGBE_TX_FLAGS_CSUM;
6399         }
6400
6401         count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first, hdr_len);
6402         if (count) {
6403                 /* add the ATR filter if ATR is on */
6404                 if (tx_ring->atr_sample_rate) {
6405                         ++tx_ring->atr_count;
6406                         if ((tx_ring->atr_count >= tx_ring->atr_sample_rate) &&
6407                              test_bit(__IXGBE_FDIR_INIT_DONE,
6408                                       &tx_ring->reinit_state)) {
6409                                 ixgbe_atr(adapter, skb, tx_ring->queue_index,
6410                                           tx_flags, protocol);
6411                                 tx_ring->atr_count = 0;
6412                         }
6413                 }
6414                 txq = netdev_get_tx_queue(netdev, tx_ring->queue_index);
6415                 txq->tx_bytes += skb->len;
6416                 txq->tx_packets++;
6417                 ixgbe_tx_queue(tx_ring, tx_flags, count, skb->len, hdr_len);
6418                 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6419
6420         } else {
6421                 dev_kfree_skb_any(skb);
6422                 tx_ring->tx_buffer_info[first].time_stamp = 0;
6423                 tx_ring->next_to_use = first;
6424         }
6425
6426         return NETDEV_TX_OK;
6427 }
6428
6429 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
6430 {
6431         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6432         struct ixgbe_ring *tx_ring;
6433
6434         tx_ring = adapter->tx_ring[skb->queue_mapping];
6435         return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6436 }
6437
6438 /**
6439  * ixgbe_set_mac - Change the Ethernet Address of the NIC
6440  * @netdev: network interface device structure
6441  * @p: pointer to an address structure
6442  *
6443  * Returns 0 on success, negative on failure
6444  **/
6445 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6446 {
6447         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6448         struct ixgbe_hw *hw = &adapter->hw;
6449         struct sockaddr *addr = p;
6450
6451         if (!is_valid_ether_addr(addr->sa_data))
6452                 return -EADDRNOTAVAIL;
6453
6454         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6455         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6456
6457         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6458                             IXGBE_RAH_AV);
6459
6460         return 0;
6461 }
6462
6463 static int
6464 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6465 {
6466         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6467         struct ixgbe_hw *hw = &adapter->hw;
6468         u16 value;
6469         int rc;
6470
6471         if (prtad != hw->phy.mdio.prtad)
6472                 return -EINVAL;
6473         rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6474         if (!rc)
6475                 rc = value;
6476         return rc;
6477 }
6478
6479 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6480                             u16 addr, u16 value)
6481 {
6482         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6483         struct ixgbe_hw *hw = &adapter->hw;
6484
6485         if (prtad != hw->phy.mdio.prtad)
6486                 return -EINVAL;
6487         return hw->phy.ops.write_reg(hw, addr, devad, value);
6488 }
6489
6490 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6491 {
6492         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6493
6494         return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6495 }
6496
6497 /**
6498  * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6499  * netdev->dev_addrs
6500  * @netdev: network interface device structure
6501  *
6502  * Returns non-zero on failure
6503  **/
6504 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6505 {
6506         int err = 0;
6507         struct ixgbe_adapter *adapter = netdev_priv(dev);
6508         struct ixgbe_mac_info *mac = &adapter->hw.mac;
6509
6510         if (is_valid_ether_addr(mac->san_addr)) {
6511                 rtnl_lock();
6512                 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6513                 rtnl_unlock();
6514         }
6515         return err;
6516 }
6517
6518 /**
6519  * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6520  * netdev->dev_addrs
6521  * @netdev: network interface device structure
6522  *
6523  * Returns non-zero on failure
6524  **/
6525 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6526 {
6527         int err = 0;
6528         struct ixgbe_adapter *adapter = netdev_priv(dev);
6529         struct ixgbe_mac_info *mac = &adapter->hw.mac;
6530
6531         if (is_valid_ether_addr(mac->san_addr)) {
6532                 rtnl_lock();
6533                 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6534                 rtnl_unlock();
6535         }
6536         return err;
6537 }
6538
6539 #ifdef CONFIG_NET_POLL_CONTROLLER
6540 /*
6541  * Polling 'interrupt' - used by things like netconsole to send skbs
6542  * without having to re-enable interrupts. It's not called while
6543  * the interrupt routine is executing.
