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         [board_X540] = &ixgbe_X540_info,
  63 };
  64
  65 /* ixgbe_pci_tbl - PCI Device ID Table
  66  *
  67  * Wildcard entries (PCI_ANY_ID) should come last
  68  * Last entry must be all 0s
  69  *
  70  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
  71  *   Class, Class Mask, private data (not used) }
  72  */
  73 static DEFINE_PCI_DEVICE_TABLE(ixgbe_pci_tbl) = {
  74         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598),
  75          board_82598 },
  76         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT),
  77          board_82598 },
  78         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT),
  79          board_82598 },
  80         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT),
  81          board_82598 },
  82         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2),
  83          board_82598 },
  84         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4),
  85          board_82598 },
  86         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT),
  87          board_82598 },
  88         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT),
  89          board_82598 },
  90         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM),
  91          board_82598 },
  92         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR),
  93          board_82598 },
  94         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM),
  95          board_82598 },
  96         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX),
  97          board_82598 },
  98         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4),
  99          board_82599 },
 100         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM),
 101          board_82599 },
 102         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR),
 103          board_82599 },
 104         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP),
 105          board_82599 },
 106         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM),
 107          board_82599 },
 108         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ),
 109          board_82599 },
 110         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4),
 111          board_82599 },
 112         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM),
 113          board_82599 },
 114         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE),
 115          board_82599 },
 116         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T),
 117          board_82599 },
 118
 119         /* required last entry */
 120         {0, }
 121 };
 122 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
 123
 124 #ifdef CONFIG_IXGBE_DCA
 125 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
 126                             void *p);
 127 static struct notifier_block dca_notifier = {
 128         .notifier_call = ixgbe_notify_dca,
 129         .next          = NULL,
 130         .priority      = 0
 131 };
 132 #endif
 133
 134 #ifdef CONFIG_PCI_IOV
 135 static unsigned int max_vfs;
 136 module_param(max_vfs, uint, 0);
 137 MODULE_PARM_DESC(max_vfs,
 138                  "Maximum number of virtual functions to allocate per physical function");
 139 #endif /* CONFIG_PCI_IOV */
 140
 141 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
 142 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
 143 MODULE_LICENSE("GPL");
 144 MODULE_VERSION(DRV_VERSION);
 145
 146 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
 147
 148 static inline void ixgbe_disable_sriov(struct ixgbe_adapter *adapter)
 149 {
 150         struct ixgbe_hw *hw = &adapter->hw;
 151         u32 gcr;
 152         u32 gpie;
 153         u32 vmdctl;
 154
 155 #ifdef CONFIG_PCI_IOV
 156         /* disable iov and allow time for transactions to clear */
 157         pci_disable_sriov(adapter->pdev);
 158 #endif
 159
 160         /* turn off device IOV mode */
 161         gcr = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
 162         gcr &= ~(IXGBE_GCR_EXT_SRIOV);
 163         IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr);
 164         gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
 165         gpie &= ~IXGBE_GPIE_VTMODE_MASK;
 166         IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
 167
 168         /* set default pool back to 0 */
 169         vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
 170         vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
 171         IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
 172
 173         /* take a breather then clean up driver data */
 174         msleep(100);
 175
 176         kfree(adapter->vfinfo);
 177         adapter->vfinfo = NULL;
 178
 179         adapter->num_vfs = 0;
 180         adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
 181 }
 182
 183 struct ixgbe_reg_info {
 184         u32 ofs;
 185         char *name;
 186 };
 187
 188 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
 189
 190         /* General Registers */
 191         {IXGBE_CTRL, "CTRL"},
 192         {IXGBE_STATUS, "STATUS"},
 193         {IXGBE_CTRL_EXT, "CTRL_EXT"},
 194
 195         /* Interrupt Registers */
 196         {IXGBE_EICR, "EICR"},
 197
 198         /* RX Registers */
 199         {IXGBE_SRRCTL(0), "SRRCTL"},
 200         {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
 201         {IXGBE_RDLEN(0), "RDLEN"},
 202         {IXGBE_RDH(0), "RDH"},
 203         {IXGBE_RDT(0), "RDT"},
 204         {IXGBE_RXDCTL(0), "RXDCTL"},
 205         {IXGBE_RDBAL(0), "RDBAL"},
 206         {IXGBE_RDBAH(0), "RDBAH"},
 207
 208         /* TX Registers */
 209         {IXGBE_TDBAL(0), "TDBAL"},
 210         {IXGBE_TDBAH(0), "TDBAH"},
 211         {IXGBE_TDLEN(0), "TDLEN"},
 212         {IXGBE_TDH(0), "TDH"},
 213         {IXGBE_TDT(0), "TDT"},
 214         {IXGBE_TXDCTL(0), "TXDCTL"},
 215
 216         /* List Terminator */
 217         {}
 218 };
 219
 220
 221 /*
 222  * ixgbe_regdump - register printout routine
 223  */
 224 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
 225 {
 226         int i = 0, j = 0;
 227         char rname[16];
 228         u32 regs[64];
 229
 230         switch (reginfo->ofs) {
 231         case IXGBE_SRRCTL(0):
 232                 for (i = 0; i < 64; i++)
 233                         regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
 234                 break;
 235         case IXGBE_DCA_RXCTRL(0):
 236                 for (i = 0; i < 64; i++)
 237                         regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
 238                 break;
 239         case IXGBE_RDLEN(0):
 240                 for (i = 0; i < 64; i++)
 241                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
 242                 break;
 243         case IXGBE_RDH(0):
 244                 for (i = 0; i < 64; i++)
 245                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
 246                 break;
 247         case IXGBE_RDT(0):
 248                 for (i = 0; i < 64; i++)
 249                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
 250                 break;
 251         case IXGBE_RXDCTL(0):
 252                 for (i = 0; i < 64; i++)
 253                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
 254                 break;
 255         case IXGBE_RDBAL(0):
 256                 for (i = 0; i < 64; i++)
 257                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
 258                 break;
 259         case IXGBE_RDBAH(0):
 260                 for (i = 0; i < 64; i++)
 261                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
 262                 break;
 263         case IXGBE_TDBAL(0):
 264                 for (i = 0; i < 64; i++)
 265                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
 266                 break;
 267         case IXGBE_TDBAH(0):
 268                 for (i = 0; i < 64; i++)
 269                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
 270                 break;
 271         case IXGBE_TDLEN(0):
 272                 for (i = 0; i < 64; i++)
 273                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
 274                 break;
 275         case IXGBE_TDH(0):
 276                 for (i = 0; i < 64; i++)
 277                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
 278                 break;
 279         case IXGBE_TDT(0):
 280                 for (i = 0; i < 64; i++)
 281                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
 282                 break;
 283         case IXGBE_TXDCTL(0):
 284                 for (i = 0; i < 64; i++)
 285                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
 286                 break;
 287         default:
 288                 pr_info("%-15s %08x\n", reginfo->name,
 289                         IXGBE_READ_REG(hw, reginfo->ofs));
 290                 return;
 291         }
 292
 293         for (i = 0; i < 8; i++) {
 294                 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
 295                 pr_err("%-15s", rname);
 296                 for (j = 0; j < 8; j++)
 297                         pr_cont(" %08x", regs[i*8+j]);
 298                 pr_cont("\n");
 299         }
 300
 301 }
 302
 303 /*
 304  * ixgbe_dump - Print registers, tx-rings and rx-rings
 305  */
 306 static void ixgbe_dump(struct ixgbe_adapter *adapter)
 307 {
 308         struct net_device *netdev = adapter->netdev;
 309         struct ixgbe_hw *hw = &adapter->hw;
 310         struct ixgbe_reg_info *reginfo;
 311         int n = 0;
 312         struct ixgbe_ring *tx_ring;
 313         struct ixgbe_tx_buffer *tx_buffer_info;
 314         union ixgbe_adv_tx_desc *tx_desc;
 315         struct my_u0 { u64 a; u64 b; } *u0;
 316         struct ixgbe_ring *rx_ring;
 317         union ixgbe_adv_rx_desc *rx_desc;
 318         struct ixgbe_rx_buffer *rx_buffer_info;
 319         u32 staterr;
 320         int i = 0;
 321
 322         if (!netif_msg_hw(adapter))
 323                 return;
 324
 325         /* Print netdevice Info */
 326         if (netdev) {
 327                 dev_info(&adapter->pdev->dev, "Net device Info\n");
 328                 pr_info("Device Name     state            "
 329                         "trans_start      last_rx\n");
 330                 pr_info("%-15s %016lX %016lX %016lX\n",
 331                         netdev->name,
 332                         netdev->state,
 333                         netdev->trans_start,
 334                         netdev->last_rx);
 335         }
 336
 337         /* Print Registers */
 338         dev_info(&adapter->pdev->dev, "Register Dump\n");
 339         pr_info(" Register Name   Value\n");
 340         for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
 341              reginfo->name; reginfo++) {
 342                 ixgbe_regdump(hw, reginfo);
 343         }
 344
 345         /* Print TX Ring Summary */
 346         if (!netdev || !netif_running(netdev))
 347                 goto exit;
 348
 349         dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
 350         pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
 351         for (n = 0; n < adapter->num_tx_queues; n++) {
 352                 tx_ring = adapter->tx_ring[n];
 353                 tx_buffer_info =
 354                         &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
 355                 pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
 356                            n, tx_ring->next_to_use, tx_ring->next_to_clean,
 357                            (u64)tx_buffer_info->dma,
 358                            tx_buffer_info->length,
 359                            tx_buffer_info->next_to_watch,
 360                            (u64)tx_buffer_info->time_stamp);
 361         }
 362
 363         /* Print TX Rings */
 364         if (!netif_msg_tx_done(adapter))
 365                 goto rx_ring_summary;
 366
 367         dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
 368
 369         /* Transmit Descriptor Formats
 370          *
 371          * Advanced Transmit Descriptor
 372          *   +--------------------------------------------------------------+
 373          * 0 |         Buffer Address [63:0]                                |
 374          *   +--------------------------------------------------------------+
 375          * 8 |  PAYLEN  | PORTS  | IDX | STA | DCMD  |DTYP |  RSV |  DTALEN |
 376          *   +--------------------------------------------------------------+
 377          *   63       46 45    40 39 36 35 32 31   24 23 20 19              0
 378          */
 379
 380         for (n = 0; n < adapter->num_tx_queues; n++) {
 381                 tx_ring = adapter->tx_ring[n];
 382                 pr_info("------------------------------------\n");
 383                 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
 384                 pr_info("------------------------------------\n");
 385                 pr_info("T [desc]     [address 63:0  ] "
 386                         "[PlPOIdStDDt Ln] [bi->dma       ] "
 387                         "leng  ntw timestamp        bi->skb\n");
 388
 389                 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
 390                         tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
 391                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
 392                         u0 = (struct my_u0 *)tx_desc;
 393                         pr_info("T [0x%03X]    %016llX %016llX %016llX"
 394                                 " %04X  %3X %016llX %p", i,
 395                                 le64_to_cpu(u0->a),
 396                                 le64_to_cpu(u0->b),
 397                                 (u64)tx_buffer_info->dma,
 398                                 tx_buffer_info->length,
 399                                 tx_buffer_info->next_to_watch,
 400                                 (u64)tx_buffer_info->time_stamp,
 401                                 tx_buffer_info->skb);
 402                         if (i == tx_ring->next_to_use &&
 403                                 i == tx_ring->next_to_clean)
 404                                 pr_cont(" NTC/U\n");
 405                         else if (i == tx_ring->next_to_use)
 406                                 pr_cont(" NTU\n");
 407                         else if (i == tx_ring->next_to_clean)
 408                                 pr_cont(" NTC\n");
 409                         else
 410                                 pr_cont("\n");
 411
 412                         if (netif_msg_pktdata(adapter) &&
 413                                 tx_buffer_info->dma != 0)
 414                                 print_hex_dump(KERN_INFO, "",
 415                                         DUMP_PREFIX_ADDRESS, 16, 1,
 416                                         phys_to_virt(tx_buffer_info->dma),
 417                                         tx_buffer_info->length, true);
 418                 }
 419         }
 420
 421         /* Print RX Rings Summary */
 422 rx_ring_summary:
 423         dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
 424         pr_info("Queue [NTU] [NTC]\n");
 425         for (n = 0; n < adapter->num_rx_queues; n++) {
 426                 rx_ring = adapter->rx_ring[n];
 427                 pr_info("%5d %5X %5X\n",
 428                         n, rx_ring->next_to_use, rx_ring->next_to_clean);
 429         }
 430
 431         /* Print RX Rings */
 432         if (!netif_msg_rx_status(adapter))
 433                 goto exit;
 434
 435         dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
 436
 437         /* Advanced Receive Descriptor (Read) Format
 438          *    63                                           1        0
 439          *    +-----------------------------------------------------+
 440          *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
 441          *    +----------------------------------------------+------+
 442          *  8 |       Header Buffer Address [63:1]           |  DD  |
 443          *    +-----------------------------------------------------+
 444          *
 445          *
 446          * Advanced Receive Descriptor (Write-Back) Format
 447          *
 448          *   63       48 47    32 31  30      21 20 16 15   4 3     0
 449          *   +------------------------------------------------------+
 450          * 0 | Packet     IP     |SPH| HDR_LEN   | RSV|Packet|  RSS |
 451          *   | Checksum   Ident  |   |           |    | Type | Type |
 452          *   +------------------------------------------------------+
 453          * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 454          *   +------------------------------------------------------+
 455          *   63       48 47    32 31            20 19               0
 456          */
 457         for (n = 0; n < adapter->num_rx_queues; n++) {
 458                 rx_ring = adapter->rx_ring[n];
 459                 pr_info("------------------------------------\n");
 460                 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
 461                 pr_info("------------------------------------\n");
 462                 pr_info("R  [desc]      [ PktBuf     A0] "
 463                         "[  HeadBuf   DD] [bi->dma       ] [bi->skb] "
 464                         "<-- Adv Rx Read format\n");
 465                 pr_info("RWB[desc]      [PcsmIpSHl PtRs] "
 466                         "[vl er S cks ln] ---------------- [bi->skb] "
 467                         "<-- Adv Rx Write-Back format\n");
 468
 469                 for (i = 0; i < rx_ring->count; i++) {
 470                         rx_buffer_info = &rx_ring->rx_buffer_info[i];
 471                         rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
 472                         u0 = (struct my_u0 *)rx_desc;
 473                         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 474                         if (staterr & IXGBE_RXD_STAT_DD) {
 475                                 /* Descriptor Done */
 476                                 pr_info("RWB[0x%03X]     %016llX "
 477                                         "%016llX ---------------- %p", i,
 478                                         le64_to_cpu(u0->a),
 479                                         le64_to_cpu(u0->b),
 480                                         rx_buffer_info->skb);
 481                         } else {
 482                                 pr_info("R  [0x%03X]     %016llX "
 483                                         "%016llX %016llX %p", i,
 484                                         le64_to_cpu(u0->a),
 485                                         le64_to_cpu(u0->b),
 486                                         (u64)rx_buffer_info->dma,
 487                                         rx_buffer_info->skb);
 488
 489                                 if (netif_msg_pktdata(adapter)) {
 490                                         print_hex_dump(KERN_INFO, "",
 491                                            DUMP_PREFIX_ADDRESS, 16, 1,
 492                                            phys_to_virt(rx_buffer_info->dma),
 493                                            rx_ring->rx_buf_len, true);
 494
 495                                         if (rx_ring->rx_buf_len
 496                                                 < IXGBE_RXBUFFER_2048)
 497                                                 print_hex_dump(KERN_INFO, "",
 498                                                   DUMP_PREFIX_ADDRESS, 16, 1,
 499                                                   phys_to_virt(
 500                                                     rx_buffer_info->page_dma +
 501                                                     rx_buffer_info->page_offset
 502                                                   ),
 503                                                   PAGE_SIZE/2, true);
 504                                 }
 505                         }
 506
 507                         if (i == rx_ring->next_to_use)
 508                                 pr_cont(" NTU\n");
 509                         else if (i == rx_ring->next_to_clean)
 510                                 pr_cont(" NTC\n");
 511                         else
 512                                 pr_cont("\n");
 513
 514                 }
 515         }
 516
 517 exit:
 518         return;
 519 }
 520
 521 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
 522 {
 523         u32 ctrl_ext;
 524
 525         /* Let firmware take over control of h/w */
 526         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
 527         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
 528                         ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
 529 }
 530
 531 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
 532 {
 533         u32 ctrl_ext;
 534
 535         /* Let firmware know the driver has taken over */
 536         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
 537         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
 538                         ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
 539 }
 540
 541 /*
 542  * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
 543  * @adapter: pointer to adapter struct
 544  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
 545  * @queue: queue to map the corresponding interrupt to
 546  * @msix_vector: the vector to map to the corresponding queue
 547  *
 548  */
 549 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
 550                            u8 queue, u8 msix_vector)
 551 {
 552         u32 ivar, index;
 553         struct ixgbe_hw *hw = &adapter->hw;
 554         switch (hw->mac.type) {
 555         case ixgbe_mac_82598EB:
 556                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 557                 if (direction == -1)
 558                         direction = 0;
 559                 index = (((direction * 64) + queue) >> 2) & 0x1F;
 560                 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
 561                 ivar &= ~(0xFF << (8 * (queue & 0x3)));
 562                 ivar |= (msix_vector << (8 * (queue & 0x3)));
 563                 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
 564                 break;
 565         case ixgbe_mac_82599EB:
 566         case ixgbe_mac_X540:
 567                 if (direction == -1) {
 568                         /* other causes */
 569                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 570                         index = ((queue & 1) * 8);
 571                         ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
 572                         ivar &= ~(0xFF << index);
 573                         ivar |= (msix_vector << index);
 574                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
 575                         break;
 576                 } else {
 577                         /* tx or rx causes */
 578                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 579                         index = ((16 * (queue & 1)) + (8 * direction));
 580                         ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
 581                         ivar &= ~(0xFF << index);
 582                         ivar |= (msix_vector << index);
 583                         IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
 584                         break;
 585                 }
 586         default:
 587                 break;
 588         }
 589 }
 590
 591 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
 592                                           u64 qmask)
 593 {
 594         u32 mask;
 595
 596         switch (adapter->hw.mac.type) {
 597         case ixgbe_mac_82598EB:
 598                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
 599                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
 600                 break;
 601         case ixgbe_mac_82599EB:
 602         case ixgbe_mac_X540:
 603                 mask = (qmask & 0xFFFFFFFF);
 604                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
 605                 mask = (qmask >> 32);
 606                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
 607                 break;
 608         default:
 609                 break;
 610         }
 611 }
 612
 613 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *tx_ring,
 614                                       struct ixgbe_tx_buffer *tx_buffer_info)
 615 {
 616         if (tx_buffer_info->dma) {
 617                 if (tx_buffer_info->mapped_as_page)
 618                         dma_unmap_page(tx_ring->dev,
 619                                        tx_buffer_info->dma,
 620                                        tx_buffer_info->length,
 621                                        DMA_TO_DEVICE);
 622                 else
 623                         dma_unmap_single(tx_ring->dev,
 624                                          tx_buffer_info->dma,
 625                                          tx_buffer_info->length,
 626                                          DMA_TO_DEVICE);
 627                 tx_buffer_info->dma = 0;
 628         }
 629         if (tx_buffer_info->skb) {
 630                 dev_kfree_skb_any(tx_buffer_info->skb);
 631                 tx_buffer_info->skb = NULL;
 632         }
 633         tx_buffer_info->time_stamp = 0;
 634         /* tx_buffer_info must be completely set up in the transmit path */
 635 }
 636
 637 /**
 638  * ixgbe_dcb_txq_to_tc - convert a reg index to a traffic class
 639  * @adapter: driver private struct
 640  * @index: reg idx of queue to query (0-127)
 641  *
 642  * Helper function to determine the traffic index for a paticular
 643  * register index.
 644  *
 645  * Returns : a tc index for use in range 0-7, or 0-3
 646  */
 647 u8 ixgbe_dcb_txq_to_tc(struct ixgbe_adapter *adapter, u8 reg_idx)
 648 {
 649         int tc = -1;
 650         int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
 651
 652         /* if DCB is not enabled the queues have no TC */
 653         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
 654                 return tc;
 655
 656         /* check valid range */
 657         if (reg_idx >= adapter->hw.mac.max_tx_queues)
 658                 return tc;
 659
 660         switch (adapter->hw.mac.type) {
 661         case ixgbe_mac_82598EB:
 662                 tc = reg_idx >> 2;
 663                 break;
 664         default:
 665                 if (dcb_i != 4 && dcb_i != 8)
 666                         break;
 667
 668                 /* if VMDq is enabled the lowest order bits determine TC */
 669                 if (adapter->flags & (IXGBE_FLAG_SRIOV_ENABLED |
 670                                       IXGBE_FLAG_VMDQ_ENABLED)) {
 671                         tc = reg_idx & (dcb_i - 1);
 672                         break;
 673                 }
 674
 675                 /*
 676                  * Convert the reg_idx into the correct TC. This bitmask
 677                  * targets the last full 32 ring traffic class and assigns
 678                  * it a value of 1. From there the rest of the rings are
 679                  * based on shifting the mask further up to include the
 680                  * reg_idx / 16 and then reg_idx / 8. It assumes dcB_i
 681                  * will only ever be 8 or 4 and that reg_idx will never
 682                  * be greater then 128. The code without the power of 2
 683                  * optimizations would be:
 684                  * (((reg_idx % 32) + 32) * dcb_i) >> (9 - reg_idx / 32)
 685                  */
 686                 tc = ((reg_idx & 0X1F) + 0x20) * dcb_i;
 687                 tc >>= 9 - (reg_idx >> 5);
 688         }
 689
 690         return tc;
 691 }
 692
 693 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
 694 {
 695         struct ixgbe_hw *hw = &adapter->hw;
 696         struct ixgbe_hw_stats *hwstats = &adapter->stats;
 697         u32 data = 0;
 698         u32 xoff[8] = {0};
 699         int i;
 700
 701         if ((hw->fc.current_mode == ixgbe_fc_full) ||
 702             (hw->fc.current_mode == ixgbe_fc_rx_pause)) {
 703                 switch (hw->mac.type) {
 704                 case ixgbe_mac_82598EB:
 705                         data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
 706                         break;
 707                 default:
 708                         data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
 709                 }
 710                 hwstats->lxoffrxc += data;
 711
 712                 /* refill credits (no tx hang) if we received xoff */
 713                 if (!data)
 714                         return;
 715
 716                 for (i = 0; i < adapter->num_tx_queues; i++)
 717                         clear_bit(__IXGBE_HANG_CHECK_ARMED,
 718                                   &adapter->tx_ring[i]->state);
 719                 return;
 720         } else if (!(adapter->dcb_cfg.pfc_mode_enable))
 721                 return;
 722
 723         /* update stats for each tc, only valid with PFC enabled */
 724         for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
 725                 switch (hw->mac.type) {
 726                 case ixgbe_mac_82598EB:
 727                         xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
 728                         break;
 729                 default:
 730                         xoff[i] = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
 731                 }
 732                 hwstats->pxoffrxc[i] += xoff[i];
 733         }
 734
 735         /* disarm tx queues that have received xoff frames */
 736         for (i = 0; i < adapter->num_tx_queues; i++) {
 737                 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
 738                 u32 tc = ixgbe_dcb_txq_to_tc(adapter, tx_ring->reg_idx);
 739
 740                 if (xoff[tc])
 741                         clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
 742         }
 743 }
 744
 745 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
 746 {
 747         return ring->tx_stats.completed;
 748 }
 749
 750 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
 751 {
 752         struct ixgbe_adapter *adapter = netdev_priv(ring->netdev);
 753         struct ixgbe_hw *hw = &adapter->hw;
 754
 755         u32 head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
 756         u32 tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
 757
 758         if (head != tail)
 759                 return (head < tail) ?
 760                         tail - head : (tail + ring->count - head);
 761
 762         return 0;
 763 }
 764
 765 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
 766 {
 767         u32 tx_done = ixgbe_get_tx_completed(tx_ring);
 768         u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
 769         u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
 770         bool ret = false;
 771
 772         clear_check_for_tx_hang(tx_ring);
 773
 774         /*
 775          * Check for a hung queue, but be thorough. This verifies
 776          * that a transmit has been completed since the previous
 777          * check AND there is at least one packet pending. The
 778          * ARMED bit is set to indicate a potential hang. The
 779          * bit is cleared if a pause frame is received to remove
 780          * false hang detection due to PFC or 802.3x frames. By
 781          * requiring this to fail twice we avoid races with
 782          * pfc clearing the ARMED bit and conditions where we
 783          * run the check_tx_hang logic with a transmit completion
 784          * pending but without time to complete it yet.
 785          */
 786         if ((tx_done_old == tx_done) && tx_pending) {
 787                 /* make sure it is true for two checks in a row */
 788                 ret = test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
 789                                        &tx_ring->state);
 790         } else {
 791                 /* update completed stats and continue */
 792                 tx_ring->tx_stats.tx_done_old = tx_done;
 793                 /* reset the countdown */
 794                 clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
 795         }
 796
 797         return ret;
 798 }
 799
 800 #define IXGBE_MAX_TXD_PWR       14
 801 #define IXGBE_MAX_DATA_PER_TXD  (1 << IXGBE_MAX_TXD_PWR)
 802
 803 /* Tx Descriptors needed, worst case */
 804 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
 805                          (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
 806 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
 807         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
 808
 809 static void ixgbe_tx_timeout(struct net_device *netdev);
 810
 811 /**
 812  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
 813  * @q_vector: structure containing interrupt and ring information
 814  * @tx_ring: tx ring to clean
 815  **/
 816 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
 817                                struct ixgbe_ring *tx_ring)
 818 {
 819         struct ixgbe_adapter *adapter = q_vector->adapter;
 820         union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
 821         struct ixgbe_tx_buffer *tx_buffer_info;
 822         unsigned int total_bytes = 0, total_packets = 0;
 823         u16 i, eop, count = 0;
 824
 825         i = tx_ring->next_to_clean;
 826         eop = tx_ring->tx_buffer_info[i].next_to_watch;
 827         eop_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
 828
 829         while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
 830                (count < tx_ring->work_limit)) {
 831                 bool cleaned = false;
 832                 rmb(); /* read buffer_info after eop_desc */
 833                 for ( ; !cleaned; count++) {
 834                         tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
 835                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
 836
 837                         tx_desc->wb.status = 0;
 838                         cleaned = (i == eop);
 839
 840                         i++;
 841                         if (i == tx_ring->count)
 842                                 i = 0;
 843
 844                         if (cleaned && tx_buffer_info->skb) {
 845                                 total_bytes += tx_buffer_info->bytecount;
 846                                 total_packets += tx_buffer_info->gso_segs;
 847                         }
 848
 849                         ixgbe_unmap_and_free_tx_resource(tx_ring,
 850                                                          tx_buffer_info);
 851                 }
 852
 853                 tx_ring->tx_stats.completed++;
 854                 eop = tx_ring->tx_buffer_info[i].next_to_watch;
 855                 eop_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
 856         }
 857
 858         tx_ring->next_to_clean = i;
 859         tx_ring->total_bytes += total_bytes;
 860         tx_ring->total_packets += total_packets;
 861         u64_stats_update_begin(&tx_ring->syncp);
 862         tx_ring->stats.packets += total_packets;
 863         tx_ring->stats.bytes += total_bytes;
 864         u64_stats_update_end(&tx_ring->syncp);
 865
 866         if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
 867                 /* schedule immediate reset if we believe we hung */
 868                 struct ixgbe_hw *hw = &adapter->hw;
 869                 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, eop);
 870                 e_err(drv, "Detected Tx Unit Hang\n"
 871                         "  Tx Queue             <%d>\n"
 872                         "  TDH, TDT             <%x>, <%x>\n"
 873                         "  next_to_use          <%x>\n"
 874                         "  next_to_clean        <%x>\n"
 875                         "tx_buffer_info[next_to_clean]\n"
 876                         "  time_stamp           <%lx>\n"
 877                         "  jiffies              <%lx>\n",
 878                         tx_ring->queue_index,
 879                         IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
 880                         IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
 881                         tx_ring->next_to_use, eop,
 882                         tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
 883
 884                 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
 885
 886                 e_info(probe,
 887                        "tx hang %d detected on queue %d, resetting adapter\n",
 888                         adapter->tx_timeout_count + 1, tx_ring->queue_index);
 889
 890                 /* schedule immediate reset if we believe we hung */
 891                 ixgbe_tx_timeout(adapter->netdev);
 892
 893                 /* the adapter is about to reset, no point in enabling stuff */
 894                 return true;
 895         }
 896
 897 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
 898         if (unlikely(count && netif_carrier_ok(tx_ring->netdev) &&
 899                      (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
 900                 /* Make sure that anybody stopping the queue after this
 901                  * sees the new next_to_clean.
