Pileus Git - ~andy/linux/blob - drivers/net/ethernet/intel/e1000e/netdev.c

   1 /*******************************************************************************
   2
   3   Intel PRO/1000 Linux driver
   4   Copyright(c) 1999 - 2012 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   Linux NICS <linux.nics@intel.com>
  24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  26
  27 *******************************************************************************/
  28
  29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  30
  31 #include <linux/module.h>
  32 #include <linux/types.h>
  33 #include <linux/init.h>
  34 #include <linux/pci.h>
  35 #include <linux/vmalloc.h>
  36 #include <linux/pagemap.h>
  37 #include <linux/delay.h>
  38 #include <linux/netdevice.h>
  39 #include <linux/interrupt.h>
  40 #include <linux/tcp.h>
  41 #include <linux/ipv6.h>
  42 #include <linux/slab.h>
  43 #include <net/checksum.h>
  44 #include <net/ip6_checksum.h>
  45 #include <linux/mii.h>
  46 #include <linux/ethtool.h>
  47 #include <linux/if_vlan.h>
  48 #include <linux/cpu.h>
  49 #include <linux/smp.h>
  50 #include <linux/pm_qos.h>
  51 #include <linux/pm_runtime.h>
  52 #include <linux/aer.h>
  53 #include <linux/prefetch.h>
  54
  55 #include "e1000.h"
  56
  57 #define DRV_EXTRAVERSION "-k"
  58
  59 #define DRV_VERSION "2.0.0" DRV_EXTRAVERSION
  60 char e1000e_driver_name[] = "e1000e";
  61 const char e1000e_driver_version[] = DRV_VERSION;
  62
  63 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
  64 static int debug = -1;
  65 module_param(debug, int, 0);
  66 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  67
  68 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state);
  69
  70 static const struct e1000_info *e1000_info_tbl[] = {
  71         [board_82571]           = &e1000_82571_info,
  72         [board_82572]           = &e1000_82572_info,
  73         [board_82573]           = &e1000_82573_info,
  74         [board_82574]           = &e1000_82574_info,
  75         [board_82583]           = &e1000_82583_info,
  76         [board_80003es2lan]     = &e1000_es2_info,
  77         [board_ich8lan]         = &e1000_ich8_info,
  78         [board_ich9lan]         = &e1000_ich9_info,
  79         [board_ich10lan]        = &e1000_ich10_info,
  80         [board_pchlan]          = &e1000_pch_info,
  81         [board_pch2lan]         = &e1000_pch2_info,
  82         [board_pch_lpt]         = &e1000_pch_lpt_info,
  83 };
  84
  85 struct e1000_reg_info {
  86         u32 ofs;
  87         char *name;
  88 };
  89
  90 #define E1000_RDFH      0x02410 /* Rx Data FIFO Head - RW */
  91 #define E1000_RDFT      0x02418 /* Rx Data FIFO Tail - RW */
  92 #define E1000_RDFHS     0x02420 /* Rx Data FIFO Head Saved - RW */
  93 #define E1000_RDFTS     0x02428 /* Rx Data FIFO Tail Saved - RW */
  94 #define E1000_RDFPC     0x02430 /* Rx Data FIFO Packet Count - RW */
  95
  96 #define E1000_TDFH      0x03410 /* Tx Data FIFO Head - RW */
  97 #define E1000_TDFT      0x03418 /* Tx Data FIFO Tail - RW */
  98 #define E1000_TDFHS     0x03420 /* Tx Data FIFO Head Saved - RW */
  99 #define E1000_TDFTS     0x03428 /* Tx Data FIFO Tail Saved - RW */
 100 #define E1000_TDFPC     0x03430 /* Tx Data FIFO Packet Count - RW */
 101
 102 static const struct e1000_reg_info e1000_reg_info_tbl[] = {
 103
 104         /* General Registers */
 105         {E1000_CTRL, "CTRL"},
 106         {E1000_STATUS, "STATUS"},
 107         {E1000_CTRL_EXT, "CTRL_EXT"},
 108
 109         /* Interrupt Registers */
 110         {E1000_ICR, "ICR"},
 111
 112         /* Rx Registers */
 113         {E1000_RCTL, "RCTL"},
 114         {E1000_RDLEN(0), "RDLEN"},
 115         {E1000_RDH(0), "RDH"},
 116         {E1000_RDT(0), "RDT"},
 117         {E1000_RDTR, "RDTR"},
 118         {E1000_RXDCTL(0), "RXDCTL"},
 119         {E1000_ERT, "ERT"},
 120         {E1000_RDBAL(0), "RDBAL"},
 121         {E1000_RDBAH(0), "RDBAH"},
 122         {E1000_RDFH, "RDFH"},
 123         {E1000_RDFT, "RDFT"},
 124         {E1000_RDFHS, "RDFHS"},
 125         {E1000_RDFTS, "RDFTS"},
 126         {E1000_RDFPC, "RDFPC"},
 127
 128         /* Tx Registers */
 129         {E1000_TCTL, "TCTL"},
 130         {E1000_TDBAL(0), "TDBAL"},
 131         {E1000_TDBAH(0), "TDBAH"},
 132         {E1000_TDLEN(0), "TDLEN"},
 133         {E1000_TDH(0), "TDH"},
 134         {E1000_TDT(0), "TDT"},
 135         {E1000_TIDV, "TIDV"},
 136         {E1000_TXDCTL(0), "TXDCTL"},
 137         {E1000_TADV, "TADV"},
 138         {E1000_TARC(0), "TARC"},
 139         {E1000_TDFH, "TDFH"},
 140         {E1000_TDFT, "TDFT"},
 141         {E1000_TDFHS, "TDFHS"},
 142         {E1000_TDFTS, "TDFTS"},
 143         {E1000_TDFPC, "TDFPC"},
 144
 145         /* List Terminator */
 146         {0, NULL}
 147 };
 148
 149 /*
 150  * e1000_regdump - register printout routine
 151  */
 152 static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
 153 {
 154         int n = 0;
 155         char rname[16];
 156         u32 regs[8];
 157
 158         switch (reginfo->ofs) {
 159         case E1000_RXDCTL(0):
 160                 for (n = 0; n < 2; n++)
 161                         regs[n] = __er32(hw, E1000_RXDCTL(n));
 162                 break;
 163         case E1000_TXDCTL(0):
 164                 for (n = 0; n < 2; n++)
 165                         regs[n] = __er32(hw, E1000_TXDCTL(n));
 166                 break;
 167         case E1000_TARC(0):
 168                 for (n = 0; n < 2; n++)
 169                         regs[n] = __er32(hw, E1000_TARC(n));
 170                 break;
 171         default:
 172                 pr_info("%-15s %08x\n",
 173                         reginfo->name, __er32(hw, reginfo->ofs));
 174                 return;
 175         }
 176
 177         snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
 178         pr_info("%-15s %08x %08x\n", rname, regs[0], regs[1]);
 179 }
 180
 181 static void e1000e_dump_ps_pages(struct e1000_adapter *adapter,
 182                                  struct e1000_buffer *bi)
 183 {
 184         int i;
 185         struct e1000_ps_page *ps_page;
 186
 187         for (i = 0; i < adapter->rx_ps_pages; i++) {
 188                 ps_page = &bi->ps_pages[i];
 189
 190                 if (ps_page->page) {
 191                         pr_info("packet dump for ps_page %d:\n", i);
 192                         print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
 193                                        16, 1, page_address(ps_page->page),
 194                                        PAGE_SIZE, true);
 195                 }
 196         }
 197 }
 198
 199 /*
 200  * e1000e_dump - Print registers, Tx-ring and Rx-ring
 201  */
 202 static void e1000e_dump(struct e1000_adapter *adapter)
 203 {
 204         struct net_device *netdev = adapter->netdev;
 205         struct e1000_hw *hw = &adapter->hw;
 206         struct e1000_reg_info *reginfo;
 207         struct e1000_ring *tx_ring = adapter->tx_ring;
 208         struct e1000_tx_desc *tx_desc;
 209         struct my_u0 {
 210                 __le64 a;
 211                 __le64 b;
 212         } *u0;
 213         struct e1000_buffer *buffer_info;
 214         struct e1000_ring *rx_ring = adapter->rx_ring;
 215         union e1000_rx_desc_packet_split *rx_desc_ps;
 216         union e1000_rx_desc_extended *rx_desc;
 217         struct my_u1 {
 218                 __le64 a;
 219                 __le64 b;
 220                 __le64 c;
 221                 __le64 d;
 222         } *u1;
 223         u32 staterr;
 224         int i = 0;
 225
 226         if (!netif_msg_hw(adapter))
 227                 return;
 228
 229         /* Print netdevice Info */
 230         if (netdev) {
 231                 dev_info(&adapter->pdev->dev, "Net device Info\n");
 232                 pr_info("Device Name     state            trans_start      last_rx\n");
 233                 pr_info("%-15s %016lX %016lX %016lX\n",
 234                         netdev->name, netdev->state, netdev->trans_start,
 235                         netdev->last_rx);
 236         }
 237
 238         /* Print Registers */
 239         dev_info(&adapter->pdev->dev, "Register Dump\n");
 240         pr_info(" Register Name   Value\n");
 241         for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
 242              reginfo->name; reginfo++) {
 243                 e1000_regdump(hw, reginfo);
 244         }
 245
 246         /* Print Tx Ring Summary */
 247         if (!netdev || !netif_running(netdev))
 248                 return;
 249
 250         dev_info(&adapter->pdev->dev, "Tx Ring Summary\n");
 251         pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
 252         buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
 253         pr_info(" %5d %5X %5X %016llX %04X %3X %016llX\n",
 254                 0, tx_ring->next_to_use, tx_ring->next_to_clean,
 255                 (unsigned long long)buffer_info->dma,
 256                 buffer_info->length,
 257                 buffer_info->next_to_watch,
 258                 (unsigned long long)buffer_info->time_stamp);
 259
 260         /* Print Tx Ring */
 261         if (!netif_msg_tx_done(adapter))
 262                 goto rx_ring_summary;
 263
 264         dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
 265
 266         /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
 267          *
 268          * Legacy Transmit Descriptor
 269          *   +--------------------------------------------------------------+
 270          * 0 |         Buffer Address [63:0] (Reserved on Write Back)       |
 271          *   +--------------------------------------------------------------+
 272          * 8 | Special  |    CSS     | Status |  CMD    |  CSO   |  Length  |
 273          *   +--------------------------------------------------------------+
 274          *   63       48 47        36 35    32 31     24 23    16 15        0
 275          *
 276          * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
 277          *   63      48 47    40 39       32 31             16 15    8 7      0
 278          *   +----------------------------------------------------------------+
 279          * 0 |  TUCSE  | TUCS0  |   TUCSS   |     IPCSE       | IPCS0 | IPCSS |
 280          *   +----------------------------------------------------------------+
 281          * 8 |   MSS   | HDRLEN | RSV | STA | TUCMD | DTYP |      PAYLEN      |
 282          *   +----------------------------------------------------------------+
 283          *   63      48 47    40 39 36 35 32 31   24 23  20 19                0
 284          *
 285          * Extended Data Descriptor (DTYP=0x1)
 286          *   +----------------------------------------------------------------+
 287          * 0 |                     Buffer Address [63:0]                      |
 288          *   +----------------------------------------------------------------+
 289          * 8 | VLAN tag |  POPTS  | Rsvd | Status | Command | DTYP |  DTALEN  |
 290          *   +----------------------------------------------------------------+
 291          *   63       48 47     40 39  36 35    32 31     24 23  20 19        0
 292          */
 293         pr_info("Tl[desc]     [address 63:0  ] [SpeCssSCmCsLen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Legacy format\n");
 294         pr_info("Tc[desc]     [Ce CoCsIpceCoS] [MssHlRSCm0Plen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Ext Context format\n");
 295         pr_info("Td[desc]     [address 63:0  ] [VlaPoRSCm1Dlen] [bi->dma       ] leng  ntw timestamp        bi->skb <-- Ext Data format\n");
 296         for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
 297                 const char *next_desc;
 298                 tx_desc = E1000_TX_DESC(*tx_ring, i);
 299                 buffer_info = &tx_ring->buffer_info[i];
 300                 u0 = (struct my_u0 *)tx_desc;
 301                 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
 302                         next_desc = " NTC/U";
 303                 else if (i == tx_ring->next_to_use)
 304                         next_desc = " NTU";
 305                 else if (i == tx_ring->next_to_clean)
 306                         next_desc = " NTC";
 307                 else
 308                         next_desc = "";
 309                 pr_info("T%c[0x%03X]    %016llX %016llX %016llX %04X  %3X %016llX %p%s\n",
 310                         (!(le64_to_cpu(u0->b) & (1 << 29)) ? 'l' :
 311                          ((le64_to_cpu(u0->b) & (1 << 20)) ? 'd' : 'c')),
 312                         i,
 313                         (unsigned long long)le64_to_cpu(u0->a),
 314                         (unsigned long long)le64_to_cpu(u0->b),
 315                         (unsigned long long)buffer_info->dma,
 316                         buffer_info->length, buffer_info->next_to_watch,
 317                         (unsigned long long)buffer_info->time_stamp,
 318                         buffer_info->skb, next_desc);
 319
 320                 if (netif_msg_pktdata(adapter) && buffer_info->skb)
 321                         print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
 322                                        16, 1, buffer_info->skb->data,
 323                                        buffer_info->skb->len, true);
 324         }
 325
 326         /* Print Rx Ring Summary */
 327 rx_ring_summary:
 328         dev_info(&adapter->pdev->dev, "Rx Ring Summary\n");
 329         pr_info("Queue [NTU] [NTC]\n");
 330         pr_info(" %5d %5X %5X\n",
 331                 0, rx_ring->next_to_use, rx_ring->next_to_clean);
 332
 333         /* Print Rx Ring */
 334         if (!netif_msg_rx_status(adapter))
 335                 return;
 336
 337         dev_info(&adapter->pdev->dev, "Rx Ring Dump\n");
 338         switch (adapter->rx_ps_pages) {
 339         case 1:
 340         case 2:
 341         case 3:
 342                 /* [Extended] Packet Split Receive Descriptor Format
 343                  *
 344                  *    +-----------------------------------------------------+
 345                  *  0 |                Buffer Address 0 [63:0]              |
 346                  *    +-----------------------------------------------------+
 347                  *  8 |                Buffer Address 1 [63:0]              |
 348                  *    +-----------------------------------------------------+
 349                  * 16 |                Buffer Address 2 [63:0]              |
 350                  *    +-----------------------------------------------------+
 351                  * 24 |                Buffer Address 3 [63:0]              |
 352                  *    +-----------------------------------------------------+
 353                  */
 354                 pr_info("R  [desc]      [buffer 0 63:0 ] [buffer 1 63:0 ] [buffer 2 63:0 ] [buffer 3 63:0 ] [bi->dma       ] [bi->skb] <-- Ext Pkt Split format\n");
 355                 /* [Extended] Receive Descriptor (Write-Back) Format
 356                  *
 357                  *   63       48 47    32 31     13 12    8 7    4 3        0
 358                  *   +------------------------------------------------------+
 359                  * 0 | Packet   | IP     |  Rsvd   | MRQ   | Rsvd | MRQ RSS |
 360                  *   | Checksum | Ident  |         | Queue |      |  Type   |
 361                  *   +------------------------------------------------------+
 362                  * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 363                  *   +------------------------------------------------------+
 364                  *   63       48 47    32 31            20 19               0
 365                  */
 366                 pr_info("RWB[desc]      [ck ipid mrqhsh] [vl   l0 ee  es] [ l3  l2  l1 hs] [reserved      ] ---------------- [bi->skb] <-- Ext Rx Write-Back format\n");
 367                 for (i = 0; i < rx_ring->count; i++) {
 368                         const char *next_desc;
 369                         buffer_info = &rx_ring->buffer_info[i];
 370                         rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
 371                         u1 = (struct my_u1 *)rx_desc_ps;
 372                         staterr =
 373                             le32_to_cpu(rx_desc_ps->wb.middle.status_error);
 374
 375                         if (i == rx_ring->next_to_use)
 376                                 next_desc = " NTU";
 377                         else if (i == rx_ring->next_to_clean)
 378                                 next_desc = " NTC";
 379                         else
 380                                 next_desc = "";
 381
 382                         if (staterr & E1000_RXD_STAT_DD) {
 383                                 /* Descriptor Done */
 384                                 pr_info("%s[0x%03X]     %016llX %016llX %016llX %016llX ---------------- %p%s\n",
 385                                         "RWB", i,