6544  */
6545 static void ixgbe_netpoll(struct net_device *netdev)
6546 {
6547         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6548         int i;
6549
6550         /* if interface is down do nothing */
6551         if (test_bit(__IXGBE_DOWN, &adapter->state))
6552                 return;
6553
6554         adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6555         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6556                 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
6557                 for (i = 0; i < num_q_vectors; i++) {
6558                         struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
6559                         ixgbe_msix_clean_many(0, q_vector);
6560                 }
6561         } else {
6562                 ixgbe_intr(adapter->pdev->irq, netdev);
6563         }
6564         adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6565 }
6566 #endif
6567
6568 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
6569                                                    struct rtnl_link_stats64 *stats)
6570 {
6571         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6572         int i;
6573
6574         /* accurate rx/tx bytes/packets stats */
6575         dev_txq_stats_fold(netdev, stats);
6576         rcu_read_lock();
6577         for (i = 0; i < adapter->num_rx_queues; i++) {
6578                 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
6579                 u64 bytes, packets;
6580                 unsigned int start;
6581
6582                 if (ring) {
6583                         do {
6584                                 start = u64_stats_fetch_begin_bh(&ring->syncp);
6585                                 packets = ring->stats.packets;
6586                                 bytes   = ring->stats.bytes;
6587                         } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6588                         stats->rx_packets += packets;
6589                         stats->rx_bytes   += bytes;
6590                 }
6591         }
6592         rcu_read_unlock();
6593         /* following stats updated by ixgbe_watchdog_task() */
6594         stats->multicast        = netdev->stats.multicast;
6595         stats->rx_errors        = netdev->stats.rx_errors;
6596         stats->rx_length_errors = netdev->stats.rx_length_errors;
6597         stats->rx_crc_errors    = netdev->stats.rx_crc_errors;
6598         stats->rx_missed_errors = netdev->stats.rx_missed_errors;
6599         return stats;
6600 }
6601
6602
6603 static const struct net_device_ops ixgbe_netdev_ops = {
6604         .ndo_open               = ixgbe_open,
6605         .ndo_stop               = ixgbe_close,
6606         .ndo_start_xmit         = ixgbe_xmit_frame,
6607         .ndo_select_queue       = ixgbe_select_queue,
6608         .ndo_set_rx_mode        = ixgbe_set_rx_mode,
6609         .ndo_set_multicast_list = ixgbe_set_rx_mode,
6610         .ndo_validate_addr      = eth_validate_addr,
6611         .ndo_set_mac_address    = ixgbe_set_mac,
6612         .ndo_change_mtu         = ixgbe_change_mtu,
6613         .ndo_tx_timeout         = ixgbe_tx_timeout,
6614         .ndo_vlan_rx_add_vid    = ixgbe_vlan_rx_add_vid,
6615         .ndo_vlan_rx_kill_vid   = ixgbe_vlan_rx_kill_vid,
6616         .ndo_do_ioctl           = ixgbe_ioctl,
6617         .ndo_set_vf_mac         = ixgbe_ndo_set_vf_mac,
6618         .ndo_set_vf_vlan        = ixgbe_ndo_set_vf_vlan,
6619         .ndo_set_vf_tx_rate     = ixgbe_ndo_set_vf_bw,
6620         .ndo_get_vf_config      = ixgbe_ndo_get_vf_config,
6621         .ndo_get_stats64        = ixgbe_get_stats64,
6622 #ifdef CONFIG_NET_POLL_CONTROLLER
6623         .ndo_poll_controller    = ixgbe_netpoll,
6624 #endif
6625 #ifdef IXGBE_FCOE
6626         .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
6627         .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
6628         .ndo_fcoe_enable = ixgbe_fcoe_enable,
6629         .ndo_fcoe_disable = ixgbe_fcoe_disable,
6630         .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
6631 #endif /* IXGBE_FCOE */
6632 };
6633
6634 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
6635                            const struct ixgbe_info *ii)
6636 {
6637 #ifdef CONFIG_PCI_IOV
6638         struct ixgbe_hw *hw = &adapter->hw;
6639         int err;
6640
6641         if (hw->mac.type != ixgbe_mac_82599EB || !max_vfs)
6642                 return;
6643
6644         /* The 82599 supports up to 64 VFs per physical function
6645          * but this implementation limits allocation to 63 so that
6646          * basic networking resources are still available to the
6647          * physical function
6648          */
6649         adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
6650         adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
6651         err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
6652         if (err) {
6653                 e_err(probe, "Failed to enable PCI sriov: %d\n", err);
6654                 goto err_novfs;
6655         }
6656         /* If call to enable VFs succeeded then allocate memory
6657          * for per VF control structures.