 902                  */
 903                 smp_mb();
 904                 if (__netif_subqueue_stopped(tx_ring->netdev, tx_ring->queue_index) &&
 905                     !test_bit(__IXGBE_DOWN, &adapter->state)) {
 906                         netif_wake_subqueue(tx_ring->netdev, tx_ring->queue_index);
 907                         ++tx_ring->tx_stats.restart_queue;
 908                 }
 909         }
 910
 911         return count < tx_ring->work_limit;
 912 }
 913
 914 #ifdef CONFIG_IXGBE_DCA
 915 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
 916                                 struct ixgbe_ring *rx_ring,
 917                                 int cpu)
 918 {
 919         struct ixgbe_hw *hw = &adapter->hw;
 920         u32 rxctrl;
 921         u8 reg_idx = rx_ring->reg_idx;
 922
 923         rxctrl = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(reg_idx));
 924         switch (hw->mac.type) {
 925         case ixgbe_mac_82598EB:
 926                 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
 927                 rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
 928                 break;
 929         case ixgbe_mac_82599EB:
 930         case ixgbe_mac_X540:
 931                 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599;
 932                 rxctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
 933                            IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599);
 934                 break;
 935         default:
 936                 break;
 937         }
 938         rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
 939         rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
 940         rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN);
 941         rxctrl &= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN |
 942                     IXGBE_DCA_RXCTRL_DESC_HSRO_EN);
 943         IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
 944 }
 945
 946 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
 947                                 struct ixgbe_ring *tx_ring,
 948                                 int cpu)
 949 {
 950         struct ixgbe_hw *hw = &adapter->hw;
 951         u32 txctrl;
 952         u8 reg_idx = tx_ring->reg_idx;
 953
 954         switch (hw->mac.type) {
 955         case ixgbe_mac_82598EB:
 956                 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(reg_idx));
 957                 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
 958                 txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
 959                 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
 960                 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
 961                 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(reg_idx), txctrl);
 962                 break;
 963         case ixgbe_mac_82599EB:
 964         case ixgbe_mac_X540:
 965                 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx));
 966                 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599;
 967                 txctrl |= (dca3_get_tag(&adapter->pdev->dev, cpu) <<
 968                            IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599);
 969                 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
 970                 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
 971                 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(reg_idx), txctrl);
 972                 break;
 973         default:
 974                 break;
 975         }
 976 }
 977
 978 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
 979 {
 980         struct ixgbe_adapter *adapter = q_vector->adapter;
 981         int cpu = get_cpu();
 982         long r_idx;
 983         int i;
 984
 985         if (q_vector->cpu == cpu)
 986                 goto out_no_update;
 987
 988         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
 989         for (i = 0; i < q_vector->txr_count; i++) {
 990                 ixgbe_update_tx_dca(adapter, adapter->tx_ring[r_idx], cpu);
 991                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
 992                                       r_idx + 1);
 993         }
 994
 995         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
 996         for (i = 0; i < q_vector->rxr_count; i++) {
 997                 ixgbe_update_rx_dca(adapter, adapter->rx_ring[r_idx], cpu);
 998                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
 999                                       r_idx + 1);
1000         }
1001
1002         q_vector->cpu = cpu;
1003 out_no_update:
1004         put_cpu();
1005 }
1006
1007 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1008 {
1009         int num_q_vectors;
1010         int i;
1011
1012         if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
1013                 return;
1014
1015         /* always use CB2 mode, difference is masked in the CB driver */
1016         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
1017
1018         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
1019                 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1020         else
1021                 num_q_vectors = 1;
1022
1023         for (i = 0; i < num_q_vectors; i++) {
1024                 adapter->q_vector[i]->cpu = -1;
1025                 ixgbe_update_dca(adapter->q_vector[i]);
1026         }
1027 }
1028
1029 static int __ixgbe_notify_dca(struct device *dev, void *data)
1030 {
1031         struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1032         unsigned long event = *(unsigned long *)data;
1033
1034         if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
1035                 return 0;
1036
1037         switch (event) {
1038         case DCA_PROVIDER_ADD:
1039                 /* if we're already enabled, don't do it again */
1040                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1041                         break;
1042                 if (dca_add_requester(dev) == 0) {
1043                         adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1044                         ixgbe_setup_dca(adapter);
1045                         break;
1046                 }
1047                 /* Fall Through since DCA is disabled. */
1048         case DCA_PROVIDER_REMOVE:
1049                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1050                         dca_remove_requester(dev);
1051                         adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1052                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
1053                 }
1054                 break;
1055         }
1056
1057         return 0;
1058 }
1059
1060 #endif /* CONFIG_IXGBE_DCA */
1061 /**
1062  * ixgbe_receive_skb - Send a completed packet up the stack
1063  * @adapter: board private structure
1064  * @skb: packet to send up
1065  * @status: hardware indication of status of receive
1066  * @rx_ring: rx descriptor ring (for a specific queue) to setup
1067  * @rx_desc: rx descriptor
1068  **/
1069 static void ixgbe_receive_skb(struct ixgbe_q_vector *q_vector,
1070                               struct sk_buff *skb, u8 status,
1071                               struct ixgbe_ring *ring,
1072                               union ixgbe_adv_rx_desc *rx_desc)
1073 {
1074         struct ixgbe_adapter *adapter = q_vector->adapter;
1075         struct napi_struct *napi = &q_vector->napi;
1076         bool is_vlan = (status & IXGBE_RXD_STAT_VP);
1077         u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
1078
1079         if (is_vlan && (tag & VLAN_VID_MASK))
1080                 __vlan_hwaccel_put_tag(skb, tag);
1081
1082         if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1083                 napi_gro_receive(napi, skb);
1084         else
1085                 netif_rx(skb);
1086 }
1087
1088 /**
1089  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1090  * @adapter: address of board private structure
1091  * @status_err: hardware indication of status of receive
1092  * @skb: skb currently being received and modified
1093  **/
1094 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
1095                                      union ixgbe_adv_rx_desc *rx_desc,
1096                                      struct sk_buff *skb)
1097 {
1098         u32 status_err = le32_to_cpu(rx_desc->wb.upper.status_error);
1099
1100         skb_checksum_none_assert(skb);
1101
1102         /* Rx csum disabled */
1103         if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
1104                 return;
1105
1106         /* if IP and error */
1107         if ((status_err & IXGBE_RXD_STAT_IPCS) &&
1108             (status_err & IXGBE_RXDADV_ERR_IPE)) {
1109                 adapter->hw_csum_rx_error++;
1110                 return;
1111         }
1112
1113         if (!(status_err & IXGBE_RXD_STAT_L4CS))
1114                 return;
1115
1116         if (status_err & IXGBE_RXDADV_ERR_TCPE) {
1117                 u16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1118
1119                 /*
1120                  * 82599 errata, UDP frames with a 0 checksum can be marked as
1121                  * checksum errors.
1122                  */
1123                 if ((pkt_info & IXGBE_RXDADV_PKTTYPE_UDP) &&
1124                     (adapter->hw.mac.type == ixgbe_mac_82599EB))
1125                         return;
1126
1127                 adapter->hw_csum_rx_error++;
1128                 return;
1129         }
1130
1131         /* It must be a TCP or UDP packet with a valid checksum */
1132         skb->ip_summed = CHECKSUM_UNNECESSARY;
1133 }
1134
1135 static inline void ixgbe_release_rx_desc(struct ixgbe_ring *rx_ring, u32 val)
1136 {
1137         /*
1138          * Force memory writes to complete before letting h/w
1139          * know there are new descriptors to fetch.  (Only
1140          * applicable for weak-ordered memory model archs,
1141          * such as IA-64).
1142          */
1143         wmb();
1144         writel(val, rx_ring->tail);
1145 }
1146
1147 /**
1148  * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
1149  * @rx_ring: ring to place buffers on
1150  * @cleaned_count: number of buffers to replace
1151  **/
1152 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1153 {
1154         union ixgbe_adv_rx_desc *rx_desc;
1155         struct ixgbe_rx_buffer *bi;
1156         struct sk_buff *skb;
1157         u16 i = rx_ring->next_to_use;
1158
1159         /* do nothing if no valid netdev defined */
1160         if (!rx_ring->netdev)
1161                 return;
1162
1163         while (cleaned_count--) {
1164                 rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1165                 bi = &rx_ring->rx_buffer_info[i];
1166                 skb = bi->skb;
1167
1168                 if (!skb) {
1169                         skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1170                                                         rx_ring->rx_buf_len);
1171                         if (!skb) {
1172                                 rx_ring->rx_stats.alloc_rx_buff_failed++;
1173                                 goto no_buffers;
1174                         }
1175                         /* initialize queue mapping */
1176                         skb_record_rx_queue(skb, rx_ring->queue_index);
1177                         bi->skb = skb;
1178                 }
1179
1180                 if (!bi->dma) {
1181                         bi->dma = dma_map_single(rx_ring->dev,
1182                                                  skb->data,
1183                                                  rx_ring->rx_buf_len,
1184                                                  DMA_FROM_DEVICE);
1185                         if (dma_mapping_error(rx_ring->dev, bi->dma)) {
1186                                 rx_ring->rx_stats.alloc_rx_buff_failed++;
1187                                 bi->dma = 0;
1188                                 goto no_buffers;
1189                         }
1190                 }
1191
1192                 if (ring_is_ps_enabled(rx_ring)) {
1193                         if (!bi->page) {
1194                                 bi->page = netdev_alloc_page(rx_ring->netdev);
1195                                 if (!bi->page) {
1196                                         rx_ring->rx_stats.alloc_rx_page_failed++;
1197                                         goto no_buffers;
1198                                 }
1199                         }
1200
1201                         if (!bi->page_dma) {
1202                                 /* use a half page if we're re-using */
1203                                 bi->page_offset ^= PAGE_SIZE / 2;
1204                                 bi->page_dma = dma_map_page(rx_ring->dev,
1205                                                             bi->page,
1206                                                             bi->page_offset,
1207                                                             PAGE_SIZE / 2,
1208                                                             DMA_FROM_DEVICE);
1209                                 if (dma_mapping_error(rx_ring->dev,
1210                                                       bi->page_dma)) {
1211                                         rx_ring->rx_stats.alloc_rx_page_failed++;
1212                                         bi->page_dma = 0;
1213                                         goto no_buffers;
1214                                 }
1215                         }
1216
1217                         /* Refresh the desc even if buffer_addrs didn't change
1218                          * because each write-back erases this info. */
1219                         rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
1220                         rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
1221                 } else {
1222                         rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
1223                         rx_desc->read.hdr_addr = 0;
1224                 }
1225
1226                 i++;
1227                 if (i == rx_ring->count)
1228                         i = 0;
1229         }
1230
1231 no_buffers:
1232         if (rx_ring->next_to_use != i) {
1233                 rx_ring->next_to_use = i;
1234                 ixgbe_release_rx_desc(rx_ring, i);
1235         }
1236 }
1237
1238 static inline u16 ixgbe_get_hlen(union ixgbe_adv_rx_desc *rx_desc)
1239 {
1240         /* HW will not DMA in data larger than the given buffer, even if it
1241          * parses the (NFS, of course) header to be larger.  In that case, it
1242          * fills the header buffer and spills the rest into the page.
1243          */
1244         u16 hdr_info = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info);
1245         u16 hlen = (hdr_info &  IXGBE_RXDADV_HDRBUFLEN_MASK) >>
1246                     IXGBE_RXDADV_HDRBUFLEN_SHIFT;
1247         if (hlen > IXGBE_RX_HDR_SIZE)
1248                 hlen = IXGBE_RX_HDR_SIZE;
1249         return hlen;
1250 }
1251
1252 /**
1253  * ixgbe_transform_rsc_queue - change rsc queue into a full packet
1254  * @skb: pointer to the last skb in the rsc queue
1255  *
1256  * This function changes a queue full of hw rsc buffers into a completed
1257  * packet.  It uses the ->prev pointers to find the first packet and then
1258  * turns it into the frag list owner.
1259  **/
1260 static inline struct sk_buff *ixgbe_transform_rsc_queue(struct sk_buff *skb)
1261 {
1262         unsigned int frag_list_size = 0;
1263         unsigned int skb_cnt = 1;
1264
1265         while (skb->prev) {
1266                 struct sk_buff *prev = skb->prev;
1267                 frag_list_size += skb->len;
1268                 skb->prev = NULL;
1269                 skb = prev;
1270                 skb_cnt++;
1271         }
1272
1273         skb_shinfo(skb)->frag_list = skb->next;
1274         skb->next = NULL;
1275         skb->len += frag_list_size;
1276         skb->data_len += frag_list_size;
1277         skb->truesize += frag_list_size;
1278         IXGBE_RSC_CB(skb)->skb_cnt = skb_cnt;
1279
1280         return skb;
1281 }
1282
1283 static inline bool ixgbe_get_rsc_state(union ixgbe_adv_rx_desc *rx_desc)
1284 {
1285         return !!(le32_to_cpu(rx_desc->wb.lower.lo_dword.data) &
1286                 IXGBE_RXDADV_RSCCNT_MASK);
1287 }
1288
1289 static void ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
1290                                struct ixgbe_ring *rx_ring,
1291                                int *work_done, int work_to_do)
1292 {
1293         struct ixgbe_adapter *adapter = q_vector->adapter;
1294         union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
1295         struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
1296         struct sk_buff *skb;
1297         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1298         const int current_node = numa_node_id();
1299 #ifdef IXGBE_FCOE
1300         int ddp_bytes = 0;
1301 #endif /* IXGBE_FCOE */
1302         u32 staterr;
1303         u16 i;
1304         u16 cleaned_count = 0;
1305         bool pkt_is_rsc = false;
1306
1307         i = rx_ring->next_to_clean;
1308         rx_desc = IXGBE_RX_DESC_ADV(rx_ring, i);
1309         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1310
1311         while (staterr & IXGBE_RXD_STAT_DD) {
1312                 u32 upper_len = 0;
1313
1314                 rmb(); /* read descriptor and rx_buffer_info after status DD */
1315
1316                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1317
1318                 skb = rx_buffer_info->skb;
1319                 rx_buffer_info->skb = NULL;
1320                 prefetch(skb->data);
1321
1322                 if (ring_is_rsc_enabled(rx_ring))
1323                         pkt_is_rsc = ixgbe_get_rsc_state(rx_desc);
1324
1325                 /* if this is a skb from previous receive DMA will be 0 */
1326                 if (rx_buffer_info->dma) {
1327                         u16 hlen;
1328                         if (pkt_is_rsc &&
1329                             !(staterr & IXGBE_RXD_STAT_EOP) &&
1330                             !skb->prev) {
1331                                 /*
1332                                  * When HWRSC is enabled, delay unmapping
1333                                  * of the first packet. It carries the
1334                                  * header information, HW may still
1335                                  * access the header after the writeback.
1336                                  * Only unmap it when EOP is reached
1337                                  */
1338                                 IXGBE_RSC_CB(skb)->delay_unmap = true;
1339                                 IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
1340                         } else {
1341                                 dma_unmap_single(rx_ring->dev,
1342                                                  rx_buffer_info->dma,
1343                                                  rx_ring->rx_buf_len,
1344                                                  DMA_FROM_DEVICE);
1345                         }
1346                         rx_buffer_info->dma = 0;
1347
1348                         if (ring_is_ps_enabled(rx_ring)) {
1349                                 hlen = ixgbe_get_hlen(rx_desc);
1350                                 upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1351                         } else {
1352                                 hlen = le16_to_cpu(rx_desc->wb.upper.length);
1353                         }
1354
1355                         skb_put(skb, hlen);
1356                 } else {
1357                         /* assume packet split since header is unmapped */
1358                         upper_len = le16_to_cpu(rx_desc->wb.upper.length);
1359                 }
1360
1361                 if (upper_len) {
1362                         dma_unmap_page(rx_ring->dev,
1363                                        rx_buffer_info->page_dma,
1364                                        PAGE_SIZE / 2,
1365                                        DMA_FROM_DEVICE);
1366                         rx_buffer_info->page_dma = 0;
1367                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1368                                            rx_buffer_info->page,
1369                                            rx_buffer_info->page_offset,
1370                                            upper_len);
1371
1372                         if ((page_count(rx_buffer_info->page) == 1) &&
1373                             (page_to_nid(rx_buffer_info->page) == current_node))
1374                                 get_page(rx_buffer_info->page);
1375                         else
1376                                 rx_buffer_info->page = NULL;
1377
1378                         skb->len += upper_len;
1379                         skb->data_len += upper_len;
1380                         skb->truesize += upper_len;
1381                 }
1382
1383                 i++;
1384                 if (i == rx_ring->count)
1385                         i = 0;
1386
1387                 next_rxd = IXGBE_RX_DESC_ADV(rx_ring, i);
1388                 prefetch(next_rxd);
1389                 cleaned_count++;
1390
1391                 if (pkt_is_rsc) {
1392                         u32 nextp = (staterr & IXGBE_RXDADV_NEXTP_MASK) >>
1393                                      IXGBE_RXDADV_NEXTP_SHIFT;
1394                         next_buffer = &rx_ring->rx_buffer_info[nextp];
1395                 } else {
1396                         next_buffer = &rx_ring->rx_buffer_info[i];
1397                 }
1398
1399                 if (!(staterr & IXGBE_RXD_STAT_EOP)) {
1400                         if (ring_is_ps_enabled(rx_ring)) {
1401                                 rx_buffer_info->skb = next_buffer->skb;
1402                                 rx_buffer_info->dma = next_buffer->dma;
1403                                 next_buffer->skb = skb;
1404                                 next_buffer->dma = 0;
1405                         } else {
1406                                 skb->next = next_buffer->skb;
1407                                 skb->next->prev = skb;
1408                         }
1409                         rx_ring->rx_stats.non_eop_descs++;
1410                         goto next_desc;
1411                 }
1412
1413                 if (skb->prev) {
1414                         skb = ixgbe_transform_rsc_queue(skb);
1415                         /* if we got here without RSC the packet is invalid */
1416                         if (!pkt_is_rsc) {
1417                                 __pskb_trim(skb, 0);
1418                                 rx_buffer_info->skb = skb;
1419                                 goto next_desc;
1420                         }
1421                 }
1422
1423                 if (ring_is_rsc_enabled(rx_ring)) {
1424                         if (IXGBE_RSC_CB(skb)->delay_unmap) {
1425                                 dma_unmap_single(rx_ring->dev,
1426                                                  IXGBE_RSC_CB(skb)->dma,
1427                                                  rx_ring->rx_buf_len,
1428                                                  DMA_FROM_DEVICE);
1429                                 IXGBE_RSC_CB(skb)->dma = 0;
1430                                 IXGBE_RSC_CB(skb)->delay_unmap = false;
1431                         }
1432                 }
1433                 if (pkt_is_rsc) {
1434                         if (ring_is_ps_enabled(rx_ring))
1435                                 rx_ring->rx_stats.rsc_count +=
1436                                         skb_shinfo(skb)->nr_frags;
1437                         else
1438                                 rx_ring->rx_stats.rsc_count +=
1439                                         IXGBE_RSC_CB(skb)->skb_cnt;
1440                         rx_ring->rx_stats.rsc_flush++;
1441                 }
1442
1443                 /* ERR_MASK will only have valid bits if EOP set */
1444                 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
1445                         /* trim packet back to size 0 and recycle it */
1446                         __pskb_trim(skb, 0);
1447                         rx_buffer_info->skb = skb;
1448                         goto next_desc;
1449                 }
1450
1451                 ixgbe_rx_checksum(adapter, rx_desc, skb);
1452
1453                 /* probably a little skewed due to removing CRC */
1454                 total_rx_bytes += skb->len;
1455                 total_rx_packets++;
1456
1457                 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1458 #ifdef IXGBE_FCOE
1459                 /* if ddp, not passing to ULD unless for FCP_RSP or error */
1460                 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
1461                         ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
1462                         if (!ddp_bytes)
1463                                 goto next_desc;
1464                 }
1465 #endif /* IXGBE_FCOE */
1466                 ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
1467
1468 next_desc:
1469                 rx_desc->wb.upper.status_error = 0;
1470
1471                 (*work_done)++;
1472                 if (*work_done >= work_to_do)
1473                         break;
1474
1475                 /* return some buffers to hardware, one at a time is too slow */
1476                 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
1477                         ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1478                         cleaned_count = 0;
1479                 }
1480
1481                 /* use prefetched values */
1482                 rx_desc = next_rxd;
1483                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1484         }
1485
1486         rx_ring->next_to_clean = i;
1487         cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
1488
1489         if (cleaned_count)
1490                 ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
1491
1492 #ifdef IXGBE_FCOE
1493         /* include DDPed FCoE data */
1494         if (ddp_bytes > 0) {
1495                 unsigned int mss;
1496
1497                 mss = rx_ring->netdev->mtu - sizeof(struct fcoe_hdr) -
1498                         sizeof(struct fc_frame_header) -
1499                         sizeof(struct fcoe_crc_eof);
1500                 if (mss > 512)
1501                         mss &= ~511;
1502                 total_rx_bytes += ddp_bytes;
1503                 total_rx_packets += DIV_ROUND_UP(ddp_bytes, mss);
1504         }
1505 #endif /* IXGBE_FCOE */
1506
1507         rx_ring->total_packets += total_rx_packets;
1508         rx_ring->total_bytes += total_rx_bytes;
1509         u64_stats_update_begin(&rx_ring->syncp);
1510         rx_ring->stats.packets += total_rx_packets;
1511         rx_ring->stats.bytes += total_rx_bytes;
1512         u64_stats_update_end(&rx_ring->syncp);
1513 }
1514
1515 static int ixgbe_clean_rxonly(struct napi_struct *, int);
1516 /**
1517  * ixgbe_configure_msix - Configure MSI-X hardware
1518  * @adapter: board private structure
1519  *
1520  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
1521  * interrupts.
1522  **/
1523 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
1524 {
1525         struct ixgbe_q_vector *q_vector;
1526         int i, q_vectors, v_idx, r_idx;
1527         u32 mask;
1528
1529         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1530
1531         /*
1532          * Populate the IVAR table and set the ITR values to the
1533          * corresponding register.
1534          */
1535         for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1536                 q_vector = adapter->q_vector[v_idx];
1537                 /* XXX for_each_set_bit(...) */
1538                 r_idx = find_first_bit(q_vector->rxr_idx,
1539                                        adapter->num_rx_queues);
1540
1541                 for (i = 0; i < q_vector->rxr_count; i++) {
1542                         u8 reg_idx = adapter->rx_ring[r_idx]->reg_idx;
1543                         ixgbe_set_ivar(adapter, 0, reg_idx, v_idx);
1544                         r_idx = find_next_bit(q_vector->rxr_idx,
1545                                               adapter->num_rx_queues,
1546                                               r_idx + 1);
1547                 }
1548                 r_idx = find_first_bit(q_vector->txr_idx,
1549                                        adapter->num_tx_queues);
1550
1551                 for (i = 0; i < q_vector->txr_count; i++) {
1552                         u8 reg_idx = adapter->tx_ring[r_idx]->reg_idx;
1553                         ixgbe_set_ivar(adapter, 1, reg_idx, v_idx);
1554                         r_idx = find_next_bit(q_vector->txr_idx,
1555                                               adapter->num_tx_queues,
1556                                               r_idx + 1);
1557                 }
1558
1559                 if (q_vector->txr_count && !q_vector->rxr_count)
1560                         /* tx only */
1561                         q_vector->eitr = adapter->tx_eitr_param;
1562                 else if (q_vector->rxr_count)
1563                         /* rx or mixed */
1564                         q_vector->eitr = adapter->rx_eitr_param;
1565
1566                 ixgbe_write_eitr(q_vector);
1567                 /* If Flow Director is enabled, set interrupt affinity */
1568                 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
1569                     (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)) {
1570                         /*
1571                          * Allocate the affinity_hint cpumask, assign the mask
1572                          * for this vector, and set our affinity_hint for
1573                          * this irq.
1574                          */
1575                         if (!alloc_cpumask_var(&q_vector->affinity_mask,
1576                                                GFP_KERNEL))
1577                                 return;
1578                         cpumask_set_cpu(v_idx, q_vector->affinity_mask);
1579                         irq_set_affinity_hint(adapter->msix_entries[v_idx].vector,
1580                                               q_vector->affinity_mask);
1581                 }
1582         }
1583
1584         switch (adapter->hw.mac.type) {
1585         case ixgbe_mac_82598EB:
1586                 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
1587                                v_idx);
1588                 break;
1589         case ixgbe_mac_82599EB:
1590         case ixgbe_mac_X540:
1591                 ixgbe_set_ivar(adapter, -1, 1, v_idx);
1592                 break;
1593
1594         default:
1595                 break;
1596         }
1597         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
1598
1599         /* set up to autoclear timer, and the vectors */
1600         mask = IXGBE_EIMS_ENABLE_MASK;
1601         if (adapter->num_vfs)
1602                 mask &= ~(IXGBE_EIMS_OTHER |
1603                           IXGBE_EIMS_MAILBOX |
1604                           IXGBE_EIMS_LSC);
1605         else
1606                 mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
1607         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
1608 }
1609
1610 enum latency_range {
1611         lowest_latency = 0,
1612         low_latency = 1,
1613         bulk_latency = 2,
1614         latency_invalid = 255
1615 };
1616
1617 /**
1618  * ixgbe_update_itr - update the dynamic ITR value based on statistics
1619  * @adapter: pointer to adapter
1620  * @eitr: eitr setting (ints per sec) to give last timeslice
1621  * @itr_setting: current throttle rate in ints/second
1622  * @packets: the number of packets during this measurement interval
1623  * @bytes: the number of bytes during this measurement interval
1624  *
1625  *      Stores a new ITR value based on packets and byte
1626  *      counts during the last interrupt.  The advantage of per interrupt
1627  *      computation is faster updates and more accurate ITR for the current
1628  *      traffic pattern.  Constants in this function were computed
1629  *      based on theoretical maximum wire speed and thresholds were set based
1630  *      on testing data as well as attempting to minimize response time
1631  *      while increasing bulk throughput.
1632  *      this functionality is controlled by the InterruptThrottleRate module
1633  *      parameter (see ixgbe_param.c)
1634  **/
1635 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
1636                            u32 eitr, u8 itr_setting,
1637                            int packets, int bytes)
1638 {
1639         unsigned int retval = itr_setting;
1640         u32 timepassed_us;
1641         u64 bytes_perint;
1642
1643         if (packets == 0)
1644                 goto update_itr_done;
1645
1646
1647         /* simple throttlerate management
1648          *    0-20MB/s lowest (100000 ints/s)
1649          *   20-100MB/s low   (20000 ints/s)
1650          *  100-1249MB/s bulk (8000 ints/s)
1651          */
1652         /* what was last interrupt timeslice? */
1653         timepassed_us = 1000000/eitr;
1654         bytes_perint = bytes / timepassed_us; /* bytes/usec */
1655
1656         switch (itr_setting) {
1657         case lowest_latency:
1658                 if (bytes_perint > adapter->eitr_low)
1659                         retval = low_latency;
1660                 break;
1661         case low_latency:
1662                 if (bytes_perint > adapter->eitr_high)
1663                         retval = bulk_latency;
1664                 else if (bytes_perint <= adapter->eitr_low)
1665                         retval = lowest_latency;
1666                 break;
1667         case bulk_latency:
1668                 if (bytes_perint <= adapter->eitr_high)
1669                         retval = low_latency;
1670                 break;
1671         }
1672
1673 update_itr_done:
1674         return retval;
1675 }
1676
1677 /**
1678  * ixgbe_write_eitr - write EITR register in hardware specific way
1679  * @q_vector: structure containing interrupt and ring information
1680  *
1681  * This function is made to be called by ethtool and by the driver
1682  * when it needs to update EITR registers at runtime.  Hardware
1683  * specific quirks/differences are taken care of here.
1684  */
1685 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
1686 {
1687         struct ixgbe_adapter *adapter = q_vector->adapter;
1688         struct ixgbe_hw *hw = &adapter->hw;
1689         int v_idx = q_vector->v_idx;
1690         u32 itr_reg = EITR_INTS_PER_SEC_TO_REG(q_vector->eitr);
1691
1692         switch (adapter->hw.mac.type) {
1693         case ixgbe_mac_82598EB:
1694                 /* must write high and low 16 bits to reset counter */
1695                 itr_reg |= (itr_reg << 16);
1696                 break;
1697         case ixgbe_mac_82599EB:
1698         case ixgbe_mac_X540:
1699                 /*
1700                  * 82599 and X540 can support a value of zero, so allow it for
1701                  * max interrupt rate, but there is an errata where it can
1702                  * not be zero with RSC
1703                  */
1704                 if (itr_reg == 8 &&
1705                     !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
1706                         itr_reg = 0;
1707
1708                 /*
1709                  * set the WDIS bit to not clear the timer bits and cause an
1710                  * immediate assertion of the interrupt
1711                  */
1712                 itr_reg |= IXGBE_EITR_CNT_WDIS;
1713                 break;
1714         default:
1715                 break;
1716         }
1717         IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
1718 }
1719
1720 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
1721 {
1722         struct ixgbe_adapter *adapter = q_vector->adapter;
1723         int i, r_idx;
1724         u32 new_itr;
1725         u8 current_itr, ret_itr;
1726
1727         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1728         for (i = 0; i < q_vector->txr_count; i++) {
1729                 struct ixgbe_ring *tx_ring = adapter->tx_ring[r_idx];
1730                 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1731                                            q_vector->tx_itr,
1732                                            tx_ring->total_packets,
1733                                            tx_ring->total_bytes);
1734                 /* if the result for this queue would decrease interrupt
1735                  * rate for this vector then use that result */
1736                 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
1737                                     q_vector->tx_itr - 1 : ret_itr);
1738                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1739                                       r_idx + 1);
1740         }
1741
1742         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1743         for (i = 0; i < q_vector->rxr_count; i++) {
1744                 struct ixgbe_ring *rx_ring = adapter->rx_ring[r_idx];
1745                 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
1746                                            q_vector->rx_itr,
1747                                            rx_ring->total_packets,
1748                                            rx_ring->total_bytes);
1749                 /* if the result for this queue would decrease interrupt
1750                  * rate for this vector then use that result */
1751                 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
1752                                     q_vector->rx_itr - 1 : ret_itr);
1753                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1754                                       r_idx + 1);
1755         }
1756
1757         current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1758
1759         switch (current_itr) {
1760         /* counts and packets in update_itr are dependent on these numbers */
1761         case lowest_latency:
1762                 new_itr = 100000;
1763                 break;
1764         case low_latency:
1765                 new_itr = 20000; /* aka hwitr = ~200 */
1766                 break;
1767         case bulk_latency:
1768         default:
1769                 new_itr = 8000;
1770                 break;
1771         }
1772
1773         if (new_itr != q_vector->eitr) {
1774                 /* do an exponential smoothing */
1775                 new_itr = ((q_vector->eitr * 9) + new_itr)/10;
1776
1777                 /* save the algorithm value here, not the smoothed one */
1778                 q_vector->eitr = new_itr;
1779
1780                 ixgbe_write_eitr(q_vector);
1781         }
1782 }
1783
1784 /**
1785  * ixgbe_check_overtemp_task - worker thread to check over tempurature
1786  * @work: pointer to work_struct containing our data
1787  **/
1788 static void ixgbe_check_overtemp_task(struct work_struct *work)
1789 {
1790         struct ixgbe_adapter *adapter = container_of(work,
1791                                                      struct ixgbe_adapter,
1792                                                      check_overtemp_task);
1793         struct ixgbe_hw *hw = &adapter->hw;
1794         u32 eicr = adapter->interrupt_event;
1795
1796         if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
1797                 return;
1798
1799         switch (hw->device_id) {
1800         case IXGBE_DEV_ID_82599_T3_LOM: {
1801                 u32 autoneg;
1802                 bool link_up = false;
1803
1804                 if (hw->mac.ops.check_link)
1805                         hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
1806
1807                 if (((eicr & IXGBE_EICR_GPI_SDP0) && (!link_up)) ||
1808                     (eicr & IXGBE_EICR_LSC))
1809                         /* Check if this is due to overtemp */
1810                         if (hw->phy.ops.check_overtemp(hw) == IXGBE_ERR_OVERTEMP)
1811                                 break;
1812                 return;
1813         }
1814         default:
1815                 if (!(eicr & IXGBE_EICR_GPI_SDP0))
1816                         return;
1817                 break;
1818         }
1819         e_crit(drv,
1820                "Network adapter has been stopped because it has over heated. "
1821                "Restart the computer. If the problem persists, "
1822                "power off the system and replace the adapter\n");
1823         /* write to clear the interrupt */
1824         IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP0);
1825 }
1826
1827 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
1828 {
1829         struct ixgbe_hw *hw = &adapter->hw;
1830
1831         if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
1832             (eicr & IXGBE_EICR_GPI_SDP1)) {
1833                 e_crit(probe, "Fan has stopped, replace the adapter\n");
1834                 /* write to clear the interrupt */
1835                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1836         }
1837 }
1838
1839 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
1840 {
1841         struct ixgbe_hw *hw = &adapter->hw;
1842
1843         if (eicr & IXGBE_EICR_GPI_SDP2) {
1844                 /* Clear the interrupt */
1845                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
1846                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1847                         schedule_work(&adapter->sfp_config_module_task);
1848         }
1849
1850         if (eicr & IXGBE_EICR_GPI_SDP1) {
1851                 /* Clear the interrupt */
1852                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
1853                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1854                         schedule_work(&adapter->multispeed_fiber_task);
1855         }
1856 }
1857
1858 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
1859 {
1860         struct ixgbe_hw *hw = &adapter->hw;
1861
1862         adapter->lsc_int++;
1863         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
1864         adapter->link_check_timeout = jiffies;
1865         if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1866                 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
1867                 IXGBE_WRITE_FLUSH(hw);
1868                 schedule_work(&adapter->watchdog_task);
1869         }
1870 }
1871
1872 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
1873 {
1874         struct net_device *netdev = data;
1875         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1876         struct ixgbe_hw *hw = &adapter->hw;
1877         u32 eicr;
1878
1879         /*
1880          * Workaround for Silicon errata.  Use clear-by-write instead
1881          * of clear-by-read.  Reading with EICS will return the
1882          * interrupt causes without clearing, which later be done
1883          * with the write to EICR.