 386                                         (unsigned long long)le64_to_cpu(u1->a),
 387                                         (unsigned long long)le64_to_cpu(u1->b),
 388                                         (unsigned long long)le64_to_cpu(u1->c),
 389                                         (unsigned long long)le64_to_cpu(u1->d),
 390                                         buffer_info->skb, next_desc);
 391                         } else {
 392                                 pr_info("%s[0x%03X]     %016llX %016llX %016llX %016llX %016llX %p%s\n",
 393                                         "R  ", i,
 394                                         (unsigned long long)le64_to_cpu(u1->a),
 395                                         (unsigned long long)le64_to_cpu(u1->b),
 396                                         (unsigned long long)le64_to_cpu(u1->c),
 397                                         (unsigned long long)le64_to_cpu(u1->d),
 398                                         (unsigned long long)buffer_info->dma,
 399                                         buffer_info->skb, next_desc);
 400
 401                                 if (netif_msg_pktdata(adapter))
 402                                         e1000e_dump_ps_pages(adapter,
 403                                                              buffer_info);
 404                         }
 405                 }
 406                 break;
 407         default:
 408         case 0:
 409                 /* Extended Receive Descriptor (Read) Format
 410                  *
 411                  *   +-----------------------------------------------------+
 412                  * 0 |                Buffer Address [63:0]                |
 413                  *   +-----------------------------------------------------+
 414                  * 8 |                      Reserved                       |
 415                  *   +-----------------------------------------------------+
 416                  */
 417                 pr_info("R  [desc]      [buf addr 63:0 ] [reserved 63:0 ] [bi->dma       ] [bi->skb] <-- Ext (Read) format\n");
 418                 /* Extended Receive Descriptor (Write-Back) Format
 419                  *
 420                  *   63       48 47    32 31    24 23            4 3        0
 421                  *   +------------------------------------------------------+
 422                  *   |     RSS Hash      |        |               |         |
 423                  * 0 +-------------------+  Rsvd  |   Reserved    | MRQ RSS |
 424                  *   | Packet   | IP     |        |               |  Type   |
 425                  *   | Checksum | Ident  |        |               |         |
 426                  *   +------------------------------------------------------+
 427                  * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 428                  *   +------------------------------------------------------+
 429                  *   63       48 47    32 31            20 19               0
 430                  */
 431                 pr_info("RWB[desc]      [cs ipid    mrq] [vt   ln xe  xs] [bi->skb] <-- Ext (Write-Back) format\n");
 432
 433                 for (i = 0; i < rx_ring->count; i++) {
 434                         const char *next_desc;
 435
 436                         buffer_info = &rx_ring->buffer_info[i];
 437                         rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 438                         u1 = (struct my_u1 *)rx_desc;
 439                         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 440
 441                         if (i == rx_ring->next_to_use)
 442                                 next_desc = " NTU";
 443                         else if (i == rx_ring->next_to_clean)
 444                                 next_desc = " NTC";
 445                         else
 446                                 next_desc = "";
 447
 448                         if (staterr & E1000_RXD_STAT_DD) {
 449                                 /* Descriptor Done */
 450                                 pr_info("%s[0x%03X]     %016llX %016llX ---------------- %p%s\n",
 451                                         "RWB", i,
 452                                         (unsigned long long)le64_to_cpu(u1->a),
 453                                         (unsigned long long)le64_to_cpu(u1->b),
 454                                         buffer_info->skb, next_desc);
 455                         } else {
 456                                 pr_info("%s[0x%03X]     %016llX %016llX %016llX %p%s\n",
 457                                         "R  ", i,
 458                                         (unsigned long long)le64_to_cpu(u1->a),
 459                                         (unsigned long long)le64_to_cpu(u1->b),
 460                                         (unsigned long long)buffer_info->dma,
 461                                         buffer_info->skb, next_desc);
 462
 463                                 if (netif_msg_pktdata(adapter) &&
 464                                     buffer_info->skb)
 465                                         print_hex_dump(KERN_INFO, "",
 466                                                        DUMP_PREFIX_ADDRESS, 16,
 467                                                        1,
 468                                                        buffer_info->skb->data,
 469                                                        adapter->rx_buffer_len,
 470                                                        true);
 471                         }
 472                 }
 473         }
 474 }
 475
 476 /**
 477  * e1000_desc_unused - calculate if we have unused descriptors
 478  **/
 479 static int e1000_desc_unused(struct e1000_ring *ring)
 480 {
 481         if (ring->next_to_clean > ring->next_to_use)
 482                 return ring->next_to_clean - ring->next_to_use - 1;
 483
 484         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
 485 }
 486
 487 /**
 488  * e1000_receive_skb - helper function to handle Rx indications
 489  * @adapter: board private structure
 490  * @status: descriptor status field as written by hardware
 491  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
 492  * @skb: pointer to sk_buff to be indicated to stack
 493  **/
 494 static void e1000_receive_skb(struct e1000_adapter *adapter,
 495                               struct net_device *netdev, struct sk_buff *skb,
 496                               u8 status, __le16 vlan)
 497 {
 498         u16 tag = le16_to_cpu(vlan);
 499         skb->protocol = eth_type_trans(skb, netdev);
 500
 501         if (status & E1000_RXD_STAT_VP)
 502                 __vlan_hwaccel_put_tag(skb, tag);
 503
 504         napi_gro_receive(&adapter->napi, skb);
 505 }
 506
 507 /**
 508  * e1000_rx_checksum - Receive Checksum Offload
 509  * @adapter: board private structure
 510  * @status_err: receive descriptor status and error fields
 511  * @csum: receive descriptor csum field
 512  * @sk_buff: socket buffer with received data
 513  **/
 514 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
 515                               struct sk_buff *skb)
 516 {
 517         u16 status = (u16)status_err;
 518         u8 errors = (u8)(status_err >> 24);
 519
 520         skb_checksum_none_assert(skb);
 521
 522         /* Rx checksum disabled */
 523         if (!(adapter->netdev->features & NETIF_F_RXCSUM))
 524                 return;
 525
 526         /* Ignore Checksum bit is set */
 527         if (status & E1000_RXD_STAT_IXSM)
 528                 return;
 529
 530         /* TCP/UDP checksum error bit or IP checksum error bit is set */
 531         if (errors & (E1000_RXD_ERR_TCPE | E1000_RXD_ERR_IPE)) {
 532                 /* let the stack verify checksum errors */
 533                 adapter->hw_csum_err++;
 534                 return;
 535         }
 536
 537         /* TCP/UDP Checksum has not been calculated */
 538         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
 539                 return;
 540
 541         /* It must be a TCP or UDP packet with a valid checksum */
 542         skb->ip_summed = CHECKSUM_UNNECESSARY;
 543         adapter->hw_csum_good++;
 544 }
 545
 546 static void e1000e_update_rdt_wa(struct e1000_ring *rx_ring, unsigned int i)
 547 {
 548         struct e1000_adapter *adapter = rx_ring->adapter;
 549         struct e1000_hw *hw = &adapter->hw;
 550         s32 ret_val = __ew32_prepare(hw);
 551
 552         writel(i, rx_ring->tail);
 553
 554         if (unlikely(!ret_val && (i != readl(rx_ring->tail)))) {
 555                 u32 rctl = er32(RCTL);
 556                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
 557                 e_err("ME firmware caused invalid RDT - resetting\n");
 558                 schedule_work(&adapter->reset_task);
 559         }
 560 }
 561
 562 static void e1000e_update_tdt_wa(struct e1000_ring *tx_ring, unsigned int i)
 563 {
 564         struct e1000_adapter *adapter = tx_ring->adapter;
 565         struct e1000_hw *hw = &adapter->hw;
 566         s32 ret_val = __ew32_prepare(hw);
 567
 568         writel(i, tx_ring->tail);
 569
 570         if (unlikely(!ret_val && (i != readl(tx_ring->tail)))) {
 571                 u32 tctl = er32(TCTL);
 572                 ew32(TCTL, tctl & ~E1000_TCTL_EN);
 573                 e_err("ME firmware caused invalid TDT - resetting\n");
 574                 schedule_work(&adapter->reset_task);
 575         }
 576 }
 577
 578 /**
 579  * e1000_alloc_rx_buffers - Replace used receive buffers
 580  * @rx_ring: Rx descriptor ring
 581  **/
 582 static void e1000_alloc_rx_buffers(struct e1000_ring *rx_ring,
 583                                    int cleaned_count, gfp_t gfp)
 584 {
 585         struct e1000_adapter *adapter = rx_ring->adapter;
 586         struct net_device *netdev = adapter->netdev;
 587         struct pci_dev *pdev = adapter->pdev;
 588         union e1000_rx_desc_extended *rx_desc;
 589         struct e1000_buffer *buffer_info;
 590         struct sk_buff *skb;
 591         unsigned int i;
 592         unsigned int bufsz = adapter->rx_buffer_len;
 593
 594         i = rx_ring->next_to_use;
 595         buffer_info = &rx_ring->buffer_info[i];
 596
 597         while (cleaned_count--) {
 598                 skb = buffer_info->skb;
 599                 if (skb) {
 600                         skb_trim(skb, 0);
 601                         goto map_skb;
 602                 }
 603
 604                 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
 605                 if (!skb) {
 606                         /* Better luck next round */
 607                         adapter->alloc_rx_buff_failed++;
 608                         break;
 609                 }
 610
 611                 buffer_info->skb = skb;
 612 map_skb:
 613                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 614                                                   adapter->rx_buffer_len,
 615                                                   DMA_FROM_DEVICE);
 616                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 617                         dev_err(&pdev->dev, "Rx DMA map failed\n");
 618                         adapter->rx_dma_failed++;
 619                         break;
 620                 }
 621
 622                 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 623                 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
 624
 625                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 626                         /*
 627                          * Force memory writes to complete before letting h/w
 628                          * know there are new descriptors to fetch.  (Only
 629                          * applicable for weak-ordered memory model archs,
 630                          * such as IA-64).
 631                          */
 632                         wmb();
 633                         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
 634                                 e1000e_update_rdt_wa(rx_ring, i);
 635                         else
 636                                 writel(i, rx_ring->tail);
 637                 }
 638                 i++;
 639                 if (i == rx_ring->count)
 640                         i = 0;
 641                 buffer_info = &rx_ring->buffer_info[i];
 642         }
 643
 644         rx_ring->next_to_use = i;
 645 }
 646
 647 /**
 648  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
 649  * @rx_ring: Rx descriptor ring
 650  **/
 651 static void e1000_alloc_rx_buffers_ps(struct e1000_ring *rx_ring,
 652                                       int cleaned_count, gfp_t gfp)
 653 {
 654         struct e1000_adapter *adapter = rx_ring->adapter;
 655         struct net_device *netdev = adapter->netdev;
 656         struct pci_dev *pdev = adapter->pdev;
 657         union e1000_rx_desc_packet_split *rx_desc;
 658         struct e1000_buffer *buffer_info;
 659         struct e1000_ps_page *ps_page;
 660         struct sk_buff *skb;
 661         unsigned int i, j;
 662
 663         i = rx_ring->next_to_use;
 664         buffer_info = &rx_ring->buffer_info[i];
 665
 666         while (cleaned_count--) {
 667                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 668
 669                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 670                         ps_page = &buffer_info->ps_pages[j];
 671                         if (j >= adapter->rx_ps_pages) {
 672                                 /* all unused desc entries get hw null ptr */
 673                                 rx_desc->read.buffer_addr[j + 1] =
 674                                     ~cpu_to_le64(0);
 675                                 continue;
 676                         }
 677                         if (!ps_page->page) {
 678                                 ps_page->page = alloc_page(gfp);
 679                                 if (!ps_page->page) {
 680                                         adapter->alloc_rx_buff_failed++;
 681                                         goto no_buffers;
 682                                 }
 683                                 ps_page->dma = dma_map_page(&pdev->dev,
 684                                                             ps_page->page,
 685                                                             0, PAGE_SIZE,
 686                                                             DMA_FROM_DEVICE);
 687                                 if (dma_mapping_error(&pdev->dev,
 688                                                       ps_page->dma)) {
 689                                         dev_err(&adapter->pdev->dev,
 690                                                 "Rx DMA page map failed\n");
 691                                         adapter->rx_dma_failed++;
 692                                         goto no_buffers;
 693                                 }
 694                         }
 695                         /*
 696                          * Refresh the desc even if buffer_addrs
 697                          * didn't change because each write-back
 698                          * erases this info.
 699                          */
 700                         rx_desc->read.buffer_addr[j + 1] =
 701                             cpu_to_le64(ps_page->dma);
 702                 }
 703
 704                 skb = __netdev_alloc_skb_ip_align(netdev,
 705                                                   adapter->rx_ps_bsize0,
 706                                                   gfp);
 707
 708                 if (!skb) {
 709                         adapter->alloc_rx_buff_failed++;
 710                         break;
 711                 }
 712
 713                 buffer_info->skb = skb;
 714                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 715                                                   adapter->rx_ps_bsize0,
 716                                                   DMA_FROM_DEVICE);
 717                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 718                         dev_err(&pdev->dev, "Rx DMA map failed\n");
 719                         adapter->rx_dma_failed++;
 720                         /* cleanup skb */
 721                         dev_kfree_skb_any(skb);
 722                         buffer_info->skb = NULL;
 723                         break;
 724                 }
 725
 726                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 727
 728                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 729                         /*
 730                          * Force memory writes to complete before letting h/w
 731                          * know there are new descriptors to fetch.  (Only
 732                          * applicable for weak-ordered memory model archs,
 733                          * such as IA-64).