6658          */
6659         adapter->vfinfo =
6660                 kcalloc(adapter->num_vfs,
6661                         sizeof(struct vf_data_storage), GFP_KERNEL);
6662         if (adapter->vfinfo) {
6663                 /* Now that we're sure SR-IOV is enabled
6664                  * and memory allocated set up the mailbox parameters
6665                  */
6666                 ixgbe_init_mbx_params_pf(hw);
6667                 memcpy(&hw->mbx.ops, ii->mbx_ops,
6668                        sizeof(hw->mbx.ops));
6669
6670                 /* Disable RSC when in SR-IOV mode */
6671                 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
6672                                      IXGBE_FLAG2_RSC_ENABLED);
6673                 return;
6674         }
6675
6676         /* Oh oh */
6677         e_err(probe, "Unable to allocate memory for VF Data Storage - "
6678               "SRIOV disabled\n");
6679         pci_disable_sriov(adapter->pdev);
6680
6681 err_novfs:
6682         adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
6683         adapter->num_vfs = 0;
6684 #endif /* CONFIG_PCI_IOV */
6685 }
6686
6687 /**
6688  * ixgbe_probe - Device Initialization Routine
6689  * @pdev: PCI device information struct
6690  * @ent: entry in ixgbe_pci_tbl
6691  *
6692  * Returns 0 on success, negative on failure
6693  *
6694  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
6695  * The OS initialization, configuring of the adapter private structure,
6696  * and a hardware reset occur.
6697  **/
6698 static int __devinit ixgbe_probe(struct pci_dev *pdev,
6699                                  const struct pci_device_id *ent)
6700 {
6701         struct net_device *netdev;
6702         struct ixgbe_adapter *adapter = NULL;
6703         struct ixgbe_hw *hw;
6704         const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
6705         static int cards_found;
6706         int i, err, pci_using_dac;
6707         unsigned int indices = num_possible_cpus();
6708 #ifdef IXGBE_FCOE
6709         u16 device_caps;
6710 #endif
6711         u32 part_num, eec;
6712
6713         /* Catch broken hardware that put the wrong VF device ID in
6714          * the PCIe SR-IOV capability.