1884          */
1885         eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
1886         IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
1887
1888         if (eicr & IXGBE_EICR_LSC)
1889                 ixgbe_check_lsc(adapter);
1890
1891         if (eicr & IXGBE_EICR_MAILBOX)
1892                 ixgbe_msg_task(adapter);
1893
1894         switch (hw->mac.type) {
1895         case ixgbe_mac_82599EB:
1896         case ixgbe_mac_X540:
1897                 /* Handle Flow Director Full threshold interrupt */
1898                 if (eicr & IXGBE_EICR_FLOW_DIR) {
1899                         int i;
1900                         IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
1901                         /* Disable transmits before FDIR Re-initialization */
1902                         netif_tx_stop_all_queues(netdev);
1903                         for (i = 0; i < adapter->num_tx_queues; i++) {
1904                                 struct ixgbe_ring *tx_ring =
1905                                                             adapter->tx_ring[i];
1906                                 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
1907                                                        &tx_ring->state))
1908                                         schedule_work(&adapter->fdir_reinit_task);
1909                         }
1910                 }
1911                 ixgbe_check_sfp_event(adapter, eicr);
1912                 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
1913                     ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC))) {
1914                         adapter->interrupt_event = eicr;
1915                         schedule_work(&adapter->check_overtemp_task);
1916                 }
1917                 break;
1918         default:
1919                 break;
1920         }
1921
1922         ixgbe_check_fan_failure(adapter, eicr);
1923
1924         if (!test_bit(__IXGBE_DOWN, &adapter->state))
1925                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
1926
1927         return IRQ_HANDLED;
1928 }
1929
1930 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
1931                                            u64 qmask)
1932 {
1933         u32 mask;
1934         struct ixgbe_hw *hw = &adapter->hw;
1935
1936         switch (hw->mac.type) {
1937         case ixgbe_mac_82598EB:
1938                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1939                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
1940                 break;
1941         case ixgbe_mac_82599EB:
1942         case ixgbe_mac_X540:
1943                 mask = (qmask & 0xFFFFFFFF);
1944                 if (mask)
1945                         IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
1946                 mask = (qmask >> 32);
1947                 if (mask)
1948                         IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
1949                 break;
1950         default:
1951                 break;
1952         }
1953         /* skip the flush */
1954 }
1955
1956 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
1957                                             u64 qmask)
1958 {
1959         u32 mask;
1960         struct ixgbe_hw *hw = &adapter->hw;
1961
1962         switch (hw->mac.type) {
1963         case ixgbe_mac_82598EB:
1964                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
1965                 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
1966                 break;
1967         case ixgbe_mac_82599EB:
1968         case ixgbe_mac_X540:
1969                 mask = (qmask & 0xFFFFFFFF);
1970                 if (mask)
1971                         IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
1972                 mask = (qmask >> 32);
1973                 if (mask)
1974                         IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
1975                 break;
1976         default:
1977                 break;
1978         }
1979         /* skip the flush */
1980 }
1981
1982 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
1983 {
1984         struct ixgbe_q_vector *q_vector = data;
1985         struct ixgbe_adapter  *adapter = q_vector->adapter;
1986         struct ixgbe_ring     *tx_ring;
1987         int i, r_idx;
1988
1989         if (!q_vector->txr_count)
1990                 return IRQ_HANDLED;
1991
1992         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
1993         for (i = 0; i < q_vector->txr_count; i++) {
1994                 tx_ring = adapter->tx_ring[r_idx];
1995                 tx_ring->total_bytes = 0;
1996                 tx_ring->total_packets = 0;
1997                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
1998                                       r_idx + 1);
1999         }
2000
2001         /* EIAM disabled interrupts (on this vector) for us */
2002         napi_schedule(&q_vector->napi);
2003
2004         return IRQ_HANDLED;
2005 }
2006
2007 /**
2008  * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
2009  * @irq: unused
2010  * @data: pointer to our q_vector struct for this interrupt vector
2011  **/
2012 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
2013 {
2014         struct ixgbe_q_vector *q_vector = data;
2015         struct ixgbe_adapter  *adapter = q_vector->adapter;
2016         struct ixgbe_ring  *rx_ring;
2017         int r_idx;
2018         int i;
2019
2020 #ifdef CONFIG_IXGBE_DCA
2021         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2022                 ixgbe_update_dca(q_vector);
2023 #endif
2024
2025         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2026         for (i = 0; i < q_vector->rxr_count; i++) {
2027                 rx_ring = adapter->rx_ring[r_idx];
2028                 rx_ring->total_bytes = 0;
2029                 rx_ring->total_packets = 0;
2030                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
2031                                       r_idx + 1);
2032         }
2033
2034         if (!q_vector->rxr_count)
2035                 return IRQ_HANDLED;
2036
2037         /* EIAM disabled interrupts (on this vector) for us */
2038         napi_schedule(&q_vector->napi);
2039
2040         return IRQ_HANDLED;
2041 }
2042
2043 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
2044 {
2045         struct ixgbe_q_vector *q_vector = data;
2046         struct ixgbe_adapter  *adapter = q_vector->adapter;
2047         struct ixgbe_ring  *ring;
2048         int r_idx;
2049         int i;
2050
2051         if (!q_vector->txr_count && !q_vector->rxr_count)
2052                 return IRQ_HANDLED;
2053
2054         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2055         for (i = 0; i < q_vector->txr_count; i++) {
2056                 ring = adapter->tx_ring[r_idx];
2057                 ring->total_bytes = 0;
2058                 ring->total_packets = 0;
2059                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
2060                                       r_idx + 1);
2061         }
2062
2063         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2064         for (i = 0; i < q_vector->rxr_count; i++) {
2065                 ring = adapter->rx_ring[r_idx];
2066                 ring->total_bytes = 0;
2067                 ring->total_packets = 0;
2068                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
2069                                       r_idx + 1);
2070         }
2071
2072         /* EIAM disabled interrupts (on this vector) for us */
2073         napi_schedule(&q_vector->napi);
2074
2075         return IRQ_HANDLED;
2076 }
2077
2078 /**
2079  * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
2080  * @napi: napi struct with our devices info in it
2081  * @budget: amount of work driver is allowed to do this pass, in packets
2082  *
2083  * This function is optimized for cleaning one queue only on a single
2084  * q_vector!!!
2085  **/
2086 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
2087 {
2088         struct ixgbe_q_vector *q_vector =
2089                                container_of(napi, struct ixgbe_q_vector, napi);
2090         struct ixgbe_adapter *adapter = q_vector->adapter;
2091         struct ixgbe_ring *rx_ring = NULL;
2092         int work_done = 0;
2093         long r_idx;
2094
2095 #ifdef CONFIG_IXGBE_DCA
2096         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2097                 ixgbe_update_dca(q_vector);
2098 #endif
2099
2100         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2101         rx_ring = adapter->rx_ring[r_idx];
2102
2103         ixgbe_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
2104
2105         /* If all Rx work done, exit the polling mode */
2106         if (work_done < budget) {
2107                 napi_complete(napi);
2108                 if (adapter->rx_itr_setting & 1)
2109                         ixgbe_set_itr_msix(q_vector);
2110                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2111                         ixgbe_irq_enable_queues(adapter,
2112                                                 ((u64)1 << q_vector->v_idx));
2113         }
2114
2115         return work_done;
2116 }
2117
2118 /**
2119  * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
2120  * @napi: napi struct with our devices info in it
2121  * @budget: amount of work driver is allowed to do this pass, in packets
2122  *
2123  * This function will clean more than one rx queue associated with a
2124  * q_vector.
2125  **/
2126 static int ixgbe_clean_rxtx_many(struct napi_struct *napi, int budget)
2127 {
2128         struct ixgbe_q_vector *q_vector =
2129                                container_of(napi, struct ixgbe_q_vector, napi);
2130         struct ixgbe_adapter *adapter = q_vector->adapter;
2131         struct ixgbe_ring *ring = NULL;
2132         int work_done = 0, i;
2133         long r_idx;
2134         bool tx_clean_complete = true;
2135
2136 #ifdef CONFIG_IXGBE_DCA
2137         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2138                 ixgbe_update_dca(q_vector);
2139 #endif
2140
2141         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2142         for (i = 0; i < q_vector->txr_count; i++) {
2143                 ring = adapter->tx_ring[r_idx];
2144                 tx_clean_complete &= ixgbe_clean_tx_irq(q_vector, ring);
2145                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
2146                                       r_idx + 1);
2147         }
2148
2149         /* attempt to distribute budget to each queue fairly, but don't allow
2150          * the budget to go below 1 because we'll exit polling */
2151         budget /= (q_vector->rxr_count ?: 1);
2152         budget = max(budget, 1);
2153         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2154         for (i = 0; i < q_vector->rxr_count; i++) {
2155                 ring = adapter->rx_ring[r_idx];
2156                 ixgbe_clean_rx_irq(q_vector, ring, &work_done, budget);
2157                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
2158                                       r_idx + 1);
2159         }
2160
2161         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
2162         ring = adapter->rx_ring[r_idx];
2163         /* If all Rx work done, exit the polling mode */
2164         if (work_done < budget) {
2165                 napi_complete(napi);
2166                 if (adapter->rx_itr_setting & 1)
2167                         ixgbe_set_itr_msix(q_vector);
2168                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2169                         ixgbe_irq_enable_queues(adapter,
2170                                                 ((u64)1 << q_vector->v_idx));
2171                 return 0;
2172         }
2173
2174         return work_done;
2175 }
2176
2177 /**
2178  * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
2179  * @napi: napi struct with our devices info in it
2180  * @budget: amount of work driver is allowed to do this pass, in packets
2181  *
2182  * This function is optimized for cleaning one queue only on a single
2183  * q_vector!!!
2184  **/
2185 static int ixgbe_clean_txonly(struct napi_struct *napi, int budget)
2186 {
2187         struct ixgbe_q_vector *q_vector =
2188                                container_of(napi, struct ixgbe_q_vector, napi);
2189         struct ixgbe_adapter *adapter = q_vector->adapter;
2190         struct ixgbe_ring *tx_ring = NULL;
2191         int work_done = 0;
2192         long r_idx;
2193
2194 #ifdef CONFIG_IXGBE_DCA
2195         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2196                 ixgbe_update_dca(q_vector);
2197 #endif
2198
2199         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
2200         tx_ring = adapter->tx_ring[r_idx];
2201
2202         if (!ixgbe_clean_tx_irq(q_vector, tx_ring))
2203                 work_done = budget;
2204
2205         /* If all Tx work done, exit the polling mode */
2206         if (work_done < budget) {
2207                 napi_complete(napi);
2208                 if (adapter->tx_itr_setting & 1)
2209                         ixgbe_set_itr_msix(q_vector);
2210                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2211                         ixgbe_irq_enable_queues(adapter,
2212                                                 ((u64)1 << q_vector->v_idx));
2213         }
2214
2215         return work_done;
2216 }
2217
2218 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
2219                                      int r_idx)
2220 {
2221         struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2222         struct ixgbe_ring *rx_ring = a->rx_ring[r_idx];
2223
2224         set_bit(r_idx, q_vector->rxr_idx);
2225         q_vector->rxr_count++;
2226         rx_ring->q_vector = q_vector;
2227 }
2228
2229 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
2230                                      int t_idx)
2231 {
2232         struct ixgbe_q_vector *q_vector = a->q_vector[v_idx];
2233         struct ixgbe_ring *tx_ring = a->tx_ring[t_idx];
2234
2235         set_bit(t_idx, q_vector->txr_idx);
2236         q_vector->txr_count++;
2237         tx_ring->q_vector = q_vector;
2238 }
2239
2240 /**
2241  * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
2242  * @adapter: board private structure to initialize
2243  *
2244  * This function maps descriptor rings to the queue-specific vectors
2245  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
2246  * one vector per ring/queue, but on a constrained vector budget, we
2247  * group the rings as "efficiently" as possible.  You would add new
2248  * mapping configurations in here.
2249  **/
2250 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter)
2251 {
2252         int q_vectors;
2253         int v_start = 0;
2254         int rxr_idx = 0, txr_idx = 0;
2255         int rxr_remaining = adapter->num_rx_queues;
2256         int txr_remaining = adapter->num_tx_queues;
2257         int i, j;
2258         int rqpv, tqpv;
2259         int err = 0;
2260
2261         /* No mapping required if MSI-X is disabled. */
2262         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2263                 goto out;
2264
2265         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2266
2267         /*
2268          * The ideal configuration...
2269          * We have enough vectors to map one per queue.
2270          */
2271         if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
2272                 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
2273                         map_vector_to_rxq(adapter, v_start, rxr_idx);
2274
2275                 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
2276                         map_vector_to_txq(adapter, v_start, txr_idx);
2277
2278                 goto out;
2279         }
2280
2281         /*
2282          * If we don't have enough vectors for a 1-to-1
2283          * mapping, we'll have to group them so there are
2284          * multiple queues per vector.
2285          */
2286         /* Re-adjusting *qpv takes care of the remainder. */
2287         for (i = v_start; i < q_vectors; i++) {
2288                 rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
2289                 for (j = 0; j < rqpv; j++) {
2290                         map_vector_to_rxq(adapter, i, rxr_idx);
2291                         rxr_idx++;
2292                         rxr_remaining--;
2293                 }
2294                 tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
2295                 for (j = 0; j < tqpv; j++) {
2296                         map_vector_to_txq(adapter, i, txr_idx);
2297                         txr_idx++;
2298                         txr_remaining--;
2299                 }
2300         }
2301 out:
2302         return err;
2303 }
2304
2305 /**
2306  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2307  * @adapter: board private structure
2308  *
2309  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2310  * interrupts from the kernel.
2311  **/
2312 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2313 {
2314         struct net_device *netdev = adapter->netdev;
2315         irqreturn_t (*handler)(int, void *);
2316         int i, vector, q_vectors, err;
2317         int ri = 0, ti = 0;
2318
2319         /* Decrement for Other and TCP Timer vectors */
2320         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2321
2322         err = ixgbe_map_rings_to_vectors(adapter);
2323         if (err)
2324                 return err;
2325
2326 #define SET_HANDLER(_v) (((_v)->rxr_count && (_v)->txr_count)        \
2327                                           ? &ixgbe_msix_clean_many : \
2328                           (_v)->rxr_count ? &ixgbe_msix_clean_rx   : \
2329                           (_v)->txr_count ? &ixgbe_msix_clean_tx   : \
2330                           NULL)
2331         for (vector = 0; vector < q_vectors; vector++) {
2332                 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2333                 handler = SET_HANDLER(q_vector);
2334
2335                 if (handler == &ixgbe_msix_clean_rx) {
2336                         sprintf(q_vector->name, "%s-%s-%d",
2337                                 netdev->name, "rx", ri++);
2338                 } else if (handler == &ixgbe_msix_clean_tx) {
2339                         sprintf(q_vector->name, "%s-%s-%d",
2340                                 netdev->name, "tx", ti++);
2341                 } else if (handler == &ixgbe_msix_clean_many) {
2342                         sprintf(q_vector->name, "%s-%s-%d",
2343                                 netdev->name, "TxRx", ri++);
2344                         ti++;
2345                 } else {
2346                         /* skip this unused q_vector */
2347                         continue;
2348                 }
2349                 err = request_irq(adapter->msix_entries[vector].vector,
2350                                   handler, 0, q_vector->name,
2351                                   q_vector);
2352                 if (err) {
2353                         e_err(probe, "request_irq failed for MSIX interrupt "
2354                               "Error: %d\n", err);
2355                         goto free_queue_irqs;
2356                 }
2357         }
2358
2359         sprintf(adapter->lsc_int_name, "%s:lsc", netdev->name);
2360         err = request_irq(adapter->msix_entries[vector].vector,
2361                           ixgbe_msix_lsc, 0, adapter->lsc_int_name, netdev);
2362         if (err) {
2363                 e_err(probe, "request_irq for msix_lsc failed: %d\n", err);
2364                 goto free_queue_irqs;
2365         }
2366
2367         return 0;
2368
2369 free_queue_irqs:
2370         for (i = vector - 1; i >= 0; i--)
2371                 free_irq(adapter->msix_entries[--vector].vector,
2372                          adapter->q_vector[i]);
2373         adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2374         pci_disable_msix(adapter->pdev);
2375         kfree(adapter->msix_entries);
2376         adapter->msix_entries = NULL;
2377         return err;
2378 }
2379
2380 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
2381 {
2382         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2383         struct ixgbe_ring *rx_ring = adapter->rx_ring[0];
2384         struct ixgbe_ring *tx_ring = adapter->tx_ring[0];
2385         u32 new_itr = q_vector->eitr;
2386         u8 current_itr;
2387
2388         q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
2389                                             q_vector->tx_itr,
2390                                             tx_ring->total_packets,
2391                                             tx_ring->total_bytes);
2392         q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
2393                                             q_vector->rx_itr,
2394                                             rx_ring->total_packets,
2395                                             rx_ring->total_bytes);
2396
2397         current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
2398
2399         switch (current_itr) {
2400         /* counts and packets in update_itr are dependent on these numbers */
2401         case lowest_latency:
2402                 new_itr = 100000;
2403                 break;
2404         case low_latency:
2405                 new_itr = 20000; /* aka hwitr = ~200 */
2406                 break;
2407         case bulk_latency:
2408                 new_itr = 8000;
2409                 break;
2410         default:
2411                 break;
2412         }
2413
2414         if (new_itr != q_vector->eitr) {
2415                 /* do an exponential smoothing */
2416                 new_itr = ((q_vector->eitr * 9) + new_itr)/10;
2417
2418                 /* save the algorithm value here */
2419                 q_vector->eitr = new_itr;
2420
2421                 ixgbe_write_eitr(q_vector);
2422         }
2423 }
2424
2425 /**
2426  * ixgbe_irq_enable - Enable default interrupt generation settings
2427  * @adapter: board private structure
2428  **/
2429 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2430                                     bool flush)
2431 {
2432         u32 mask;
2433
2434         mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2435         if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2436                 mask |= IXGBE_EIMS_GPI_SDP0;
2437         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2438                 mask |= IXGBE_EIMS_GPI_SDP1;
2439         switch (adapter->hw.mac.type) {
2440         case ixgbe_mac_82599EB:
2441         case ixgbe_mac_X540:
2442                 mask |= IXGBE_EIMS_ECC;
2443                 mask |= IXGBE_EIMS_GPI_SDP1;
2444                 mask |= IXGBE_EIMS_GPI_SDP2;
2445                 if (adapter->num_vfs)
2446                         mask |= IXGBE_EIMS_MAILBOX;
2447                 break;
2448         default:
2449                 break;
2450         }
2451         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
2452             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
2453                 mask |= IXGBE_EIMS_FLOW_DIR;
2454
2455         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2456         if (queues)
2457                 ixgbe_irq_enable_queues(adapter, ~0);
2458         if (flush)
2459                 IXGBE_WRITE_FLUSH(&adapter->hw);
2460
2461         if (adapter->num_vfs > 32) {
2462                 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2463                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2464         }
2465 }
2466
2467 /**
2468  * ixgbe_intr - legacy mode Interrupt Handler
2469  * @irq: interrupt number
2470  * @data: pointer to a network interface device structure
2471  **/
2472 static irqreturn_t ixgbe_intr(int irq, void *data)
2473 {
2474         struct net_device *netdev = data;
2475         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2476         struct ixgbe_hw *hw = &adapter->hw;
2477         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2478         u32 eicr;
2479
2480         /*
2481          * Workaround for silicon errata on 82598.  Mask the interrupts
2482          * before the read of EICR.
2483          */
2484         IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2485
2486         /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2487          * therefore no explict interrupt disable is necessary */
2488         eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2489         if (!eicr) {
2490                 /*
2491                  * shared interrupt alert!
2492                  * make sure interrupts are enabled because the read will
2493                  * have disabled interrupts due to EIAM
2494                  * finish the workaround of silicon errata on 82598.  Unmask
2495                  * the interrupt that we masked before the EICR read.
2496                  */
2497                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2498                         ixgbe_irq_enable(adapter, true, true);
2499                 return IRQ_NONE;        /* Not our interrupt */
2500         }
2501
2502         if (eicr & IXGBE_EICR_LSC)
2503                 ixgbe_check_lsc(adapter);
2504
2505         switch (hw->mac.type) {
2506         case ixgbe_mac_82599EB:
2507         case ixgbe_mac_X540:
2508                 ixgbe_check_sfp_event(adapter, eicr);
2509                 if ((adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2510                     ((eicr & IXGBE_EICR_GPI_SDP0) || (eicr & IXGBE_EICR_LSC))) {
2511                         adapter->interrupt_event = eicr;
2512                         schedule_work(&adapter->check_overtemp_task);
2513                 }
2514                 break;
2515         default:
2516                 break;
2517         }
2518
2519         ixgbe_check_fan_failure(adapter, eicr);
2520
2521         if (napi_schedule_prep(&(q_vector->napi))) {
2522                 adapter->tx_ring[0]->total_packets = 0;
2523                 adapter->tx_ring[0]->total_bytes = 0;
2524                 adapter->rx_ring[0]->total_packets = 0;
2525                 adapter->rx_ring[0]->total_bytes = 0;
2526                 /* would disable interrupts here but EIAM disabled it */
2527                 __napi_schedule(&(q_vector->napi));
2528         }
2529
2530         /*
2531          * re-enable link(maybe) and non-queue interrupts, no flush.
2532          * ixgbe_poll will re-enable the queue interrupts
2533          */
2534
2535         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2536                 ixgbe_irq_enable(adapter, false, false);
2537
2538         return IRQ_HANDLED;
2539 }
2540
2541 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
2542 {
2543         int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2544
2545         for (i = 0; i < q_vectors; i++) {
2546                 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
2547                 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
2548                 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
2549                 q_vector->rxr_count = 0;
2550                 q_vector->txr_count = 0;
2551         }
2552 }
2553
2554 /**
2555  * ixgbe_request_irq - initialize interrupts
2556  * @adapter: board private structure
2557  *
2558  * Attempts to configure interrupts using the best available
2559  * capabilities of the hardware and kernel.
2560  **/
2561 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2562 {
2563         struct net_device *netdev = adapter->netdev;
2564         int err;
2565
2566         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2567                 err = ixgbe_request_msix_irqs(adapter);
2568         } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
2569                 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2570                                   netdev->name, netdev);
2571         } else {
2572                 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2573                                   netdev->name, netdev);
2574         }
2575
2576         if (err)
2577                 e_err(probe, "request_irq failed, Error %d\n", err);
2578
2579         return err;
2580 }
2581
2582 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2583 {
2584         struct net_device *netdev = adapter->netdev;
2585
2586         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2587                 int i, q_vectors;
2588
2589                 q_vectors = adapter->num_msix_vectors;
2590
2591                 i = q_vectors - 1;
2592                 free_irq(adapter->msix_entries[i].vector, netdev);
2593
2594                 i--;
2595                 for (; i >= 0; i--) {
2596                         free_irq(adapter->msix_entries[i].vector,
2597                                  adapter->q_vector[i]);
2598                 }
2599
2600                 ixgbe_reset_q_vectors(adapter);
2601         } else {
2602                 free_irq(adapter->pdev->irq, netdev);
2603         }
2604 }
2605
2606 /**
2607  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2608  * @adapter: board private structure
2609  **/
2610 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2611 {
2612         switch (adapter->hw.mac.type) {
2613         case ixgbe_mac_82598EB:
2614                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
2615                 break;
2616         case ixgbe_mac_82599EB:
2617         case ixgbe_mac_X540:
2618                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
2619                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
2620                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
2621                 if (adapter->num_vfs > 32)
2622                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
2623                 break;
2624         default:
2625                 break;
2626         }
2627         IXGBE_WRITE_FLUSH(&adapter->hw);
2628         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2629                 int i;
2630                 for (i = 0; i < adapter->num_msix_vectors; i++)
2631                         synchronize_irq(adapter->msix_entries[i].vector);
2632         } else {
2633                 synchronize_irq(adapter->pdev->irq);
2634         }
2635 }
2636
2637 /**
2638  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
2639  *
2640  **/
2641 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
2642 {
2643         struct ixgbe_hw *hw = &adapter->hw;
2644
2645         IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
2646                         EITR_INTS_PER_SEC_TO_REG(adapter->rx_eitr_param));
2647
2648         ixgbe_set_ivar(adapter, 0, 0, 0);
2649         ixgbe_set_ivar(adapter, 1, 0, 0);
2650
2651         map_vector_to_rxq(adapter, 0, 0);
2652         map_vector_to_txq(adapter, 0, 0);
2653
2654         e_info(hw, "Legacy interrupt IVAR setup done\n");
2655 }
2656
2657 /**
2658  * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
2659  * @adapter: board private structure
2660  * @ring: structure containing ring specific data
2661  *
2662  * Configure the Tx descriptor ring after a reset.
2663  **/
2664 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
2665                              struct ixgbe_ring *ring)
2666 {
2667         struct ixgbe_hw *hw = &adapter->hw;
2668         u64 tdba = ring->dma;
2669         int wait_loop = 10;
2670         u32 txdctl;
2671         u8 reg_idx = ring->reg_idx;
2672
2673         /* disable queue to avoid issues while updating state */
2674         txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2675         IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx),
2676                         txdctl & ~IXGBE_TXDCTL_ENABLE);
2677         IXGBE_WRITE_FLUSH(hw);
2678
2679         IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
2680                         (tdba & DMA_BIT_MASK(32)));
2681         IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
2682         IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
2683                         ring->count * sizeof(union ixgbe_adv_tx_desc));
2684         IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
2685         IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
2686         ring->tail = hw->hw_addr + IXGBE_TDT(reg_idx);
2687
2688         /* configure fetching thresholds */
2689         if (adapter->rx_itr_setting == 0) {
2690                 /* cannot set wthresh when itr==0 */
2691                 txdctl &= ~0x007F0000;
2692         } else {
2693                 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2694                 txdctl |= (8 << 16);
2695         }
2696         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
2697                 /* PThresh workaround for Tx hang with DFP enabled. */
2698                 txdctl |= 32;
2699         }
2700
2701         /* reinitialize flowdirector state */
2702         if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2703             adapter->atr_sample_rate) {
2704                 ring->atr_sample_rate = adapter->atr_sample_rate;
2705                 ring->atr_count = 0;
2706                 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
2707         } else {
2708                 ring->atr_sample_rate = 0;
2709         }
2710
2711         clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
2712
2713         /* enable queue */
2714         txdctl |= IXGBE_TXDCTL_ENABLE;
2715         IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
2716
2717         /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
2718         if (hw->mac.type == ixgbe_mac_82598EB &&
2719             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
2720                 return;
2721
2722         /* poll to verify queue is enabled */
2723         do {
2724                 msleep(1);
2725                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
2726         } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
2727         if (!wait_loop)
2728                 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
2729 }
2730
2731 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
2732 {
2733         struct ixgbe_hw *hw = &adapter->hw;
2734         u32 rttdcs;
2735         u32 mask;
2736
2737         if (hw->mac.type == ixgbe_mac_82598EB)
2738                 return;
2739
2740         /* disable the arbiter while setting MTQC */
2741         rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
2742         rttdcs |= IXGBE_RTTDCS_ARBDIS;
2743         IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2744
2745         /* set transmit pool layout */
2746         mask = (IXGBE_FLAG_SRIOV_ENABLED | IXGBE_FLAG_DCB_ENABLED);
2747         switch (adapter->flags & mask) {
2748
2749         case (IXGBE_FLAG_SRIOV_ENABLED):
2750                 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2751                                 (IXGBE_MTQC_VT_ENA | IXGBE_MTQC_64VF));
2752                 break;
2753
2754         case (IXGBE_FLAG_DCB_ENABLED):
2755                 /* We enable 8 traffic classes, DCB only */
2756                 IXGBE_WRITE_REG(hw, IXGBE_MTQC,
2757                               (IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ));
2758                 break;
2759
2760         default:
2761                 IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
2762                 break;
2763         }
2764
2765         /* re-enable the arbiter */
2766         rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
2767         IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
2768 }
2769
2770 /**
2771  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
2772  * @adapter: board private structure
2773  *
2774  * Configure the Tx unit of the MAC after a reset.
2775  **/
2776 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
2777 {
2778         struct ixgbe_hw *hw = &adapter->hw;
2779         u32 dmatxctl;
2780         u32 i;
2781
2782         ixgbe_setup_mtqc(adapter);
2783
2784         if (hw->mac.type != ixgbe_mac_82598EB) {
2785                 /* DMATXCTL.EN must be before Tx queues are enabled */
2786                 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
2787                 dmatxctl |= IXGBE_DMATXCTL_TE;
2788                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
2789         }
2790
2791         /* Setup the HW Tx Head and Tail descriptor pointers */
2792         for (i = 0; i < adapter->num_tx_queues; i++)
2793                 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
2794 }
2795
2796 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
2797
2798 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
2799                                    struct ixgbe_ring *rx_ring)
2800 {
2801         u32 srrctl;
2802         u8 reg_idx = rx_ring->reg_idx;
2803
2804         switch (adapter->hw.mac.type) {
2805         case ixgbe_mac_82598EB: {
2806                 struct ixgbe_ring_feature *feature = adapter->ring_feature;
2807                 const int mask = feature[RING_F_RSS].mask;
2808                 reg_idx = reg_idx & mask;
2809         }
2810                 break;
2811         case ixgbe_mac_82599EB:
2812         case ixgbe_mac_X540:
2813         default:
2814                 break;
2815         }
2816
2817         srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx));
2818
2819         srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
2820         srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
2821         if (adapter->num_vfs)
2822                 srrctl |= IXGBE_SRRCTL_DROP_EN;
2823
2824         srrctl |= (IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
2825                   IXGBE_SRRCTL_BSIZEHDR_MASK;
2826
2827         if (ring_is_ps_enabled(rx_ring)) {
2828 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
2829                 srrctl |= IXGBE_MAX_RXBUFFER >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2830 #else
2831                 srrctl |= (PAGE_SIZE / 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2832 #endif
2833                 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2834         } else {
2835                 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
2836                           IXGBE_SRRCTL_BSIZEPKT_SHIFT;
2837                 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
2838         }
2839
2840         IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(reg_idx), srrctl);
2841 }
2842
2843 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
2844 {
2845         struct ixgbe_hw *hw = &adapter->hw;
2846         static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
2847                           0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
2848                           0x6A3E67EA, 0x14364D17, 0x3BED200D};
2849         u32 mrqc = 0, reta = 0;
2850         u32 rxcsum;
2851         int i, j;
2852         int mask;
2853
2854         /* Fill out hash function seeds */
2855         for (i = 0; i < 10; i++)
2856                 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
2857
2858         /* Fill out redirection table */
2859         for (i = 0, j = 0; i < 128; i++, j++) {
2860                 if (j == adapter->ring_feature[RING_F_RSS].indices)
2861                         j = 0;
2862                 /* reta = 4-byte sliding window of
2863                  * 0x00..(indices-1)(indices-1)00..etc. */
2864                 reta = (reta << 8) | (j * 0x11);
2865                 if ((i & 3) == 3)
2866                         IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
2867         }
2868
2869         /* Disable indicating checksum in descriptor, enables RSS hash */
2870         rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
2871         rxcsum |= IXGBE_RXCSUM_PCSD;
2872         IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
2873
2874         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
2875                 mask = adapter->flags & IXGBE_FLAG_RSS_ENABLED;
2876         else
2877                 mask = adapter->flags & (IXGBE_FLAG_RSS_ENABLED
2878 #ifdef CONFIG_IXGBE_DCB
2879                                          | IXGBE_FLAG_DCB_ENABLED
2880 #endif
2881                                          | IXGBE_FLAG_SRIOV_ENABLED
2882                                         );
2883
2884         switch (mask) {
2885         case (IXGBE_FLAG_RSS_ENABLED):
2886                 mrqc = IXGBE_MRQC_RSSEN;
2887                 break;
2888         case (IXGBE_FLAG_SRIOV_ENABLED):
2889                 mrqc = IXGBE_MRQC_VMDQEN;
2890                 break;
2891 #ifdef CONFIG_IXGBE_DCB
2892         case (IXGBE_FLAG_DCB_ENABLED):
2893                 mrqc = IXGBE_MRQC_RT8TCEN;
2894                 break;
2895 #endif /* CONFIG_IXGBE_DCB */
2896         default:
2897                 break;
2898         }
2899
2900         /* Perform hash on these packet types */
2901         mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4
2902               | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2903               | IXGBE_MRQC_RSS_FIELD_IPV6
2904               | IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
2905
2906         IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
2907 }
2908
2909 /**
2910  * ixgbe_clear_rscctl - disable RSC for the indicated ring
2911  * @adapter: address of board private structure
2912  * @ring: structure containing ring specific data
2913  **/
2914 void ixgbe_clear_rscctl(struct ixgbe_adapter *adapter,
2915                         struct ixgbe_ring *ring)
2916 {
2917         struct ixgbe_hw *hw = &adapter->hw;
2918         u32 rscctrl;
2919         u8 reg_idx = ring->reg_idx;
2920
2921         rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2922         rscctrl &= ~IXGBE_RSCCTL_RSCEN;
2923         IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2924 }
2925
2926 /**
2927  * ixgbe_configure_rscctl - enable RSC for the indicated ring
2928  * @adapter:    address of board private structure
2929  * @index:      index of ring to set
2930  **/
2931 void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
2932                                    struct ixgbe_ring *ring)
2933 {
2934         struct ixgbe_hw *hw = &adapter->hw;
2935         u32 rscctrl;
2936         int rx_buf_len;
2937         u8 reg_idx = ring->reg_idx;
2938
2939         if (!ring_is_rsc_enabled(ring))
2940                 return;
2941
2942         rx_buf_len = ring->rx_buf_len;
2943         rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
2944         rscctrl |= IXGBE_RSCCTL_RSCEN;
2945         /*
2946          * we must limit the number of descriptors so that the
2947          * total size of max desc * buf_len is not greater
2948          * than 65535
2949          */
2950         if (ring_is_ps_enabled(ring)) {
2951 #if (MAX_SKB_FRAGS > 16)
2952                 rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2953 #elif (MAX_SKB_FRAGS > 8)
2954                 rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2955 #elif (MAX_SKB_FRAGS > 4)
2956                 rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2957 #else
2958                 rscctrl |= IXGBE_RSCCTL_MAXDESC_1;
2959 #endif
2960         } else {
2961                 if (rx_buf_len < IXGBE_RXBUFFER_4096)
2962                         rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
2963                 else if (rx_buf_len < IXGBE_RXBUFFER_8192)
2964                         rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
2965                 else
2966                         rscctrl |= IXGBE_RSCCTL_MAXDESC_4;
2967         }
2968         IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
2969 }
2970
2971 /**
2972  *  ixgbe_set_uta - Set unicast filter table address
2973  *  @adapter: board private structure
2974  *
2975  *  The unicast table address is a register array of 32-bit registers.