 734                          */
 735                         wmb();
 736                         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
 737                                 e1000e_update_rdt_wa(rx_ring, i << 1);
 738                         else
 739                                 writel(i << 1, rx_ring->tail);
 740                 }
 741
 742                 i++;
 743                 if (i == rx_ring->count)
 744                         i = 0;
 745                 buffer_info = &rx_ring->buffer_info[i];
 746         }
 747
 748 no_buffers:
 749         rx_ring->next_to_use = i;
 750 }
 751
 752 /**
 753  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
 754  * @rx_ring: Rx descriptor ring
 755  * @cleaned_count: number of buffers to allocate this pass
 756  **/
 757
 758 static void e1000_alloc_jumbo_rx_buffers(struct e1000_ring *rx_ring,
 759                                          int cleaned_count, gfp_t gfp)
 760 {
 761         struct e1000_adapter *adapter = rx_ring->adapter;
 762         struct net_device *netdev = adapter->netdev;
 763         struct pci_dev *pdev = adapter->pdev;
 764         union e1000_rx_desc_extended *rx_desc;
 765         struct e1000_buffer *buffer_info;
 766         struct sk_buff *skb;
 767         unsigned int i;
 768         unsigned int bufsz = 256 - 16 /* for skb_reserve */;
 769
 770         i = rx_ring->next_to_use;
 771         buffer_info = &rx_ring->buffer_info[i];
 772
 773         while (cleaned_count--) {
 774                 skb = buffer_info->skb;
 775                 if (skb) {
 776                         skb_trim(skb, 0);
 777                         goto check_page;
 778                 }
 779
 780                 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
 781                 if (unlikely(!skb)) {
 782                         /* Better luck next round */
 783                         adapter->alloc_rx_buff_failed++;
 784                         break;
 785                 }
 786
 787                 buffer_info->skb = skb;
 788 check_page:
 789                 /* allocate a new page if necessary */
 790                 if (!buffer_info->page) {
 791                         buffer_info->page = alloc_page(gfp);
 792                         if (unlikely(!buffer_info->page)) {
 793                                 adapter->alloc_rx_buff_failed++;
 794                                 break;
 795                         }
 796                 }
 797
 798                 if (!buffer_info->dma)
 799                         buffer_info->dma = dma_map_page(&pdev->dev,
 800                                                         buffer_info->page, 0,
 801                                                         PAGE_SIZE,
 802                                                         DMA_FROM_DEVICE);
 803
 804                 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 805                 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
 806
 807                 if (unlikely(++i == rx_ring->count))
 808                         i = 0;
 809                 buffer_info = &rx_ring->buffer_info[i];
 810         }
 811
 812         if (likely(rx_ring->next_to_use != i)) {
 813                 rx_ring->next_to_use = i;
 814                 if (unlikely(i-- == 0))
 815                         i = (rx_ring->count - 1);
 816
 817                 /* Force memory writes to complete before letting h/w
 818                  * know there are new descriptors to fetch.  (Only
 819                  * applicable for weak-ordered memory model archs,
 820                  * such as IA-64). */
 821                 wmb();
 822                 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
 823                         e1000e_update_rdt_wa(rx_ring, i);
 824                 else
 825                         writel(i, rx_ring->tail);
 826         }
 827 }
 828
 829 static inline void e1000_rx_hash(struct net_device *netdev, __le32 rss,
 830                                  struct sk_buff *skb)
 831 {
 832         if (netdev->features & NETIF_F_RXHASH)
 833                 skb->rxhash = le32_to_cpu(rss);
 834 }
 835
 836 /**
 837  * e1000_clean_rx_irq - Send received data up the network stack
 838  * @rx_ring: Rx descriptor ring
 839  *
 840  * the return value indicates whether actual cleaning was done, there
 841  * is no guarantee that everything was cleaned
 842  **/
 843 static bool e1000_clean_rx_irq(struct e1000_ring *rx_ring, int *work_done,
 844                                int work_to_do)
 845 {
 846         struct e1000_adapter *adapter = rx_ring->adapter;
 847         struct net_device *netdev = adapter->netdev;
 848         struct pci_dev *pdev = adapter->pdev;
 849         struct e1000_hw *hw = &adapter->hw;
 850         union e1000_rx_desc_extended *rx_desc, *next_rxd;
 851         struct e1000_buffer *buffer_info, *next_buffer;
 852         u32 length, staterr;
 853         unsigned int i;
 854         int cleaned_count = 0;
 855         bool cleaned = false;
 856         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 857
 858         i = rx_ring->next_to_clean;
 859         rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 860         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 861         buffer_info = &rx_ring->buffer_info[i];
 862
 863         while (staterr & E1000_RXD_STAT_DD) {
 864                 struct sk_buff *skb;
 865
 866                 if (*work_done >= work_to_do)
 867                         break;
 868                 (*work_done)++;
 869                 rmb();  /* read descriptor and rx_buffer_info after status DD */
 870
 871                 skb = buffer_info->skb;
 872                 buffer_info->skb = NULL;
 873
 874                 prefetch(skb->data - NET_IP_ALIGN);
 875
 876                 i++;
 877                 if (i == rx_ring->count)
 878                         i = 0;
 879                 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
 880                 prefetch(next_rxd);
 881
 882                 next_buffer = &rx_ring->buffer_info[i];
 883
 884                 cleaned = true;
 885                 cleaned_count++;
 886                 dma_unmap_single(&pdev->dev,
 887                                  buffer_info->dma,
 888                                  adapter->rx_buffer_len,
 889                                  DMA_FROM_DEVICE);
 890                 buffer_info->dma = 0;
 891
 892                 length = le16_to_cpu(rx_desc->wb.upper.length);
 893
 894                 /*
 895                  * !EOP means multiple descriptors were used to store a single
 896                  * packet, if that's the case we need to toss it.  In fact, we
 897                  * need to toss every packet with the EOP bit clear and the
 898                  * next frame that _does_ have the EOP bit set, as it is by
 899                  * definition only a frame fragment
 900                  */
 901                 if (unlikely(!(staterr & E1000_RXD_STAT_EOP)))
 902                         adapter->flags2 |= FLAG2_IS_DISCARDING;
 903
 904                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
 905                         /* All receives must fit into a single buffer */
 906                         e_dbg("Receive packet consumed multiple buffers\n");
 907                         /* recycle */
 908                         buffer_info->skb = skb;
 909                         if (staterr & E1000_RXD_STAT_EOP)
 910                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
 911                         goto next_desc;
 912                 }
 913
 914                 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
 915                              !(netdev->features & NETIF_F_RXALL))) {
 916                         /* recycle */
 917                         buffer_info->skb = skb;
 918                         goto next_desc;
 919                 }
 920
 921                 /* adjust length to remove Ethernet CRC */
 922                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
 923                         /* If configured to store CRC, don't subtract FCS,
 924                          * but keep the FCS bytes out of the total_rx_bytes
 925                          * counter
 926                          */
 927                         if (netdev->features & NETIF_F_RXFCS)
 928                                 total_rx_bytes -= 4;
 929                         else
 930                                 length -= 4;
 931                 }
 932
 933                 total_rx_bytes += length;
 934                 total_rx_packets++;
 935
 936                 /*
 937                  * code added for copybreak, this should improve
 938                  * performance for small packets with large amounts
 939                  * of reassembly being done in the stack
 940                  */
 941                 if (length < copybreak) {
 942                         struct sk_buff *new_skb =
 943                             netdev_alloc_skb_ip_align(netdev, length);
 944                         if (new_skb) {
 945                                 skb_copy_to_linear_data_offset(new_skb,
 946                                                                -NET_IP_ALIGN,
 947                                                                (skb->data -
 948                                                                 NET_IP_ALIGN),
 949                                                                (length +
 950                                                                 NET_IP_ALIGN));
 951                                 /* save the skb in buffer_info as good */
 952                                 buffer_info->skb = skb;
 953                                 skb = new_skb;
 954                         }
 955                         /* else just continue with the old one */
 956                 }
 957                 /* end copybreak code */
 958                 skb_put(skb, length);
 959
 960                 /* Receive Checksum Offload */
 961                 e1000_rx_checksum(adapter, staterr, skb);
 962
 963                 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
 964
 965                 e1000_receive_skb(adapter, netdev, skb, staterr,
 966                                   rx_desc->wb.upper.vlan);
 967
 968 next_desc:
 969                 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
 970
 971                 /* return some buffers to hardware, one at a time is too slow */
 972                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 973                         adapter->alloc_rx_buf(rx_ring, cleaned_count,
 974                                               GFP_ATOMIC);
 975                         cleaned_count = 0;
 976                 }
 977
 978                 /* use prefetched values */
 979                 rx_desc = next_rxd;
 980                 buffer_info = next_buffer;
 981
 982                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 983         }
 984         rx_ring->next_to_clean = i;
 985
 986         cleaned_count = e1000_desc_unused(rx_ring);
 987         if (cleaned_count)
 988                 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
 989
 990         adapter->total_rx_bytes += total_rx_bytes;
 991         adapter->total_rx_packets += total_rx_packets;
 992         return cleaned;
 993 }
 994
 995 static void e1000_put_txbuf(struct e1000_ring *tx_ring,
 996                             struct e1000_buffer *buffer_info)
 997 {
 998         struct e1000_adapter *adapter = tx_ring->adapter;
 999
1000         if (buffer_info->dma) {
1001                 if (buffer_info->mapped_as_page)
1002                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1003                                        buffer_info->length, DMA_TO_DEVICE);
1004                 else
1005                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1006                                          buffer_info->length, DMA_TO_DEVICE);
1007                 buffer_info->dma = 0;
1008         }
1009         if (buffer_info->skb) {
1010                 dev_kfree_skb_any(buffer_info->skb);
1011                 buffer_info->skb = NULL;
1012         }
1013         buffer_info->time_stamp = 0;
1014 }
1015
1016 static void e1000_print_hw_hang(struct work_struct *work)
1017 {
1018         struct e1000_adapter *adapter = container_of(work,
1019                                                      struct e1000_adapter,
1020                                                      print_hang_task);
1021         struct net_device *netdev = adapter->netdev;
1022         struct e1000_ring *tx_ring = adapter->tx_ring;
1023         unsigned int i = tx_ring->next_to_clean;
1024         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
1025         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
1026         struct e1000_hw *hw = &adapter->hw;
1027         u16 phy_status, phy_1000t_status, phy_ext_status;
1028         u16 pci_status;
1029
1030         if (test_bit(__E1000_DOWN, &adapter->state))
1031                 return;
1032
1033         if (!adapter->tx_hang_recheck &&
1034             (adapter->flags2 & FLAG2_DMA_BURST)) {
1035                 /*
1036                  * May be block on write-back, flush and detect again
1037                  * flush pending descriptor writebacks to memory
1038                  */
1039                 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1040                 /* execute the writes immediately */
1041                 e1e_flush();
1042                 /*
1043                  * Due to rare timing issues, write to TIDV again to ensure
1044                  * the write is successful
1045                  */
1046                 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1047                 /* execute the writes immediately */
1048                 e1e_flush();
1049                 adapter->tx_hang_recheck = true;
1050                 return;
1051         }
1052         /* Real hang detected */
1053         adapter->tx_hang_recheck = false;
1054         netif_stop_queue(netdev);
1055
1056         e1e_rphy(hw, PHY_STATUS, &phy_status);
1057         e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
1058         e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
1059
1060         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
1061
1062         /* detected Hardware unit hang */
1063         e_err("Detected Hardware Unit Hang:\n"
1064               "  TDH                  <%x>\n"
1065               "  TDT                  <%x>\n"
1066               "  next_to_use          <%x>\n"
1067               "  next_to_clean        <%x>\n"
1068               "buffer_info[next_to_clean]:\n"
1069               "  time_stamp           <%lx>\n"
1070               "  next_to_watch        <%x>\n"
1071               "  jiffies              <%lx>\n"
1072               "  next_to_watch.status <%x>\n"
1073               "MAC Status             <%x>\n"
1074               "PHY Status             <%x>\n"
1075               "PHY 1000BASE-T Status  <%x>\n"
1076               "PHY Extended Status    <%x>\n"
1077               "PCI Status             <%x>\n",
1078               readl(tx_ring->head),
1079               readl(tx_ring->tail),
1080               tx_ring->next_to_use,
1081               tx_ring->next_to_clean,
1082               tx_ring->buffer_info[eop].time_stamp,
1083               eop,
1084               jiffies,
1085               eop_desc->upper.fields.status,
1086               er32(STATUS),
1087               phy_status,
1088               phy_1000t_status,
1089               phy_ext_status,
1090               pci_status);
1091
1092         /* Suggest workaround for known h/w issue */
1093         if ((hw->mac.type == e1000_pchlan) && (er32(CTRL) & E1000_CTRL_TFCE))
1094                 e_err("Try turning off Tx pause (flow control) via ethtool\n");
1095 }
1096
1097 /**
1098  * e1000_clean_tx_irq - Reclaim resources after transmit completes
1099  * @tx_ring: Tx descriptor ring
1100  *
1101  * the return value indicates whether actual cleaning was done, there
1102  * is no guarantee that everything was cleaned
1103  **/
1104 static bool e1000_clean_tx_irq(struct e1000_ring *tx_ring)
1105 {
1106         struct e1000_adapter *adapter = tx_ring->adapter;
1107         struct net_device *netdev = adapter->netdev;
1108         struct e1000_hw *hw = &adapter->hw;
1109         struct e1000_tx_desc *tx_desc, *eop_desc;
1110         struct e1000_buffer *buffer_info;
1111         unsigned int i, eop;
1112         unsigned int count = 0;
1113         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
1114         unsigned int bytes_compl = 0, pkts_compl = 0;
1115
1116         i = tx_ring->next_to_clean;
1117         eop = tx_ring->buffer_info[i].next_to_watch;
1118         eop_desc = E1000_TX_DESC(*tx_ring, eop);
1119
1120         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1121                (count < tx_ring->count)) {
1122                 bool cleaned = false;
1123                 rmb(); /* read buffer_info after eop_desc */
1124                 for (; !cleaned; count++) {
1125                         tx_desc = E1000_TX_DESC(*tx_ring, i);
1126                         buffer_info = &tx_ring->buffer_info[i];
1127                         cleaned = (i == eop);
1128
1129                         if (cleaned) {
1130                                 total_tx_packets += buffer_info->segs;
1131                                 total_tx_bytes += buffer_info->bytecount;
1132                                 if (buffer_info->skb) {
1133                                         bytes_compl += buffer_info->skb->len;
1134                                         pkts_compl++;
1135                                 }
1136                         }
1137
1138                         e1000_put_txbuf(tx_ring, buffer_info);
1139                         tx_desc->upper.data = 0;
1140
1141                         i++;
1142                         if (i == tx_ring->count)
1143                                 i = 0;
1144                 }
1145
1146                 if (i == tx_ring->next_to_use)
1147                         break;
1148                 eop = tx_ring->buffer_info[i].next_to_watch;
1149                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1150         }
1151
1152         tx_ring->next_to_clean = i;
1153
1154         netdev_completed_queue(netdev, pkts_compl, bytes_compl);
1155
1156 #define TX_WAKE_THRESHOLD 32
1157         if (count && netif_carrier_ok(netdev) &&
1158             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1159                 /* Make sure that anybody stopping the queue after this
1160                  * sees the new next_to_clean.