6715          */
6716         if (pdev->is_virtfn) {
6717                 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
6718                      pci_name(pdev), pdev->vendor, pdev->device);
6719                 return -EINVAL;
6720         }
6721
6722         err = pci_enable_device_mem(pdev);
6723         if (err)
6724                 return err;
6725
6726         if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
6727             !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
6728                 pci_using_dac = 1;
6729         } else {
6730                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6731                 if (err) {
6732                         err = dma_set_coherent_mask(&pdev->dev,
6733                                                     DMA_BIT_MASK(32));
6734                         if (err) {
6735                                 dev_err(&pdev->dev,
6736                                         "No usable DMA configuration, aborting\n");
6737                                 goto err_dma;
6738                         }
6739                 }
6740                 pci_using_dac = 0;
6741         }
6742
6743         err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
6744                                            IORESOURCE_MEM), ixgbe_driver_name);
6745         if (err) {
6746                 dev_err(&pdev->dev,
6747                         "pci_request_selected_regions failed 0x%x\n", err);
6748                 goto err_pci_reg;
6749         }
6750
6751         pci_enable_pcie_error_reporting(pdev);
6752
6753         pci_set_master(pdev);
6754         pci_save_state(pdev);
6755
6756         if (ii->mac == ixgbe_mac_82598EB)
6757                 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
6758         else
6759                 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
6760
6761         indices = max_t(unsigned int, indices, IXGBE_MAX_DCB_INDICES);
6762 #ifdef IXGBE_FCOE
6763         indices += min_t(unsigned int, num_possible_cpus(),
6764                          IXGBE_MAX_FCOE_INDICES);
6765 #endif
6766         netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
6767         if (!netdev) {
6768                 err = -ENOMEM;
6769                 goto err_alloc_etherdev;
6770         }
6771
6772         SET_NETDEV_DEV(netdev, &pdev->dev);
6773
6774         pci_set_drvdata(pdev, netdev);
6775         adapter = netdev_priv(netdev);
6776
6777         adapter->netdev = netdev;
6778         adapter->pdev = pdev;
6779         hw = &adapter->hw;
6780         hw->back = adapter;
6781         adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
6782
6783         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
6784                               pci_resource_len(pdev, 0));
6785         if (!hw->hw_addr) {
6786                 err = -EIO;
6787                 goto err_ioremap;
6788         }
6789
6790         for (i = 1; i <= 5; i++) {
6791                 if (pci_resource_len(pdev, i) == 0)
6792                         continue;
6793         }
6794
6795         netdev->netdev_ops = &ixgbe_netdev_ops;
6796         ixgbe_set_ethtool_ops(netdev);
6797         netdev->watchdog_timeo = 5 * HZ;
6798         strcpy(netdev->name, pci_name(pdev));
6799
6800         adapter->bd_number = cards_found;
6801
6802         /* Setup hw api */
6803         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
6804         hw->mac.type  = ii->mac;
6805
6806         /* EEPROM */
6807         memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
6808         eec = IXGBE_READ_REG(hw, IXGBE_EEC);
6809         /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
6810         if (!(eec & (1 << 8)))
6811                 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
6812
6813         /* PHY */
6814         memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
6815         hw->phy.sfp_type = ixgbe_sfp_type_unknown;
6816         /* ixgbe_identify_phy_generic will set prtad and mmds properly */
6817         hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
6818         hw->phy.mdio.mmds = 0;
6819         hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
6820         hw->phy.mdio.dev = netdev;
6821         hw->phy.mdio.mdio_read = ixgbe_mdio_read;
6822         hw->phy.mdio.mdio_write = ixgbe_mdio_write;
6823
6824         /* set up this timer and work struct before calling get_invariants
6825          * which might start the timer
6826          */
6827         init_timer(&adapter->sfp_timer);
6828         adapter->sfp_timer.function = ixgbe_sfp_timer;
6829         adapter->sfp_timer.data = (unsigned long) adapter;
6830
6831         INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
6832
6833         /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
6834         INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
6835
6836         /* a new SFP+ module arrival, called from GPI SDP2 context */
6837         INIT_WORK(&adapter->sfp_config_module_task,
6838                   ixgbe_sfp_config_module_task);
6839
6840         ii->get_invariants(hw);
6841
6842         /* setup the private structure */
6843         err = ixgbe_sw_init(adapter);
6844         if (err)
6845                 goto err_sw_init;
6846
6847         /* Make it possible the adapter to be woken up via WOL */
6848         if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6849                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6850
6851         /*
6852          * If there is a fan on this device and it has failed log the
6853          * failure.
6854          */
6855         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
6856                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
6857                 if (esdp & IXGBE_ESDP_SDP1)
6858                         e_crit(probe, "Fan has stopped, replace the adapter\n");
6859         }
6860
6861         /* reset_hw fills in the perm_addr as well */
6862         hw->phy.reset_if_overtemp = true;
6863         err = hw->mac.ops.reset_hw(hw);
6864         hw->phy.reset_if_overtemp = false;
6865         if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
6866             hw->mac.type == ixgbe_mac_82598EB) {
6867                 /*
6868                  * Start a kernel thread to watch for a module to arrive.