2976  *  The table is meant to be used in a way similar to how the MTA is used
2977  *  however due to certain limitations in the hardware it is necessary to
2978  *  set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
2979  *  enable bit to allow vlan tag stripping when promiscuous mode is enabled
2980  **/
2981 static void ixgbe_set_uta(struct ixgbe_adapter *adapter)
2982 {
2983         struct ixgbe_hw *hw = &adapter->hw;
2984         int i;
2985
2986         /* The UTA table only exists on 82599 hardware and newer */
2987         if (hw->mac.type < ixgbe_mac_82599EB)
2988                 return;
2989
2990         /* we only need to do this if VMDq is enabled */
2991         if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
2992                 return;
2993
2994         for (i = 0; i < 128; i++)
2995                 IXGBE_WRITE_REG(hw, IXGBE_UTA(i), ~0);
2996 }
2997
2998 #define IXGBE_MAX_RX_DESC_POLL 10
2999 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
3000                                        struct ixgbe_ring *ring)
3001 {
3002         struct ixgbe_hw *hw = &adapter->hw;
3003         int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3004         u32 rxdctl;
3005         u8 reg_idx = ring->reg_idx;
3006
3007         /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3008         if (hw->mac.type == ixgbe_mac_82598EB &&
3009             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3010                 return;
3011
3012         do {
3013                 msleep(1);
3014                 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3015         } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
3016
3017         if (!wait_loop) {
3018                 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
3019                       "the polling period\n", reg_idx);
3020         }
3021 }
3022
3023 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
3024                              struct ixgbe_ring *ring)
3025 {
3026         struct ixgbe_hw *hw = &adapter->hw;
3027         u64 rdba = ring->dma;
3028         u32 rxdctl;
3029         u8 reg_idx = ring->reg_idx;
3030
3031         /* disable queue to avoid issues while updating state */
3032         rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3033         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx),
3034                         rxdctl & ~IXGBE_RXDCTL_ENABLE);
3035         IXGBE_WRITE_FLUSH(hw);
3036
3037         IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
3038         IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
3039         IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
3040                         ring->count * sizeof(union ixgbe_adv_rx_desc));
3041         IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
3042         IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
3043         ring->tail = hw->hw_addr + IXGBE_RDT(reg_idx);
3044
3045         ixgbe_configure_srrctl(adapter, ring);
3046         ixgbe_configure_rscctl(adapter, ring);
3047
3048         if (hw->mac.type == ixgbe_mac_82598EB) {
3049                 /*
3050                  * enable cache line friendly hardware writes:
3051                  * PTHRESH=32 descriptors (half the internal cache),
3052                  * this also removes ugly rx_no_buffer_count increment
3053                  * HTHRESH=4 descriptors (to minimize latency on fetch)
3054                  * WTHRESH=8 burst writeback up to two cache lines
3055                  */
3056                 rxdctl &= ~0x3FFFFF;
3057                 rxdctl |=  0x080420;
3058         }
3059
3060         /* enable receive descriptor ring */
3061         rxdctl |= IXGBE_RXDCTL_ENABLE;
3062         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3063
3064         ixgbe_rx_desc_queue_enable(adapter, ring);
3065         ixgbe_alloc_rx_buffers(ring, IXGBE_DESC_UNUSED(ring));
3066 }
3067
3068 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3069 {
3070         struct ixgbe_hw *hw = &adapter->hw;
3071         int p;
3072
3073         /* PSRTYPE must be initialized in non 82598 adapters */
3074         u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3075                       IXGBE_PSRTYPE_UDPHDR |
3076                       IXGBE_PSRTYPE_IPV4HDR |
3077                       IXGBE_PSRTYPE_L2HDR |
3078                       IXGBE_PSRTYPE_IPV6HDR;
3079
3080         if (hw->mac.type == ixgbe_mac_82598EB)
3081                 return;
3082
3083         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED)
3084                 psrtype |= (adapter->num_rx_queues_per_pool << 29);
3085
3086         for (p = 0; p < adapter->num_rx_pools; p++)
3087                 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(adapter->num_vfs + p),
3088                                 psrtype);
3089 }
3090
3091 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3092 {
3093         struct ixgbe_hw *hw = &adapter->hw;
3094         u32 gcr_ext;
3095         u32 vt_reg_bits;
3096         u32 reg_offset, vf_shift;
3097         u32 vmdctl;
3098
3099         if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3100                 return;
3101
3102         vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3103         vt_reg_bits = IXGBE_VMD_CTL_VMDQ_EN | IXGBE_VT_CTL_REPLEN;
3104         vt_reg_bits |= (adapter->num_vfs << IXGBE_VT_CTL_POOL_SHIFT);
3105         IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl | vt_reg_bits);
3106
3107         vf_shift = adapter->num_vfs % 32;
3108         reg_offset = (adapter->num_vfs > 32) ? 1 : 0;
3109
3110         /* Enable only the PF's pool for Tx/Rx */
3111         IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (1 << vf_shift));
3112         IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), 0);
3113         IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (1 << vf_shift));
3114         IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), 0);
3115         IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3116
3117         /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3118         hw->mac.ops.set_vmdq(hw, 0, adapter->num_vfs);
3119
3120         /*
3121          * Set up VF register offsets for selected VT Mode,
3122          * i.e. 32 or 64 VFs for SR-IOV
3123          */
3124         gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
3125         gcr_ext |= IXGBE_GCR_EXT_MSIX_EN;
3126         gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
3127         IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3128
3129         /* enable Tx loopback for VF/PF communication */
3130         IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3131 }
3132
3133 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3134 {
3135         struct ixgbe_hw *hw = &adapter->hw;
3136         struct net_device *netdev = adapter->netdev;
3137         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3138         int rx_buf_len;
3139         struct ixgbe_ring *rx_ring;
3140         int i;
3141         u32 mhadd, hlreg0;
3142
3143         /* Decide whether to use packet split mode or not */
3144         /* Do not use packet split if we're in SR-IOV Mode */
3145         if (!adapter->num_vfs)
3146                 adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
3147
3148         /* Set the RX buffer length according to the mode */
3149         if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
3150                 rx_buf_len = IXGBE_RX_HDR_SIZE;
3151         } else {
3152                 if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
3153                     (netdev->mtu <= ETH_DATA_LEN))
3154                         rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3155                 else
3156                         rx_buf_len = ALIGN(max_frame + VLAN_HLEN, 1024);
3157         }
3158
3159 #ifdef IXGBE_FCOE
3160         /* adjust max frame to be able to do baby jumbo for FCoE */
3161         if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3162             (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3163                 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3164
3165 #endif /* IXGBE_FCOE */
3166         mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3167         if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3168                 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3169                 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3170
3171                 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3172         }
3173
3174         hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3175         /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3176         hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3177         IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3178
3179         /*
3180          * Setup the HW Rx Head and Tail Descriptor Pointers and
3181          * the Base and Length of the Rx Descriptor Ring
3182          */
3183         for (i = 0; i < adapter->num_rx_queues; i++) {
3184                 rx_ring = adapter->rx_ring[i];
3185                 rx_ring->rx_buf_len = rx_buf_len;
3186
3187                 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)
3188                         set_ring_ps_enabled(rx_ring);
3189                 else
3190                         clear_ring_ps_enabled(rx_ring);
3191
3192                 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3193                         set_ring_rsc_enabled(rx_ring);
3194                 else
3195                         clear_ring_rsc_enabled(rx_ring);
3196
3197 #ifdef IXGBE_FCOE
3198                 if (netdev->features & NETIF_F_FCOE_MTU) {
3199                         struct ixgbe_ring_feature *f;
3200                         f = &adapter->ring_feature[RING_F_FCOE];
3201                         if ((i >= f->mask) && (i < f->mask + f->indices)) {
3202                                 clear_ring_ps_enabled(rx_ring);
3203                                 if (rx_buf_len < IXGBE_FCOE_JUMBO_FRAME_SIZE)
3204                                         rx_ring->rx_buf_len =
3205                                                 IXGBE_FCOE_JUMBO_FRAME_SIZE;
3206                         } else if (!ring_is_rsc_enabled(rx_ring) &&
3207                                    !ring_is_ps_enabled(rx_ring)) {
3208                                 rx_ring->rx_buf_len =
3209                                                 IXGBE_FCOE_JUMBO_FRAME_SIZE;
3210                         }
3211                 }
3212 #endif /* IXGBE_FCOE */
3213         }
3214 }
3215
3216 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3217 {
3218         struct ixgbe_hw *hw = &adapter->hw;
3219         u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3220
3221         switch (hw->mac.type) {
3222         case ixgbe_mac_82598EB:
3223                 /*
3224                  * For VMDq support of different descriptor types or
3225                  * buffer sizes through the use of multiple SRRCTL
3226                  * registers, RDRXCTL.MVMEN must be set to 1
3227                  *
3228                  * also, the manual doesn't mention it clearly but DCA hints
3229                  * will only use queue 0's tags unless this bit is set.  Side
3230                  * effects of setting this bit are only that SRRCTL must be
3231                  * fully programmed [0..15]
3232                  */
3233                 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3234                 break;
3235         case ixgbe_mac_82599EB:
3236         case ixgbe_mac_X540:
3237                 /* Disable RSC for ACK packets */
3238                 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3239                    (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3240                 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3241                 /* hardware requires some bits to be set by default */
3242                 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3243                 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3244                 break;
3245         default:
3246                 /* We should do nothing since we don't know this hardware */
3247                 return;
3248         }
3249
3250         IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3251 }
3252
3253 /**
3254  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3255  * @adapter: board private structure
3256  *
3257  * Configure the Rx unit of the MAC after a reset.
3258  **/
3259 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3260 {
3261         struct ixgbe_hw *hw = &adapter->hw;
3262         int i;
3263         u32 rxctrl;
3264
3265         /* disable receives while setting up the descriptors */
3266         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3267         IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
3268
3269         ixgbe_setup_psrtype(adapter);
3270         ixgbe_setup_rdrxctl(adapter);
3271
3272         /* Program registers for the distribution of queues */
3273         ixgbe_setup_mrqc(adapter);
3274
3275         ixgbe_set_uta(adapter);
3276
3277         /* set_rx_buffer_len must be called before ring initialization */
3278         ixgbe_set_rx_buffer_len(adapter);
3279
3280         /*
3281          * Setup the HW Rx Head and Tail Descriptor Pointers and
3282          * the Base and Length of the Rx Descriptor Ring
3283          */
3284         for (i = 0; i < adapter->num_rx_queues; i++)
3285                 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3286
3287         /* disable drop enable for 82598 parts */
3288         if (hw->mac.type == ixgbe_mac_82598EB)
3289                 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3290
3291         /* enable all receives */
3292         rxctrl |= IXGBE_RXCTRL_RXEN;
3293         hw->mac.ops.enable_rx_dma(hw, rxctrl);
3294 }
3295
3296 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3297 {
3298         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3299         struct ixgbe_hw *hw = &adapter->hw;
3300         int pool_ndx = adapter->num_vfs;
3301
3302         /* add VID to filter table */
3303         hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, true);
3304         set_bit(vid, adapter->active_vlans);
3305 }
3306
3307 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3308 {
3309         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3310         struct ixgbe_hw *hw = &adapter->hw;
3311         int pool_ndx = adapter->num_vfs;
3312
3313         /* remove VID from filter table */
3314         hw->mac.ops.set_vfta(&adapter->hw, vid, pool_ndx, false);
3315         clear_bit(vid, adapter->active_vlans);
3316 }
3317
3318 /**
3319  * ixgbe_vlan_filter_disable - helper to disable hw vlan filtering
3320  * @adapter: driver data
3321  */
3322 static void ixgbe_vlan_filter_disable(struct ixgbe_adapter *adapter)
3323 {
3324         struct ixgbe_hw *hw = &adapter->hw;
3325         u32 vlnctrl;
3326
3327         vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3328         vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
3329         IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3330 }
3331
3332 /**
3333  * ixgbe_vlan_filter_enable - helper to enable hw vlan filtering
3334  * @adapter: driver data
3335  */
3336 static void ixgbe_vlan_filter_enable(struct ixgbe_adapter *adapter)
3337 {
3338         struct ixgbe_hw *hw = &adapter->hw;
3339         u32 vlnctrl;
3340
3341         vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3342         vlnctrl |= IXGBE_VLNCTRL_VFE;
3343         vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
3344         IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3345 }
3346
3347 /**
3348  * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3349  * @adapter: driver data
3350  */
3351 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3352 {
3353         struct ixgbe_hw *hw = &adapter->hw;
3354         u32 vlnctrl;
3355         int i, j;
3356
3357         switch (hw->mac.type) {
3358         case ixgbe_mac_82598EB:
3359                 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3360                 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3361                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3362                 break;
3363         case ixgbe_mac_82599EB:
3364         case ixgbe_mac_X540:
3365                 for (i = 0; i < adapter->num_rx_queues; i++) {
3366                         j = adapter->rx_ring[i]->reg_idx;
3367                         vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3368                         vlnctrl &= ~IXGBE_RXDCTL_VME;
3369                         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3370                 }
3371                 break;
3372         default:
3373                 break;
3374         }
3375 }
3376
3377 /**
3378  * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3379  * @adapter: driver data
3380  */
3381 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3382 {
3383         struct ixgbe_hw *hw = &adapter->hw;
3384         u32 vlnctrl;
3385         int i, j;
3386
3387         switch (hw->mac.type) {
3388         case ixgbe_mac_82598EB:
3389                 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3390                 vlnctrl |= IXGBE_VLNCTRL_VME;
3391                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3392                 break;
3393         case ixgbe_mac_82599EB:
3394         case ixgbe_mac_X540:
3395                 for (i = 0; i < adapter->num_rx_queues; i++) {
3396                         j = adapter->rx_ring[i]->reg_idx;
3397                         vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3398                         vlnctrl |= IXGBE_RXDCTL_VME;
3399                         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3400                 }
3401                 break;
3402         default:
3403                 break;
3404         }
3405 }
3406
3407 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3408 {
3409         u16 vid;
3410
3411         ixgbe_vlan_rx_add_vid(adapter->netdev, 0);
3412
3413         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3414                 ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
3415 }
3416
3417 /**
3418  * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
3419  * @netdev: network interface device structure
3420  *
3421  * Writes unicast address list to the RAR table.
3422  * Returns: -ENOMEM on failure/insufficient address space
3423  *                0 on no addresses written
3424  *                X on writing X addresses to the RAR table
3425  **/
3426 static int ixgbe_write_uc_addr_list(struct net_device *netdev)
3427 {
3428         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3429         struct ixgbe_hw *hw = &adapter->hw;
3430         unsigned int vfn = adapter->num_vfs;
3431         unsigned int rar_entries = hw->mac.num_rar_entries - (vfn + 1);
3432         int count = 0;
3433
3434         /* return ENOMEM indicating insufficient memory for addresses */
3435         if (netdev_uc_count(netdev) > rar_entries)
3436                 return -ENOMEM;
3437
3438         if (!netdev_uc_empty(netdev) && rar_entries) {
3439                 struct netdev_hw_addr *ha;
3440                 /* return error if we do not support writing to RAR table */
3441                 if (!hw->mac.ops.set_rar)
3442                         return -ENOMEM;
3443
3444                 netdev_for_each_uc_addr(ha, netdev) {
3445                         if (!rar_entries)
3446                                 break;
3447                         hw->mac.ops.set_rar(hw, rar_entries--, ha->addr,
3448                                             vfn, IXGBE_RAH_AV);
3449                         count++;
3450                 }
3451         }
3452         /* write the addresses in reverse order to avoid write combining */
3453         for (; rar_entries > 0 ; rar_entries--)
3454                 hw->mac.ops.clear_rar(hw, rar_entries);
3455
3456         return count;
3457 }
3458
3459 /**
3460  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
3461  * @netdev: network interface device structure
3462  *
3463  * The set_rx_method entry point is called whenever the unicast/multicast
3464  * address list or the network interface flags are updated.  This routine is
3465  * responsible for configuring the hardware for proper unicast, multicast and
3466  * promiscuous mode.
3467  **/
3468 void ixgbe_set_rx_mode(struct net_device *netdev)
3469 {
3470         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3471         struct ixgbe_hw *hw = &adapter->hw;
3472         u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
3473         int count;
3474
3475         /* Check for Promiscuous and All Multicast modes */
3476
3477         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3478
3479         /* set all bits that we expect to always be set */
3480         fctrl |= IXGBE_FCTRL_BAM;
3481         fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
3482         fctrl |= IXGBE_FCTRL_PMCF;
3483
3484         /* clear the bits we are changing the status of */
3485         fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3486
3487         if (netdev->flags & IFF_PROMISC) {
3488                 hw->addr_ctrl.user_set_promisc = true;
3489                 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
3490                 vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
3491                 /* don't hardware filter vlans in promisc mode */
3492                 ixgbe_vlan_filter_disable(adapter);
3493         } else {
3494                 if (netdev->flags & IFF_ALLMULTI) {
3495                         fctrl |= IXGBE_FCTRL_MPE;
3496                         vmolr |= IXGBE_VMOLR_MPE;
3497                 } else {
3498                         /*
3499                          * Write addresses to the MTA, if the attempt fails
3500                          * then we should just turn on promiscous mode so
3501                          * that we can at least receive multicast traffic
3502                          */
3503                         hw->mac.ops.update_mc_addr_list(hw, netdev);
3504                         vmolr |= IXGBE_VMOLR_ROMPE;
3505                 }
3506                 ixgbe_vlan_filter_enable(adapter);
3507                 hw->addr_ctrl.user_set_promisc = false;
3508                 /*
3509                  * Write addresses to available RAR registers, if there is not
3510                  * sufficient space to store all the addresses then enable
3511                  * unicast promiscous mode
3512                  */
3513                 count = ixgbe_write_uc_addr_list(netdev);
3514                 if (count < 0) {
3515                         fctrl |= IXGBE_FCTRL_UPE;
3516                         vmolr |= IXGBE_VMOLR_ROPE;
3517                 }
3518         }
3519
3520         if (adapter->num_vfs) {
3521                 ixgbe_restore_vf_multicasts(adapter);
3522                 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(adapter->num_vfs)) &
3523                          ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
3524                            IXGBE_VMOLR_ROPE);
3525                 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(adapter->num_vfs), vmolr);
3526         }
3527
3528         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
3529
3530         if (netdev->features & NETIF_F_HW_VLAN_RX)
3531                 ixgbe_vlan_strip_enable(adapter);
3532         else
3533                 ixgbe_vlan_strip_disable(adapter);
3534 }
3535
3536 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
3537 {
3538         int q_idx;
3539         struct ixgbe_q_vector *q_vector;
3540         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3541
3542         /* legacy and MSI only use one vector */
3543         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3544                 q_vectors = 1;
3545
3546         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3547                 struct napi_struct *napi;
3548                 q_vector = adapter->q_vector[q_idx];
3549                 napi = &q_vector->napi;
3550                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3551                         if (!q_vector->rxr_count || !q_vector->txr_count) {
3552                                 if (q_vector->txr_count == 1)
3553                                         napi->poll = &ixgbe_clean_txonly;
3554                                 else if (q_vector->rxr_count == 1)
3555                                         napi->poll = &ixgbe_clean_rxonly;
3556                         }
3557                 }
3558
3559                 napi_enable(napi);
3560         }
3561 }
3562
3563 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
3564 {
3565         int q_idx;
3566         struct ixgbe_q_vector *q_vector;
3567         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
3568
3569         /* legacy and MSI only use one vector */
3570         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
3571                 q_vectors = 1;
3572
3573         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3574                 q_vector = adapter->q_vector[q_idx];
3575                 napi_disable(&q_vector->napi);
3576         }
3577 }
3578
3579 #ifdef CONFIG_IXGBE_DCB
3580 /*
3581  * ixgbe_configure_dcb - Configure DCB hardware
3582  * @adapter: ixgbe adapter struct
3583  *
3584  * This is called by the driver on open to configure the DCB hardware.
3585  * This is also called by the gennetlink interface when reconfiguring
3586  * the DCB state.
3587  */
3588 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
3589 {
3590         struct ixgbe_hw *hw = &adapter->hw;
3591         int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3592
3593         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
3594                 if (hw->mac.type == ixgbe_mac_82598EB)
3595                         netif_set_gso_max_size(adapter->netdev, 65536);
3596                 return;
3597         }
3598
3599         if (hw->mac.type == ixgbe_mac_82598EB)
3600                 netif_set_gso_max_size(adapter->netdev, 32768);
3601
3602 #ifdef CONFIG_FCOE
3603         if (adapter->netdev->features & NETIF_F_FCOE_MTU)
3604                 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
3605 #endif
3606
3607         ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3608                                         DCB_TX_CONFIG);
3609         ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
3610                                         DCB_RX_CONFIG);
3611
3612         /* Enable VLAN tag insert/strip */
3613         adapter->netdev->features |= NETIF_F_HW_VLAN_RX;
3614
3615         hw->mac.ops.set_vfta(&adapter->hw, 0, 0, true);
3616
3617         /* reconfigure the hardware */
3618         ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
3619 }
3620
3621 #endif
3622 static void ixgbe_configure(struct ixgbe_adapter *adapter)
3623 {
3624         struct net_device *netdev = adapter->netdev;
3625         struct ixgbe_hw *hw = &adapter->hw;
3626         int i;
3627
3628 #ifdef CONFIG_IXGBE_DCB
3629         ixgbe_configure_dcb(adapter);
3630 #endif
3631
3632         ixgbe_set_rx_mode(netdev);
3633         ixgbe_restore_vlan(adapter);
3634
3635 #ifdef IXGBE_FCOE
3636         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
3637                 ixgbe_configure_fcoe(adapter);
3638
3639 #endif /* IXGBE_FCOE */
3640         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3641                 for (i = 0; i < adapter->num_tx_queues; i++)
3642                         adapter->tx_ring[i]->atr_sample_rate =
3643                                                        adapter->atr_sample_rate;
3644                 ixgbe_init_fdir_signature_82599(hw, adapter->fdir_pballoc);
3645         } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
3646                 ixgbe_init_fdir_perfect_82599(hw, adapter->fdir_pballoc);
3647         }
3648         ixgbe_configure_virtualization(adapter);
3649
3650         ixgbe_configure_tx(adapter);
3651         ixgbe_configure_rx(adapter);
3652 }
3653
3654 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
3655 {
3656         switch (hw->phy.type) {
3657         case ixgbe_phy_sfp_avago:
3658         case ixgbe_phy_sfp_ftl:
3659         case ixgbe_phy_sfp_intel:
3660         case ixgbe_phy_sfp_unknown:
3661         case ixgbe_phy_sfp_passive_tyco:
3662         case ixgbe_phy_sfp_passive_unknown:
3663         case ixgbe_phy_sfp_active_unknown:
3664         case ixgbe_phy_sfp_ftl_active:
3665                 return true;
3666         default:
3667                 return false;
3668         }
3669 }
3670
3671 /**
3672  * ixgbe_sfp_link_config - set up SFP+ link
3673  * @adapter: pointer to private adapter struct
3674  **/
3675 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
3676 {
3677         struct ixgbe_hw *hw = &adapter->hw;
3678
3679                 if (hw->phy.multispeed_fiber) {
3680                         /*
3681                          * In multispeed fiber setups, the device may not have
3682                          * had a physical connection when the driver loaded.
3683                          * If that's the case, the initial link configuration
3684                          * couldn't get the MAC into 10G or 1G mode, so we'll
3685                          * never have a link status change interrupt fire.
3686                          * We need to try and force an autonegotiation
3687                          * session, then bring up link.
3688                          */
3689                         hw->mac.ops.setup_sfp(hw);
3690                         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
3691                                 schedule_work(&adapter->multispeed_fiber_task);
3692                 } else {
3693                         /*
3694                          * Direct Attach Cu and non-multispeed fiber modules
3695                          * still need to be configured properly prior to
3696                          * attempting link.
3697                          */
3698                         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_MOD_TASK))
3699                                 schedule_work(&adapter->sfp_config_module_task);
3700                 }
3701 }
3702
3703 /**
3704  * ixgbe_non_sfp_link_config - set up non-SFP+ link
3705  * @hw: pointer to private hardware struct
3706  *
3707  * Returns 0 on success, negative on failure
3708  **/
3709 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
3710 {
3711         u32 autoneg;
3712         bool negotiation, link_up = false;
3713         u32 ret = IXGBE_ERR_LINK_SETUP;
3714
3715         if (hw->mac.ops.check_link)
3716                 ret = hw->mac.ops.check_link(hw, &autoneg, &link_up, false);
3717
3718         if (ret)
3719                 goto link_cfg_out;
3720
3721         if (hw->mac.ops.get_link_capabilities)
3722                 ret = hw->mac.ops.get_link_capabilities(hw, &autoneg,
3723                                                         &negotiation);
3724         if (ret)
3725                 goto link_cfg_out;
3726
3727         if (hw->mac.ops.setup_link)
3728                 ret = hw->mac.ops.setup_link(hw, autoneg, negotiation, link_up);
3729 link_cfg_out:
3730         return ret;
3731 }
3732
3733 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
3734 {
3735         struct ixgbe_hw *hw = &adapter->hw;
3736         u32 gpie = 0;
3737
3738         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3739                 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
3740                        IXGBE_GPIE_OCD;
3741                 gpie |= IXGBE_GPIE_EIAME;
3742                 /*
3743                  * use EIAM to auto-mask when MSI-X interrupt is asserted
3744                  * this saves a register write for every interrupt
3745                  */
3746                 switch (hw->mac.type) {
3747                 case ixgbe_mac_82598EB:
3748                         IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3749                         break;
3750                 case ixgbe_mac_82599EB:
3751                 case ixgbe_mac_X540:
3752                 default:
3753                         IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
3754                         IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
3755                         break;
3756                 }
3757         } else {
3758                 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
3759                  * specifically only auto mask tx and rx interrupts */
3760                 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
3761         }
3762
3763         /* XXX: to interrupt immediately for EICS writes, enable this */
3764         /* gpie |= IXGBE_GPIE_EIMEN; */
3765
3766         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3767                 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
3768                 gpie |= IXGBE_GPIE_VTMODE_64;
3769         }
3770
3771         /* Enable fan failure interrupt */
3772         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3773                 gpie |= IXGBE_SDP1_GPIEN;
3774
3775         if (hw->mac.type == ixgbe_mac_82599EB)
3776                 gpie |= IXGBE_SDP1_GPIEN;
3777                 gpie |= IXGBE_SDP2_GPIEN;
3778
3779         IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
3780 }
3781
3782 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
3783 {
3784         struct ixgbe_hw *hw = &adapter->hw;
3785         int err;
3786         u32 ctrl_ext;
3787
3788         ixgbe_get_hw_control(adapter);
3789         ixgbe_setup_gpie(adapter);
3790
3791         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3792                 ixgbe_configure_msix(adapter);
3793         else
3794                 ixgbe_configure_msi_and_legacy(adapter);
3795
3796         /* enable the optics */
3797         if (hw->phy.multispeed_fiber && hw->mac.ops.enable_tx_laser)
3798                 hw->mac.ops.enable_tx_laser(hw);
3799
3800         clear_bit(__IXGBE_DOWN, &adapter->state);
3801         ixgbe_napi_enable_all(adapter);
3802
3803         if (ixgbe_is_sfp(hw)) {
3804                 ixgbe_sfp_link_config(adapter);
3805         } else {
3806                 err = ixgbe_non_sfp_link_config(hw);
3807                 if (err)
3808                         e_err(probe, "link_config FAILED %d\n", err);
3809         }
3810
3811         /* clear any pending interrupts, may auto mask */
3812         IXGBE_READ_REG(hw, IXGBE_EICR);
3813         ixgbe_irq_enable(adapter, true, true);
3814
3815         /*
3816          * If this adapter has a fan, check to see if we had a failure
3817          * before we enabled the interrupt.
3818          */
3819         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
3820                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
3821                 if (esdp & IXGBE_ESDP_SDP1)
3822                         e_crit(drv, "Fan has stopped, replace the adapter\n");
3823         }
3824
3825         /*
3826          * For hot-pluggable SFP+ devices, a new SFP+ module may have
3827          * arrived before interrupts were enabled but after probe.  Such
3828          * devices wouldn't have their type identified yet. We need to
3829          * kick off the SFP+ module setup first, then try to bring up link.
3830          * If we're not hot-pluggable SFP+, we just need to configure link
3831          * and bring it up.
3832          */
3833         if (hw->phy.type == ixgbe_phy_unknown)
3834                 schedule_work(&adapter->sfp_config_module_task);
3835
3836         /* enable transmits */
3837         netif_tx_start_all_queues(adapter->netdev);
3838
3839         /* bring the link up in the watchdog, this could race with our first
3840          * link up interrupt but shouldn't be a problem */
3841         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
3842         adapter->link_check_timeout = jiffies;
3843         mod_timer(&adapter->watchdog_timer, jiffies);
3844
3845         /* Set PF Reset Done bit so PF/VF Mail Ops can work */
3846         ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
3847         ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
3848         IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
3849
3850         return 0;
3851 }
3852
3853 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
3854 {
3855         WARN_ON(in_interrupt());
3856         while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
3857                 msleep(1);
3858         ixgbe_down(adapter);
3859         /*
3860          * If SR-IOV enabled then wait a bit before bringing the adapter
3861          * back up to give the VFs time to respond to the reset.  The
3862          * two second wait is based upon the watchdog timer cycle in
3863          * the VF driver.