1161                  */
1162                 smp_mb();
1163
1164                 if (netif_queue_stopped(netdev) &&
1165                     !(test_bit(__E1000_DOWN, &adapter->state))) {
1166                         netif_wake_queue(netdev);
1167                         ++adapter->restart_queue;
1168                 }
1169         }
1170
1171         if (adapter->detect_tx_hung) {
1172                 /*
1173                  * Detect a transmit hang in hardware, this serializes the
1174                  * check with the clearing of time_stamp and movement of i
1175                  */
1176                 adapter->detect_tx_hung = false;
1177                 if (tx_ring->buffer_info[i].time_stamp &&
1178                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1179                                + (adapter->tx_timeout_factor * HZ)) &&
1180                     !(er32(STATUS) & E1000_STATUS_TXOFF))
1181                         schedule_work(&adapter->print_hang_task);
1182                 else
1183                         adapter->tx_hang_recheck = false;
1184         }
1185         adapter->total_tx_bytes += total_tx_bytes;
1186         adapter->total_tx_packets += total_tx_packets;
1187         return count < tx_ring->count;
1188 }
1189
1190 /**
1191  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1192  * @rx_ring: Rx descriptor ring
1193  *
1194  * the return value indicates whether actual cleaning was done, there
1195  * is no guarantee that everything was cleaned
1196  **/
1197 static bool e1000_clean_rx_irq_ps(struct e1000_ring *rx_ring, int *work_done,
1198                                   int work_to_do)
1199 {
1200         struct e1000_adapter *adapter = rx_ring->adapter;
1201         struct e1000_hw *hw = &adapter->hw;
1202         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1203         struct net_device *netdev = adapter->netdev;
1204         struct pci_dev *pdev = adapter->pdev;
1205         struct e1000_buffer *buffer_info, *next_buffer;
1206         struct e1000_ps_page *ps_page;
1207         struct sk_buff *skb;
1208         unsigned int i, j;
1209         u32 length, staterr;
1210         int cleaned_count = 0;
1211         bool cleaned = false;
1212         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1213
1214         i = rx_ring->next_to_clean;
1215         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1216         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1217         buffer_info = &rx_ring->buffer_info[i];
1218
1219         while (staterr & E1000_RXD_STAT_DD) {
1220                 if (*work_done >= work_to_do)
1221                         break;
1222                 (*work_done)++;
1223                 skb = buffer_info->skb;
1224                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1225
1226                 /* in the packet split case this is header only */
1227                 prefetch(skb->data - NET_IP_ALIGN);
1228
1229                 i++;
1230                 if (i == rx_ring->count)
1231                         i = 0;
1232                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1233                 prefetch(next_rxd);
1234
1235                 next_buffer = &rx_ring->buffer_info[i];
1236
1237                 cleaned = true;
1238                 cleaned_count++;
1239                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1240                                  adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1241                 buffer_info->dma = 0;
1242
1243                 /* see !EOP comment in other Rx routine */
1244                 if (!(staterr & E1000_RXD_STAT_EOP))
1245                         adapter->flags2 |= FLAG2_IS_DISCARDING;
1246
1247                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1248                         e_dbg("Packet Split buffers didn't pick up the full packet\n");
1249                         dev_kfree_skb_irq(skb);
1250                         if (staterr & E1000_RXD_STAT_EOP)
1251                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1252                         goto next_desc;
1253                 }
1254
1255                 if (unlikely((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1256                              !(netdev->features & NETIF_F_RXALL))) {
1257                         dev_kfree_skb_irq(skb);
1258                         goto next_desc;
1259                 }
1260
1261                 length = le16_to_cpu(rx_desc->wb.middle.length0);
1262
1263                 if (!length) {
1264                         e_dbg("Last part of the packet spanning multiple descriptors\n");
1265                         dev_kfree_skb_irq(skb);
1266                         goto next_desc;
1267                 }
1268
1269                 /* Good Receive */
1270                 skb_put(skb, length);
1271
1272                 {
1273                         /*
1274                          * this looks ugly, but it seems compiler issues make
1275                          * it more efficient than reusing j
1276                          */
1277                         int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1278
1279                         /*
1280                          * page alloc/put takes too long and effects small
1281                          * packet throughput, so unsplit small packets and
1282                          * save the alloc/put only valid in softirq (napi)
1283                          * context to call kmap_*
1284                          */
1285                         if (l1 && (l1 <= copybreak) &&
1286                             ((length + l1) <= adapter->rx_ps_bsize0)) {
1287                                 u8 *vaddr;
1288
1289                                 ps_page = &buffer_info->ps_pages[0];
1290
1291                                 /*
1292                                  * there is no documentation about how to call
1293                                  * kmap_atomic, so we can't hold the mapping
1294                                  * very long
1295                                  */
1296                                 dma_sync_single_for_cpu(&pdev->dev,
1297                                                         ps_page->dma,
1298                                                         PAGE_SIZE,
1299                                                         DMA_FROM_DEVICE);
1300                                 vaddr = kmap_atomic(ps_page->page);
1301                                 memcpy(skb_tail_pointer(skb), vaddr, l1);
1302                                 kunmap_atomic(vaddr);
1303                                 dma_sync_single_for_device(&pdev->dev,
1304                                                            ps_page->dma,
1305                                                            PAGE_SIZE,
1306                                                            DMA_FROM_DEVICE);
1307
1308                                 /* remove the CRC */
1309                                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1310                                         if (!(netdev->features & NETIF_F_RXFCS))
1311                                                 l1 -= 4;
1312                                 }
1313
1314                                 skb_put(skb, l1);
1315                                 goto copydone;
1316                         } /* if */
1317                 }
1318
1319                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1320                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1321                         if (!length)
1322                                 break;
1323
1324                         ps_page = &buffer_info->ps_pages[j];
1325                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1326                                        DMA_FROM_DEVICE);
1327                         ps_page->dma = 0;
1328                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1329                         ps_page->page = NULL;
1330                         skb->len += length;
1331                         skb->data_len += length;
1332                         skb->truesize += PAGE_SIZE;
1333                 }
1334
1335                 /* strip the ethernet crc, problem is we're using pages now so
1336                  * this whole operation can get a little cpu intensive
1337                  */
1338                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING)) {
1339                         if (!(netdev->features & NETIF_F_RXFCS))
1340                                 pskb_trim(skb, skb->len - 4);
1341                 }
1342
1343 copydone:
1344                 total_rx_bytes += skb->len;
1345                 total_rx_packets++;
1346
1347                 e1000_rx_checksum(adapter, staterr, skb);
1348
1349                 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1350
1351                 if (rx_desc->wb.upper.header_status &
1352                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1353                         adapter->rx_hdr_split++;
1354
1355                 e1000_receive_skb(adapter, netdev, skb,
1356                                   staterr, rx_desc->wb.middle.vlan);
1357
1358 next_desc:
1359                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1360                 buffer_info->skb = NULL;
1361
1362                 /* return some buffers to hardware, one at a time is too slow */
1363                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1364                         adapter->alloc_rx_buf(rx_ring, cleaned_count,
1365                                               GFP_ATOMIC);
1366                         cleaned_count = 0;
1367                 }
1368
1369                 /* use prefetched values */
1370                 rx_desc = next_rxd;
1371                 buffer_info = next_buffer;
1372
1373                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1374         }
1375         rx_ring->next_to_clean = i;
1376
1377         cleaned_count = e1000_desc_unused(rx_ring);
1378         if (cleaned_count)
1379                 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1380
1381         adapter->total_rx_bytes += total_rx_bytes;
1382         adapter->total_rx_packets += total_rx_packets;
1383         return cleaned;
1384 }
1385
1386 /**
1387  * e1000_consume_page - helper function
1388  **/
1389 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1390                                u16 length)
1391 {
1392         bi->page = NULL;
1393         skb->len += length;
1394         skb->data_len += length;
1395         skb->truesize += PAGE_SIZE;
1396 }
1397
1398 /**
1399  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1400  * @adapter: board private structure
1401  *
1402  * the return value indicates whether actual cleaning was done, there
1403  * is no guarantee that everything was cleaned
1404  **/
1405 static bool e1000_clean_jumbo_rx_irq(struct e1000_ring *rx_ring, int *work_done,
1406                                      int work_to_do)
1407 {
1408         struct e1000_adapter *adapter = rx_ring->adapter;
1409         struct net_device *netdev = adapter->netdev;
1410         struct pci_dev *pdev = adapter->pdev;
1411         union e1000_rx_desc_extended *rx_desc, *next_rxd;
1412         struct e1000_buffer *buffer_info, *next_buffer;
1413         u32 length, staterr;
1414         unsigned int i;
1415         int cleaned_count = 0;
1416         bool cleaned = false;
1417         unsigned int total_rx_bytes=0, total_rx_packets=0;
1418
1419         i = rx_ring->next_to_clean;
1420         rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
1421         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1422         buffer_info = &rx_ring->buffer_info[i];
1423
1424         while (staterr & E1000_RXD_STAT_DD) {
1425                 struct sk_buff *skb;
1426
1427                 if (*work_done >= work_to_do)
1428                         break;
1429                 (*work_done)++;
1430                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1431
1432                 skb = buffer_info->skb;
1433                 buffer_info->skb = NULL;
1434
1435                 ++i;
1436                 if (i == rx_ring->count)
1437                         i = 0;
1438                 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
1439                 prefetch(next_rxd);
1440
1441                 next_buffer = &rx_ring->buffer_info[i];
1442
1443                 cleaned = true;
1444                 cleaned_count++;
1445                 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1446                                DMA_FROM_DEVICE);
1447                 buffer_info->dma = 0;
1448
1449                 length = le16_to_cpu(rx_desc->wb.upper.length);
1450
1451                 /* errors is only valid for DD + EOP descriptors */
1452                 if (unlikely((staterr & E1000_RXD_STAT_EOP) &&
1453                              ((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) &&
1454                               !(netdev->features & NETIF_F_RXALL)))) {
1455                         /* recycle both page and skb */
1456                         buffer_info->skb = skb;
1457                         /* an error means any chain goes out the window too */
1458                         if (rx_ring->rx_skb_top)
1459                                 dev_kfree_skb_irq(rx_ring->rx_skb_top);
1460                         rx_ring->rx_skb_top = NULL;
1461                         goto next_desc;
1462                 }
1463
1464 #define rxtop (rx_ring->rx_skb_top)
1465                 if (!(staterr & E1000_RXD_STAT_EOP)) {
1466                         /* this descriptor is only the beginning (or middle) */
1467                         if (!rxtop) {
1468                                 /* this is the beginning of a chain */
1469                                 rxtop = skb;
1470                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1471                                                    0, length);
1472                         } else {
1473                                 /* this is the middle of a chain */
1474                                 skb_fill_page_desc(rxtop,
1475                                     skb_shinfo(rxtop)->nr_frags,
1476                                     buffer_info->page, 0, length);
1477                                 /* re-use the skb, only consumed the page */
1478                                 buffer_info->skb = skb;
1479                         }
1480                         e1000_consume_page(buffer_info, rxtop, length);
1481                         goto next_desc;
1482                 } else {
1483                         if (rxtop) {
1484                                 /* end of the chain */
1485                                 skb_fill_page_desc(rxtop,
1486                                     skb_shinfo(rxtop)->nr_frags,
1487                                     buffer_info->page, 0, length);
1488                                 /* re-use the current skb, we only consumed the
1489                                  * page */
1490                                 buffer_info->skb = skb;
1491                                 skb = rxtop;
1492                                 rxtop = NULL;
1493                                 e1000_consume_page(buffer_info, skb, length);
1494                         } else {
1495                                 /* no chain, got EOP, this buf is the packet
1496                                  * copybreak to save the put_page/alloc_page */
1497                                 if (length <= copybreak &&
1498                                     skb_tailroom(skb) >= length) {
1499                                         u8 *vaddr;
1500                                         vaddr = kmap_atomic(buffer_info->page);
1501                                         memcpy(skb_tail_pointer(skb), vaddr,
1502                                                length);
1503                                         kunmap_atomic(vaddr);
1504                                         /* re-use the page, so don't erase
1505                                          * buffer_info->page */
1506                                         skb_put(skb, length);
1507                                 } else {
1508                                         skb_fill_page_desc(skb, 0,
1509                                                            buffer_info->page, 0,
1510                                                            length);
1511                                         e1000_consume_page(buffer_info, skb,
1512                                                            length);
1513                                 }
1514                         }
1515                 }
1516
1517                 /* Receive Checksum Offload */
1518                 e1000_rx_checksum(adapter, staterr, skb);
1519
1520                 e1000_rx_hash(netdev, rx_desc->wb.lower.hi_dword.rss, skb);
1521
1522                 /* probably a little skewed due to removing CRC */
1523                 total_rx_bytes += skb->len;
1524                 total_rx_packets++;
1525
1526                 /* eth type trans needs skb->data to point to something */
1527                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1528                         e_err("pskb_may_pull failed.\n");
1529                         dev_kfree_skb_irq(skb);
1530                         goto next_desc;
1531                 }
1532
1533                 e1000_receive_skb(adapter, netdev, skb, staterr,
1534                                   rx_desc->wb.upper.vlan);
1535
1536 next_desc:
1537                 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1538
1539                 /* return some buffers to hardware, one at a time is too slow */
1540                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1541                         adapter->alloc_rx_buf(rx_ring, cleaned_count,
1542                                               GFP_ATOMIC);
1543                         cleaned_count = 0;
1544                 }
1545
1546                 /* use prefetched values */
1547                 rx_desc = next_rxd;
1548                 buffer_info = next_buffer;
1549
1550                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1551         }
1552         rx_ring->next_to_clean = i;
1553
1554         cleaned_count = e1000_desc_unused(rx_ring);
1555         if (cleaned_count)
1556                 adapter->alloc_rx_buf(rx_ring, cleaned_count, GFP_ATOMIC);
1557
1558         adapter->total_rx_bytes += total_rx_bytes;
1559         adapter->total_rx_packets += total_rx_packets;
1560         return cleaned;
1561 }
1562
1563 /**
1564  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1565  * @rx_ring: Rx descriptor ring
1566  **/
1567 static void e1000_clean_rx_ring(struct e1000_ring *rx_ring)
1568 {
1569         struct e1000_adapter *adapter = rx_ring->adapter;
1570         struct e1000_buffer *buffer_info;
1571         struct e1000_ps_page *ps_page;
1572         struct pci_dev *pdev = adapter->pdev;
1573         unsigned int i, j;
1574
1575         /* Free all the Rx ring sk_buffs */
1576         for (i = 0; i < rx_ring->count; i++) {
1577                 buffer_info = &rx_ring->buffer_info[i];
1578                 if (buffer_info->dma) {
1579                         if (adapter->clean_rx == e1000_clean_rx_irq)
1580                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1581                                                  adapter->rx_buffer_len,
1582                                                  DMA_FROM_DEVICE);
1583                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1584                                 dma_unmap_page(&pdev->dev, buffer_info->dma,
1585                                                PAGE_SIZE,
1586                                                DMA_FROM_DEVICE);
1587                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1588                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1589                                                  adapter->rx_ps_bsize0,
1590                                                  DMA_FROM_DEVICE);
1591                         buffer_info->dma = 0;
1592                 }
1593
1594                 if (buffer_info->page) {
1595                         put_page(buffer_info->page);
1596                         buffer_info->page = NULL;
1597                 }
1598
1599                 if (buffer_info->skb) {
1600                         dev_kfree_skb(buffer_info->skb);
1601                         buffer_info->skb = NULL;
1602                 }
1603
1604                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1605                         ps_page = &buffer_info->ps_pages[j];
1606                         if (!ps_page->page)