6869                  * Only do this for 82598, since 82599 will generate
6870                  * interrupts on module arrival.
6871                  */
6872                 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
6873                 mod_timer(&adapter->sfp_timer,
6874                           round_jiffies(jiffies + (2 * HZ)));
6875                 err = 0;
6876         } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
6877                 e_dev_err("failed to initialize because an unsupported SFP+ "
6878                           "module type was detected.\n");
6879                 e_dev_err("Reload the driver after installing a supported "
6880                           "module.\n");
6881                 goto err_sw_init;
6882         } else if (err) {
6883                 e_dev_err("HW Init failed: %d\n", err);
6884                 goto err_sw_init;
6885         }
6886
6887         ixgbe_probe_vf(adapter, ii);
6888
6889         netdev->features = NETIF_F_SG |
6890                            NETIF_F_IP_CSUM |
6891                            NETIF_F_HW_VLAN_TX |
6892                            NETIF_F_HW_VLAN_RX |
6893                            NETIF_F_HW_VLAN_FILTER;
6894
6895         netdev->features |= NETIF_F_IPV6_CSUM;
6896         netdev->features |= NETIF_F_TSO;
6897         netdev->features |= NETIF_F_TSO6;
6898         netdev->features |= NETIF_F_GRO;
6899
6900         if (adapter->hw.mac.type == ixgbe_mac_82599EB)
6901                 netdev->features |= NETIF_F_SCTP_CSUM;
6902
6903         netdev->vlan_features |= NETIF_F_TSO;
6904         netdev->vlan_features |= NETIF_F_TSO6;
6905         netdev->vlan_features |= NETIF_F_IP_CSUM;
6906         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
6907         netdev->vlan_features |= NETIF_F_SG;
6908
6909         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
6910                 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
6911                                     IXGBE_FLAG_DCB_ENABLED);
6912         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
6913                 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
6914
6915 #ifdef CONFIG_IXGBE_DCB
6916         netdev->dcbnl_ops = &dcbnl_ops;
6917 #endif
6918
6919 #ifdef IXGBE_FCOE
6920         if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6921                 if (hw->mac.ops.get_device_caps) {
6922                         hw->mac.ops.get_device_caps(hw, &device_caps);
6923                         if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
6924                                 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6925                 }
6926         }
6927         if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
6928                 netdev->vlan_features |= NETIF_F_FCOE_CRC;
6929                 netdev->vlan_features |= NETIF_F_FSO;
6930                 netdev->vlan_features |= NETIF_F_FCOE_MTU;
6931         }
6932 #endif /* IXGBE_FCOE */
6933         if (pci_using_dac) {
6934                 netdev->features |= NETIF_F_HIGHDMA;
6935                 netdev->vlan_features |= NETIF_F_HIGHDMA;
6936         }
6937
6938         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
6939                 netdev->features |= NETIF_F_LRO;
6940
6941         /* make sure the EEPROM is good */
6942         if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
6943                 e_dev_err("The EEPROM Checksum Is Not Valid\n");
6944                 err = -EIO;
6945                 goto err_eeprom;
6946         }
6947
6948         memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
6949         memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
6950
6951         if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
6952                 e_dev_err("invalid MAC address\n");
6953                 err = -EIO;
6954                 goto err_eeprom;
6955         }
6956
6957         /* power down the optics */
6958         if (hw->phy.multispeed_fiber)
6959                 hw->mac.ops.disable_tx_laser(hw);
6960
6961         init_timer(&adapter->watchdog_timer);
6962         adapter->watchdog_timer.function = ixgbe_watchdog;
6963         adapter->watchdog_timer.data = (unsigned long)adapter;
6964
6965         INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
6966         INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
6967
6968         err = ixgbe_init_interrupt_scheme(adapter);
6969         if (err)
6970                 goto err_sw_init;
6971
6972         switch (pdev->device) {
6973         case IXGBE_DEV_ID_82599_KX4:
6974                 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
6975                                 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
6976                 break;
6977         default:
6978                 adapter->wol = 0;
6979                 break;
6980         }
6981         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6982
6983         /* pick up the PCI bus settings for reporting later */
6984         hw->mac.ops.get_bus_info(hw);
6985
6986         /* print bus type/speed/width info */