3864          */
3865         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3866                 msleep(2000);
3867         ixgbe_up(adapter);
3868         clear_bit(__IXGBE_RESETTING, &adapter->state);
3869 }
3870
3871 int ixgbe_up(struct ixgbe_adapter *adapter)
3872 {
3873         /* hardware has been reset, we need to reload some things */
3874         ixgbe_configure(adapter);
3875
3876         return ixgbe_up_complete(adapter);
3877 }
3878
3879 void ixgbe_reset(struct ixgbe_adapter *adapter)
3880 {
3881         struct ixgbe_hw *hw = &adapter->hw;
3882         int err;
3883
3884         err = hw->mac.ops.init_hw(hw);
3885         switch (err) {
3886         case 0:
3887         case IXGBE_ERR_SFP_NOT_PRESENT:
3888                 break;
3889         case IXGBE_ERR_MASTER_REQUESTS_PENDING:
3890                 e_dev_err("master disable timed out\n");
3891                 break;
3892         case IXGBE_ERR_EEPROM_VERSION:
3893                 /* We are running on a pre-production device, log a warning */
3894                 e_dev_warn("This device is a pre-production adapter/LOM. "
3895                            "Please be aware there may be issuesassociated with "
3896                            "your hardware.  If you are experiencing problems "
3897                            "please contact your Intel or hardware "
3898                            "representative who provided you with this "
3899                            "hardware.\n");
3900                 break;
3901         default:
3902                 e_dev_err("Hardware Error: %d\n", err);
3903         }
3904
3905         /* reprogram the RAR[0] in case user changed it. */
3906         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
3907                             IXGBE_RAH_AV);
3908 }
3909
3910 /**
3911  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
3912  * @rx_ring: ring to free buffers from
3913  **/
3914 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
3915 {
3916         struct device *dev = rx_ring->dev;
3917         unsigned long size;
3918         u16 i;
3919
3920         /* ring already cleared, nothing to do */
3921         if (!rx_ring->rx_buffer_info)
3922                 return;
3923
3924         /* Free all the Rx ring sk_buffs */
3925         for (i = 0; i < rx_ring->count; i++) {
3926                 struct ixgbe_rx_buffer *rx_buffer_info;
3927
3928                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
3929                 if (rx_buffer_info->dma) {
3930                         dma_unmap_single(rx_ring->dev, rx_buffer_info->dma,
3931                                          rx_ring->rx_buf_len,
3932                                          DMA_FROM_DEVICE);
3933                         rx_buffer_info->dma = 0;
3934                 }
3935                 if (rx_buffer_info->skb) {
3936                         struct sk_buff *skb = rx_buffer_info->skb;
3937                         rx_buffer_info->skb = NULL;
3938                         do {
3939                                 struct sk_buff *this = skb;
3940                                 if (IXGBE_RSC_CB(this)->delay_unmap) {
3941                                         dma_unmap_single(dev,
3942                                                          IXGBE_RSC_CB(this)->dma,
3943                                                          rx_ring->rx_buf_len,
3944                                                          DMA_FROM_DEVICE);
3945                                         IXGBE_RSC_CB(this)->dma = 0;
3946                                         IXGBE_RSC_CB(skb)->delay_unmap = false;
3947                                 }
3948                                 skb = skb->prev;
3949                                 dev_kfree_skb(this);
3950                         } while (skb);
3951                 }
3952                 if (!rx_buffer_info->page)
3953                         continue;
3954                 if (rx_buffer_info->page_dma) {
3955                         dma_unmap_page(dev, rx_buffer_info->page_dma,
3956                                        PAGE_SIZE / 2, DMA_FROM_DEVICE);
3957                         rx_buffer_info->page_dma = 0;
3958                 }
3959                 put_page(rx_buffer_info->page);
3960                 rx_buffer_info->page = NULL;
3961                 rx_buffer_info->page_offset = 0;
3962         }
3963
3964         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
3965         memset(rx_ring->rx_buffer_info, 0, size);
3966
3967         /* Zero out the descriptor ring */
3968         memset(rx_ring->desc, 0, rx_ring->size);
3969
3970         rx_ring->next_to_clean = 0;
3971         rx_ring->next_to_use = 0;
3972 }
3973
3974 /**
3975  * ixgbe_clean_tx_ring - Free Tx Buffers
3976  * @tx_ring: ring to be cleaned
3977  **/
3978 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
3979 {
3980         struct ixgbe_tx_buffer *tx_buffer_info;
3981         unsigned long size;
3982         u16 i;
3983
3984         /* ring already cleared, nothing to do */
3985         if (!tx_ring->tx_buffer_info)
3986                 return;
3987
3988         /* Free all the Tx ring sk_buffs */
3989         for (i = 0; i < tx_ring->count; i++) {
3990                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3991                 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
3992         }
3993
3994         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
3995         memset(tx_ring->tx_buffer_info, 0, size);
3996
3997         /* Zero out the descriptor ring */
3998         memset(tx_ring->desc, 0, tx_ring->size);
3999
4000         tx_ring->next_to_use = 0;
4001         tx_ring->next_to_clean = 0;
4002 }
4003
4004 /**
4005  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
4006  * @adapter: board private structure
4007  **/
4008 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
4009 {
4010         int i;
4011
4012         for (i = 0; i < adapter->num_rx_queues; i++)
4013                 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
4014 }
4015
4016 /**
4017  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
4018  * @adapter: board private structure
4019  **/
4020 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
4021 {
4022         int i;
4023
4024         for (i = 0; i < adapter->num_tx_queues; i++)
4025                 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
4026 }
4027
4028 void ixgbe_down(struct ixgbe_adapter *adapter)
4029 {
4030         struct net_device *netdev = adapter->netdev;
4031         struct ixgbe_hw *hw = &adapter->hw;
4032         u32 rxctrl;
4033         u32 txdctl;
4034         int i;
4035         int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4036
4037         /* signal that we are down to the interrupt handler */
4038         set_bit(__IXGBE_DOWN, &adapter->state);
4039
4040         /* disable receive for all VFs and wait one second */
4041         if (adapter->num_vfs) {
4042                 /* ping all the active vfs to let them know we are going down */
4043                 ixgbe_ping_all_vfs(adapter);
4044
4045                 /* Disable all VFTE/VFRE TX/RX */
4046                 ixgbe_disable_tx_rx(adapter);
4047
4048                 /* Mark all the VFs as inactive */
4049                 for (i = 0 ; i < adapter->num_vfs; i++)
4050                         adapter->vfinfo[i].clear_to_send = 0;
4051         }
4052
4053         /* disable receives */
4054         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4055         IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
4056
4057         IXGBE_WRITE_FLUSH(hw);
4058         msleep(10);
4059
4060         netif_tx_stop_all_queues(netdev);
4061
4062         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
4063         del_timer_sync(&adapter->sfp_timer);
4064         del_timer_sync(&adapter->watchdog_timer);
4065         cancel_work_sync(&adapter->watchdog_task);
4066
4067         netif_carrier_off(netdev);
4068         netif_tx_disable(netdev);
4069
4070         ixgbe_irq_disable(adapter);
4071
4072         ixgbe_napi_disable_all(adapter);
4073
4074         /* Cleanup the affinity_hint CPU mask memory and callback */
4075         for (i = 0; i < num_q_vectors; i++) {
4076                 struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
4077                 /* clear the affinity_mask in the IRQ descriptor */
4078                 irq_set_affinity_hint(adapter->msix_entries[i]. vector, NULL);
4079                 /* release the CPU mask memory */
4080                 free_cpumask_var(q_vector->affinity_mask);
4081         }
4082
4083         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4084             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
4085                 cancel_work_sync(&adapter->fdir_reinit_task);
4086
4087         if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
4088                 cancel_work_sync(&adapter->check_overtemp_task);
4089
4090         /* disable transmits in the hardware now that interrupts are off */
4091         for (i = 0; i < adapter->num_tx_queues; i++) {
4092                 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
4093                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
4094                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx),
4095                                 (txdctl & ~IXGBE_TXDCTL_ENABLE));
4096         }
4097         /* Disable the Tx DMA engine on 82599 */
4098         switch (hw->mac.type) {
4099         case ixgbe_mac_82599EB:
4100         case ixgbe_mac_X540:
4101                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
4102                                 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
4103                                  ~IXGBE_DMATXCTL_TE));
4104                 break;
4105         default:
4106                 break;
4107         }
4108
4109         /* power down the optics */
4110         if (hw->phy.multispeed_fiber && hw->mac.ops.disable_tx_laser)
4111                 hw->mac.ops.disable_tx_laser(hw);
4112
4113         /* clear n-tuple filters that are cached */
4114         ethtool_ntuple_flush(netdev);
4115
4116         if (!pci_channel_offline(adapter->pdev))
4117                 ixgbe_reset(adapter);
4118         ixgbe_clean_all_tx_rings(adapter);
4119         ixgbe_clean_all_rx_rings(adapter);
4120
4121 #ifdef CONFIG_IXGBE_DCA
4122         /* since we reset the hardware DCA settings were cleared */
4123         ixgbe_setup_dca(adapter);
4124 #endif
4125 }
4126
4127 /**
4128  * ixgbe_poll - NAPI Rx polling callback
4129  * @napi: structure for representing this polling device
4130  * @budget: how many packets driver is allowed to clean
4131  *
4132  * This function is used for legacy and MSI, NAPI mode
4133  **/
4134 static int ixgbe_poll(struct napi_struct *napi, int budget)
4135 {
4136         struct ixgbe_q_vector *q_vector =
4137                                 container_of(napi, struct ixgbe_q_vector, napi);
4138         struct ixgbe_adapter *adapter = q_vector->adapter;
4139         int tx_clean_complete, work_done = 0;
4140
4141 #ifdef CONFIG_IXGBE_DCA
4142         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
4143                 ixgbe_update_dca(q_vector);
4144 #endif
4145
4146         tx_clean_complete = ixgbe_clean_tx_irq(q_vector, adapter->tx_ring[0]);
4147         ixgbe_clean_rx_irq(q_vector, adapter->rx_ring[0], &work_done, budget);
4148
4149         if (!tx_clean_complete)
4150                 work_done = budget;
4151
4152         /* If budget not fully consumed, exit the polling mode */
4153         if (work_done < budget) {
4154                 napi_complete(napi);
4155                 if (adapter->rx_itr_setting & 1)
4156                         ixgbe_set_itr(adapter);
4157                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
4158                         ixgbe_irq_enable_queues(adapter, IXGBE_EIMS_RTX_QUEUE);
4159         }
4160         return work_done;
4161 }
4162
4163 /**
4164  * ixgbe_tx_timeout - Respond to a Tx Hang
4165  * @netdev: network interface device structure
4166  **/
4167 static void ixgbe_tx_timeout(struct net_device *netdev)
4168 {
4169         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4170
4171         adapter->tx_timeout_count++;
4172
4173         /* Do the reset outside of interrupt context */
4174         schedule_work(&adapter->reset_task);
4175 }
4176
4177 static void ixgbe_reset_task(struct work_struct *work)
4178 {
4179         struct ixgbe_adapter *adapter;
4180         adapter = container_of(work, struct ixgbe_adapter, reset_task);
4181
4182         /* If we're already down or resetting, just bail */
4183         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
4184             test_bit(__IXGBE_RESETTING, &adapter->state))
4185                 return;
4186
4187         ixgbe_dump(adapter);
4188         netdev_err(adapter->netdev, "Reset adapter\n");
4189         ixgbe_reinit_locked(adapter);
4190 }
4191
4192 #ifdef CONFIG_IXGBE_DCB
4193 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter *adapter)
4194 {
4195         bool ret = false;
4196         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_DCB];
4197
4198         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
4199                 return ret;
4200
4201         f->mask = 0x7 << 3;
4202         adapter->num_rx_queues = f->indices;
4203         adapter->num_tx_queues = f->indices;
4204         ret = true;
4205
4206         return ret;
4207 }
4208 #endif
4209
4210 /**
4211  * ixgbe_set_rss_queues: Allocate queues for RSS
4212  * @adapter: board private structure to initialize
4213  *
4214  * This is our "base" multiqueue mode.  RSS (Receive Side Scaling) will try
4215  * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
4216  *
4217  **/
4218 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter *adapter)
4219 {
4220         bool ret = false;
4221         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_RSS];
4222
4223         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4224                 f->mask = 0xF;
4225                 adapter->num_rx_queues = f->indices;
4226                 adapter->num_tx_queues = f->indices;
4227                 ret = true;
4228         } else {
4229                 ret = false;
4230         }
4231
4232         return ret;
4233 }
4234
4235 /**
4236  * ixgbe_set_fdir_queues: Allocate queues for Flow Director
4237  * @adapter: board private structure to initialize
4238  *
4239  * Flow Director is an advanced Rx filter, attempting to get Rx flows back
4240  * to the original CPU that initiated the Tx session.  This runs in addition
4241  * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
4242  * Rx load across CPUs using RSS.
4243  *
4244  **/
4245 static inline bool ixgbe_set_fdir_queues(struct ixgbe_adapter *adapter)
4246 {
4247         bool ret = false;
4248         struct ixgbe_ring_feature *f_fdir = &adapter->ring_feature[RING_F_FDIR];
4249
4250         f_fdir->indices = min((int)num_online_cpus(), f_fdir->indices);
4251         f_fdir->mask = 0;
4252
4253         /* Flow Director must have RSS enabled */
4254         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4255             ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4256              (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)))) {
4257                 adapter->num_tx_queues = f_fdir->indices;
4258                 adapter->num_rx_queues = f_fdir->indices;
4259                 ret = true;
4260         } else {
4261                 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4262                 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4263         }
4264         return ret;
4265 }
4266
4267 #ifdef IXGBE_FCOE
4268 /**
4269  * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
4270  * @adapter: board private structure to initialize
4271  *
4272  * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
4273  * The ring feature mask is not used as a mask for FCoE, as it can take any 8
4274  * rx queues out of the max number of rx queues, instead, it is used as the
4275  * index of the first rx queue used by FCoE.
4276  *
4277  **/
4278 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter *adapter)
4279 {
4280         bool ret = false;
4281         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4282
4283         f->indices = min((int)num_online_cpus(), f->indices);
4284         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
4285                 adapter->num_rx_queues = 1;
4286                 adapter->num_tx_queues = 1;
4287 #ifdef CONFIG_IXGBE_DCB
4288                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4289                         e_info(probe, "FCoE enabled with DCB\n");
4290                         ixgbe_set_dcb_queues(adapter);
4291                 }
4292 #endif
4293                 if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4294                         e_info(probe, "FCoE enabled with RSS\n");
4295                         if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4296                             (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4297                                 ixgbe_set_fdir_queues(adapter);
4298                         else
4299                                 ixgbe_set_rss_queues(adapter);
4300                 }
4301                 /* adding FCoE rx rings to the end */
4302                 f->mask = adapter->num_rx_queues;
4303                 adapter->num_rx_queues += f->indices;
4304                 adapter->num_tx_queues += f->indices;
4305
4306                 ret = true;
4307         }
4308
4309         return ret;
4310 }
4311
4312 #endif /* IXGBE_FCOE */
4313 /**
4314  * ixgbe_set_sriov_queues: Allocate queues for IOV use
4315  * @adapter: board private structure to initialize
4316  *
4317  * IOV doesn't actually use anything, so just NAK the
4318  * request for now and let the other queue routines
4319  * figure out what to do.
4320  */
4321 static inline bool ixgbe_set_sriov_queues(struct ixgbe_adapter *adapter)
4322 {
4323         return false;
4324 }
4325
4326 /*
4327  * ixgbe_set_num_queues: Allocate queues for device, feature dependant
4328  * @adapter: board private structure to initialize
4329  *
4330  * This is the top level queue allocation routine.  The order here is very
4331  * important, starting with the "most" number of features turned on at once,
4332  * and ending with the smallest set of features.  This way large combinations
4333  * can be allocated if they're turned on, and smaller combinations are the
4334  * fallthrough conditions.
4335  *
4336  **/
4337 static int ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
4338 {
4339         /* Start with base case */
4340         adapter->num_rx_queues = 1;
4341         adapter->num_tx_queues = 1;
4342         adapter->num_rx_pools = adapter->num_rx_queues;
4343         adapter->num_rx_queues_per_pool = 1;
4344
4345         if (ixgbe_set_sriov_queues(adapter))
4346                 goto done;
4347
4348 #ifdef IXGBE_FCOE
4349         if (ixgbe_set_fcoe_queues(adapter))
4350                 goto done;
4351
4352 #endif /* IXGBE_FCOE */
4353 #ifdef CONFIG_IXGBE_DCB
4354         if (ixgbe_set_dcb_queues(adapter))
4355                 goto done;
4356
4357 #endif
4358         if (ixgbe_set_fdir_queues(adapter))
4359                 goto done;
4360
4361         if (ixgbe_set_rss_queues(adapter))
4362                 goto done;
4363
4364         /* fallback to base case */
4365         adapter->num_rx_queues = 1;
4366         adapter->num_tx_queues = 1;
4367
4368 done:
4369         /* Notify the stack of the (possibly) reduced queue counts. */
4370         netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
4371         return netif_set_real_num_rx_queues(adapter->netdev,
4372                                             adapter->num_rx_queues);
4373 }
4374
4375 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
4376                                        int vectors)
4377 {
4378         int err, vector_threshold;
4379
4380         /* We'll want at least 3 (vector_threshold):
4381          * 1) TxQ[0] Cleanup
4382          * 2) RxQ[0] Cleanup
4383          * 3) Other (Link Status Change, etc.)
4384          * 4) TCP Timer (optional)
4385          */
4386         vector_threshold = MIN_MSIX_COUNT;
4387
4388         /* The more we get, the more we will assign to Tx/Rx Cleanup
4389          * for the separate queues...where Rx Cleanup >= Tx Cleanup.
4390          * Right now, we simply care about how many we'll get; we'll
4391          * set them up later while requesting irq's.
4392          */
4393         while (vectors >= vector_threshold) {
4394                 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
4395                                       vectors);
4396                 if (!err) /* Success in acquiring all requested vectors. */
4397                         break;
4398                 else if (err < 0)
4399                         vectors = 0; /* Nasty failure, quit now */
4400                 else /* err == number of vectors we should try again with */
4401                         vectors = err;
4402         }
4403
4404         if (vectors < vector_threshold) {
4405                 /* Can't allocate enough MSI-X interrupts?  Oh well.
4406                  * This just means we'll go with either a single MSI
4407                  * vector or fall back to legacy interrupts.
4408                  */
4409                 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4410                              "Unable to allocate MSI-X interrupts\n");
4411                 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4412                 kfree(adapter->msix_entries);
4413                 adapter->msix_entries = NULL;
4414         } else {
4415                 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
4416                 /*
4417                  * Adjust for only the vectors we'll use, which is minimum
4418                  * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
4419                  * vectors we were allocated.
4420                  */
4421                 adapter->num_msix_vectors = min(vectors,
4422                                    adapter->max_msix_q_vectors + NON_Q_VECTORS);
4423         }
4424 }
4425
4426 /**
4427  * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
4428  * @adapter: board private structure to initialize
4429  *
4430  * Cache the descriptor ring offsets for RSS to the assigned rings.
4431  *
4432  **/
4433 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter *adapter)
4434 {
4435         int i;
4436
4437         if (!(adapter->flags & IXGBE_FLAG_RSS_ENABLED))
4438                 return false;
4439
4440         for (i = 0; i < adapter->num_rx_queues; i++)
4441                 adapter->rx_ring[i]->reg_idx = i;
4442         for (i = 0; i < adapter->num_tx_queues; i++)
4443                 adapter->tx_ring[i]->reg_idx = i;
4444
4445         return true;
4446 }
4447
4448 #ifdef CONFIG_IXGBE_DCB
4449 /**
4450  * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
4451  * @adapter: board private structure to initialize
4452  *
4453  * Cache the descriptor ring offsets for DCB to the assigned rings.
4454  *
4455  **/
4456 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter *adapter)
4457 {
4458         int i;
4459         bool ret = false;
4460         int dcb_i = adapter->ring_feature[RING_F_DCB].indices;
4461
4462         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED))
4463                 return false;
4464
4465         /* the number of queues is assumed to be symmetric */
4466         switch (adapter->hw.mac.type) {
4467         case ixgbe_mac_82598EB:
4468                 for (i = 0; i < dcb_i; i++) {
4469                         adapter->rx_ring[i]->reg_idx = i << 3;
4470                         adapter->tx_ring[i]->reg_idx = i << 2;
4471                 }
4472                 ret = true;
4473                 break;
4474         case ixgbe_mac_82599EB:
4475         case ixgbe_mac_X540:
4476                 if (dcb_i == 8) {
4477                         /*
4478                          * Tx TC0 starts at: descriptor queue 0
4479                          * Tx TC1 starts at: descriptor queue 32
4480                          * Tx TC2 starts at: descriptor queue 64
4481                          * Tx TC3 starts at: descriptor queue 80
4482                          * Tx TC4 starts at: descriptor queue 96
4483                          * Tx TC5 starts at: descriptor queue 104
4484                          * Tx TC6 starts at: descriptor queue 112
4485                          * Tx TC7 starts at: descriptor queue 120
4486                          *
4487                          * Rx TC0-TC7 are offset by 16 queues each
4488                          */
4489                         for (i = 0; i < 3; i++) {
4490                                 adapter->tx_ring[i]->reg_idx = i << 5;
4491                                 adapter->rx_ring[i]->reg_idx = i << 4;
4492                         }
4493                         for ( ; i < 5; i++) {
4494                                 adapter->tx_ring[i]->reg_idx = ((i + 2) << 4);
4495                                 adapter->rx_ring[i]->reg_idx = i << 4;
4496                         }
4497                         for ( ; i < dcb_i; i++) {
4498                                 adapter->tx_ring[i]->reg_idx = ((i + 8) << 3);
4499                                 adapter->rx_ring[i]->reg_idx = i << 4;
4500                         }
4501                         ret = true;
4502                 } else if (dcb_i == 4) {
4503                         /*
4504                          * Tx TC0 starts at: descriptor queue 0
4505                          * Tx TC1 starts at: descriptor queue 64
4506                          * Tx TC2 starts at: descriptor queue 96
4507                          * Tx TC3 starts at: descriptor queue 112
4508                          *
4509                          * Rx TC0-TC3 are offset by 32 queues each
4510                          */
4511                         adapter->tx_ring[0]->reg_idx = 0;
4512                         adapter->tx_ring[1]->reg_idx = 64;
4513                         adapter->tx_ring[2]->reg_idx = 96;
4514                         adapter->tx_ring[3]->reg_idx = 112;
4515                         for (i = 0 ; i < dcb_i; i++)
4516                                 adapter->rx_ring[i]->reg_idx = i << 5;
4517                         ret = true;
4518                 }
4519                 break;
4520         default:
4521                 break;
4522         }
4523         return ret;
4524 }
4525 #endif
4526
4527 /**
4528  * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
4529  * @adapter: board private structure to initialize
4530  *
4531  * Cache the descriptor ring offsets for Flow Director to the assigned rings.
4532  *
4533  **/
4534 static inline bool ixgbe_cache_ring_fdir(struct ixgbe_adapter *adapter)
4535 {
4536         int i;
4537         bool ret = false;
4538
4539         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED &&
4540             ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4541              (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))) {
4542                 for (i = 0; i < adapter->num_rx_queues; i++)
4543                         adapter->rx_ring[i]->reg_idx = i;
4544                 for (i = 0; i < adapter->num_tx_queues; i++)
4545                         adapter->tx_ring[i]->reg_idx = i;
4546                 ret = true;
4547         }
4548
4549         return ret;
4550 }
4551
4552 #ifdef IXGBE_FCOE
4553 /**
4554  * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
4555  * @adapter: board private structure to initialize
4556  *
4557  * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
4558  *
4559  */
4560 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter *adapter)
4561 {
4562         struct ixgbe_ring_feature *f = &adapter->ring_feature[RING_F_FCOE];
4563         int i;
4564         u8 fcoe_rx_i = 0, fcoe_tx_i = 0;
4565
4566         if (!(adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
4567                 return false;
4568
4569 #ifdef CONFIG_IXGBE_DCB
4570         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
4571                 struct ixgbe_fcoe *fcoe = &adapter->fcoe;
4572
4573                 ixgbe_cache_ring_dcb(adapter);
4574                 /* find out queues in TC for FCoE */
4575                 fcoe_rx_i = adapter->rx_ring[fcoe->tc]->reg_idx + 1;
4576                 fcoe_tx_i = adapter->tx_ring[fcoe->tc]->reg_idx + 1;
4577                 /*
4578                  * In 82599, the number of Tx queues for each traffic
4579                  * class for both 8-TC and 4-TC modes are:
4580                  * TCs  : TC0 TC1 TC2 TC3 TC4 TC5 TC6 TC7
4581                  * 8 TCs:  32  32  16  16   8   8   8   8
4582                  * 4 TCs:  64  64  32  32
4583                  * We have max 8 queues for FCoE, where 8 the is
4584                  * FCoE redirection table size. If TC for FCoE is
4585                  * less than or equal to TC3, we have enough queues
4586                  * to add max of 8 queues for FCoE, so we start FCoE
4587                  * Tx queue from the next one, i.e., reg_idx + 1.
4588                  * If TC for FCoE is above TC3, implying 8 TC mode,
4589                  * and we need 8 for FCoE, we have to take all queues
4590                  * in that traffic class for FCoE.
4591                  */
4592                 if ((f->indices == IXGBE_FCRETA_SIZE) && (fcoe->tc > 3))
4593                         fcoe_tx_i--;
4594         }
4595 #endif /* CONFIG_IXGBE_DCB */
4596         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
4597                 if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) ||
4598                     (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
4599                         ixgbe_cache_ring_fdir(adapter);
4600                 else
4601                         ixgbe_cache_ring_rss(adapter);
4602
4603                 fcoe_rx_i = f->mask;
4604                 fcoe_tx_i = f->mask;
4605         }
4606         for (i = 0; i < f->indices; i++, fcoe_rx_i++, fcoe_tx_i++) {
4607                 adapter->rx_ring[f->mask + i]->reg_idx = fcoe_rx_i;
4608                 adapter->tx_ring[f->mask + i]->reg_idx = fcoe_tx_i;
4609         }
4610         return true;
4611 }
4612
4613 #endif /* IXGBE_FCOE */
4614 /**
4615  * ixgbe_cache_ring_sriov - Descriptor ring to register mapping for sriov
4616  * @adapter: board private structure to initialize
4617  *
4618  * SR-IOV doesn't use any descriptor rings but changes the default if
4619  * no other mapping is used.
4620  *
4621  */
4622 static inline bool ixgbe_cache_ring_sriov(struct ixgbe_adapter *adapter)
4623 {
4624         adapter->rx_ring[0]->reg_idx = adapter->num_vfs * 2;
4625         adapter->tx_ring[0]->reg_idx = adapter->num_vfs * 2;
4626         if (adapter->num_vfs)
4627                 return true;
4628         else
4629                 return false;
4630 }
4631
4632 /**
4633  * ixgbe_cache_ring_register - Descriptor ring to register mapping
4634  * @adapter: board private structure to initialize
4635  *
4636  * Once we know the feature-set enabled for the device, we'll cache
4637  * the register offset the descriptor ring is assigned to.
4638  *
4639  * Note, the order the various feature calls is important.  It must start with
4640  * the "most" features enabled at the same time, then trickle down to the
4641  * least amount of features turned on at once.
4642  **/
4643 static void ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
4644 {
4645         /* start with default case */
4646         adapter->rx_ring[0]->reg_idx = 0;
4647         adapter->tx_ring[0]->reg_idx = 0;
4648
4649         if (ixgbe_cache_ring_sriov(adapter))
4650                 return;
4651
4652 #ifdef IXGBE_FCOE
4653         if (ixgbe_cache_ring_fcoe(adapter))
4654                 return;
4655
4656 #endif /* IXGBE_FCOE */
4657 #ifdef CONFIG_IXGBE_DCB
4658         if (ixgbe_cache_ring_dcb(adapter))
4659                 return;
4660
4661 #endif
4662         if (ixgbe_cache_ring_fdir(adapter))
4663                 return;
4664
4665         if (ixgbe_cache_ring_rss(adapter))
4666                 return;
4667 }
4668
4669 /**
4670  * ixgbe_alloc_queues - Allocate memory for all rings
4671  * @adapter: board private structure to initialize
4672  *
4673  * We allocate one ring per queue at run-time since we don't know the
4674  * number of queues at compile-time.  The polling_netdev array is
4675  * intended for Multiqueue, but should work fine with a single queue.
4676  **/
4677 static int ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
4678 {
4679         int i;
4680         int rx_count;
4681         int orig_node = adapter->node;
4682
4683         for (i = 0; i < adapter->num_tx_queues; i++) {
4684                 struct ixgbe_ring *ring = adapter->tx_ring[i];
4685                 if (orig_node == -1) {
4686                         int cur_node = next_online_node(adapter->node);
4687                         if (cur_node == MAX_NUMNODES)
4688                                 cur_node = first_online_node;
4689                         adapter->node = cur_node;
4690                 }
4691                 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4692                                     adapter->node);
4693                 if (!ring)
4694                         ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4695                 if (!ring)
4696                         goto err_tx_ring_allocation;
4697                 ring->count = adapter->tx_ring_count;
4698                 ring->queue_index = i;
4699                 ring->dev = &adapter->pdev->dev;
4700                 ring->netdev = adapter->netdev;
4701                 ring->numa_node = adapter->node;
4702
4703                 adapter->tx_ring[i] = ring;
4704         }
4705
4706         /* Restore the adapter's original node */
4707         adapter->node = orig_node;
4708
4709         rx_count = adapter->rx_ring_count;
4710         for (i = 0; i < adapter->num_rx_queues; i++) {
4711                 struct ixgbe_ring *ring = adapter->rx_ring[i];
4712                 if (orig_node == -1) {
4713                         int cur_node = next_online_node(adapter->node);
4714                         if (cur_node == MAX_NUMNODES)
4715                                 cur_node = first_online_node;
4716                         adapter->node = cur_node;
4717                 }
4718                 ring = kzalloc_node(sizeof(struct ixgbe_ring), GFP_KERNEL,
4719                                     adapter->node);
4720                 if (!ring)
4721                         ring = kzalloc(sizeof(struct ixgbe_ring), GFP_KERNEL);
4722                 if (!ring)
4723                         goto err_rx_ring_allocation;
4724                 ring->count = rx_count;
4725                 ring->queue_index = i;
4726                 ring->dev = &adapter->pdev->dev;
4727                 ring->netdev = adapter->netdev;
4728                 ring->numa_node = adapter->node;
4729
4730                 adapter->rx_ring[i] = ring;
4731         }
4732
4733         /* Restore the adapter's original node */
4734         adapter->node = orig_node;
4735
4736         ixgbe_cache_ring_register(adapter);
4737
4738         return 0;
4739
4740 err_rx_ring_allocation:
4741         for (i = 0; i < adapter->num_tx_queues; i++)
4742                 kfree(adapter->tx_ring[i]);
4743 err_tx_ring_allocation:
4744         return -ENOMEM;
4745 }
4746
4747 /**
4748  * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
4749  * @adapter: board private structure to initialize
4750  *
4751  * Attempt to configure the interrupts using the best available
4752  * capabilities of the hardware and the kernel.
4753  **/
4754 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter *adapter)
4755 {
4756         struct ixgbe_hw *hw = &adapter->hw;
4757         int err = 0;
4758         int vector, v_budget;
4759
4760         /*
4761          * It's easy to be greedy for MSI-X vectors, but it really
4762          * doesn't do us much good if we have a lot more vectors
4763          * than CPU's.  So let's be conservative and only ask for
4764          * (roughly) the same number of vectors as there are CPU's.
4765          */
4766         v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
4767                        (int)num_online_cpus()) + NON_Q_VECTORS;
4768
4769         /*
4770          * At the same time, hardware can only support a maximum of
4771          * hw.mac->max_msix_vectors vectors.  With features
4772          * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
4773          * descriptor queues supported by our device.  Thus, we cap it off in
4774          * those rare cases where the cpu count also exceeds our vector limit.