1607                                 break;
1608                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1609                                        DMA_FROM_DEVICE);
1610                         ps_page->dma = 0;
1611                         put_page(ps_page->page);
1612                         ps_page->page = NULL;
1613                 }
1614         }
1615
1616         /* there also may be some cached data from a chained receive */
1617         if (rx_ring->rx_skb_top) {
1618                 dev_kfree_skb(rx_ring->rx_skb_top);
1619                 rx_ring->rx_skb_top = NULL;
1620         }
1621
1622         /* Zero out the descriptor ring */
1623         memset(rx_ring->desc, 0, rx_ring->size);
1624
1625         rx_ring->next_to_clean = 0;
1626         rx_ring->next_to_use = 0;
1627         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1628
1629         writel(0, rx_ring->head);
1630         if (rx_ring->adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
1631                 e1000e_update_rdt_wa(rx_ring, 0);
1632         else
1633                 writel(0, rx_ring->tail);
1634 }
1635
1636 static void e1000e_downshift_workaround(struct work_struct *work)
1637 {
1638         struct e1000_adapter *adapter = container_of(work,
1639                                         struct e1000_adapter, downshift_task);
1640
1641         if (test_bit(__E1000_DOWN, &adapter->state))
1642                 return;
1643
1644         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1645 }
1646
1647 /**
1648  * e1000_intr_msi - Interrupt Handler
1649  * @irq: interrupt number
1650  * @data: pointer to a network interface device structure
1651  **/
1652 static irqreturn_t e1000_intr_msi(int irq, void *data)
1653 {
1654         struct net_device *netdev = data;
1655         struct e1000_adapter *adapter = netdev_priv(netdev);
1656         struct e1000_hw *hw = &adapter->hw;
1657         u32 icr = er32(ICR);
1658
1659         /*
1660          * read ICR disables interrupts using IAM
1661          */
1662
1663         if (icr & E1000_ICR_LSC) {
1664                 hw->mac.get_link_status = true;
1665                 /*
1666                  * ICH8 workaround-- Call gig speed drop workaround on cable
1667                  * disconnect (LSC) before accessing any PHY registers
1668                  */
1669                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1670                     (!(er32(STATUS) & E1000_STATUS_LU)))
1671                         schedule_work(&adapter->downshift_task);
1672
1673                 /*
1674                  * 80003ES2LAN workaround-- For packet buffer work-around on
1675                  * link down event; disable receives here in the ISR and reset
1676                  * adapter in watchdog
1677                  */
1678                 if (netif_carrier_ok(netdev) &&
1679                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1680                         /* disable receives */
1681                         u32 rctl = er32(RCTL);
1682                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1683                         adapter->flags |= FLAG_RX_RESTART_NOW;
1684                 }
1685                 /* guard against interrupt when we're going down */
1686                 if (!test_bit(__E1000_DOWN, &adapter->state))
1687                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1688         }
1689
1690         if (napi_schedule_prep(&adapter->napi)) {
1691                 adapter->total_tx_bytes = 0;
1692                 adapter->total_tx_packets = 0;
1693                 adapter->total_rx_bytes = 0;
1694                 adapter->total_rx_packets = 0;
1695                 __napi_schedule(&adapter->napi);
1696         }
1697
1698         return IRQ_HANDLED;
1699 }
1700
1701 /**
1702  * e1000_intr - Interrupt Handler
1703  * @irq: interrupt number
1704  * @data: pointer to a network interface device structure
1705  **/
1706 static irqreturn_t e1000_intr(int irq, void *data)
1707 {
1708         struct net_device *netdev = data;
1709         struct e1000_adapter *adapter = netdev_priv(netdev);
1710         struct e1000_hw *hw = &adapter->hw;
1711         u32 rctl, icr = er32(ICR);
1712
1713         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1714                 return IRQ_NONE;  /* Not our interrupt */
1715
1716         /*
1717          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1718          * not set, then the adapter didn't send an interrupt
1719          */
1720         if (!(icr & E1000_ICR_INT_ASSERTED))
1721                 return IRQ_NONE;
1722
1723         /*
1724          * Interrupt Auto-Mask...upon reading ICR,
1725          * interrupts are masked.  No need for the
1726          * IMC write
1727          */
1728
1729         if (icr & E1000_ICR_LSC) {
1730                 hw->mac.get_link_status = true;
1731                 /*
1732                  * ICH8 workaround-- Call gig speed drop workaround on cable
1733                  * disconnect (LSC) before accessing any PHY registers
1734                  */
1735                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1736                     (!(er32(STATUS) & E1000_STATUS_LU)))
1737                         schedule_work(&adapter->downshift_task);
1738
1739                 /*
1740                  * 80003ES2LAN workaround--
1741                  * For packet buffer work-around on link down event;
1742                  * disable receives here in the ISR and
1743                  * reset adapter in watchdog
1744                  */
1745                 if (netif_carrier_ok(netdev) &&
1746                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1747                         /* disable receives */
1748                         rctl = er32(RCTL);
1749                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1750                         adapter->flags |= FLAG_RX_RESTART_NOW;
1751                 }
1752                 /* guard against interrupt when we're going down */
1753                 if (!test_bit(__E1000_DOWN, &adapter->state))
1754                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1755         }
1756
1757         if (napi_schedule_prep(&adapter->napi)) {
1758                 adapter->total_tx_bytes = 0;
1759                 adapter->total_tx_packets = 0;
1760                 adapter->total_rx_bytes = 0;
1761                 adapter->total_rx_packets = 0;
1762                 __napi_schedule(&adapter->napi);
1763         }
1764
1765         return IRQ_HANDLED;
1766 }
1767
1768 static irqreturn_t e1000_msix_other(int irq, void *data)
1769 {
1770         struct net_device *netdev = data;
1771         struct e1000_adapter *adapter = netdev_priv(netdev);
1772         struct e1000_hw *hw = &adapter->hw;
1773         u32 icr = er32(ICR);
1774
1775         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1776                 if (!test_bit(__E1000_DOWN, &adapter->state))
1777                         ew32(IMS, E1000_IMS_OTHER);
1778                 return IRQ_NONE;
1779         }
1780
1781         if (icr & adapter->eiac_mask)
1782                 ew32(ICS, (icr & adapter->eiac_mask));
1783
1784         if (icr & E1000_ICR_OTHER) {
1785                 if (!(icr & E1000_ICR_LSC))
1786                         goto no_link_interrupt;
1787                 hw->mac.get_link_status = true;
1788                 /* guard against interrupt when we're going down */
1789                 if (!test_bit(__E1000_DOWN, &adapter->state))
1790                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1791         }
1792
1793 no_link_interrupt:
1794         if (!test_bit(__E1000_DOWN, &adapter->state))
1795                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1796
1797         return IRQ_HANDLED;
1798 }
1799
1800
1801 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1802 {
1803         struct net_device *netdev = data;
1804         struct e1000_adapter *adapter = netdev_priv(netdev);
1805         struct e1000_hw *hw = &adapter->hw;
1806         struct e1000_ring *tx_ring = adapter->tx_ring;
1807
1808
1809         adapter->total_tx_bytes = 0;
1810         adapter->total_tx_packets = 0;
1811
1812         if (!e1000_clean_tx_irq(tx_ring))
1813                 /* Ring was not completely cleaned, so fire another interrupt */
1814                 ew32(ICS, tx_ring->ims_val);
1815
1816         return IRQ_HANDLED;
1817 }
1818
1819 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1820 {
1821         struct net_device *netdev = data;
1822         struct e1000_adapter *adapter = netdev_priv(netdev);
1823         struct e1000_ring *rx_ring = adapter->rx_ring;
1824
1825         /* Write the ITR value calculated at the end of the
1826          * previous interrupt.
1827          */
1828         if (rx_ring->set_itr) {
1829                 writel(1000000000 / (rx_ring->itr_val * 256),
1830                        rx_ring->itr_register);
1831                 rx_ring->set_itr = 0;
1832         }
1833
1834         if (napi_schedule_prep(&adapter->napi)) {
1835                 adapter->total_rx_bytes = 0;
1836                 adapter->total_rx_packets = 0;
1837                 __napi_schedule(&adapter->napi);
1838         }
1839         return IRQ_HANDLED;
1840 }
1841
1842 /**
1843  * e1000_configure_msix - Configure MSI-X hardware
1844  *
1845  * e1000_configure_msix sets up the hardware to properly
1846  * generate MSI-X interrupts.
1847  **/
1848 static void e1000_configure_msix(struct e1000_adapter *adapter)
1849 {
1850         struct e1000_hw *hw = &adapter->hw;
1851         struct e1000_ring *rx_ring = adapter->rx_ring;
1852         struct e1000_ring *tx_ring = adapter->tx_ring;
1853         int vector = 0;
1854         u32 ctrl_ext, ivar = 0;
1855
1856         adapter->eiac_mask = 0;
1857
1858         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1859         if (hw->mac.type == e1000_82574) {
1860                 u32 rfctl = er32(RFCTL);
1861                 rfctl |= E1000_RFCTL_ACK_DIS;
1862                 ew32(RFCTL, rfctl);
1863         }
1864
1865 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1866         /* Configure Rx vector */
1867         rx_ring->ims_val = E1000_IMS_RXQ0;
1868         adapter->eiac_mask |= rx_ring->ims_val;
1869         if (rx_ring->itr_val)
1870                 writel(1000000000 / (rx_ring->itr_val * 256),
1871                        rx_ring->itr_register);
1872         else
1873                 writel(1, rx_ring->itr_register);
1874         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1875
1876         /* Configure Tx vector */
1877         tx_ring->ims_val = E1000_IMS_TXQ0;
1878         vector++;
1879         if (tx_ring->itr_val)
1880                 writel(1000000000 / (tx_ring->itr_val * 256),
1881                        tx_ring->itr_register);
1882         else
1883                 writel(1, tx_ring->itr_register);
1884         adapter->eiac_mask |= tx_ring->ims_val;
1885         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1886
1887         /* set vector for Other Causes, e.g. link changes */
1888         vector++;
1889         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1890         if (rx_ring->itr_val)
1891                 writel(1000000000 / (rx_ring->itr_val * 256),
1892                        hw->hw_addr + E1000_EITR_82574(vector));
1893         else
1894                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1895
1896         /* Cause Tx interrupts on every write back */
1897         ivar |= (1 << 31);
1898
1899         ew32(IVAR, ivar);
1900
1901         /* enable MSI-X PBA support */
1902         ctrl_ext = er32(CTRL_EXT);
1903         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1904
1905         /* Auto-Mask Other interrupts upon ICR read */
1906 #define E1000_EIAC_MASK_82574   0x01F00000
1907         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1908         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1909         ew32(CTRL_EXT, ctrl_ext);
1910         e1e_flush();
1911 }
1912
1913 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1914 {
1915         if (adapter->msix_entries) {
1916                 pci_disable_msix(adapter->pdev);
1917                 kfree(adapter->msix_entries);
1918                 adapter->msix_entries = NULL;
1919         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1920                 pci_disable_msi(adapter->pdev);
1921                 adapter->flags &= ~FLAG_MSI_ENABLED;
1922         }
1923 }
1924
1925 /**
1926  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1927  *
1928  * Attempt to configure interrupts using the best available
1929  * capabilities of the hardware and kernel.
1930  **/
1931 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1932 {
1933         int err;
1934         int i;
1935
1936         switch (adapter->int_mode) {
1937         case E1000E_INT_MODE_MSIX:
1938                 if (adapter->flags & FLAG_HAS_MSIX) {
1939                         adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1940                         adapter->msix_entries = kcalloc(adapter->num_vectors,
1941                                                       sizeof(struct msix_entry),
1942                                                       GFP_KERNEL);
1943                         if (adapter->msix_entries) {
1944                                 for (i = 0; i < adapter->num_vectors; i++)
1945                                         adapter->msix_entries[i].entry = i;
1946
1947                                 err = pci_enable_msix(adapter->pdev,
1948                                                       adapter->msix_entries,
1949                                                       adapter->num_vectors);
1950                                 if (err == 0)
1951                                         return;
1952                         }
1953                         /* MSI-X failed, so fall through and try MSI */
1954                         e_err("Failed to initialize MSI-X interrupts.  Falling back to MSI interrupts.\n");
1955                         e1000e_reset_interrupt_capability(adapter);
1956                 }
1957                 adapter->int_mode = E1000E_INT_MODE_MSI;
1958                 /* Fall through */
1959         case E1000E_INT_MODE_MSI:
1960                 if (!pci_enable_msi(adapter->pdev)) {
1961                         adapter->flags |= FLAG_MSI_ENABLED;
1962                 } else {
1963                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1964                         e_err("Failed to initialize MSI interrupts.  Falling back to legacy interrupts.\n");
1965                 }
1966                 /* Fall through */
1967         case E1000E_INT_MODE_LEGACY:
1968                 /* Don't do anything; this is the system default */
1969                 break;
1970         }
1971
1972         /* store the number of vectors being used */
1973         adapter->num_vectors = 1;
1974 }
1975
1976 /**
1977  * e1000_request_msix - Initialize MSI-X interrupts
1978  *
1979  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1980  * kernel.
1981  **/
1982 static int e1000_request_msix(struct e1000_adapter *adapter)
1983 {
1984         struct net_device *netdev = adapter->netdev;
1985         int err = 0, vector = 0;
1986
1987         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1988                 snprintf(adapter->rx_ring->name,
1989                          sizeof(adapter->rx_ring->name) - 1,
1990                          "%s-rx-0", netdev->name);
1991         else
1992                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1993         err = request_irq(adapter->msix_entries[vector].vector,
1994                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1995                           netdev);
1996         if (err)
1997                 return err;
1998         adapter->rx_ring->itr_register = adapter->hw.hw_addr +
1999             E1000_EITR_82574(vector);
2000         adapter->rx_ring->itr_val = adapter->itr;
2001         vector++;
2002
2003         if (strlen(netdev->name) < (IFNAMSIZ - 5))
2004                 snprintf(adapter->tx_ring->name,
2005                          sizeof(adapter->tx_ring->name) - 1,
2006                          "%s-tx-0", netdev->name);
2007         else
2008                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
2009         err = request_irq(adapter->msix_entries[vector].vector,
2010                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
2011                           netdev);
2012         if (err)
2013                 return err;
2014         adapter->tx_ring->itr_register = adapter->hw.hw_addr +
2015             E1000_EITR_82574(vector);
2016         adapter->tx_ring->itr_val = adapter->itr;
2017         vector++;
2018
2019         err = request_irq(adapter->msix_entries[vector].vector,
2020                           e1000_msix_other, 0, netdev->name, netdev);
2021         if (err)
2022                 return err;
2023
2024         e1000_configure_msix(adapter);
2025
2026         return 0;
2027 }
2028
2029 /**
2030  * e1000_request_irq - initialize interrupts
2031  *
2032  * Attempts to configure interrupts using the best available
2033  * capabilities of the hardware and kernel.
2034  **/
2035 static int e1000_request_irq(struct e1000_adapter *adapter)
2036 {
2037         struct net_device *netdev = adapter->netdev;
2038         int err;
2039
2040         if (adapter->msix_entries) {
2041                 err = e1000_request_msix(adapter);
2042                 if (!err)
2043                         return err;
2044                 /* fall back to MSI */
2045                 e1000e_reset_interrupt_capability(adapter);
2046                 adapter->int_mode = E1000E_INT_MODE_MSI;
2047                 e1000e_set_interrupt_capability(adapter);
2048         }
2049         if (adapter->flags & FLAG_MSI_ENABLED) {
2050                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
2051                                   netdev->name, netdev);
2052                 if (!err)
2053                         return err;
2054
2055                 /* fall back to legacy interrupt */
2056                 e1000e_reset_interrupt_capability(adapter);
2057                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2058         }
2059
2060         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
2061                           netdev->name, netdev);
2062         if (err)
2063                 e_err("Unable to allocate interrupt, Error: %d\n", err);
2064
2065         return err;
2066 }
2067
2068 static void e1000_free_irq(struct e1000_adapter *adapter)
2069 {
2070         struct net_device *netdev = adapter->netdev;
2071
2072         if (adapter->msix_entries) {
2073                 int vector = 0;
2074
2075                 free_irq(adapter->msix_entries[vector].vector, netdev);
2076                 vector++;
2077
2078                 free_irq(adapter->msix_entries[vector].vector, netdev);
2079                 vector++;
2080
2081                 /* Other Causes interrupt vector */
2082                 free_irq(adapter->msix_entries[vector].vector, netdev);
2083                 return;
2084         }
2085
2086         free_irq(adapter->pdev->irq, netdev);
2087 }
2088
2089 /**
2090  * e1000_irq_disable - Mask off interrupt generation on the NIC
2091  **/
2092 static void e1000_irq_disable(struct e1000_adapter *adapter)
2093 {
2094         struct e1000_hw *hw = &adapter->hw;
2095
2096         ew32(IMC, ~0);
2097         if (adapter->msix_entries)
2098                 ew32(EIAC_82574, 0);
2099         e1e_flush();
2100
2101         if (adapter->msix_entries) {
2102                 int i;
2103                 for (i = 0; i < adapter->num_vectors; i++)
2104                         synchronize_irq(adapter->msix_entries[i].vector);
2105         } else {
2106                 synchronize_irq(adapter->pdev->irq);
2107         }
2108 }
2109
2110 /**
2111  * e1000_irq_enable - Enable default interrupt generation settings
2112  **/
2113 static void e1000_irq_enable(struct e1000_adapter *adapter)
2114 {
2115         struct e1000_hw *hw = &adapter->hw;
2116
2117         if (adapter->msix_entries) {
2118                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
2119                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
2120         } else {
2121                 ew32(IMS, IMS_ENABLE_MASK);
2122         }
2123         e1e_flush();
2124 }
2125
2126 /**
2127  * e1000e_get_hw_control - get control of the h/w from f/w
2128  * @adapter: address of board private structure
2129  *
2130  * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2131  * For ASF and Pass Through versions of f/w this means that
2132  * the driver is loaded. For AMT version (only with 82573)
2133  * of the f/w this means that the network i/f is open.