6987         e_dev_info("(PCI Express:%s:%s) %pM\n",
6988                    (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0Gb/s" :
6989                     hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5Gb/s" :
6990                     "Unknown"),
6991                    (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
6992                     hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
6993                     hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
6994                     "Unknown"),
6995                    netdev->dev_addr);
6996         ixgbe_read_pba_num_generic(hw, &part_num);
6997         if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
6998                 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, "
6999                            "PBA No: %06x-%03x\n",
7000                            hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7001                            (part_num >> 8), (part_num & 0xff));
7002         else
7003                 e_dev_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
7004                            hw->mac.type, hw->phy.type,
7005                            (part_num >> 8), (part_num & 0xff));
7006
7007         if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7008                 e_dev_warn("PCI-Express bandwidth available for this card is "
7009                            "not sufficient for optimal performance.\n");
7010                 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7011                            "is required.\n");
7012         }
7013
7014         /* save off EEPROM version number */
7015         hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
7016
7017         /* reset the hardware with the new settings */
7018         err = hw->mac.ops.start_hw(hw);
7019
7020         if (err == IXGBE_ERR_EEPROM_VERSION) {
7021                 /* We are running on a pre-production device, log a warning */
7022                 e_dev_warn("This device is a pre-production adapter/LOM. "
7023                            "Please be aware there may be issues associated "
7024                            "with your hardware.  If you are experiencing "
7025                            "problems please contact your Intel or hardware "
7026                            "representative who provided you with this "
7027                            "hardware.\n");
7028         }
7029         strcpy(netdev->name, "eth%d");
7030         err = register_netdev(netdev);
7031         if (err)
7032                 goto err_register;
7033
7034         /* carrier off reporting is important to ethtool even BEFORE open */
7035         netif_carrier_off(netdev);
7036
7037         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7038             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7039                 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
7040
7041         if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
7042                 INIT_WORK(&adapter->check_overtemp_task,
7043                           ixgbe_check_overtemp_task);
7044 #ifdef CONFIG_IXGBE_DCA
7045         if (dca_add_requester(&pdev->dev) == 0) {
7046                 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7047                 ixgbe_setup_dca(adapter);
7048         }
7049 #endif
7050         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7051                 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7052                 for (i = 0; i < adapter->num_vfs; i++)
7053                         ixgbe_vf_configuration(pdev, (i | 0x10000000));
7054         }
7055
7056         /* add san mac addr to netdev */
7057         ixgbe_add_sanmac_netdev(netdev);
7058
7059         e_dev_info("Intel(R) 10 Gigabit Network Connection\n");
7060         cards_found++;
7061         return 0;
7062
7063 err_register:
7064         ixgbe_release_hw_control(adapter);
7065         ixgbe_clear_interrupt_scheme(adapter);
7066 err_sw_init:
7067 err_eeprom:
7068         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7069                 ixgbe_disable_sriov(adapter);
7070         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7071         del_timer_sync(&adapter->sfp_timer);
7072         cancel_work_sync(&adapter->sfp_task);
7073         cancel_work_sync(&adapter->multispeed_fiber_task);
7074         cancel_work_sync(&adapter->sfp_config_module_task);
7075         iounmap(hw->hw_addr);
7076 err_ioremap:
7077         free_netdev(netdev);
7078 err_alloc_etherdev:
7079         pci_release_selected_regions(pdev,
7080                                      pci_select_bars(pdev, IORESOURCE_MEM));
7081 err_pci_reg:
7082 err_dma:
7083         pci_disable_device(pdev);
7084         return err;
7085 }
7086
7087 /**
7088  * ixgbe_remove - Device Removal Routine
7089  * @pdev: PCI device information struct
7090  *
7091  * ixgbe_remove is called by the PCI subsystem to alert the driver
7092  * that it should release a PCI device.  The could be caused by a
7093  * Hot-Plug event, or because the driver is going to be removed from
7094  * memory.