4775          */
4776         v_budget = min(v_budget, (int)hw->mac.max_msix_vectors);
4777
4778         /* A failure in MSI-X entry allocation isn't fatal, but it does
4779          * mean we disable MSI-X capabilities of the adapter. */
4780         adapter->msix_entries = kcalloc(v_budget,
4781                                         sizeof(struct msix_entry), GFP_KERNEL);
4782         if (adapter->msix_entries) {
4783                 for (vector = 0; vector < v_budget; vector++)
4784                         adapter->msix_entries[vector].entry = vector;
4785
4786                 ixgbe_acquire_msix_vectors(adapter, v_budget);
4787
4788                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4789                         goto out;
4790         }
4791
4792         adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
4793         adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
4794         adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
4795         adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
4796         adapter->atr_sample_rate = 0;
4797         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4798                 ixgbe_disable_sriov(adapter);
4799
4800         err = ixgbe_set_num_queues(adapter);
4801         if (err)
4802                 return err;
4803
4804         err = pci_enable_msi(adapter->pdev);
4805         if (!err) {
4806                 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
4807         } else {
4808                 netif_printk(adapter, hw, KERN_DEBUG, adapter->netdev,
4809                              "Unable to allocate MSI interrupt, "
4810                              "falling back to legacy.  Error: %d\n", err);
4811                 /* reset err */
4812                 err = 0;
4813         }
4814
4815 out:
4816         return err;
4817 }
4818
4819 /**
4820  * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
4821  * @adapter: board private structure to initialize
4822  *
4823  * We allocate one q_vector per queue interrupt.  If allocation fails we
4824  * return -ENOMEM.
4825  **/
4826 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter *adapter)
4827 {
4828         int q_idx, num_q_vectors;
4829         struct ixgbe_q_vector *q_vector;
4830         int napi_vectors;
4831         int (*poll)(struct napi_struct *, int);
4832
4833         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4834                 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4835                 napi_vectors = adapter->num_rx_queues;
4836                 poll = &ixgbe_clean_rxtx_many;
4837         } else {
4838                 num_q_vectors = 1;
4839                 napi_vectors = 1;
4840                 poll = &ixgbe_poll;
4841         }
4842
4843         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4844                 q_vector = kzalloc_node(sizeof(struct ixgbe_q_vector),
4845                                         GFP_KERNEL, adapter->node);
4846                 if (!q_vector)
4847                         q_vector = kzalloc(sizeof(struct ixgbe_q_vector),
4848                                            GFP_KERNEL);
4849                 if (!q_vector)
4850                         goto err_out;
4851                 q_vector->adapter = adapter;
4852                 if (q_vector->txr_count && !q_vector->rxr_count)
4853                         q_vector->eitr = adapter->tx_eitr_param;
4854                 else
4855                         q_vector->eitr = adapter->rx_eitr_param;
4856                 q_vector->v_idx = q_idx;
4857                 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
4858                 adapter->q_vector[q_idx] = q_vector;
4859         }
4860
4861         return 0;
4862
4863 err_out:
4864         while (q_idx) {
4865                 q_idx--;
4866                 q_vector = adapter->q_vector[q_idx];
4867                 netif_napi_del(&q_vector->napi);
4868                 kfree(q_vector);
4869                 adapter->q_vector[q_idx] = NULL;
4870         }
4871         return -ENOMEM;
4872 }
4873
4874 /**
4875  * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
4876  * @adapter: board private structure to initialize
4877  *
4878  * This function frees the memory allocated to the q_vectors.  In addition if
4879  * NAPI is enabled it will delete any references to the NAPI struct prior
4880  * to freeing the q_vector.
4881  **/
4882 static void ixgbe_free_q_vectors(struct ixgbe_adapter *adapter)
4883 {
4884         int q_idx, num_q_vectors;
4885
4886         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4887                 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
4888         else
4889                 num_q_vectors = 1;
4890
4891         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
4892                 struct ixgbe_q_vector *q_vector = adapter->q_vector[q_idx];
4893                 adapter->q_vector[q_idx] = NULL;
4894                 netif_napi_del(&q_vector->napi);
4895                 kfree(q_vector);
4896         }
4897 }
4898
4899 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
4900 {
4901         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4902                 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
4903                 pci_disable_msix(adapter->pdev);
4904                 kfree(adapter->msix_entries);
4905                 adapter->msix_entries = NULL;
4906         } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
4907                 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
4908                 pci_disable_msi(adapter->pdev);
4909         }
4910 }
4911
4912 /**
4913  * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
4914  * @adapter: board private structure to initialize
4915  *
4916  * We determine which interrupt scheme to use based on...
4917  * - Kernel support (MSI, MSI-X)
4918  *   - which can be user-defined (via MODULE_PARAM)
4919  * - Hardware queue count (num_*_queues)
4920  *   - defined by miscellaneous hardware support/features (RSS, etc.)
4921  **/
4922 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
4923 {
4924         int err;
4925
4926         /* Number of supported queues */
4927         err = ixgbe_set_num_queues(adapter);
4928         if (err)
4929                 return err;
4930
4931         err = ixgbe_set_interrupt_capability(adapter);
4932         if (err) {
4933                 e_dev_err("Unable to setup interrupt capabilities\n");
4934                 goto err_set_interrupt;
4935         }
4936
4937         err = ixgbe_alloc_q_vectors(adapter);
4938         if (err) {
4939                 e_dev_err("Unable to allocate memory for queue vectors\n");
4940                 goto err_alloc_q_vectors;
4941         }
4942
4943         err = ixgbe_alloc_queues(adapter);
4944         if (err) {
4945                 e_dev_err("Unable to allocate memory for queues\n");
4946                 goto err_alloc_queues;
4947         }
4948
4949         e_dev_info("Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
4950                    (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
4951                    adapter->num_rx_queues, adapter->num_tx_queues);
4952
4953         set_bit(__IXGBE_DOWN, &adapter->state);
4954
4955         return 0;
4956
4957 err_alloc_queues:
4958         ixgbe_free_q_vectors(adapter);
4959 err_alloc_q_vectors:
4960         ixgbe_reset_interrupt_capability(adapter);
4961 err_set_interrupt:
4962         return err;
4963 }
4964
4965 static void ring_free_rcu(struct rcu_head *head)
4966 {
4967         kfree(container_of(head, struct ixgbe_ring, rcu));
4968 }
4969
4970 /**
4971  * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
4972  * @adapter: board private structure to clear interrupt scheme on
4973  *
4974  * We go through and clear interrupt specific resources and reset the structure
4975  * to pre-load conditions
4976  **/
4977 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter *adapter)
4978 {
4979         int i;
4980
4981         for (i = 0; i < adapter->num_tx_queues; i++) {
4982                 kfree(adapter->tx_ring[i]);
4983                 adapter->tx_ring[i] = NULL;
4984         }
4985         for (i = 0; i < adapter->num_rx_queues; i++) {
4986                 struct ixgbe_ring *ring = adapter->rx_ring[i];
4987
4988                 /* ixgbe_get_stats64() might access this ring, we must wait
4989                  * a grace period before freeing it.
4990                  */
4991                 call_rcu(&ring->rcu, ring_free_rcu);
4992                 adapter->rx_ring[i] = NULL;
4993         }
4994
4995         ixgbe_free_q_vectors(adapter);
4996         ixgbe_reset_interrupt_capability(adapter);
4997 }
4998
4999 /**
5000  * ixgbe_sfp_timer - worker thread to find a missing module
5001  * @data: pointer to our adapter struct
5002  **/
5003 static void ixgbe_sfp_timer(unsigned long data)
5004 {
5005         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5006
5007         /*
5008          * Do the sfp_timer outside of interrupt context due to the
5009          * delays that sfp+ detection requires
5010          */
5011         schedule_work(&adapter->sfp_task);
5012 }
5013
5014 /**
5015  * ixgbe_sfp_task - worker thread to find a missing module
5016  * @work: pointer to work_struct containing our data
5017  **/
5018 static void ixgbe_sfp_task(struct work_struct *work)
5019 {
5020         struct ixgbe_adapter *adapter = container_of(work,
5021                                                      struct ixgbe_adapter,
5022                                                      sfp_task);
5023         struct ixgbe_hw *hw = &adapter->hw;
5024
5025         if ((hw->phy.type == ixgbe_phy_nl) &&
5026             (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
5027                 s32 ret = hw->phy.ops.identify_sfp(hw);
5028                 if (ret == IXGBE_ERR_SFP_NOT_PRESENT)
5029                         goto reschedule;
5030                 ret = hw->phy.ops.reset(hw);
5031                 if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5032                         e_dev_err("failed to initialize because an unsupported "
5033                                   "SFP+ module type was detected.\n");
5034                         e_dev_err("Reload the driver after installing a "
5035                                   "supported module.\n");
5036                         unregister_netdev(adapter->netdev);
5037                 } else {
5038                         e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
5039                 }
5040                 /* don't need this routine any more */
5041                 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
5042         }
5043         return;
5044 reschedule:
5045         if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state))
5046                 mod_timer(&adapter->sfp_timer,
5047                           round_jiffies(jiffies + (2 * HZ)));
5048 }
5049
5050 /**
5051  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
5052  * @adapter: board private structure to initialize
5053  *
5054  * ixgbe_sw_init initializes the Adapter private data structure.
5055  * Fields are initialized based on PCI device information and
5056  * OS network device settings (MTU size).
5057  **/
5058 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
5059 {
5060         struct ixgbe_hw *hw = &adapter->hw;
5061         struct pci_dev *pdev = adapter->pdev;
5062         struct net_device *dev = adapter->netdev;
5063         unsigned int rss;
5064 #ifdef CONFIG_IXGBE_DCB
5065         int j;
5066         struct tc_configuration *tc;
5067 #endif
5068         int max_frame = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
5069
5070         /* PCI config space info */
5071
5072         hw->vendor_id = pdev->vendor;
5073         hw->device_id = pdev->device;
5074         hw->revision_id = pdev->revision;
5075         hw->subsystem_vendor_id = pdev->subsystem_vendor;
5076         hw->subsystem_device_id = pdev->subsystem_device;
5077
5078         /* Set capability flags */
5079         rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
5080         adapter->ring_feature[RING_F_RSS].indices = rss;
5081         adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
5082         adapter->ring_feature[RING_F_DCB].indices = IXGBE_MAX_DCB_INDICES;
5083         switch (hw->mac.type) {
5084         case ixgbe_mac_82598EB:
5085                 if (hw->device_id == IXGBE_DEV_ID_82598AT)
5086                         adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
5087                 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82598;
5088                 break;
5089         case ixgbe_mac_82599EB:
5090         case ixgbe_mac_X540:
5091                 adapter->max_msix_q_vectors = MAX_MSIX_Q_VECTORS_82599;
5092                 adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
5093                 adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
5094                 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
5095                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5096                 if (dev->features & NETIF_F_NTUPLE) {
5097                         /* Flow Director perfect filter enabled */
5098                         adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
5099                         adapter->atr_sample_rate = 0;
5100                         spin_lock_init(&adapter->fdir_perfect_lock);
5101                 } else {
5102                         /* Flow Director hash filters enabled */
5103                         adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
5104                         adapter->atr_sample_rate = 20;
5105                 }
5106                 adapter->ring_feature[RING_F_FDIR].indices =
5107                                                          IXGBE_MAX_FDIR_INDICES;
5108                 adapter->fdir_pballoc = 0;
5109 #ifdef IXGBE_FCOE
5110                 adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
5111                 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
5112                 adapter->ring_feature[RING_F_FCOE].indices = 0;
5113 #ifdef CONFIG_IXGBE_DCB
5114                 /* Default traffic class to use for FCoE */
5115                 adapter->fcoe.tc = IXGBE_FCOE_DEFTC;
5116                 adapter->fcoe.up = IXGBE_FCOE_DEFTC;
5117 #endif
5118 #endif /* IXGBE_FCOE */
5119                 break;
5120         default:
5121                 break;
5122         }
5123
5124 #ifdef CONFIG_IXGBE_DCB
5125         /* Configure DCB traffic classes */
5126         for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
5127                 tc = &adapter->dcb_cfg.tc_config[j];
5128                 tc->path[DCB_TX_CONFIG].bwg_id = 0;
5129                 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
5130                 tc->path[DCB_RX_CONFIG].bwg_id = 0;
5131                 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
5132                 tc->dcb_pfc = pfc_disabled;
5133         }
5134         adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
5135         adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
5136         adapter->dcb_cfg.rx_pba_cfg = pba_equal;
5137         adapter->dcb_cfg.pfc_mode_enable = false;
5138         adapter->dcb_cfg.round_robin_enable = false;
5139         adapter->dcb_set_bitmap = 0x00;
5140         ixgbe_copy_dcb_cfg(&adapter->dcb_cfg, &adapter->temp_dcb_cfg,
5141                            adapter->ring_feature[RING_F_DCB].indices);
5142
5143 #endif
5144
5145         /* default flow control settings */
5146         hw->fc.requested_mode = ixgbe_fc_full;
5147         hw->fc.current_mode = ixgbe_fc_full;    /* init for ethtool output */
5148 #ifdef CONFIG_DCB
5149         adapter->last_lfc_mode = hw->fc.current_mode;
5150 #endif
5151         hw->fc.high_water = FC_HIGH_WATER(max_frame);
5152         hw->fc.low_water = FC_LOW_WATER(max_frame);
5153         hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
5154         hw->fc.send_xon = true;
5155         hw->fc.disable_fc_autoneg = false;
5156
5157         /* enable itr by default in dynamic mode */
5158         adapter->rx_itr_setting = 1;
5159         adapter->rx_eitr_param = 20000;
5160         adapter->tx_itr_setting = 1;
5161         adapter->tx_eitr_param = 10000;
5162
5163         /* set defaults for eitr in MegaBytes */
5164         adapter->eitr_low = 10;
5165         adapter->eitr_high = 20;
5166
5167         /* set default ring sizes */
5168         adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
5169         adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
5170
5171         /* initialize eeprom parameters */
5172         if (ixgbe_init_eeprom_params_generic(hw)) {
5173                 e_dev_err("EEPROM initialization failed\n");
5174                 return -EIO;
5175         }
5176
5177         /* enable rx csum by default */
5178         adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
5179
5180         /* get assigned NUMA node */
5181         adapter->node = dev_to_node(&pdev->dev);
5182
5183         set_bit(__IXGBE_DOWN, &adapter->state);
5184
5185         return 0;
5186 }
5187
5188 /**
5189  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
5190  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
5191  *
5192  * Return 0 on success, negative on failure
5193  **/
5194 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
5195 {
5196         struct device *dev = tx_ring->dev;
5197         int size;
5198
5199         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5200         tx_ring->tx_buffer_info = vmalloc_node(size, tx_ring->numa_node);
5201         if (!tx_ring->tx_buffer_info)
5202                 tx_ring->tx_buffer_info = vmalloc(size);
5203         if (!tx_ring->tx_buffer_info)
5204                 goto err;
5205         memset(tx_ring->tx_buffer_info, 0, size);
5206
5207         /* round up to nearest 4K */
5208         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
5209         tx_ring->size = ALIGN(tx_ring->size, 4096);
5210
5211         tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
5212                                            &tx_ring->dma, GFP_KERNEL);
5213         if (!tx_ring->desc)
5214                 goto err;
5215
5216         tx_ring->next_to_use = 0;
5217         tx_ring->next_to_clean = 0;
5218         tx_ring->work_limit = tx_ring->count;
5219         return 0;
5220
5221 err:
5222         vfree(tx_ring->tx_buffer_info);
5223         tx_ring->tx_buffer_info = NULL;
5224         dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
5225         return -ENOMEM;
5226 }
5227
5228 /**
5229  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
5230  * @adapter: board private structure
5231  *
5232  * If this function returns with an error, then it's possible one or
5233  * more of the rings is populated (while the rest are not).  It is the
5234  * callers duty to clean those orphaned rings.
5235  *
5236  * Return 0 on success, negative on failure
5237  **/
5238 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
5239 {
5240         int i, err = 0;
5241
5242         for (i = 0; i < adapter->num_tx_queues; i++) {
5243                 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
5244                 if (!err)
5245                         continue;
5246                 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5247                 break;
5248         }
5249
5250         return err;
5251 }
5252
5253 /**
5254  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5255  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
5256  *
5257  * Returns 0 on success, negative on failure
5258  **/
5259 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5260 {
5261         struct device *dev = rx_ring->dev;
5262         int size;
5263
5264         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5265         rx_ring->rx_buffer_info = vmalloc_node(size, rx_ring->numa_node);
5266         if (!rx_ring->rx_buffer_info)
5267                 rx_ring->rx_buffer_info = vmalloc(size);
5268         if (!rx_ring->rx_buffer_info)
5269                 goto err;
5270         memset(rx_ring->rx_buffer_info, 0, size);
5271
5272         /* Round up to nearest 4K */
5273         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5274         rx_ring->size = ALIGN(rx_ring->size, 4096);
5275
5276         rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5277                                            &rx_ring->dma, GFP_KERNEL);
5278
5279         if (!rx_ring->desc)
5280                 goto err;
5281
5282         rx_ring->next_to_clean = 0;
5283         rx_ring->next_to_use = 0;
5284
5285         return 0;
5286 err:
5287         vfree(rx_ring->rx_buffer_info);
5288         rx_ring->rx_buffer_info = NULL;
5289         dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5290         return -ENOMEM;
5291 }
5292
5293 /**
5294  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5295  * @adapter: board private structure
5296  *
5297  * If this function returns with an error, then it's possible one or
5298  * more of the rings is populated (while the rest are not).  It is the
5299  * callers duty to clean those orphaned rings.
5300  *
5301  * Return 0 on success, negative on failure
5302  **/
5303 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5304 {
5305         int i, err = 0;
5306
5307         for (i = 0; i < adapter->num_rx_queues; i++) {
5308                 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5309                 if (!err)
5310                         continue;
5311                 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5312                 break;
5313         }
5314
5315         return err;
5316 }
5317
5318 /**
5319  * ixgbe_free_tx_resources - Free Tx Resources per Queue
5320  * @tx_ring: Tx descriptor ring for a specific queue
5321  *
5322  * Free all transmit software resources
5323  **/
5324 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5325 {
5326         ixgbe_clean_tx_ring(tx_ring);
5327
5328         vfree(tx_ring->tx_buffer_info);
5329         tx_ring->tx_buffer_info = NULL;
5330
5331         /* if not set, then don't free */
5332         if (!tx_ring->desc)
5333                 return;
5334
5335         dma_free_coherent(tx_ring->dev, tx_ring->size,
5336                           tx_ring->desc, tx_ring->dma);
5337
5338         tx_ring->desc = NULL;
5339 }
5340
5341 /**
5342  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5343  * @adapter: board private structure
5344  *
5345  * Free all transmit software resources
5346  **/
5347 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5348 {
5349         int i;
5350
5351         for (i = 0; i < adapter->num_tx_queues; i++)
5352                 if (adapter->tx_ring[i]->desc)
5353                         ixgbe_free_tx_resources(adapter->tx_ring[i]);
5354 }
5355
5356 /**
5357  * ixgbe_free_rx_resources - Free Rx Resources
5358  * @rx_ring: ring to clean the resources from
5359  *
5360  * Free all receive software resources
5361  **/
5362 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5363 {
5364         ixgbe_clean_rx_ring(rx_ring);
5365
5366         vfree(rx_ring->rx_buffer_info);
5367         rx_ring->rx_buffer_info = NULL;
5368
5369         /* if not set, then don't free */
5370         if (!rx_ring->desc)
5371                 return;
5372
5373         dma_free_coherent(rx_ring->dev, rx_ring->size,
5374                           rx_ring->desc, rx_ring->dma);
5375
5376         rx_ring->desc = NULL;
5377 }
5378
5379 /**
5380  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5381  * @adapter: board private structure
5382  *
5383  * Free all receive software resources
5384  **/
5385 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5386 {
5387         int i;
5388
5389         for (i = 0; i < adapter->num_rx_queues; i++)
5390                 if (adapter->rx_ring[i]->desc)
5391                         ixgbe_free_rx_resources(adapter->rx_ring[i]);
5392 }
5393
5394 /**
5395  * ixgbe_change_mtu - Change the Maximum Transfer Unit
5396  * @netdev: network interface device structure
5397  * @new_mtu: new value for maximum frame size
5398  *
5399  * Returns 0 on success, negative on failure
5400  **/
5401 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5402 {
5403         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5404         struct ixgbe_hw *hw = &adapter->hw;
5405         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5406
5407         /* MTU < 68 is an error and causes problems on some kernels */
5408         if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5409                 return -EINVAL;
5410
5411         e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5412         /* must set new MTU before calling down or up */
5413         netdev->mtu = new_mtu;
5414
5415         hw->fc.high_water = FC_HIGH_WATER(max_frame);
5416         hw->fc.low_water = FC_LOW_WATER(max_frame);
5417
5418         if (netif_running(netdev))
5419                 ixgbe_reinit_locked(adapter);
5420
5421         return 0;
5422 }
5423
5424 /**
5425  * ixgbe_open - Called when a network interface is made active
5426  * @netdev: network interface device structure
5427  *
5428  * Returns 0 on success, negative value on failure
5429  *
5430  * The open entry point is called when a network interface is made
5431  * active by the system (IFF_UP).  At this point all resources needed
5432  * for transmit and receive operations are allocated, the interrupt
5433  * handler is registered with the OS, the watchdog timer is started,
5434  * and the stack is notified that the interface is ready.
5435  **/
5436 static int ixgbe_open(struct net_device *netdev)
5437 {
5438         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5439         int err;
5440
5441         /* disallow open during test */
5442         if (test_bit(__IXGBE_TESTING, &adapter->state))
5443                 return -EBUSY;
5444
5445         netif_carrier_off(netdev);
5446
5447         /* allocate transmit descriptors */
5448         err = ixgbe_setup_all_tx_resources(adapter);
5449         if (err)
5450                 goto err_setup_tx;
5451
5452         /* allocate receive descriptors */
5453         err = ixgbe_setup_all_rx_resources(adapter);
5454         if (err)
5455                 goto err_setup_rx;
5456
5457         ixgbe_configure(adapter);
5458
5459         err = ixgbe_request_irq(adapter);
5460         if (err)
5461                 goto err_req_irq;
5462
5463         err = ixgbe_up_complete(adapter);
5464         if (err)
5465                 goto err_up;
5466
5467         netif_tx_start_all_queues(netdev);
5468
5469         return 0;
5470
5471 err_up:
5472         ixgbe_release_hw_control(adapter);
5473         ixgbe_free_irq(adapter);
5474 err_req_irq:
5475 err_setup_rx:
5476         ixgbe_free_all_rx_resources(adapter);
5477 err_setup_tx:
5478         ixgbe_free_all_tx_resources(adapter);
5479         ixgbe_reset(adapter);
5480
5481         return err;
5482 }
5483
5484 /**
5485  * ixgbe_close - Disables a network interface
5486  * @netdev: network interface device structure
5487  *
5488  * Returns 0, this is not allowed to fail
5489  *
5490  * The close entry point is called when an interface is de-activated
5491  * by the OS.  The hardware is still under the drivers control, but
5492  * needs to be disabled.  A global MAC reset is issued to stop the
5493  * hardware, and all transmit and receive resources are freed.
5494  **/
5495 static int ixgbe_close(struct net_device *netdev)
5496 {
5497         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5498
5499         ixgbe_down(adapter);
5500         ixgbe_free_irq(adapter);
5501
5502         ixgbe_free_all_tx_resources(adapter);
5503         ixgbe_free_all_rx_resources(adapter);
5504
5505         ixgbe_release_hw_control(adapter);
5506
5507         return 0;
5508 }
5509
5510 #ifdef CONFIG_PM
5511 static int ixgbe_resume(struct pci_dev *pdev)
5512 {
5513         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5514         struct net_device *netdev = adapter->netdev;
5515         u32 err;
5516
5517         pci_set_power_state(pdev, PCI_D0);
5518         pci_restore_state(pdev);
5519         /*
5520          * pci_restore_state clears dev->state_saved so call
5521          * pci_save_state to restore it.
5522          */
5523         pci_save_state(pdev);
5524
5525         err = pci_enable_device_mem(pdev);
5526         if (err) {
5527                 e_dev_err("Cannot enable PCI device from suspend\n");
5528                 return err;
5529         }
5530         pci_set_master(pdev);
5531
5532         pci_wake_from_d3(pdev, false);
5533
5534         err = ixgbe_init_interrupt_scheme(adapter);
5535         if (err) {
5536                 e_dev_err("Cannot initialize interrupts for device\n");
5537                 return err;
5538         }
5539
5540         ixgbe_reset(adapter);
5541
5542         IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5543
5544         if (netif_running(netdev)) {
5545                 err = ixgbe_open(netdev);
5546                 if (err)
5547                         return err;
5548         }
5549
5550         netif_device_attach(netdev);
5551
5552         return 0;
5553 }
5554 #endif /* CONFIG_PM */
5555
5556 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5557 {
5558         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5559         struct net_device *netdev = adapter->netdev;
5560         struct ixgbe_hw *hw = &adapter->hw;
5561         u32 ctrl, fctrl;
5562         u32 wufc = adapter->wol;
5563 #ifdef CONFIG_PM
5564         int retval = 0;
5565 #endif
5566
5567         netif_device_detach(netdev);
5568
5569         if (netif_running(netdev)) {
5570                 ixgbe_down(adapter);
5571                 ixgbe_free_irq(adapter);
5572                 ixgbe_free_all_tx_resources(adapter);
5573                 ixgbe_free_all_rx_resources(adapter);
5574         }
5575
5576         ixgbe_clear_interrupt_scheme(adapter);
5577
5578 #ifdef CONFIG_PM
5579         retval = pci_save_state(pdev);
5580         if (retval)
5581                 return retval;
5582
5583 #endif
5584         if (wufc) {
5585                 ixgbe_set_rx_mode(netdev);
5586
5587                 /* turn on all-multi mode if wake on multicast is enabled */
5588                 if (wufc & IXGBE_WUFC_MC) {
5589                         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5590                         fctrl |= IXGBE_FCTRL_MPE;
5591                         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5592                 }
5593
5594                 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5595                 ctrl |= IXGBE_CTRL_GIO_DIS;
5596                 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5597
5598                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5599         } else {
5600                 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5601                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5602         }
5603
5604         switch (hw->mac.type) {
5605         case ixgbe_mac_82598EB:
5606                 pci_wake_from_d3(pdev, false);
5607                 break;
5608         case ixgbe_mac_82599EB:
5609         case ixgbe_mac_X540:
5610                 pci_wake_from_d3(pdev, !!wufc);
5611                 break;
5612         default:
5613                 break;
5614         }
5615
5616         *enable_wake = !!wufc;
5617
5618         ixgbe_release_hw_control(adapter);
5619
5620         pci_disable_device(pdev);
5621
5622         return 0;
5623 }
5624
5625 #ifdef CONFIG_PM
5626 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5627 {
5628         int retval;
5629         bool wake;
5630
5631         retval = __ixgbe_shutdown(pdev, &wake);
5632         if (retval)
5633                 return retval;
5634
5635         if (wake) {
5636                 pci_prepare_to_sleep(pdev);
5637         } else {
5638                 pci_wake_from_d3(pdev, false);
5639                 pci_set_power_state(pdev, PCI_D3hot);
5640         }
5641
5642         return 0;
5643 }
5644 #endif /* CONFIG_PM */
5645
5646 static void ixgbe_shutdown(struct pci_dev *pdev)
5647 {
5648         bool wake;
5649
5650         __ixgbe_shutdown(pdev, &wake);
5651
5652         if (system_state == SYSTEM_POWER_OFF) {
5653                 pci_wake_from_d3(pdev, wake);
5654                 pci_set_power_state(pdev, PCI_D3hot);
5655         }
5656 }
5657
5658 /**
5659  * ixgbe_update_stats - Update the board statistics counters.