2134  **/
2135 void e1000e_get_hw_control(struct e1000_adapter *adapter)
2136 {
2137         struct e1000_hw *hw = &adapter->hw;
2138         u32 ctrl_ext;
2139         u32 swsm;
2140
2141         /* Let firmware know the driver has taken over */
2142         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2143                 swsm = er32(SWSM);
2144                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2145         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2146                 ctrl_ext = er32(CTRL_EXT);
2147                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2148         }
2149 }
2150
2151 /**
2152  * e1000e_release_hw_control - release control of the h/w to f/w
2153  * @adapter: address of board private structure
2154  *
2155  * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2156  * For ASF and Pass Through versions of f/w this means that the
2157  * driver is no longer loaded. For AMT version (only with 82573) i
2158  * of the f/w this means that the network i/f is closed.
2159  *
2160  **/
2161 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2162 {
2163         struct e1000_hw *hw = &adapter->hw;
2164         u32 ctrl_ext;
2165         u32 swsm;
2166
2167         /* Let firmware taken over control of h/w */
2168         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2169                 swsm = er32(SWSM);
2170                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2171         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2172                 ctrl_ext = er32(CTRL_EXT);
2173                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2174         }
2175 }
2176
2177 /**
2178  * e1000_alloc_ring_dma - allocate memory for a ring structure
2179  **/
2180 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2181                                 struct e1000_ring *ring)
2182 {
2183         struct pci_dev *pdev = adapter->pdev;
2184
2185         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2186                                         GFP_KERNEL);
2187         if (!ring->desc)
2188                 return -ENOMEM;
2189
2190         return 0;
2191 }
2192
2193 /**
2194  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2195  * @tx_ring: Tx descriptor ring
2196  *
2197  * Return 0 on success, negative on failure
2198  **/
2199 int e1000e_setup_tx_resources(struct e1000_ring *tx_ring)
2200 {
2201         struct e1000_adapter *adapter = tx_ring->adapter;
2202         int err = -ENOMEM, size;
2203
2204         size = sizeof(struct e1000_buffer) * tx_ring->count;
2205         tx_ring->buffer_info = vzalloc(size);
2206         if (!tx_ring->buffer_info)
2207                 goto err;
2208
2209         /* round up to nearest 4K */
2210         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2211         tx_ring->size = ALIGN(tx_ring->size, 4096);
2212
2213         err = e1000_alloc_ring_dma(adapter, tx_ring);
2214         if (err)
2215                 goto err;
2216
2217         tx_ring->next_to_use = 0;
2218         tx_ring->next_to_clean = 0;
2219
2220         return 0;
2221 err:
2222         vfree(tx_ring->buffer_info);
2223         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2224         return err;
2225 }
2226
2227 /**
2228  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2229  * @rx_ring: Rx descriptor ring
2230  *
2231  * Returns 0 on success, negative on failure
2232  **/
2233 int e1000e_setup_rx_resources(struct e1000_ring *rx_ring)
2234 {
2235         struct e1000_adapter *adapter = rx_ring->adapter;
2236         struct e1000_buffer *buffer_info;
2237         int i, size, desc_len, err = -ENOMEM;
2238
2239         size = sizeof(struct e1000_buffer) * rx_ring->count;
2240         rx_ring->buffer_info = vzalloc(size);
2241         if (!rx_ring->buffer_info)
2242                 goto err;
2243
2244         for (i = 0; i < rx_ring->count; i++) {
2245                 buffer_info = &rx_ring->buffer_info[i];
2246                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2247                                                 sizeof(struct e1000_ps_page),
2248                                                 GFP_KERNEL);
2249                 if (!buffer_info->ps_pages)
2250                         goto err_pages;
2251         }
2252
2253         desc_len = sizeof(union e1000_rx_desc_packet_split);
2254
2255         /* Round up to nearest 4K */
2256         rx_ring->size = rx_ring->count * desc_len;
2257         rx_ring->size = ALIGN(rx_ring->size, 4096);
2258
2259         err = e1000_alloc_ring_dma(adapter, rx_ring);
2260         if (err)
2261                 goto err_pages;
2262
2263         rx_ring->next_to_clean = 0;
2264         rx_ring->next_to_use = 0;
2265         rx_ring->rx_skb_top = NULL;
2266
2267         return 0;
2268
2269 err_pages:
2270         for (i = 0; i < rx_ring->count; i++) {
2271                 buffer_info = &rx_ring->buffer_info[i];
2272                 kfree(buffer_info->ps_pages);
2273         }
2274 err:
2275         vfree(rx_ring->buffer_info);
2276         e_err("Unable to allocate memory for the receive descriptor ring\n");
2277         return err;
2278 }
2279
2280 /**
2281  * e1000_clean_tx_ring - Free Tx Buffers
2282  * @tx_ring: Tx descriptor ring
2283  **/
2284 static void e1000_clean_tx_ring(struct e1000_ring *tx_ring)
2285 {
2286         struct e1000_adapter *adapter = tx_ring->adapter;
2287         struct e1000_buffer *buffer_info;
2288         unsigned long size;
2289         unsigned int i;
2290
2291         for (i = 0; i < tx_ring->count; i++) {
2292                 buffer_info = &tx_ring->buffer_info[i];
2293                 e1000_put_txbuf(tx_ring, buffer_info);
2294         }
2295
2296         netdev_reset_queue(adapter->netdev);
2297         size = sizeof(struct e1000_buffer) * tx_ring->count;
2298         memset(tx_ring->buffer_info, 0, size);
2299
2300         memset(tx_ring->desc, 0, tx_ring->size);
2301
2302         tx_ring->next_to_use = 0;
2303         tx_ring->next_to_clean = 0;
2304
2305         writel(0, tx_ring->head);
2306         if (tx_ring->adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
2307                 e1000e_update_tdt_wa(tx_ring, 0);
2308         else
2309                 writel(0, tx_ring->tail);
2310 }
2311
2312 /**
2313  * e1000e_free_tx_resources - Free Tx Resources per Queue
2314  * @tx_ring: Tx descriptor ring
2315  *
2316  * Free all transmit software resources
2317  **/
2318 void e1000e_free_tx_resources(struct e1000_ring *tx_ring)
2319 {
2320         struct e1000_adapter *adapter = tx_ring->adapter;
2321         struct pci_dev *pdev = adapter->pdev;
2322
2323         e1000_clean_tx_ring(tx_ring);
2324
2325         vfree(tx_ring->buffer_info);
2326         tx_ring->buffer_info = NULL;
2327
2328         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2329                           tx_ring->dma);
2330         tx_ring->desc = NULL;
2331 }
2332
2333 /**
2334  * e1000e_free_rx_resources - Free Rx Resources
2335  * @rx_ring: Rx descriptor ring
2336  *
2337  * Free all receive software resources
2338  **/
2339 void e1000e_free_rx_resources(struct e1000_ring *rx_ring)
2340 {
2341         struct e1000_adapter *adapter = rx_ring->adapter;
2342         struct pci_dev *pdev = adapter->pdev;
2343         int i;
2344
2345         e1000_clean_rx_ring(rx_ring);
2346
2347         for (i = 0; i < rx_ring->count; i++)
2348                 kfree(rx_ring->buffer_info[i].ps_pages);
2349
2350         vfree(rx_ring->buffer_info);
2351         rx_ring->buffer_info = NULL;
2352
2353         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2354                           rx_ring->dma);
2355         rx_ring->desc = NULL;
2356 }
2357
2358 /**
2359  * e1000_update_itr - update the dynamic ITR value based on statistics
2360  * @adapter: pointer to adapter
2361  * @itr_setting: current adapter->itr
2362  * @packets: the number of packets during this measurement interval
2363  * @bytes: the number of bytes during this measurement interval
2364  *
2365  *      Stores a new ITR value based on packets and byte
2366  *      counts during the last interrupt.  The advantage of per interrupt
2367  *      computation is faster updates and more accurate ITR for the current
2368  *      traffic pattern.  Constants in this function were computed
2369  *      based on theoretical maximum wire speed and thresholds were set based
2370  *      on testing data as well as attempting to minimize response time
2371  *      while increasing bulk throughput.  This functionality is controlled
2372  *      by the InterruptThrottleRate module parameter.
2373  **/
2374 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2375                                      u16 itr_setting, int packets,
2376                                      int bytes)
2377 {
2378         unsigned int retval = itr_setting;
2379
2380         if (packets == 0)
2381                 return itr_setting;
2382
2383         switch (itr_setting) {
2384         case lowest_latency:
2385                 /* handle TSO and jumbo frames */
2386                 if (bytes/packets > 8000)
2387                         retval = bulk_latency;
2388                 else if ((packets < 5) && (bytes > 512))
2389                         retval = low_latency;
2390                 break;
2391         case low_latency:  /* 50 usec aka 20000 ints/s */
2392                 if (bytes > 10000) {
2393                         /* this if handles the TSO accounting */
2394                         if (bytes/packets > 8000)
2395                                 retval = bulk_latency;
2396                         else if ((packets < 10) || ((bytes/packets) > 1200))
2397                                 retval = bulk_latency;
2398                         else if ((packets > 35))
2399                                 retval = lowest_latency;
2400                 } else if (bytes/packets > 2000) {
2401                         retval = bulk_latency;
2402                 } else if (packets <= 2 && bytes < 512) {
2403                         retval = lowest_latency;
2404                 }
2405                 break;
2406         case bulk_latency: /* 250 usec aka 4000 ints/s */
2407                 if (bytes > 25000) {
2408                         if (packets > 35)
2409                                 retval = low_latency;
2410                 } else if (bytes < 6000) {
2411                         retval = low_latency;
2412                 }
2413                 break;
2414         }
2415
2416         return retval;
2417 }
2418
2419 static void e1000_set_itr(struct e1000_adapter *adapter)
2420 {
2421         struct e1000_hw *hw = &adapter->hw;
2422         u16 current_itr;
2423         u32 new_itr = adapter->itr;
2424
2425         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2426         if (adapter->link_speed != SPEED_1000) {
2427                 current_itr = 0;
2428                 new_itr = 4000;
2429                 goto set_itr_now;
2430         }
2431
2432         if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2433                 new_itr = 0;
2434                 goto set_itr_now;
2435         }
2436
2437         adapter->tx_itr = e1000_update_itr(adapter,
2438                                     adapter->tx_itr,
2439                                     adapter->total_tx_packets,
2440                                     adapter->total_tx_bytes);
2441         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2442         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2443                 adapter->tx_itr = low_latency;
2444
2445         adapter->rx_itr = e1000_update_itr(adapter,
2446                                     adapter->rx_itr,
2447                                     adapter->total_rx_packets,
2448                                     adapter->total_rx_bytes);
2449         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2450         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2451                 adapter->rx_itr = low_latency;
2452
2453         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2454
2455         switch (current_itr) {
2456         /* counts and packets in update_itr are dependent on these numbers */
2457         case lowest_latency:
2458                 new_itr = 70000;
2459                 break;
2460         case low_latency:
2461                 new_itr = 20000; /* aka hwitr = ~200 */
2462                 break;
2463         case bulk_latency:
2464                 new_itr = 4000;
2465                 break;
2466         default:
2467                 break;
2468         }
2469
2470 set_itr_now:
2471         if (new_itr != adapter->itr) {
2472                 /*
2473                  * this attempts to bias the interrupt rate towards Bulk
2474                  * by adding intermediate steps when interrupt rate is
2475                  * increasing
2476                  */
2477                 new_itr = new_itr > adapter->itr ?
2478                              min(adapter->itr + (new_itr >> 2), new_itr) :
2479                              new_itr;
2480                 adapter->itr = new_itr;
2481                 adapter->rx_ring->itr_val = new_itr;
2482                 if (adapter->msix_entries)
2483                         adapter->rx_ring->set_itr = 1;
2484                 else
2485                         if (new_itr)
2486                                 ew32(ITR, 1000000000 / (new_itr * 256));
2487                         else
2488                                 ew32(ITR, 0);
2489         }
2490 }
2491
2492 /**
2493  * e1000e_write_itr - write the ITR value to the appropriate registers
2494  * @adapter: address of board private structure
2495  * @itr: new ITR value to program
2496  *
2497  * e1000e_write_itr determines if the adapter is in MSI-X mode
2498  * and, if so, writes the EITR registers with the ITR value.
2499  * Otherwise, it writes the ITR value into the ITR register.