7095  **/
7096 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7097 {
7098         struct net_device *netdev = pci_get_drvdata(pdev);
7099         struct ixgbe_adapter *adapter = netdev_priv(netdev);
7100
7101         set_bit(__IXGBE_DOWN, &adapter->state);
7102         /* clear the module not found bit to make sure the worker won't
7103          * reschedule
7104          */
7105         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7106         del_timer_sync(&adapter->watchdog_timer);
7107
7108         del_timer_sync(&adapter->sfp_timer);
7109         cancel_work_sync(&adapter->watchdog_task);
7110         cancel_work_sync(&adapter->sfp_task);
7111         cancel_work_sync(&adapter->multispeed_fiber_task);
7112         cancel_work_sync(&adapter->sfp_config_module_task);
7113         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7114             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7115                 cancel_work_sync(&adapter->fdir_reinit_task);
7116         flush_scheduled_work();
7117
7118 #ifdef CONFIG_IXGBE_DCA
7119         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7120                 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7121                 dca_remove_requester(&pdev->dev);
7122                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7123         }
7124
7125 #endif
7126 #ifdef IXGBE_FCOE
7127         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7128                 ixgbe_cleanup_fcoe(adapter);
7129
7130 #endif /* IXGBE_FCOE */
7131
7132         /* remove the added san mac */
7133         ixgbe_del_sanmac_netdev(netdev);
7134
7135         if (netdev->reg_state == NETREG_REGISTERED)
7136                 unregister_netdev(netdev);
7137
7138         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7139                 ixgbe_disable_sriov(adapter);
7140
7141         ixgbe_clear_interrupt_scheme(adapter);
7142
7143         ixgbe_release_hw_control(adapter);
7144
7145         iounmap(adapter->hw.hw_addr);
7146         pci_release_selected_regions(pdev, pci_select_bars(pdev,
7147                                      IORESOURCE_MEM));
7148
7149         e_dev_info("complete\n");
7150
7151         free_netdev(netdev);
7152
7153         pci_disable_pcie_error_reporting(pdev);
7154
7155         pci_disable_device(pdev);
7156 }
7157
7158 /**
7159  * ixgbe_io_error_detected - called when PCI error is detected
7160  * @pdev: Pointer to PCI device
7161  * @state: The current pci connection state
7162  *
7163  * This function is called after a PCI bus error affecting
7164  * this device has been detected.
7165  */
7166 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7167                                                 pci_channel_state_t state)
7168 {
7169         struct net_device *netdev = pci_get_drvdata(pdev);
7170         struct ixgbe_adapter *adapter = netdev_priv(netdev);
7171
7172         netif_device_detach(netdev);
7173
7174         if (state == pci_channel_io_perm_failure)
7175                 return PCI_ERS_RESULT_DISCONNECT;
7176
7177         if (netif_running(netdev))
7178                 ixgbe_down(adapter);
7179         pci_disable_device(pdev);
7180
7181         /* Request a slot reset. */
7182         return PCI_ERS_RESULT_NEED_RESET;
7183 }
7184
7185 /**
7186  * ixgbe_io_slot_reset - called after the pci bus has been reset.
7187  * @pdev: Pointer to PCI device
7188  *
7189  * Restart the card from scratch, as if from a cold-boot.