5660  * @adapter: board private structure
5661  **/
5662 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
5663 {
5664         struct net_device *netdev = adapter->netdev;
5665         struct ixgbe_hw *hw = &adapter->hw;
5666         struct ixgbe_hw_stats *hwstats = &adapter->stats;
5667         u64 total_mpc = 0;
5668         u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
5669         u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
5670         u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
5671         u64 bytes = 0, packets = 0;
5672
5673         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
5674             test_bit(__IXGBE_RESETTING, &adapter->state))
5675                 return;
5676
5677         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
5678                 u64 rsc_count = 0;
5679                 u64 rsc_flush = 0;
5680                 for (i = 0; i < 16; i++)
5681                         adapter->hw_rx_no_dma_resources +=
5682                                 IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
5683                 for (i = 0; i < adapter->num_rx_queues; i++) {
5684                         rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
5685                         rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
5686                 }
5687                 adapter->rsc_total_count = rsc_count;
5688                 adapter->rsc_total_flush = rsc_flush;
5689         }
5690
5691         for (i = 0; i < adapter->num_rx_queues; i++) {
5692                 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
5693                 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
5694                 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
5695                 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
5696                 bytes += rx_ring->stats.bytes;
5697                 packets += rx_ring->stats.packets;
5698         }
5699         adapter->non_eop_descs = non_eop_descs;
5700         adapter->alloc_rx_page_failed = alloc_rx_page_failed;
5701         adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
5702         netdev->stats.rx_bytes = bytes;
5703         netdev->stats.rx_packets = packets;
5704
5705         bytes = 0;
5706         packets = 0;
5707         /* gather some stats to the adapter struct that are per queue */
5708         for (i = 0; i < adapter->num_tx_queues; i++) {
5709                 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
5710                 restart_queue += tx_ring->tx_stats.restart_queue;
5711                 tx_busy += tx_ring->tx_stats.tx_busy;
5712                 bytes += tx_ring->stats.bytes;
5713                 packets += tx_ring->stats.packets;
5714         }
5715         adapter->restart_queue = restart_queue;
5716         adapter->tx_busy = tx_busy;
5717         netdev->stats.tx_bytes = bytes;
5718         netdev->stats.tx_packets = packets;
5719
5720         hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
5721         for (i = 0; i < 8; i++) {
5722                 /* for packet buffers not used, the register should read 0 */
5723                 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
5724                 missed_rx += mpc;
5725                 hwstats->mpc[i] += mpc;
5726                 total_mpc += hwstats->mpc[i];
5727                 if (hw->mac.type == ixgbe_mac_82598EB)
5728                         hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
5729                 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
5730                 hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
5731                 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
5732                 hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
5733                 switch (hw->mac.type) {
5734                 case ixgbe_mac_82598EB:
5735                         hwstats->pxonrxc[i] +=
5736                                 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
5737                         break;
5738                 case ixgbe_mac_82599EB:
5739                 case ixgbe_mac_X540:
5740                         hwstats->pxonrxc[i] +=
5741                                 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
5742                         break;
5743                 default:
5744                         break;
5745                 }
5746                 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
5747                 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
5748         }
5749         hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
5750         /* work around hardware counting issue */
5751         hwstats->gprc -= missed_rx;
5752
5753         ixgbe_update_xoff_received(adapter);
5754
5755         /* 82598 hardware only has a 32 bit counter in the high register */
5756         switch (hw->mac.type) {
5757         case ixgbe_mac_82598EB:
5758                 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
5759                 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
5760                 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
5761                 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
5762                 break;
5763         case ixgbe_mac_82599EB:
5764         case ixgbe_mac_X540:
5765                 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
5766                 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
5767                 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
5768                 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
5769                 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
5770                 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
5771                 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
5772                 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
5773                 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
5774 #ifdef IXGBE_FCOE
5775                 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
5776                 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
5777                 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
5778                 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
5779                 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
5780                 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
5781 #endif /* IXGBE_FCOE */
5782                 break;
5783         default:
5784                 break;
5785         }
5786         bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
5787         hwstats->bprc += bprc;
5788         hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
5789         if (hw->mac.type == ixgbe_mac_82598EB)
5790                 hwstats->mprc -= bprc;
5791         hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
5792         hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
5793         hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
5794         hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
5795         hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
5796         hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
5797         hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
5798         hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
5799         lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
5800         hwstats->lxontxc += lxon;
5801         lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
5802         hwstats->lxofftxc += lxoff;
5803         hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5804         hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
5805         hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
5806         /*
5807          * 82598 errata - tx of flow control packets is included in tx counters
5808          */
5809         xon_off_tot = lxon + lxoff;
5810         hwstats->gptc -= xon_off_tot;
5811         hwstats->mptc -= xon_off_tot;
5812         hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
5813         hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
5814         hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
5815         hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
5816         hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
5817         hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
5818         hwstats->ptc64 -= xon_off_tot;
5819         hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
5820         hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
5821         hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
5822         hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
5823         hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
5824         hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
5825
5826         /* Fill out the OS statistics structure */
5827         netdev->stats.multicast = hwstats->mprc;
5828
5829         /* Rx Errors */
5830         netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
5831         netdev->stats.rx_dropped = 0;
5832         netdev->stats.rx_length_errors = hwstats->rlec;
5833         netdev->stats.rx_crc_errors = hwstats->crcerrs;
5834         netdev->stats.rx_missed_errors = total_mpc;
5835 }
5836
5837 /**
5838  * ixgbe_watchdog - Timer Call-back
5839  * @data: pointer to adapter cast into an unsigned long
5840  **/
5841 static void ixgbe_watchdog(unsigned long data)
5842 {
5843         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
5844         struct ixgbe_hw *hw = &adapter->hw;
5845         u64 eics = 0;
5846         int i;
5847
5848         /*
5849          *  Do the watchdog outside of interrupt context due to the lovely
5850          * delays that some of the newer hardware requires
5851          */
5852
5853         if (test_bit(__IXGBE_DOWN, &adapter->state))
5854                 goto watchdog_short_circuit;
5855
5856         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
5857                 /*
5858                  * for legacy and MSI interrupts don't set any bits
5859                  * that are enabled for EIAM, because this operation
5860                  * would set *both* EIMS and EICS for any bit in EIAM
5861                  */
5862                 IXGBE_WRITE_REG(hw, IXGBE_EICS,
5863                         (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
5864                 goto watchdog_reschedule;
5865         }
5866
5867         /* get one bit for every active tx/rx interrupt vector */
5868         for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
5869                 struct ixgbe_q_vector *qv = adapter->q_vector[i];
5870                 if (qv->rxr_count || qv->txr_count)
5871                         eics |= ((u64)1 << i);
5872         }
5873
5874         /* Cause software interrupt to ensure rx rings are cleaned */
5875         ixgbe_irq_rearm_queues(adapter, eics);
5876
5877 watchdog_reschedule:
5878         /* Reset the timer */
5879         mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
5880
5881 watchdog_short_circuit:
5882         schedule_work(&adapter->watchdog_task);
5883 }
5884
5885 /**
5886  * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
5887  * @work: pointer to work_struct containing our data
5888  **/
5889 static void ixgbe_multispeed_fiber_task(struct work_struct *work)
5890 {
5891         struct ixgbe_adapter *adapter = container_of(work,
5892                                                      struct ixgbe_adapter,
5893                                                      multispeed_fiber_task);
5894         struct ixgbe_hw *hw = &adapter->hw;
5895         u32 autoneg;
5896         bool negotiation;
5897
5898         adapter->flags |= IXGBE_FLAG_IN_SFP_LINK_TASK;
5899         autoneg = hw->phy.autoneg_advertised;
5900         if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
5901                 hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiation);
5902         hw->mac.autotry_restart = false;
5903         if (hw->mac.ops.setup_link)
5904                 hw->mac.ops.setup_link(hw, autoneg, negotiation, true);
5905         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5906         adapter->flags &= ~IXGBE_FLAG_IN_SFP_LINK_TASK;
5907 }
5908
5909 /**
5910  * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
5911  * @work: pointer to work_struct containing our data
5912  **/
5913 static void ixgbe_sfp_config_module_task(struct work_struct *work)
5914 {
5915         struct ixgbe_adapter *adapter = container_of(work,
5916                                                      struct ixgbe_adapter,
5917                                                      sfp_config_module_task);
5918         struct ixgbe_hw *hw = &adapter->hw;
5919         u32 err;
5920
5921         adapter->flags |= IXGBE_FLAG_IN_SFP_MOD_TASK;
5922
5923         /* Time for electrical oscillations to settle down */
5924         msleep(100);
5925         err = hw->phy.ops.identify_sfp(hw);
5926
5927         if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
5928                 e_dev_err("failed to initialize because an unsupported SFP+ "
5929                           "module type was detected.\n");
5930                 e_dev_err("Reload the driver after installing a supported "
5931                           "module.\n");
5932                 unregister_netdev(adapter->netdev);
5933                 return;
5934         }
5935         hw->mac.ops.setup_sfp(hw);
5936
5937         if (!(adapter->flags & IXGBE_FLAG_IN_SFP_LINK_TASK))
5938                 /* This will also work for DA Twinax connections */
5939                 schedule_work(&adapter->multispeed_fiber_task);
5940         adapter->flags &= ~IXGBE_FLAG_IN_SFP_MOD_TASK;
5941 }
5942
5943 /**
5944  * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
5945  * @work: pointer to work_struct containing our data
5946  **/
5947 static void ixgbe_fdir_reinit_task(struct work_struct *work)
5948 {
5949         struct ixgbe_adapter *adapter = container_of(work,
5950                                                      struct ixgbe_adapter,
5951                                                      fdir_reinit_task);
5952         struct ixgbe_hw *hw = &adapter->hw;
5953         int i;
5954
5955         if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
5956                 for (i = 0; i < adapter->num_tx_queues; i++)
5957                         set_bit(__IXGBE_TX_FDIR_INIT_DONE,
5958                                 &(adapter->tx_ring[i]->state));
5959         } else {
5960                 e_err(probe, "failed to finish FDIR re-initialization, "
5961                       "ignored adding FDIR ATR filters\n");
5962         }
5963         /* Done FDIR Re-initialization, enable transmits */
5964         netif_tx_start_all_queues(adapter->netdev);
5965 }
5966
5967 static DEFINE_MUTEX(ixgbe_watchdog_lock);
5968
5969 /**
5970  * ixgbe_watchdog_task - worker thread to bring link up
5971  * @work: pointer to work_struct containing our data
5972  **/
5973 static void ixgbe_watchdog_task(struct work_struct *work)
5974 {
5975         struct ixgbe_adapter *adapter = container_of(work,
5976                                                      struct ixgbe_adapter,
5977                                                      watchdog_task);
5978         struct net_device *netdev = adapter->netdev;
5979         struct ixgbe_hw *hw = &adapter->hw;
5980         u32 link_speed;
5981         bool link_up;
5982         int i;
5983         struct ixgbe_ring *tx_ring;
5984         int some_tx_pending = 0;
5985
5986         mutex_lock(&ixgbe_watchdog_lock);
5987
5988         link_up = adapter->link_up;
5989         link_speed = adapter->link_speed;
5990
5991         if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
5992                 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
5993                 if (link_up) {
5994 #ifdef CONFIG_DCB
5995                         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
5996                                 for (i = 0; i < MAX_TRAFFIC_CLASS; i++)
5997                                         hw->mac.ops.fc_enable(hw, i);
5998                         } else {
5999                                 hw->mac.ops.fc_enable(hw, 0);
6000                         }
6001 #else
6002                         hw->mac.ops.fc_enable(hw, 0);
6003 #endif
6004                 }
6005
6006                 if (link_up ||
6007                     time_after(jiffies, (adapter->link_check_timeout +
6008                                          IXGBE_TRY_LINK_TIMEOUT))) {
6009                         adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6010                         IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
6011                 }
6012                 adapter->link_up = link_up;
6013                 adapter->link_speed = link_speed;
6014         }
6015
6016         if (link_up) {
6017                 if (!netif_carrier_ok(netdev)) {
6018                         bool flow_rx, flow_tx;
6019
6020                         switch (hw->mac.type) {
6021                         case ixgbe_mac_82598EB: {
6022                                 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6023                                 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
6024                                 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
6025                                 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
6026                         }
6027                                 break;
6028                         case ixgbe_mac_82599EB:
6029                         case ixgbe_mac_X540: {
6030                                 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
6031                                 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
6032                                 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
6033                                 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
6034                         }
6035                                 break;
6036                         default:
6037                                 flow_tx = false;
6038                                 flow_rx = false;
6039                                 break;
6040                         }
6041
6042                         e_info(drv, "NIC Link is Up %s, Flow Control: %s\n",
6043                                (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
6044                                "10 Gbps" :
6045                                (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
6046                                "1 Gbps" : "unknown speed")),
6047                                ((flow_rx && flow_tx) ? "RX/TX" :
6048                                (flow_rx ? "RX" :
6049                                (flow_tx ? "TX" : "None"))));
6050
6051                         netif_carrier_on(netdev);
6052                 } else {
6053                         /* Force detection of hung controller */
6054                         for (i = 0; i < adapter->num_tx_queues; i++) {
6055                                 tx_ring = adapter->tx_ring[i];
6056                                 set_check_for_tx_hang(tx_ring);
6057                         }
6058                 }
6059         } else {
6060                 adapter->link_up = false;
6061                 adapter->link_speed = 0;
6062                 if (netif_carrier_ok(netdev)) {
6063                         e_info(drv, "NIC Link is Down\n");
6064                         netif_carrier_off(netdev);
6065                 }
6066         }
6067
6068         if (!netif_carrier_ok(netdev)) {
6069                 for (i = 0; i < adapter->num_tx_queues; i++) {
6070                         tx_ring = adapter->tx_ring[i];
6071                         if (tx_ring->next_to_use != tx_ring->next_to_clean) {
6072                                 some_tx_pending = 1;
6073                                 break;
6074                         }
6075                 }
6076
6077                 if (some_tx_pending) {
6078                         /* We've lost link, so the controller stops DMA,
6079                          * but we've got queued Tx work that's never going
6080                          * to get done, so reset controller to flush Tx.
6081                          * (Do the reset outside of interrupt context).
6082                          */
6083                          schedule_work(&adapter->reset_task);
6084                 }
6085         }
6086
6087         ixgbe_update_stats(adapter);
6088         mutex_unlock(&ixgbe_watchdog_lock);
6089 }
6090
6091 static int ixgbe_tso(struct ixgbe_adapter *adapter,
6092                      struct ixgbe_ring *tx_ring, struct sk_buff *skb,
6093                      u32 tx_flags, u8 *hdr_len, __be16 protocol)
6094 {
6095         struct ixgbe_adv_tx_context_desc *context_desc;
6096         unsigned int i;
6097         int err;
6098         struct ixgbe_tx_buffer *tx_buffer_info;
6099         u32 vlan_macip_lens = 0, type_tucmd_mlhl;
6100         u32 mss_l4len_idx, l4len;
6101
6102         if (skb_is_gso(skb)) {
6103                 if (skb_header_cloned(skb)) {
6104                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
6105                         if (err)
6106                                 return err;
6107                 }
6108                 l4len = tcp_hdrlen(skb);
6109                 *hdr_len += l4len;
6110
6111                 if (protocol == htons(ETH_P_IP)) {
6112                         struct iphdr *iph = ip_hdr(skb);
6113                         iph->tot_len = 0;
6114                         iph->check = 0;
6115                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
6116                                                                  iph->daddr, 0,
6117                                                                  IPPROTO_TCP,
6118                                                                  0);
6119                 } else if (skb_is_gso_v6(skb)) {
6120                         ipv6_hdr(skb)->payload_len = 0;
6121                         tcp_hdr(skb)->check =
6122                             ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
6123                                              &ipv6_hdr(skb)->daddr,
6124                                              0, IPPROTO_TCP, 0);
6125                 }
6126
6127                 i = tx_ring->next_to_use;
6128
6129                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6130                 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
6131
6132                 /* VLAN MACLEN IPLEN */
6133                 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6134                         vlan_macip_lens |=
6135                             (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
6136                 vlan_macip_lens |= ((skb_network_offset(skb)) <<
6137                                     IXGBE_ADVTXD_MACLEN_SHIFT);
6138                 *hdr_len += skb_network_offset(skb);
6139                 vlan_macip_lens |=
6140                     (skb_transport_header(skb) - skb_network_header(skb));
6141                 *hdr_len +=
6142                     (skb_transport_header(skb) - skb_network_header(skb));
6143                 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
6144                 context_desc->seqnum_seed = 0;
6145
6146                 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
6147                 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
6148                                    IXGBE_ADVTXD_DTYP_CTXT);
6149
6150                 if (protocol == htons(ETH_P_IP))
6151                         type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
6152                 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6153                 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
6154
6155                 /* MSS L4LEN IDX */
6156                 mss_l4len_idx =
6157                     (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
6158                 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
6159                 /* use index 1 for TSO */
6160                 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6161                 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
6162
6163                 tx_buffer_info->time_stamp = jiffies;
6164                 tx_buffer_info->next_to_watch = i;
6165
6166                 i++;
6167                 if (i == tx_ring->count)
6168                         i = 0;
6169                 tx_ring->next_to_use = i;
6170
6171                 return true;
6172         }
6173         return false;
6174 }
6175
6176 static u32 ixgbe_psum(struct ixgbe_adapter *adapter, struct sk_buff *skb,
6177                       __be16 protocol)
6178 {
6179         u32 rtn = 0;
6180
6181         switch (protocol) {
6182         case cpu_to_be16(ETH_P_IP):
6183                 rtn |= IXGBE_ADVTXD_TUCMD_IPV4;
6184                 switch (ip_hdr(skb)->protocol) {
6185                 case IPPROTO_TCP:
6186                         rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6187                         break;
6188                 case IPPROTO_SCTP:
6189                         rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6190                         break;
6191                 }
6192                 break;
6193         case cpu_to_be16(ETH_P_IPV6):
6194                 /* XXX what about other V6 headers?? */
6195                 switch (ipv6_hdr(skb)->nexthdr) {
6196                 case IPPROTO_TCP:
6197                         rtn |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
6198                         break;
6199                 case IPPROTO_SCTP:
6200                         rtn |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
6201                         break;
6202                 }
6203                 break;
6204         default:
6205                 if (unlikely(net_ratelimit()))
6206                         e_warn(probe, "partial checksum but proto=%x!\n",
6207                                protocol);
6208                 break;
6209         }
6210
6211         return rtn;
6212 }
6213
6214 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
6215                           struct ixgbe_ring *tx_ring,
6216                           struct sk_buff *skb, u32 tx_flags,
6217                           __be16 protocol)
6218 {
6219         struct ixgbe_adv_tx_context_desc *context_desc;
6220         unsigned int i;
6221         struct ixgbe_tx_buffer *tx_buffer_info;
6222         u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
6223
6224         if (skb->ip_summed == CHECKSUM_PARTIAL ||
6225             (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
6226                 i = tx_ring->next_to_use;
6227                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6228                 context_desc = IXGBE_TX_CTXTDESC_ADV(tx_ring, i);
6229
6230                 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6231                         vlan_macip_lens |=
6232                             (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
6233                 vlan_macip_lens |= (skb_network_offset(skb) <<
6234                                     IXGBE_ADVTXD_MACLEN_SHIFT);
6235                 if (skb->ip_summed == CHECKSUM_PARTIAL)
6236                         vlan_macip_lens |= (skb_transport_header(skb) -
6237                                             skb_network_header(skb));
6238
6239                 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
6240                 context_desc->seqnum_seed = 0;
6241
6242                 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
6243                                     IXGBE_ADVTXD_DTYP_CTXT);
6244
6245                 if (skb->ip_summed == CHECKSUM_PARTIAL)
6246                         type_tucmd_mlhl |= ixgbe_psum(adapter, skb, protocol);
6247
6248                 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
6249                 /* use index zero for tx checksum offload */
6250                 context_desc->mss_l4len_idx = 0;
6251
6252                 tx_buffer_info->time_stamp = jiffies;
6253                 tx_buffer_info->next_to_watch = i;
6254
6255                 i++;
6256                 if (i == tx_ring->count)
6257                         i = 0;
6258                 tx_ring->next_to_use = i;
6259
6260                 return true;
6261         }
6262
6263         return false;
6264 }
6265
6266 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
6267                         struct ixgbe_ring *tx_ring,
6268                         struct sk_buff *skb, u32 tx_flags,
6269                         unsigned int first, const u8 hdr_len)
6270 {
6271         struct device *dev = tx_ring->dev;
6272         struct ixgbe_tx_buffer *tx_buffer_info;
6273         unsigned int len;
6274         unsigned int total = skb->len;
6275         unsigned int offset = 0, size, count = 0, i;
6276         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
6277         unsigned int f;
6278         unsigned int bytecount = skb->len;
6279         u16 gso_segs = 1;
6280
6281         i = tx_ring->next_to_use;
6282
6283         if (tx_flags & IXGBE_TX_FLAGS_FCOE)
6284                 /* excluding fcoe_crc_eof for FCoE */
6285                 total -= sizeof(struct fcoe_crc_eof);
6286
6287         len = min(skb_headlen(skb), total);
6288         while (len) {
6289                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6290                 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6291
6292                 tx_buffer_info->length = size;
6293                 tx_buffer_info->mapped_as_page = false;
6294                 tx_buffer_info->dma = dma_map_single(dev,
6295                                                      skb->data + offset,
6296                                                      size, DMA_TO_DEVICE);
6297                 if (dma_mapping_error(dev, tx_buffer_info->dma))
6298                         goto dma_error;
6299                 tx_buffer_info->time_stamp = jiffies;
6300                 tx_buffer_info->next_to_watch = i;
6301
6302                 len -= size;
6303                 total -= size;
6304                 offset += size;
6305                 count++;
6306
6307                 if (len) {
6308                         i++;
6309                         if (i == tx_ring->count)
6310                                 i = 0;
6311                 }
6312         }
6313
6314         for (f = 0; f < nr_frags; f++) {
6315                 struct skb_frag_struct *frag;
6316
6317                 frag = &skb_shinfo(skb)->frags[f];
6318                 len = min((unsigned int)frag->size, total);
6319                 offset = frag->page_offset;
6320
6321                 while (len) {
6322                         i++;
6323                         if (i == tx_ring->count)
6324                                 i = 0;
6325
6326                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
6327                         size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
6328
6329                         tx_buffer_info->length = size;
6330                         tx_buffer_info->dma = dma_map_page(dev,
6331                                                            frag->page,
6332                                                            offset, size,
6333                                                            DMA_TO_DEVICE);
6334                         tx_buffer_info->mapped_as_page = true;
6335                         if (dma_mapping_error(dev, tx_buffer_info->dma))
6336                                 goto dma_error;
6337                         tx_buffer_info->time_stamp = jiffies;
6338                         tx_buffer_info->next_to_watch = i;
6339
6340                         len -= size;
6341                         total -= size;
6342                         offset += size;
6343                         count++;
6344                 }
6345                 if (total == 0)
6346                         break;
6347         }
6348
6349         if (tx_flags & IXGBE_TX_FLAGS_TSO)
6350                 gso_segs = skb_shinfo(skb)->gso_segs;
6351 #ifdef IXGBE_FCOE
6352         /* adjust for FCoE Sequence Offload */
6353         else if (tx_flags & IXGBE_TX_FLAGS_FSO)
6354                 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
6355                                         skb_shinfo(skb)->gso_size);
6356 #endif /* IXGBE_FCOE */
6357         bytecount += (gso_segs - 1) * hdr_len;
6358
6359         /* multiply data chunks by size of headers */
6360         tx_ring->tx_buffer_info[i].bytecount = bytecount;
6361         tx_ring->tx_buffer_info[i].gso_segs = gso_segs;
6362         tx_ring->tx_buffer_info[i].skb = skb;
6363         tx_ring->tx_buffer_info[first].next_to_watch = i;
6364
6365         return count;
6366
6367 dma_error:
6368         e_dev_err("TX DMA map failed\n");
6369
6370         /* clear timestamp and dma mappings for failed tx_buffer_info map */
6371         tx_buffer_info->dma = 0;
6372         tx_buffer_info->time_stamp = 0;
6373         tx_buffer_info->next_to_watch = 0;
6374         if (count)
6375                 count--;
6376
6377         /* clear timestamp and dma mappings for remaining portion of packet */
6378         while (count--) {
6379                 if (i == 0)
6380                         i += tx_ring->count;
6381                 i--;
6382                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6383                 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
6384         }
6385
6386         return 0;
6387 }
6388
6389 static void ixgbe_tx_queue(struct ixgbe_ring *tx_ring,
6390                            int tx_flags, int count, u32 paylen, u8 hdr_len)
6391 {
6392         union ixgbe_adv_tx_desc *tx_desc = NULL;
6393         struct ixgbe_tx_buffer *tx_buffer_info;
6394         u32 olinfo_status = 0, cmd_type_len = 0;
6395         unsigned int i;
6396         u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
6397
6398         cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
6399
6400         cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
6401
6402         if (tx_flags & IXGBE_TX_FLAGS_VLAN)
6403                 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
6404
6405         if (tx_flags & IXGBE_TX_FLAGS_TSO) {
6406                 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6407
6408                 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6409                                  IXGBE_ADVTXD_POPTS_SHIFT;
6410
6411                 /* use index 1 context for tso */
6412                 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6413                 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
6414                         olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
6415                                          IXGBE_ADVTXD_POPTS_SHIFT;
6416
6417         } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
6418                 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
6419                                  IXGBE_ADVTXD_POPTS_SHIFT;
6420
6421         if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6422                 olinfo_status |= IXGBE_ADVTXD_CC;
6423                 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
6424                 if (tx_flags & IXGBE_TX_FLAGS_FSO)
6425                         cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
6426         }
6427
6428         olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
6429
6430         i = tx_ring->next_to_use;
6431         while (count--) {
6432                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
6433                 tx_desc = IXGBE_TX_DESC_ADV(tx_ring, i);
6434                 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
6435                 tx_desc->read.cmd_type_len =
6436                         cpu_to_le32(cmd_type_len | tx_buffer_info->length);
6437                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6438                 i++;
6439                 if (i == tx_ring->count)
6440                         i = 0;
6441         }
6442
6443         tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
6444
6445         /*
6446          * Force memory writes to complete before letting h/w
6447          * know there are new descriptors to fetch.  (Only
6448          * applicable for weak-ordered memory model archs,
6449          * such as IA-64).
6450          */
6451         wmb();
6452
6453         tx_ring->next_to_use = i;
6454         writel(i, tx_ring->tail);
6455 }
6456
6457 static void ixgbe_atr(struct ixgbe_adapter *adapter, struct sk_buff *skb,
6458                       u8 queue, u32 tx_flags, __be16 protocol)
6459 {
6460         struct ixgbe_atr_input atr_input;
6461         struct iphdr *iph = ip_hdr(skb);
6462         struct ethhdr *eth = (struct ethhdr *)skb->data;
6463         struct tcphdr *th;
6464         u16 vlan_id;
6465
6466         /* Right now, we support IPv4 w/ TCP only */
6467         if (protocol != htons(ETH_P_IP) ||
6468             iph->protocol != IPPROTO_TCP)
6469                 return;
6470
6471         memset(&atr_input, 0, sizeof(struct ixgbe_atr_input));
6472
6473         vlan_id = (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK) >>
6474                    IXGBE_TX_FLAGS_VLAN_SHIFT;
6475
6476         th = tcp_hdr(skb);
6477
6478         ixgbe_atr_set_vlan_id_82599(&atr_input, vlan_id);
6479         ixgbe_atr_set_src_port_82599(&atr_input, th->dest);
6480         ixgbe_atr_set_dst_port_82599(&atr_input, th->source);
6481         ixgbe_atr_set_flex_byte_82599(&atr_input, eth->h_proto);
6482         ixgbe_atr_set_l4type_82599(&atr_input, IXGBE_ATR_L4TYPE_TCP);
6483         /* src and dst are inverted, think how the receiver sees them */
6484         ixgbe_atr_set_src_ipv4_82599(&atr_input, iph->daddr);
6485         ixgbe_atr_set_dst_ipv4_82599(&atr_input, iph->saddr);
6486
6487         /* This assumes the Rx queue and Tx queue are bound to the same CPU */
6488         ixgbe_fdir_add_signature_filter_82599(&adapter->hw, &atr_input, queue);
6489 }
6490
6491 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, int size)
6492 {
6493         netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
6494         /* Herbert's original patch had:
6495          *  smp_mb__after_netif_stop_queue();
6496          * but since that doesn't exist yet, just open code it. */
6497         smp_mb();
6498
6499         /* We need to check again in a case another CPU has just
6500          * made room available. */
6501         if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
6502                 return -EBUSY;
6503
6504         /* A reprieve! - use start_queue because it doesn't call schedule */
6505         netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
6506         ++tx_ring->tx_stats.restart_queue;
6507         return 0;
6508 }
6509
6510 static int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, int size)
6511 {
6512         if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
6513                 return 0;
6514         return __ixgbe_maybe_stop_tx(tx_ring, size);
6515 }
6516
6517 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb)
6518 {
6519         struct ixgbe_adapter *adapter = netdev_priv(dev);
6520         int txq = smp_processor_id();
6521 #ifdef IXGBE_FCOE
6522         __be16 protocol;
6523
6524         protocol = vlan_get_protocol(skb);
6525
6526         if ((protocol == htons(ETH_P_FCOE)) ||
6527             (protocol == htons(ETH_P_FIP))) {
6528                 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
6529                         txq &= (adapter->ring_feature[RING_F_FCOE].indices - 1);
6530                         txq += adapter->ring_feature[RING_F_FCOE].mask;
6531                         return txq;
6532 #ifdef CONFIG_IXGBE_DCB
6533                 } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6534                         txq = adapter->fcoe.up;
6535                         return txq;
6536 #endif
6537                 }
6538         }
6539 #endif
6540
6541         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
6542                 while (unlikely(txq >= dev->real_num_tx_queues))
6543                         txq -= dev->real_num_tx_queues;
6544                 return txq;
6545         }
6546
6547         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6548                 if (skb->priority == TC_PRIO_CONTROL)
6549                         txq = adapter->ring_feature[RING_F_DCB].indices-1;
6550                 else
6551                         txq = (skb->vlan_tci & IXGBE_TX_FLAGS_VLAN_PRIO_MASK)
6552                                >> 13;
6553                 return txq;
6554         }
6555
6556         return skb_tx_hash(dev, skb);
6557 }
6558
6559 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
6560                           struct ixgbe_adapter *adapter,
6561                           struct ixgbe_ring *tx_ring)
6562 {
6563         struct net_device *netdev = tx_ring->netdev;
6564         struct netdev_queue *txq;
6565         unsigned int first;
6566         unsigned int tx_flags = 0;
6567         u8 hdr_len = 0;
6568         int tso;
6569         int count = 0;
6570         unsigned int f;
6571         __be16 protocol;
6572
6573         protocol = vlan_get_protocol(skb);
6574
6575         if (vlan_tx_tag_present(skb)) {
6576                 tx_flags |= vlan_tx_tag_get(skb);
6577                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6578                         tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
6579                         tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6580                 }
6581                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6582                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6583         } else if (adapter->flags & IXGBE_FLAG_DCB_ENABLED &&
6584                    skb->priority != TC_PRIO_CONTROL) {
6585                 tx_flags |= ((skb->queue_mapping & 0x7) << 13);
6586                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
6587                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
6588         }
6589
6590 #ifdef IXGBE_FCOE
6591         /* for FCoE with DCB, we force the priority to what
6592          * was specified by the switch */
6593         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED &&
6594             (protocol == htons(ETH_P_FCOE) ||
6595              protocol == htons(ETH_P_FIP))) {
6596 #ifdef CONFIG_IXGBE_DCB
6597                 if (adapter->flags & IXGBE_FLAG_DCB_ENABLED) {
6598                         tx_flags &= ~(IXGBE_TX_FLAGS_VLAN_PRIO_MASK
6599                                       << IXGBE_TX_FLAGS_VLAN_SHIFT);
6600                         tx_flags |= ((adapter->fcoe.up << 13)
6601                                       << IXGBE_TX_FLAGS_VLAN_SHIFT);
6602                 }
6603 #endif
6604                 /* flag for FCoE offloads */
6605                 if (protocol == htons(ETH_P_FCOE))
6606                         tx_flags |= IXGBE_TX_FLAGS_FCOE;
6607         }
6608 #endif
6609
6610         /* four things can cause us to need a context descriptor */
6611         if (skb_is_gso(skb) ||
6612             (skb->ip_summed == CHECKSUM_PARTIAL) ||
6613             (tx_flags & IXGBE_TX_FLAGS_VLAN) ||
6614             (tx_flags & IXGBE_TX_FLAGS_FCOE))
6615                 count++;
6616
6617         count += TXD_USE_COUNT(skb_headlen(skb));
6618         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6619                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6620
6621         if (ixgbe_maybe_stop_tx(tx_ring, count)) {
6622                 tx_ring->tx_stats.tx_busy++;
6623                 return NETDEV_TX_BUSY;
6624         }
6625
6626         first = tx_ring->next_to_use;
6627         if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
6628 #ifdef IXGBE_FCOE
6629                 /* setup tx offload for FCoE */
6630                 tso = ixgbe_fso(adapter, tx_ring, skb, tx_flags, &hdr_len);
6631                 if (tso < 0) {
6632                         dev_kfree_skb_any(skb);
6633                         return NETDEV_TX_OK;
6634                 }
6635                 if (tso)
6636                         tx_flags |= IXGBE_TX_FLAGS_FSO;
6637 #endif /* IXGBE_FCOE */
6638         } else {
6639                 if (protocol == htons(ETH_P_IP))
6640                         tx_flags |= IXGBE_TX_FLAGS_IPV4;
6641                 tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len,
6642                                 protocol);
6643                 if (tso < 0) {
6644                         dev_kfree_skb_any(skb);
6645                         return NETDEV_TX_OK;
6646                 }
6647
6648                 if (tso)
6649                         tx_flags |= IXGBE_TX_FLAGS_TSO;
6650                 else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags,
6651                                        protocol) &&
6652                          (skb->ip_summed == CHECKSUM_PARTIAL))
6653                         tx_flags |= IXGBE_TX_FLAGS_CSUM;
6654         }
6655
6656         count = ixgbe_tx_map(adapter, tx_ring, skb, tx_flags, first, hdr_len);
6657         if (count) {
6658                 /* add the ATR filter if ATR is on */
6659                 if (tx_ring->atr_sample_rate) {
6660                         ++tx_ring->atr_count;
6661                         if ((tx_ring->atr_count >= tx_ring->atr_sample_rate) &&
6662                              test_bit(__IXGBE_TX_FDIR_INIT_DONE,
6663                                       &tx_ring->state)) {
6664                                 ixgbe_atr(adapter, skb, tx_ring->queue_index,
6665                                           tx_flags, protocol);
6666                                 tx_ring->atr_count = 0;
6667                         }
6668                 }
6669                 txq = netdev_get_tx_queue(netdev, tx_ring->queue_index);
6670                 txq->tx_bytes += skb->len;
6671                 txq->tx_packets++;
6672                 ixgbe_tx_queue(tx_ring, tx_flags, count, skb->len, hdr_len);
6673                 ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
6674
6675         } else {
6676                 dev_kfree_skb_any(skb);
6677                 tx_ring->tx_buffer_info[first].time_stamp = 0;
6678                 tx_ring->next_to_use = first;
6679         }
6680
6681         return NETDEV_TX_OK;
6682 }
6683
6684 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
6685 {
6686         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6687         struct ixgbe_ring *tx_ring;
6688
6689         tx_ring = adapter->tx_ring[skb->queue_mapping];
6690         return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
6691 }
6692
6693 /**
6694  * ixgbe_set_mac - Change the Ethernet Address of the NIC
6695  * @netdev: network interface device structure
6696  * @p: pointer to an address structure
6697  *
6698  * Returns 0 on success, negative on failure
6699  **/
6700 static int ixgbe_set_mac(struct net_device *netdev, void *p)
6701 {
6702         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6703         struct ixgbe_hw *hw = &adapter->hw;
6704         struct sockaddr *addr = p;
6705
6706         if (!is_valid_ether_addr(addr->sa_data))
6707                 return -EADDRNOTAVAIL;
6708
6709         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
6710         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
6711
6712         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, adapter->num_vfs,
6713                             IXGBE_RAH_AV);
6714
6715         return 0;
6716 }
6717
6718 static int
6719 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
6720 {
6721         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6722         struct ixgbe_hw *hw = &adapter->hw;
6723         u16 value;
6724         int rc;
6725
6726         if (prtad != hw->phy.mdio.prtad)
6727                 return -EINVAL;
6728         rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
6729         if (!rc)
6730                 rc = value;
6731         return rc;
6732 }
6733
6734 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
6735                             u16 addr, u16 value)
6736 {
6737         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6738         struct ixgbe_hw *hw = &adapter->hw;
6739
6740         if (prtad != hw->phy.mdio.prtad)
6741                 return -EINVAL;
6742         return hw->phy.ops.write_reg(hw, addr, devad, value);
6743 }
6744
6745 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
6746 {
6747         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6748
6749         return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
6750 }
6751
6752 /**
6753  * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
6754  * netdev->dev_addrs
6755  * @netdev: network interface device structure
6756  *
6757  * Returns non-zero on failure
6758  **/
6759 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
6760 {
6761         int err = 0;
6762         struct ixgbe_adapter *adapter = netdev_priv(dev);
6763         struct ixgbe_mac_info *mac = &adapter->hw.mac;
6764
6765         if (is_valid_ether_addr(mac->san_addr)) {
6766                 rtnl_lock();
6767                 err = dev_addr_add(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6768                 rtnl_unlock();
6769         }
6770         return err;
6771 }
6772
6773 /**
6774  * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
6775  * netdev->dev_addrs
6776  * @netdev: network interface device structure
6777  *
6778  * Returns non-zero on failure
6779  **/
6780 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
6781 {
6782         int err = 0;
6783         struct ixgbe_adapter *adapter = netdev_priv(dev);
6784         struct ixgbe_mac_info *mac = &adapter->hw.mac;
6785
6786         if (is_valid_ether_addr(mac->san_addr)) {
6787                 rtnl_lock();
6788                 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
6789                 rtnl_unlock();
6790         }
6791         return err;
6792 }
6793
6794 #ifdef CONFIG_NET_POLL_CONTROLLER
6795 /*
6796  * Polling 'interrupt' - used by things like netconsole to send skbs
6797  * without having to re-enable interrupts. It's not called while
6798  * the interrupt routine is executing.