2500  **/
2501 void e1000e_write_itr(struct e1000_adapter *adapter, u32 itr)
2502 {
2503         struct e1000_hw *hw = &adapter->hw;
2504         u32 new_itr = itr ? 1000000000 / (itr * 256) : 0;
2505
2506         if (adapter->msix_entries) {
2507                 int vector;
2508
2509                 for (vector = 0; vector < adapter->num_vectors; vector++)
2510                         writel(new_itr, hw->hw_addr + E1000_EITR_82574(vector));
2511         } else {
2512                 ew32(ITR, new_itr);
2513         }
2514 }
2515
2516 /**
2517  * e1000_alloc_queues - Allocate memory for all rings
2518  * @adapter: board private structure to initialize
2519  **/
2520 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2521 {
2522         int size = sizeof(struct e1000_ring);
2523
2524         adapter->tx_ring = kzalloc(size, GFP_KERNEL);
2525         if (!adapter->tx_ring)
2526                 goto err;
2527         adapter->tx_ring->count = adapter->tx_ring_count;
2528         adapter->tx_ring->adapter = adapter;
2529
2530         adapter->rx_ring = kzalloc(size, GFP_KERNEL);
2531         if (!adapter->rx_ring)
2532                 goto err;
2533         adapter->rx_ring->count = adapter->rx_ring_count;
2534         adapter->rx_ring->adapter = adapter;
2535
2536         return 0;
2537 err:
2538         e_err("Unable to allocate memory for queues\n");
2539         kfree(adapter->rx_ring);
2540         kfree(adapter->tx_ring);
2541         return -ENOMEM;
2542 }
2543
2544 /**
2545  * e1000e_poll - NAPI Rx polling callback
2546  * @napi: struct associated with this polling callback
2547  * @weight: number of packets driver is allowed to process this poll
2548  **/
2549 static int e1000e_poll(struct napi_struct *napi, int weight)
2550 {
2551         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter,
2552                                                      napi);
2553         struct e1000_hw *hw = &adapter->hw;
2554         struct net_device *poll_dev = adapter->netdev;
2555         int tx_cleaned = 1, work_done = 0;
2556
2557         adapter = netdev_priv(poll_dev);
2558
2559         if (!adapter->msix_entries ||
2560             (adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2561                 tx_cleaned = e1000_clean_tx_irq(adapter->tx_ring);
2562
2563         adapter->clean_rx(adapter->rx_ring, &work_done, weight);
2564
2565         if (!tx_cleaned)
2566                 work_done = weight;
2567
2568         /* If weight not fully consumed, exit the polling mode */
2569         if (work_done < weight) {
2570                 if (adapter->itr_setting & 3)
2571                         e1000_set_itr(adapter);
2572                 napi_complete(napi);
2573                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2574                         if (adapter->msix_entries)
2575                                 ew32(IMS, adapter->rx_ring->ims_val);
2576                         else
2577                                 e1000_irq_enable(adapter);
2578                 }
2579         }
2580
2581         return work_done;
2582 }
2583
2584 static int e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2585 {
2586         struct e1000_adapter *adapter = netdev_priv(netdev);
2587         struct e1000_hw *hw = &adapter->hw;
2588         u32 vfta, index;
2589
2590         /* don't update vlan cookie if already programmed */
2591         if ((adapter->hw.mng_cookie.status &
2592              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2593             (vid == adapter->mng_vlan_id))
2594                 return 0;
2595
2596         /* add VID to filter table */
2597         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2598                 index = (vid >> 5) & 0x7F;
2599                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2600                 vfta |= (1 << (vid & 0x1F));
2601                 hw->mac.ops.write_vfta(hw, index, vfta);
2602         }
2603
2604         set_bit(vid, adapter->active_vlans);
2605
2606         return 0;
2607 }
2608
2609 static int e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2610 {
2611         struct e1000_adapter *adapter = netdev_priv(netdev);
2612         struct e1000_hw *hw = &adapter->hw;
2613         u32 vfta, index;
2614
2615         if ((adapter->hw.mng_cookie.status &
2616              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2617             (vid == adapter->mng_vlan_id)) {
2618                 /* release control to f/w */
2619                 e1000e_release_hw_control(adapter);
2620                 return 0;
2621         }
2622
2623         /* remove VID from filter table */
2624         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2625                 index = (vid >> 5) & 0x7F;
2626                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2627                 vfta &= ~(1 << (vid & 0x1F));
2628                 hw->mac.ops.write_vfta(hw, index, vfta);
2629         }
2630
2631         clear_bit(vid, adapter->active_vlans);
2632
2633         return 0;
2634 }
2635
2636 /**
2637  * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2638  * @adapter: board private structure to initialize
2639  **/
2640 static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2641 {
2642         struct net_device *netdev = adapter->netdev;
2643         struct e1000_hw *hw = &adapter->hw;
2644         u32 rctl;
2645
2646         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2647                 /* disable VLAN receive filtering */
2648                 rctl = er32(RCTL);
2649                 rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2650                 ew32(RCTL, rctl);
2651
2652                 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2653                         e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2654                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2655                 }
2656         }
2657 }
2658
2659 /**
2660  * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2661  * @adapter: board private structure to initialize
2662  **/
2663 static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2664 {
2665         struct e1000_hw *hw = &adapter->hw;
2666         u32 rctl;
2667
2668         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2669                 /* enable VLAN receive filtering */
2670                 rctl = er32(RCTL);
2671                 rctl |= E1000_RCTL_VFE;
2672                 rctl &= ~E1000_RCTL_CFIEN;
2673                 ew32(RCTL, rctl);
2674         }
2675 }
2676
2677 /**
2678  * e1000e_vlan_strip_enable - helper to disable HW VLAN stripping
2679  * @adapter: board private structure to initialize
2680  **/
2681 static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2682 {
2683         struct e1000_hw *hw = &adapter->hw;
2684         u32 ctrl;
2685
2686         /* disable VLAN tag insert/strip */
2687         ctrl = er32(CTRL);
2688         ctrl &= ~E1000_CTRL_VME;
2689         ew32(CTRL, ctrl);
2690 }
2691
2692 /**
2693  * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2694  * @adapter: board private structure to initialize
2695  **/
2696 static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2697 {
2698         struct e1000_hw *hw = &adapter->hw;
2699         u32 ctrl;
2700
2701         /* enable VLAN tag insert/strip */
2702         ctrl = er32(CTRL);
2703         ctrl |= E1000_CTRL_VME;
2704         ew32(CTRL, ctrl);
2705 }
2706
2707 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2708 {
2709         struct net_device *netdev = adapter->netdev;
2710         u16 vid = adapter->hw.mng_cookie.vlan_id;
2711         u16 old_vid = adapter->mng_vlan_id;
2712
2713         if (adapter->hw.mng_cookie.status &
2714             E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2715                 e1000_vlan_rx_add_vid(netdev, vid);
2716                 adapter->mng_vlan_id = vid;
2717         }
2718
2719         if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2720                 e1000_vlan_rx_kill_vid(netdev, old_vid);
2721 }
2722
2723 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2724 {
2725         u16 vid;
2726
2727         e1000_vlan_rx_add_vid(adapter->netdev, 0);
2728
2729         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2730                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2731 }
2732
2733 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2734 {
2735         struct e1000_hw *hw = &adapter->hw;
2736         u32 manc, manc2h, mdef, i, j;
2737
2738         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2739                 return;
2740
2741         manc = er32(MANC);
2742
2743         /*
2744          * enable receiving management packets to the host. this will probably
2745          * generate destination unreachable messages from the host OS, but
2746          * the packets will be handled on SMBUS
2747          */
2748         manc |= E1000_MANC_EN_MNG2HOST;
2749         manc2h = er32(MANC2H);
2750
2751         switch (hw->mac.type) {
2752         default:
2753                 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2754                 break;
2755         case e1000_82574:
2756         case e1000_82583:
2757                 /*
2758                  * Check if IPMI pass-through decision filter already exists;
2759                  * if so, enable it.
2760                  */
2761                 for (i = 0, j = 0; i < 8; i++) {
2762                         mdef = er32(MDEF(i));
2763
2764                         /* Ignore filters with anything other than IPMI ports */
2765                         if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2766                                 continue;
2767
2768                         /* Enable this decision filter in MANC2H */
2769                         if (mdef)
2770                                 manc2h |= (1 << i);
2771
2772                         j |= mdef;
2773                 }
2774
2775                 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2776                         break;
2777
2778                 /* Create new decision filter in an empty filter */
2779                 for (i = 0, j = 0; i < 8; i++)
2780                         if (er32(MDEF(i)) == 0) {
2781                                 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2782                                                E1000_MDEF_PORT_664));
2783                                 manc2h |= (1 << 1);
2784                                 j++;
2785                                 break;
2786                         }
2787
2788                 if (!j)
2789                         e_warn("Unable to create IPMI pass-through filter\n");
2790                 break;
2791         }
2792
2793         ew32(MANC2H, manc2h);
2794         ew32(MANC, manc);
2795 }
2796
2797 /**
2798  * e1000_configure_tx - Configure Transmit Unit after Reset
2799  * @adapter: board private structure
2800  *
2801  * Configure the Tx unit of the MAC after a reset.
2802  **/
2803 static void e1000_configure_tx(struct e1000_adapter *adapter)
2804 {
2805         struct e1000_hw *hw = &adapter->hw;
2806         struct e1000_ring *tx_ring = adapter->tx_ring;
2807         u64 tdba;
2808         u32 tdlen, tarc;
2809
2810         /* Setup the HW Tx Head and Tail descriptor pointers */
2811         tdba = tx_ring->dma;
2812         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2813         ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
2814         ew32(TDBAH(0), (tdba >> 32));
2815         ew32(TDLEN(0), tdlen);
2816         ew32(TDH(0), 0);
2817         ew32(TDT(0), 0);
2818         tx_ring->head = adapter->hw.hw_addr + E1000_TDH(0);
2819         tx_ring->tail = adapter->hw.hw_addr + E1000_TDT(0);
2820
2821         /* Set the Tx Interrupt Delay register */
2822         ew32(TIDV, adapter->tx_int_delay);
2823         /* Tx irq moderation */
2824         ew32(TADV, adapter->tx_abs_int_delay);
2825
2826         if (adapter->flags2 & FLAG2_DMA_BURST) {
2827                 u32 txdctl = er32(TXDCTL(0));
2828                 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2829                             E1000_TXDCTL_WTHRESH);
2830                 /*
2831                  * set up some performance related parameters to encourage the
2832                  * hardware to use the bus more efficiently in bursts, depends
2833                  * on the tx_int_delay to be enabled,
2834                  * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
2835                  * hthresh = 1 ==> prefetch when one or more available
2836                  * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2837                  * BEWARE: this seems to work but should be considered first if
2838                  * there are Tx hangs or other Tx related bugs
2839                  */
2840                 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2841                 ew32(TXDCTL(0), txdctl);
2842         }
2843         /* erratum work around: set txdctl the same for both queues */
2844         ew32(TXDCTL(1), er32(TXDCTL(0)));
2845
2846         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2847                 tarc = er32(TARC(0));
2848                 /*
2849                  * set the speed mode bit, we'll clear it if we're not at
2850                  * gigabit link later
2851                  */
2852 #define SPEED_MODE_BIT (1 << 21)
2853                 tarc |= SPEED_MODE_BIT;
2854                 ew32(TARC(0), tarc);
2855         }
2856
2857         /* errata: program both queues to unweighted RR */
2858         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2859                 tarc = er32(TARC(0));
2860                 tarc |= 1;
2861                 ew32(TARC(0), tarc);
2862                 tarc = er32(TARC(1));
2863                 tarc |= 1;
2864                 ew32(TARC(1), tarc);
2865         }
2866
2867         /* Setup Transmit Descriptor Settings for eop descriptor */
2868         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2869
2870         /* only set IDE if we are delaying interrupts using the timers */
2871         if (adapter->tx_int_delay)
2872                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2873
2874         /* enable Report Status bit */
2875         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2876
2877         hw->mac.ops.config_collision_dist(hw);
2878 }
2879
2880 /**
2881  * e1000_setup_rctl - configure the receive control registers
2882  * @adapter: Board private structure
2883  **/
2884 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2885                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2886 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2887 {
2888         struct e1000_hw *hw = &adapter->hw;
2889         u32 rctl, rfctl;
2890         u32 pages = 0;
2891
2892         /* Workaround Si errata on PCHx - configure jumbo frame flow */
2893         if (hw->mac.type >= e1000_pch2lan) {
2894                 s32 ret_val;
2895
2896                 if (adapter->netdev->mtu > ETH_DATA_LEN)
2897                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2898                 else
2899                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2900
2901                 if (ret_val)
2902                         e_dbg("failed to enable jumbo frame workaround mode\n");
2903         }
2904
2905         /* Program MC offset vector base */
2906         rctl = er32(RCTL);
2907         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2908         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2909                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2910                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2911
2912         /* Do not Store bad packets */
2913         rctl &= ~E1000_RCTL_SBP;
2914
2915         /* Enable Long Packet receive */
2916         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2917                 rctl &= ~E1000_RCTL_LPE;
2918         else
2919                 rctl |= E1000_RCTL_LPE;
2920
2921         /* Some systems expect that the CRC is included in SMBUS traffic. The
2922          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2923          * host memory when this is enabled
2924          */
2925         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2926                 rctl |= E1000_RCTL_SECRC;
2927
2928         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2929         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2930                 u16 phy_data;
2931
2932                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2933                 phy_data &= 0xfff8;
2934                 phy_data |= (1 << 2);
2935                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2936
2937                 e1e_rphy(hw, 22, &phy_data);
2938                 phy_data &= 0x0fff;
2939                 phy_data |= (1 << 14);
2940                 e1e_wphy(hw, 0x10, 0x2823);
2941                 e1e_wphy(hw, 0x11, 0x0003);
2942                 e1e_wphy(hw, 22, phy_data);
2943         }
2944
2945         /* Setup buffer sizes */
2946         rctl &= ~E1000_RCTL_SZ_4096;
2947         rctl |= E1000_RCTL_BSEX;
2948         switch (adapter->rx_buffer_len) {
2949         case 2048:
2950         default:
2951                 rctl |= E1000_RCTL_SZ_2048;
2952                 rctl &= ~E1000_RCTL_BSEX;
2953                 break;
2954         case 4096:
2955                 rctl |= E1000_RCTL_SZ_4096;
2956                 break;
2957         case 8192:
2958                 rctl |= E1000_RCTL_SZ_8192;
2959                 break;
2960         case 16384:
2961                 rctl |= E1000_RCTL_SZ_16384;
2962                 break;
2963         }
2964
2965         /* Enable Extended Status in all Receive Descriptors */
2966         rfctl = er32(RFCTL);
2967         rfctl |= E1000_RFCTL_EXTEN;
2968         ew32(RFCTL, rfctl);
2969
2970         /*
2971          * 82571 and greater support packet-split where the protocol
2972          * header is placed in skb->data and the packet data is
2973          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2974          * In the case of a non-split, skb->data is linearly filled,
2975          * followed by the page buffers.  Therefore, skb->data is
2976          * sized to hold the largest protocol header.
2977          *
2978          * allocations using alloc_page take too long for regular MTU
2979          * so only enable packet split for jumbo frames
2980          *
2981          * Using pages when the page size is greater than 16k wastes
2982          * a lot of memory, since we allocate 3 pages at all times
2983          * per packet.
2984          */
2985         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2986         if ((pages <= 3) && (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2987                 adapter->rx_ps_pages = pages;
2988         else
2989                 adapter->rx_ps_pages = 0;
2990
2991         if (adapter->rx_ps_pages) {
2992                 u32 psrctl = 0;
2993
2994                 /* Enable Packet split descriptors */
2995                 rctl |= E1000_RCTL_DTYP_PS;
2996
2997                 psrctl |= adapter->rx_ps_bsize0 >>
2998                         E1000_PSRCTL_BSIZE0_SHIFT;
2999
3000                 switch (adapter->rx_ps_pages) {
3001                 case 3:
3002                         psrctl |= PAGE_SIZE <<
3003                                 E1000_PSRCTL_BSIZE3_SHIFT;
3004                 case 2:
3005                         psrctl |= PAGE_SIZE <<
3006                                 E1000_PSRCTL_BSIZE2_SHIFT;
3007                 case 1:
3008                         psrctl |= PAGE_SIZE >>
3009                                 E1000_PSRCTL_BSIZE1_SHIFT;
3010                         break;
3011                 }
3012
3013                 ew32(PSRCTL, psrctl);
3014         }
3015
3016         /* This is useful for sniffing bad packets. */
3017         if (adapter->netdev->features & NETIF_F_RXALL) {
3018                 /* UPE and MPE will be handled by normal PROMISC logic
3019                  * in e1000e_set_rx_mode */
3020                 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
3021                          E1000_RCTL_BAM | /* RX All Bcast Pkts */
3022                          E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
3023
3024                 rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
3025                           E1000_RCTL_DPF | /* Allow filtered pause */
3026                           E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
3027                 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3028                  * and that breaks VLANs.