7190  */
7191 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7192 {
7193         struct net_device *netdev = pci_get_drvdata(pdev);
7194         struct ixgbe_adapter *adapter = netdev_priv(netdev);
7195         pci_ers_result_t result;
7196         int err;
7197
7198         if (pci_enable_device_mem(pdev)) {
7199                 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7200                 result = PCI_ERS_RESULT_DISCONNECT;
7201         } else {
7202                 pci_set_master(pdev);
7203                 pci_restore_state(pdev);
7204                 pci_save_state(pdev);
7205
7206                 pci_wake_from_d3(pdev, false);
7207
7208                 ixgbe_reset(adapter);
7209                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7210                 result = PCI_ERS_RESULT_RECOVERED;
7211         }
7212
7213         err = pci_cleanup_aer_uncorrect_error_status(pdev);
7214         if (err) {
7215                 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7216                           "failed 0x%0x\n", err);
7217                 /* non-fatal, continue */
7218         }
7219
7220         return result;
7221 }
7222
7223 /**
7224  * ixgbe_io_resume - called when traffic can start flowing again.
7225  * @pdev: Pointer to PCI device
7226  *
7227  * This callback is called when the error recovery driver tells us that
7228  * its OK to resume normal operation.
7229  */
7230 static void ixgbe_io_resume(struct pci_dev *pdev)
7231 {
7232         struct net_device *netdev = pci_get_drvdata(pdev);
7233         struct ixgbe_adapter *adapter = netdev_priv(netdev);
7234
7235         if (netif_running(netdev)) {
7236                 if (ixgbe_up(adapter)) {
7237                         e_info(probe, "ixgbe_up failed after reset\n");
7238                         return;
7239                 }
7240         }
7241
7242         netif_device_attach(netdev);
7243 }
7244
7245 static struct pci_error_handlers ixgbe_err_handler = {
7246         .error_detected = ixgbe_io_error_detected,
7247         .slot_reset = ixgbe_io_slot_reset,
7248         .resume = ixgbe_io_resume,
7249 };
7250
7251 static struct pci_driver ixgbe_driver = {
7252         .name     = ixgbe_driver_name,
7253         .id_table = ixgbe_pci_tbl,
7254         .probe    = ixgbe_probe,
7255         .remove   = __devexit_p(ixgbe_remove),
7256 #ifdef CONFIG_PM
7257         .suspend  = ixgbe_suspend,
7258         .resume   = ixgbe_resume,
7259 #endif
7260         .shutdown = ixgbe_shutdown,
7261         .err_handler = &ixgbe_err_handler
7262 };
7263
7264 /**
7265  * ixgbe_init_module - Driver Registration Routine
7266  *
7267  * ixgbe_init_module is the first routine called when the driver is
7268  * loaded. All it does is register with the PCI subsystem.
7269  **/
7270 static int __init ixgbe_init_module(void)
7271 {
7272         int ret;
7273         pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7274         pr_info("%s\n", ixgbe_copyright);
7275
7276 #ifdef CONFIG_IXGBE_DCA
7277         dca_register_notify(&dca_notifier);
7278 #endif
7279
7280         ret = pci_register_driver(&ixgbe_driver);
7281         return ret;
7282 }
7283
7284 module_init(ixgbe_init_module);
7285
7286 /**
7287  * ixgbe_exit_module - Driver Exit Cleanup Routine
7288  *
7289  * ixgbe_exit_module is called just before the driver is removed
7290  * from memory.
7291  **/
7292 static void __exit ixgbe_exit_module(void)
7293 {
7294 #ifdef CONFIG_IXGBE_DCA
7295         dca_unregister_notify(&dca_notifier);
7296 #endif
7297         pci_unregister_driver(&ixgbe_driver);
7298         rcu_barrier(); /* Wait for completion of call_rcu()'s */
7299 }
7300
7301 #ifdef CONFIG_IXGBE_DCA
7302 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7303                             void *p)
7304 {
7305         int ret_val;
7306
7307         ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7308                                          __ixgbe_notify_dca);
7309
7310         return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7311 }
7312
7313 #endif /* CONFIG_IXGBE_DCA */
7314
7315 /**
7316  * ixgbe_get_hw_dev return device
7317  * used by hardware layer to print debugging information
7318  **/
7319 struct net_device *ixgbe_get_hw_dev(struct ixgbe_hw *hw)
7320 {
7321         struct ixgbe_adapter *adapter = hw->back;
7322         return adapter->netdev;
7323 }
7324
7325 module_exit(ixgbe_exit_module);
7326
7327 /* ixgbe_main.c */