6799  */
6800 static void ixgbe_netpoll(struct net_device *netdev)
6801 {
6802         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6803         int i;
6804
6805         /* if interface is down do nothing */
6806         if (test_bit(__IXGBE_DOWN, &adapter->state))
6807                 return;
6808
6809         adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
6810         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
6811                 int num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
6812                 for (i = 0; i < num_q_vectors; i++) {
6813                         struct ixgbe_q_vector *q_vector = adapter->q_vector[i];
6814                         ixgbe_msix_clean_many(0, q_vector);
6815                 }
6816         } else {
6817                 ixgbe_intr(adapter->pdev->irq, netdev);
6818         }
6819         adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
6820 }
6821 #endif
6822
6823 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
6824                                                    struct rtnl_link_stats64 *stats)
6825 {
6826         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6827         int i;
6828
6829         /* accurate rx/tx bytes/packets stats */
6830         dev_txq_stats_fold(netdev, stats);
6831         rcu_read_lock();
6832         for (i = 0; i < adapter->num_rx_queues; i++) {
6833                 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
6834                 u64 bytes, packets;
6835                 unsigned int start;
6836
6837                 if (ring) {
6838                         do {
6839                                 start = u64_stats_fetch_begin_bh(&ring->syncp);
6840                                 packets = ring->stats.packets;
6841                                 bytes   = ring->stats.bytes;
6842                         } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
6843                         stats->rx_packets += packets;
6844                         stats->rx_bytes   += bytes;
6845                 }
6846         }
6847         rcu_read_unlock();
6848         /* following stats updated by ixgbe_watchdog_task() */
6849         stats->multicast        = netdev->stats.multicast;
6850         stats->rx_errors        = netdev->stats.rx_errors;
6851         stats->rx_length_errors = netdev->stats.rx_length_errors;
6852         stats->rx_crc_errors    = netdev->stats.rx_crc_errors;
6853         stats->rx_missed_errors = netdev->stats.rx_missed_errors;
6854         return stats;
6855 }
6856
6857
6858 static const struct net_device_ops ixgbe_netdev_ops = {
6859         .ndo_open               = ixgbe_open,
6860         .ndo_stop               = ixgbe_close,
6861         .ndo_start_xmit         = ixgbe_xmit_frame,
6862         .ndo_select_queue       = ixgbe_select_queue,
6863         .ndo_set_rx_mode        = ixgbe_set_rx_mode,
6864         .ndo_set_multicast_list = ixgbe_set_rx_mode,
6865         .ndo_validate_addr      = eth_validate_addr,
6866         .ndo_set_mac_address    = ixgbe_set_mac,
6867         .ndo_change_mtu         = ixgbe_change_mtu,
6868         .ndo_tx_timeout         = ixgbe_tx_timeout,
6869         .ndo_vlan_rx_add_vid    = ixgbe_vlan_rx_add_vid,
6870         .ndo_vlan_rx_kill_vid   = ixgbe_vlan_rx_kill_vid,
6871         .ndo_do_ioctl           = ixgbe_ioctl,
6872         .ndo_set_vf_mac         = ixgbe_ndo_set_vf_mac,
6873         .ndo_set_vf_vlan        = ixgbe_ndo_set_vf_vlan,
6874         .ndo_set_vf_tx_rate     = ixgbe_ndo_set_vf_bw,
6875         .ndo_get_vf_config      = ixgbe_ndo_get_vf_config,
6876         .ndo_get_stats64        = ixgbe_get_stats64,
6877 #ifdef CONFIG_NET_POLL_CONTROLLER
6878         .ndo_poll_controller    = ixgbe_netpoll,
6879 #endif
6880 #ifdef IXGBE_FCOE
6881         .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
6882         .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
6883         .ndo_fcoe_enable = ixgbe_fcoe_enable,
6884         .ndo_fcoe_disable = ixgbe_fcoe_disable,
6885         .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
6886 #endif /* IXGBE_FCOE */
6887 };
6888
6889 static void __devinit ixgbe_probe_vf(struct ixgbe_adapter *adapter,
6890                            const struct ixgbe_info *ii)
6891 {
6892 #ifdef CONFIG_PCI_IOV
6893         struct ixgbe_hw *hw = &adapter->hw;
6894         int err;
6895
6896         if (hw->mac.type != ixgbe_mac_82599EB || !max_vfs)
6897                 return;
6898
6899         /* The 82599 supports up to 64 VFs per physical function
6900          * but this implementation limits allocation to 63 so that
6901          * basic networking resources are still available to the
6902          * physical function
6903          */
6904         adapter->num_vfs = (max_vfs > 63) ? 63 : max_vfs;
6905         adapter->flags |= IXGBE_FLAG_SRIOV_ENABLED;
6906         err = pci_enable_sriov(adapter->pdev, adapter->num_vfs);
6907         if (err) {
6908                 e_err(probe, "Failed to enable PCI sriov: %d\n", err);
6909                 goto err_novfs;
6910         }
6911         /* If call to enable VFs succeeded then allocate memory
6912          * for per VF control structures.
6913          */
6914         adapter->vfinfo =
6915                 kcalloc(adapter->num_vfs,
6916                         sizeof(struct vf_data_storage), GFP_KERNEL);
6917         if (adapter->vfinfo) {
6918                 /* Now that we're sure SR-IOV is enabled
6919                  * and memory allocated set up the mailbox parameters
6920                  */
6921                 ixgbe_init_mbx_params_pf(hw);
6922                 memcpy(&hw->mbx.ops, ii->mbx_ops,
6923                        sizeof(hw->mbx.ops));
6924
6925                 /* Disable RSC when in SR-IOV mode */
6926                 adapter->flags2 &= ~(IXGBE_FLAG2_RSC_CAPABLE |
6927                                      IXGBE_FLAG2_RSC_ENABLED);
6928                 return;
6929         }
6930
6931         /* Oh oh */
6932         e_err(probe, "Unable to allocate memory for VF Data Storage - "
6933               "SRIOV disabled\n");
6934         pci_disable_sriov(adapter->pdev);
6935
6936 err_novfs:
6937         adapter->flags &= ~IXGBE_FLAG_SRIOV_ENABLED;
6938         adapter->num_vfs = 0;
6939 #endif /* CONFIG_PCI_IOV */
6940 }
6941
6942 /**
6943  * ixgbe_probe - Device Initialization Routine
6944  * @pdev: PCI device information struct
6945  * @ent: entry in ixgbe_pci_tbl
6946  *
6947  * Returns 0 on success, negative on failure
6948  *
6949  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
6950  * The OS initialization, configuring of the adapter private structure,
6951  * and a hardware reset occur.
6952  **/
6953 static int __devinit ixgbe_probe(struct pci_dev *pdev,
6954                                  const struct pci_device_id *ent)
6955 {
6956         struct net_device *netdev;
6957         struct ixgbe_adapter *adapter = NULL;
6958         struct ixgbe_hw *hw;
6959         const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
6960         static int cards_found;
6961         int i, err, pci_using_dac;
6962         unsigned int indices = num_possible_cpus();
6963 #ifdef IXGBE_FCOE
6964         u16 device_caps;
6965 #endif
6966         u32 part_num, eec;
6967
6968         /* Catch broken hardware that put the wrong VF device ID in
6969          * the PCIe SR-IOV capability.
6970          */
6971         if (pdev->is_virtfn) {
6972                 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
6973                      pci_name(pdev), pdev->vendor, pdev->device);
6974                 return -EINVAL;
6975         }
6976
6977         err = pci_enable_device_mem(pdev);
6978         if (err)
6979                 return err;
6980
6981         if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
6982             !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
6983                 pci_using_dac = 1;
6984         } else {
6985                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6986                 if (err) {
6987                         err = dma_set_coherent_mask(&pdev->dev,
6988                                                     DMA_BIT_MASK(32));
6989                         if (err) {
6990                                 dev_err(&pdev->dev,
6991                                         "No usable DMA configuration, aborting\n");
6992                                 goto err_dma;
6993                         }
6994                 }
6995                 pci_using_dac = 0;
6996         }
6997
6998         err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
6999                                            IORESOURCE_MEM), ixgbe_driver_name);
7000         if (err) {
7001                 dev_err(&pdev->dev,
7002                         "pci_request_selected_regions failed 0x%x\n", err);
7003                 goto err_pci_reg;
7004         }
7005
7006         pci_enable_pcie_error_reporting(pdev);
7007
7008         pci_set_master(pdev);
7009         pci_save_state(pdev);
7010
7011         if (ii->mac == ixgbe_mac_82598EB)
7012                 indices = min_t(unsigned int, indices, IXGBE_MAX_RSS_INDICES);
7013         else
7014                 indices = min_t(unsigned int, indices, IXGBE_MAX_FDIR_INDICES);
7015
7016         indices = max_t(unsigned int, indices, IXGBE_MAX_DCB_INDICES);
7017 #ifdef IXGBE_FCOE
7018         indices += min_t(unsigned int, num_possible_cpus(),
7019                          IXGBE_MAX_FCOE_INDICES);
7020 #endif
7021         netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
7022         if (!netdev) {
7023                 err = -ENOMEM;
7024                 goto err_alloc_etherdev;
7025         }
7026
7027         SET_NETDEV_DEV(netdev, &pdev->dev);
7028
7029         adapter = netdev_priv(netdev);
7030         pci_set_drvdata(pdev, adapter);
7031
7032         adapter->netdev = netdev;
7033         adapter->pdev = pdev;
7034         hw = &adapter->hw;
7035         hw->back = adapter;
7036         adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
7037
7038         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
7039                               pci_resource_len(pdev, 0));
7040         if (!hw->hw_addr) {
7041                 err = -EIO;
7042                 goto err_ioremap;
7043         }
7044
7045         for (i = 1; i <= 5; i++) {
7046                 if (pci_resource_len(pdev, i) == 0)
7047                         continue;
7048         }
7049
7050         netdev->netdev_ops = &ixgbe_netdev_ops;
7051         ixgbe_set_ethtool_ops(netdev);
7052         netdev->watchdog_timeo = 5 * HZ;
7053         strcpy(netdev->name, pci_name(pdev));
7054
7055         adapter->bd_number = cards_found;
7056
7057         /* Setup hw api */
7058         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
7059         hw->mac.type  = ii->mac;
7060
7061         /* EEPROM */
7062         memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
7063         eec = IXGBE_READ_REG(hw, IXGBE_EEC);
7064         /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
7065         if (!(eec & (1 << 8)))
7066                 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
7067
7068         /* PHY */
7069         memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
7070         hw->phy.sfp_type = ixgbe_sfp_type_unknown;
7071         /* ixgbe_identify_phy_generic will set prtad and mmds properly */
7072         hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
7073         hw->phy.mdio.mmds = 0;
7074         hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
7075         hw->phy.mdio.dev = netdev;
7076         hw->phy.mdio.mdio_read = ixgbe_mdio_read;
7077         hw->phy.mdio.mdio_write = ixgbe_mdio_write;
7078
7079         /* set up this timer and work struct before calling get_invariants
7080          * which might start the timer
7081          */
7082         init_timer(&adapter->sfp_timer);
7083         adapter->sfp_timer.function = ixgbe_sfp_timer;
7084         adapter->sfp_timer.data = (unsigned long) adapter;
7085
7086         INIT_WORK(&adapter->sfp_task, ixgbe_sfp_task);
7087
7088         /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
7089         INIT_WORK(&adapter->multispeed_fiber_task, ixgbe_multispeed_fiber_task);
7090
7091         /* a new SFP+ module arrival, called from GPI SDP2 context */
7092         INIT_WORK(&adapter->sfp_config_module_task,
7093                   ixgbe_sfp_config_module_task);
7094
7095         ii->get_invariants(hw);
7096
7097         /* setup the private structure */
7098         err = ixgbe_sw_init(adapter);
7099         if (err)
7100                 goto err_sw_init;
7101
7102         /* Make it possible the adapter to be woken up via WOL */
7103         switch (adapter->hw.mac.type) {
7104         case ixgbe_mac_82599EB:
7105         case ixgbe_mac_X540:
7106                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7107                 break;
7108         default:
7109                 break;
7110         }
7111
7112         /*
7113          * If there is a fan on this device and it has failed log the
7114          * failure.
7115          */
7116         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
7117                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
7118                 if (esdp & IXGBE_ESDP_SDP1)
7119                         e_crit(probe, "Fan has stopped, replace the adapter\n");
7120         }
7121
7122         /* reset_hw fills in the perm_addr as well */
7123         hw->phy.reset_if_overtemp = true;
7124         err = hw->mac.ops.reset_hw(hw);
7125         hw->phy.reset_if_overtemp = false;
7126         if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
7127             hw->mac.type == ixgbe_mac_82598EB) {
7128                 /*
7129                  * Start a kernel thread to watch for a module to arrive.
7130                  * Only do this for 82598, since 82599 will generate
7131                  * interrupts on module arrival.
7132                  */
7133                 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7134                 mod_timer(&adapter->sfp_timer,
7135                           round_jiffies(jiffies + (2 * HZ)));
7136                 err = 0;
7137         } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
7138                 e_dev_err("failed to initialize because an unsupported SFP+ "
7139                           "module type was detected.\n");
7140                 e_dev_err("Reload the driver after installing a supported "
7141                           "module.\n");
7142                 goto err_sw_init;
7143         } else if (err) {
7144                 e_dev_err("HW Init failed: %d\n", err);
7145                 goto err_sw_init;
7146         }
7147
7148         ixgbe_probe_vf(adapter, ii);
7149
7150         netdev->features = NETIF_F_SG |
7151                            NETIF_F_IP_CSUM |
7152                            NETIF_F_HW_VLAN_TX |
7153                            NETIF_F_HW_VLAN_RX |
7154                            NETIF_F_HW_VLAN_FILTER;
7155
7156         netdev->features |= NETIF_F_IPV6_CSUM;
7157         netdev->features |= NETIF_F_TSO;
7158         netdev->features |= NETIF_F_TSO6;
7159         netdev->features |= NETIF_F_GRO;
7160
7161         if (adapter->hw.mac.type == ixgbe_mac_82599EB)
7162                 netdev->features |= NETIF_F_SCTP_CSUM;
7163
7164         netdev->vlan_features |= NETIF_F_TSO;
7165         netdev->vlan_features |= NETIF_F_TSO6;
7166         netdev->vlan_features |= NETIF_F_IP_CSUM;
7167         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
7168         netdev->vlan_features |= NETIF_F_SG;
7169
7170         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7171                 adapter->flags &= ~(IXGBE_FLAG_RSS_ENABLED |
7172                                     IXGBE_FLAG_DCB_ENABLED);
7173         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
7174                 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
7175
7176 #ifdef CONFIG_IXGBE_DCB
7177         netdev->dcbnl_ops = &dcbnl_ops;
7178 #endif
7179
7180 #ifdef IXGBE_FCOE
7181         if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7182                 if (hw->mac.ops.get_device_caps) {
7183                         hw->mac.ops.get_device_caps(hw, &device_caps);
7184                         if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
7185                                 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
7186                 }
7187         }
7188         if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
7189                 netdev->vlan_features |= NETIF_F_FCOE_CRC;
7190                 netdev->vlan_features |= NETIF_F_FSO;
7191                 netdev->vlan_features |= NETIF_F_FCOE_MTU;
7192         }
7193 #endif /* IXGBE_FCOE */
7194         if (pci_using_dac) {
7195                 netdev->features |= NETIF_F_HIGHDMA;
7196                 netdev->vlan_features |= NETIF_F_HIGHDMA;
7197         }
7198
7199         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7200                 netdev->features |= NETIF_F_LRO;
7201
7202         /* make sure the EEPROM is good */
7203         if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
7204                 e_dev_err("The EEPROM Checksum Is Not Valid\n");
7205                 err = -EIO;
7206                 goto err_eeprom;
7207         }
7208
7209         memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
7210         memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
7211
7212         if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
7213                 e_dev_err("invalid MAC address\n");
7214                 err = -EIO;
7215                 goto err_eeprom;
7216         }
7217
7218         /* power down the optics */
7219         if (hw->phy.multispeed_fiber && hw->mac.ops.disable_tx_laser)
7220                 hw->mac.ops.disable_tx_laser(hw);
7221
7222         init_timer(&adapter->watchdog_timer);
7223         adapter->watchdog_timer.function = ixgbe_watchdog;
7224         adapter->watchdog_timer.data = (unsigned long)adapter;
7225
7226         INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
7227         INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
7228
7229         err = ixgbe_init_interrupt_scheme(adapter);
7230         if (err)
7231                 goto err_sw_init;
7232
7233         switch (pdev->device) {
7234         case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
7235                 /* All except this subdevice support WOL */
7236                 if (pdev->subsystem_device ==
7237                     IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ) {
7238                         adapter->wol = 0;
7239                         break;
7240                 }
7241         case IXGBE_DEV_ID_82599_KX4:
7242                 adapter->wol = (IXGBE_WUFC_MAG | IXGBE_WUFC_EX |
7243                                 IXGBE_WUFC_MC | IXGBE_WUFC_BC);
7244                 break;
7245         default:
7246                 adapter->wol = 0;
7247                 break;
7248         }
7249         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
7250
7251         /* pick up the PCI bus settings for reporting later */
7252         hw->mac.ops.get_bus_info(hw);
7253
7254         /* print bus type/speed/width info */
7255         e_dev_info("(PCI Express:%s:%s) %pM\n",
7256                    (hw->bus.speed == ixgbe_bus_speed_5000 ? "5.0Gb/s" :
7257                     hw->bus.speed == ixgbe_bus_speed_2500 ? "2.5Gb/s" :
7258                     "Unknown"),
7259                    (hw->bus.width == ixgbe_bus_width_pcie_x8 ? "Width x8" :
7260                     hw->bus.width == ixgbe_bus_width_pcie_x4 ? "Width x4" :
7261                     hw->bus.width == ixgbe_bus_width_pcie_x1 ? "Width x1" :
7262                     "Unknown"),
7263                    netdev->dev_addr);
7264         ixgbe_read_pba_num_generic(hw, &part_num);
7265         if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
7266                 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, "
7267                            "PBA No: %06x-%03x\n",
7268                            hw->mac.type, hw->phy.type, hw->phy.sfp_type,
7269                            (part_num >> 8), (part_num & 0xff));
7270         else
7271                 e_dev_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
7272                            hw->mac.type, hw->phy.type,
7273                            (part_num >> 8), (part_num & 0xff));
7274
7275         if (hw->bus.width <= ixgbe_bus_width_pcie_x4) {
7276                 e_dev_warn("PCI-Express bandwidth available for this card is "
7277                            "not sufficient for optimal performance.\n");
7278                 e_dev_warn("For optimal performance a x8 PCI-Express slot "
7279                            "is required.\n");
7280         }
7281
7282         /* save off EEPROM version number */
7283         hw->eeprom.ops.read(hw, 0x29, &adapter->eeprom_version);
7284
7285         /* reset the hardware with the new settings */
7286         err = hw->mac.ops.start_hw(hw);
7287
7288         if (err == IXGBE_ERR_EEPROM_VERSION) {
7289                 /* We are running on a pre-production device, log a warning */
7290                 e_dev_warn("This device is a pre-production adapter/LOM. "
7291                            "Please be aware there may be issues associated "
7292                            "with your hardware.  If you are experiencing "
7293                            "problems please contact your Intel or hardware "
7294                            "representative who provided you with this "
7295                            "hardware.\n");
7296         }
7297         strcpy(netdev->name, "eth%d");
7298         err = register_netdev(netdev);
7299         if (err)
7300                 goto err_register;
7301
7302         /* carrier off reporting is important to ethtool even BEFORE open */
7303         netif_carrier_off(netdev);
7304
7305         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7306             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7307                 INIT_WORK(&adapter->fdir_reinit_task, ixgbe_fdir_reinit_task);
7308
7309         if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
7310                 INIT_WORK(&adapter->check_overtemp_task,
7311                           ixgbe_check_overtemp_task);
7312 #ifdef CONFIG_IXGBE_DCA
7313         if (dca_add_requester(&pdev->dev) == 0) {
7314                 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
7315                 ixgbe_setup_dca(adapter);
7316         }
7317 #endif
7318         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
7319                 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
7320                 for (i = 0; i < adapter->num_vfs; i++)
7321                         ixgbe_vf_configuration(pdev, (i | 0x10000000));
7322         }
7323
7324         /* add san mac addr to netdev */
7325         ixgbe_add_sanmac_netdev(netdev);
7326
7327         e_dev_info("Intel(R) 10 Gigabit Network Connection\n");
7328         cards_found++;
7329         return 0;
7330
7331 err_register:
7332         ixgbe_release_hw_control(adapter);
7333         ixgbe_clear_interrupt_scheme(adapter);
7334 err_sw_init:
7335 err_eeprom:
7336         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7337                 ixgbe_disable_sriov(adapter);
7338         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7339         del_timer_sync(&adapter->sfp_timer);
7340         cancel_work_sync(&adapter->sfp_task);
7341         cancel_work_sync(&adapter->multispeed_fiber_task);
7342         cancel_work_sync(&adapter->sfp_config_module_task);
7343         iounmap(hw->hw_addr);
7344 err_ioremap:
7345         free_netdev(netdev);
7346 err_alloc_etherdev:
7347         pci_release_selected_regions(pdev,
7348                                      pci_select_bars(pdev, IORESOURCE_MEM));
7349 err_pci_reg:
7350 err_dma:
7351         pci_disable_device(pdev);
7352         return err;
7353 }
7354
7355 /**
7356  * ixgbe_remove - Device Removal Routine
7357  * @pdev: PCI device information struct
7358  *
7359  * ixgbe_remove is called by the PCI subsystem to alert the driver
7360  * that it should release a PCI device.  The could be caused by a
7361  * Hot-Plug event, or because the driver is going to be removed from
7362  * memory.
7363  **/
7364 static void __devexit ixgbe_remove(struct pci_dev *pdev)
7365 {
7366         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7367         struct net_device *netdev = adapter->netdev;
7368
7369         set_bit(__IXGBE_DOWN, &adapter->state);
7370         /* clear the module not found bit to make sure the worker won't
7371          * reschedule
7372          */
7373         clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND, &adapter->state);
7374         del_timer_sync(&adapter->watchdog_timer);
7375
7376         del_timer_sync(&adapter->sfp_timer);
7377         cancel_work_sync(&adapter->watchdog_task);
7378         cancel_work_sync(&adapter->sfp_task);
7379         cancel_work_sync(&adapter->multispeed_fiber_task);
7380         cancel_work_sync(&adapter->sfp_config_module_task);
7381         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
7382             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7383                 cancel_work_sync(&adapter->fdir_reinit_task);
7384         flush_scheduled_work();
7385
7386 #ifdef CONFIG_IXGBE_DCA
7387         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
7388                 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
7389                 dca_remove_requester(&pdev->dev);
7390                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
7391         }
7392
7393 #endif
7394 #ifdef IXGBE_FCOE
7395         if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7396                 ixgbe_cleanup_fcoe(adapter);
7397
7398 #endif /* IXGBE_FCOE */
7399
7400         /* remove the added san mac */
7401         ixgbe_del_sanmac_netdev(netdev);
7402
7403         if (netdev->reg_state == NETREG_REGISTERED)
7404                 unregister_netdev(netdev);
7405
7406         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7407                 ixgbe_disable_sriov(adapter);
7408
7409         ixgbe_clear_interrupt_scheme(adapter);
7410
7411         ixgbe_release_hw_control(adapter);
7412
7413         iounmap(adapter->hw.hw_addr);
7414         pci_release_selected_regions(pdev, pci_select_bars(pdev,
7415                                      IORESOURCE_MEM));
7416
7417         e_dev_info("complete\n");
7418
7419         free_netdev(netdev);
7420
7421         pci_disable_pcie_error_reporting(pdev);
7422
7423         pci_disable_device(pdev);
7424 }
7425
7426 /**
7427  * ixgbe_io_error_detected - called when PCI error is detected
7428  * @pdev: Pointer to PCI device
7429  * @state: The current pci connection state
7430  *
7431  * This function is called after a PCI bus error affecting
7432  * this device has been detected.
7433  */
7434 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
7435                                                 pci_channel_state_t state)
7436 {
7437         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7438         struct net_device *netdev = adapter->netdev;
7439
7440         netif_device_detach(netdev);
7441
7442         if (state == pci_channel_io_perm_failure)
7443                 return PCI_ERS_RESULT_DISCONNECT;
7444
7445         if (netif_running(netdev))
7446                 ixgbe_down(adapter);
7447         pci_disable_device(pdev);
7448
7449         /* Request a slot reset. */
7450         return PCI_ERS_RESULT_NEED_RESET;
7451 }
7452
7453 /**
7454  * ixgbe_io_slot_reset - called after the pci bus has been reset.
7455  * @pdev: Pointer to PCI device
7456  *
7457  * Restart the card from scratch, as if from a cold-boot.
7458  */
7459 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
7460 {
7461         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7462         pci_ers_result_t result;
7463         int err;
7464
7465         if (pci_enable_device_mem(pdev)) {
7466                 e_err(probe, "Cannot re-enable PCI device after reset.\n");
7467                 result = PCI_ERS_RESULT_DISCONNECT;
7468         } else {
7469                 pci_set_master(pdev);
7470                 pci_restore_state(pdev);
7471                 pci_save_state(pdev);
7472
7473                 pci_wake_from_d3(pdev, false);
7474
7475                 ixgbe_reset(adapter);
7476                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
7477                 result = PCI_ERS_RESULT_RECOVERED;
7478         }
7479
7480         err = pci_cleanup_aer_uncorrect_error_status(pdev);
7481         if (err) {
7482                 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
7483                           "failed 0x%0x\n", err);
7484                 /* non-fatal, continue */
7485         }
7486
7487         return result;
7488 }
7489
7490 /**
7491  * ixgbe_io_resume - called when traffic can start flowing again.
7492  * @pdev: Pointer to PCI device
7493  *
7494  * This callback is called when the error recovery driver tells us that
7495  * its OK to resume normal operation.
7496  */
7497 static void ixgbe_io_resume(struct pci_dev *pdev)
7498 {
7499         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
7500         struct net_device *netdev = adapter->netdev;
7501
7502         if (netif_running(netdev)) {
7503                 if (ixgbe_up(adapter)) {
7504                         e_info(probe, "ixgbe_up failed after reset\n");
7505                         return;
7506                 }
7507         }
7508
7509         netif_device_attach(netdev);
7510 }
7511
7512 static struct pci_error_handlers ixgbe_err_handler = {
7513         .error_detected = ixgbe_io_error_detected,
7514         .slot_reset = ixgbe_io_slot_reset,
7515         .resume = ixgbe_io_resume,
7516 };
7517
7518 static struct pci_driver ixgbe_driver = {
7519         .name     = ixgbe_driver_name,
7520         .id_table = ixgbe_pci_tbl,
7521         .probe    = ixgbe_probe,
7522         .remove   = __devexit_p(ixgbe_remove),
7523 #ifdef CONFIG_PM
7524         .suspend  = ixgbe_suspend,
7525         .resume   = ixgbe_resume,
7526 #endif
7527         .shutdown = ixgbe_shutdown,
7528         .err_handler = &ixgbe_err_handler
7529 };
7530
7531 /**
7532  * ixgbe_init_module - Driver Registration Routine
7533  *
7534  * ixgbe_init_module is the first routine called when the driver is
7535  * loaded. All it does is register with the PCI subsystem.
7536  **/
7537 static int __init ixgbe_init_module(void)
7538 {
7539         int ret;
7540         pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
7541         pr_info("%s\n", ixgbe_copyright);
7542
7543 #ifdef CONFIG_IXGBE_DCA
7544         dca_register_notify(&dca_notifier);
7545 #endif
7546
7547         ret = pci_register_driver(&ixgbe_driver);
7548         return ret;
7549 }
7550
7551 module_init(ixgbe_init_module);
7552
7553 /**
7554  * ixgbe_exit_module - Driver Exit Cleanup Routine
7555  *
7556  * ixgbe_exit_module is called just before the driver is removed
7557  * from memory.
7558  **/
7559 static void __exit ixgbe_exit_module(void)
7560 {
7561 #ifdef CONFIG_IXGBE_DCA
7562         dca_unregister_notify(&dca_notifier);
7563 #endif
7564         pci_unregister_driver(&ixgbe_driver);
7565         rcu_barrier(); /* Wait for completion of call_rcu()'s */
7566 }
7567
7568 #ifdef CONFIG_IXGBE_DCA
7569 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
7570                             void *p)
7571 {
7572         int ret_val;
7573
7574         ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
7575                                          __ixgbe_notify_dca);
7576
7577         return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
7578 }
7579
7580 #endif /* CONFIG_IXGBE_DCA */
7581
7582 /**
7583  * ixgbe_get_hw_dev return device
7584  * used by hardware layer to print debugging information
7585  **/
7586 struct net_device *ixgbe_get_hw_dev(struct ixgbe_hw *hw)
7587 {
7588         struct ixgbe_adapter *adapter = hw->back;
7589         return adapter->netdev;
7590 }
7591
7592 module_exit(ixgbe_exit_module);
7593
7594 /* ixgbe_main.c */