3029                  */
3030         }
3031
3032         ew32(RCTL, rctl);
3033         /* just started the receive unit, no need to restart */
3034         adapter->flags &= ~FLAG_RX_RESTART_NOW;
3035 }
3036
3037 /**
3038  * e1000_configure_rx - Configure Receive Unit after Reset
3039  * @adapter: board private structure
3040  *
3041  * Configure the Rx unit of the MAC after a reset.
3042  **/
3043 static void e1000_configure_rx(struct e1000_adapter *adapter)
3044 {
3045         struct e1000_hw *hw = &adapter->hw;
3046         struct e1000_ring *rx_ring = adapter->rx_ring;
3047         u64 rdba;
3048         u32 rdlen, rctl, rxcsum, ctrl_ext;
3049
3050         if (adapter->rx_ps_pages) {
3051                 /* this is a 32 byte descriptor */
3052                 rdlen = rx_ring->count *
3053                     sizeof(union e1000_rx_desc_packet_split);
3054                 adapter->clean_rx = e1000_clean_rx_irq_ps;
3055                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
3056         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
3057                 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3058                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
3059                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
3060         } else {
3061                 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3062                 adapter->clean_rx = e1000_clean_rx_irq;
3063                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
3064         }
3065
3066         /* disable receives while setting up the descriptors */
3067         rctl = er32(RCTL);
3068         if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3069                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3070         e1e_flush();
3071         usleep_range(10000, 20000);
3072
3073         if (adapter->flags2 & FLAG2_DMA_BURST) {
3074                 /*
3075                  * set the writeback threshold (only takes effect if the RDTR
3076                  * is set). set GRAN=1 and write back up to 0x4 worth, and
3077                  * enable prefetching of 0x20 Rx descriptors
3078                  * granularity = 01
3079                  * wthresh = 04,
3080                  * hthresh = 04,
3081                  * pthresh = 0x20
3082                  */
3083                 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
3084                 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
3085
3086                 /*
3087                  * override the delay timers for enabling bursting, only if
3088                  * the value was not set by the user via module options
3089                  */
3090                 if (adapter->rx_int_delay == DEFAULT_RDTR)
3091                         adapter->rx_int_delay = BURST_RDTR;
3092                 if (adapter->rx_abs_int_delay == DEFAULT_RADV)
3093                         adapter->rx_abs_int_delay = BURST_RADV;
3094         }
3095
3096         /* set the Receive Delay Timer Register */
3097         ew32(RDTR, adapter->rx_int_delay);
3098
3099         /* irq moderation */
3100         ew32(RADV, adapter->rx_abs_int_delay);
3101         if ((adapter->itr_setting != 0) && (adapter->itr != 0))
3102                 e1000e_write_itr(adapter, adapter->itr);
3103
3104         ctrl_ext = er32(CTRL_EXT);
3105         /* Auto-Mask interrupts upon ICR access */
3106         ctrl_ext |= E1000_CTRL_EXT_IAME;
3107         ew32(IAM, 0xffffffff);
3108         ew32(CTRL_EXT, ctrl_ext);
3109         e1e_flush();
3110
3111         /*
3112          * Setup the HW Rx Head and Tail Descriptor Pointers and
3113          * the Base and Length of the Rx Descriptor Ring
3114          */
3115         rdba = rx_ring->dma;
3116         ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
3117         ew32(RDBAH(0), (rdba >> 32));
3118         ew32(RDLEN(0), rdlen);
3119         ew32(RDH(0), 0);
3120         ew32(RDT(0), 0);
3121         rx_ring->head = adapter->hw.hw_addr + E1000_RDH(0);
3122         rx_ring->tail = adapter->hw.hw_addr + E1000_RDT(0);
3123
3124         /* Enable Receive Checksum Offload for TCP and UDP */
3125         rxcsum = er32(RXCSUM);
3126         if (adapter->netdev->features & NETIF_F_RXCSUM)
3127                 rxcsum |= E1000_RXCSUM_TUOFL;
3128         else
3129                 rxcsum &= ~E1000_RXCSUM_TUOFL;
3130         ew32(RXCSUM, rxcsum);
3131
3132         if (adapter->hw.mac.type == e1000_pch2lan) {
3133                 /*
3134                  * With jumbo frames, excessive C-state transition
3135                  * latencies result in dropped transactions.
3136                  */
3137                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3138                         u32 rxdctl = er32(RXDCTL(0));
3139                         ew32(RXDCTL(0), rxdctl | 0x3);
3140                         pm_qos_update_request(&adapter->netdev->pm_qos_req, 55);
3141                 } else {
3142                         pm_qos_update_request(&adapter->netdev->pm_qos_req,
3143                                               PM_QOS_DEFAULT_VALUE);
3144                 }
3145         }
3146
3147         /* Enable Receives */
3148         ew32(RCTL, rctl);
3149 }
3150
3151 /**
3152  * e1000e_write_mc_addr_list - write multicast addresses to MTA
3153  * @netdev: network interface device structure
3154  *
3155  * Writes multicast address list to the MTA hash table.
3156  * Returns: -ENOMEM on failure
3157  *                0 on no addresses written
3158  *                X on writing X addresses to MTA
3159  */
3160 static int e1000e_write_mc_addr_list(struct net_device *netdev)
3161 {
3162         struct e1000_adapter *adapter = netdev_priv(netdev);
3163         struct e1000_hw *hw = &adapter->hw;
3164         struct netdev_hw_addr *ha;
3165         u8 *mta_list;
3166         int i;
3167
3168         if (netdev_mc_empty(netdev)) {
3169                 /* nothing to program, so clear mc list */
3170                 hw->mac.ops.update_mc_addr_list(hw, NULL, 0);
3171                 return 0;
3172         }
3173
3174         mta_list = kzalloc(netdev_mc_count(netdev) * ETH_ALEN, GFP_ATOMIC);
3175         if (!mta_list)
3176                 return -ENOMEM;
3177
3178         /* update_mc_addr_list expects a packed array of only addresses. */
3179         i = 0;
3180         netdev_for_each_mc_addr(ha, netdev)
3181                 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3182
3183         hw->mac.ops.update_mc_addr_list(hw, mta_list, i);
3184         kfree(mta_list);
3185
3186         return netdev_mc_count(netdev);
3187 }
3188
3189 /**
3190  * e1000e_write_uc_addr_list - write unicast addresses to RAR table
3191  * @netdev: network interface device structure
3192  *
3193  * Writes unicast address list to the RAR table.
3194  * Returns: -ENOMEM on failure/insufficient address space
3195  *                0 on no addresses written
3196  *                X on writing X addresses to the RAR table
3197  **/
3198 static int e1000e_write_uc_addr_list(struct net_device *netdev)
3199 {
3200         struct e1000_adapter *adapter = netdev_priv(netdev);
3201         struct e1000_hw *hw = &adapter->hw;
3202         unsigned int rar_entries = hw->mac.rar_entry_count;
3203         int count = 0;
3204
3205         /* save a rar entry for our hardware address */
3206         rar_entries--;
3207
3208         /* save a rar entry for the LAA workaround */
3209         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA)
3210                 rar_entries--;
3211
3212         /* return ENOMEM indicating insufficient memory for addresses */
3213         if (netdev_uc_count(netdev) > rar_entries)
3214                 return -ENOMEM;
3215
3216         if (!netdev_uc_empty(netdev) && rar_entries) {
3217                 struct netdev_hw_addr *ha;
3218
3219                 /*
3220                  * write the addresses in reverse order to avoid write
3221                  * combining
3222                  */
3223                 netdev_for_each_uc_addr(ha, netdev) {
3224                         if (!rar_entries)
3225                                 break;
3226                         hw->mac.ops.rar_set(hw, ha->addr, rar_entries--);
3227                         count++;
3228                 }
3229         }
3230
3231         /* zero out the remaining RAR entries not used above */
3232         for (; rar_entries > 0; rar_entries--) {
3233                 ew32(RAH(rar_entries), 0);
3234                 ew32(RAL(rar_entries), 0);
3235         }
3236         e1e_flush();
3237
3238         return count;
3239 }
3240
3241 /**
3242  * e1000e_set_rx_mode - secondary unicast, Multicast and Promiscuous mode set
3243  * @netdev: network interface device structure
3244  *
3245  * The ndo_set_rx_mode entry point is called whenever the unicast or multicast
3246  * address list or the network interface flags are updated.  This routine is
3247  * responsible for configuring the hardware for proper unicast, multicast,
3248  * promiscuous mode, and all-multi behavior.
3249  **/
3250 static void e1000e_set_rx_mode(struct net_device *netdev)
3251 {
3252         struct e1000_adapter *adapter = netdev_priv(netdev);
3253         struct e1000_hw *hw = &adapter->hw;
3254         u32 rctl;
3255
3256         /* Check for Promiscuous and All Multicast modes */
3257         rctl = er32(RCTL);
3258
3259         /* clear the affected bits */
3260         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3261
3262         if (netdev->flags & IFF_PROMISC) {
3263                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3264                 /* Do not hardware filter VLANs in promisc mode */
3265                 e1000e_vlan_filter_disable(adapter);
3266         } else {
3267                 int count;
3268
3269                 if (netdev->flags & IFF_ALLMULTI) {
3270                         rctl |= E1000_RCTL_MPE;
3271                 } else {
3272                         /*
3273                          * Write addresses to the MTA, if the attempt fails
3274                          * then we should just turn on promiscuous mode so
3275                          * that we can at least receive multicast traffic
3276                          */
3277                         count = e1000e_write_mc_addr_list(netdev);
3278                         if (count < 0)
3279                                 rctl |= E1000_RCTL_MPE;
3280                 }
3281                 e1000e_vlan_filter_enable(adapter);
3282                 /*
3283                  * Write addresses to available RAR registers, if there is not
3284                  * sufficient space to store all the addresses then enable
3285                  * unicast promiscuous mode
3286                  */
3287                 count = e1000e_write_uc_addr_list(netdev);
3288                 if (count < 0)
3289                         rctl |= E1000_RCTL_UPE;
3290         }
3291
3292         ew32(RCTL, rctl);
3293
3294         if (netdev->features & NETIF_F_HW_VLAN_RX)
3295                 e1000e_vlan_strip_enable(adapter);
3296         else
3297                 e1000e_vlan_strip_disable(adapter);
3298 }
3299
3300 static void e1000e_setup_rss_hash(struct e1000_adapter *adapter)
3301 {
3302         struct e1000_hw *hw = &adapter->hw;
3303         u32 mrqc, rxcsum;
3304         int i;
3305         static const u32 rsskey[10] = {
3306                 0xda565a6d, 0xc20e5b25, 0x3d256741, 0xb08fa343, 0xcb2bcad0,
3307                 0xb4307bae, 0xa32dcb77, 0x0cf23080, 0x3bb7426a, 0xfa01acbe
3308         };
3309
3310         /* Fill out hash function seed */
3311         for (i = 0; i < 10; i++)
3312                 ew32(RSSRK(i), rsskey[i]);
3313
3314         /* Direct all traffic to queue 0 */
3315         for (i = 0; i < 32; i++)
3316                 ew32(RETA(i), 0);
3317
3318         /*
3319          * Disable raw packet checksumming so that RSS hash is placed in
3320          * descriptor on writeback.
3321          */
3322         rxcsum = er32(RXCSUM);
3323         rxcsum |= E1000_RXCSUM_PCSD;
3324
3325         ew32(RXCSUM, rxcsum);
3326
3327         mrqc = (E1000_MRQC_RSS_FIELD_IPV4 |
3328                 E1000_MRQC_RSS_FIELD_IPV4_TCP |
3329                 E1000_MRQC_RSS_FIELD_IPV6 |
3330                 E1000_MRQC_RSS_FIELD_IPV6_TCP |
3331                 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
3332
3333         ew32(MRQC, mrqc);
3334 }
3335
3336 /**
3337  * e1000_configure - configure the hardware for Rx and Tx
3338  * @adapter: private board structure
3339  **/
3340 static void e1000_configure(struct e1000_adapter *adapter)
3341 {
3342         struct e1000_ring *rx_ring = adapter->rx_ring;
3343
3344         e1000e_set_rx_mode(adapter->netdev);
3345
3346         e1000_restore_vlan(adapter);
3347         e1000_init_manageability_pt(adapter);
3348
3349         e1000_configure_tx(adapter);
3350
3351         if (adapter->netdev->features & NETIF_F_RXHASH)
3352                 e1000e_setup_rss_hash(adapter);
3353         e1000_setup_rctl(adapter);
3354         e1000_configure_rx(adapter);
3355         adapter->alloc_rx_buf(rx_ring, e1000_desc_unused(rx_ring), GFP_KERNEL);
3356 }
3357
3358 /**
3359  * e1000e_power_up_phy - restore link in case the phy was powered down
3360  * @adapter: address of board private structure
3361  *
3362  * The phy may be powered down to save power and turn off link when the
3363  * driver is unloaded and wake on lan is not enabled (among others)
3364  * *** this routine MUST be followed by a call to e1000e_reset ***
3365  **/
3366 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3367 {
3368         if (adapter->hw.phy.ops.power_up)
3369                 adapter->hw.phy.ops.power_up(&adapter->hw);
3370
3371         adapter->hw.mac.ops.setup_link(&adapter->hw);
3372 }
3373
3374 /**
3375  * e1000_power_down_phy - Power down the PHY
3376  *
3377  * Power down the PHY so no link is implied when interface is down.
3378  * The PHY cannot be powered down if management or WoL is active.
3379  */
3380 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3381 {
3382         /* WoL is enabled */
3383         if (adapter->wol)
3384                 return;
3385
3386         if (adapter->hw.phy.ops.power_down)
3387                 adapter->hw.phy.ops.power_down(&adapter->hw);
3388 }
3389
3390 /**
3391  * e1000e_reset - bring the hardware into a known good state
3392  *
3393  * This function boots the hardware and enables some settings that
3394  * require a configuration cycle of the hardware - those cannot be
3395  * set/changed during runtime. After reset the device needs to be
3396  * properly configured for Rx, Tx etc.
3397  */
3398 void e1000e_reset(struct e1000_adapter *adapter)
3399 {
3400         struct e1000_mac_info *mac = &adapter->hw.mac;
3401         struct e1000_fc_info *fc = &adapter->hw.fc;
3402         struct e1000_hw *hw = &adapter->hw;
3403         u32 tx_space, min_tx_space, min_rx_space;
3404         u32 pba = adapter->pba;
3405         u16 hwm;
3406
3407         /* reset Packet Buffer Allocation to default */
3408         ew32(PBA, pba);
3409
3410         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3411                 /*
3412                  * To maintain wire speed transmits, the Tx FIFO should be
3413                  * large enough to accommodate two full transmit packets,
3414                  * rounded up to the next 1KB and expressed in KB.  Likewise,
3415                  * the Rx FIFO should be large enough to accommodate at least
3416                  * one full receive packet and is similarly rounded up and
3417                  * expressed in KB.
3418                  */
3419                 pba = er32(PBA);
3420                 /* upper 16 bits has Tx packet buffer allocation size in KB */
3421                 tx_space = pba >> 16;
3422                 /* lower 16 bits has Rx packet buffer allocation size in KB */
3423                 pba &= 0xffff;
3424                 /*
3425                  * the Tx fifo also stores 16 bytes of information about the Tx
3426                  * but don't include ethernet FCS because hardware appends it
3427                  */
3428                 min_tx_space = (adapter->max_frame_size +
3429                                 sizeof(struct e1000_tx_desc) -
3430                                 ETH_FCS_LEN) * 2;
3431                 min_tx_space = ALIGN(min_tx_space, 1024);
3432                 min_tx_space >>= 10;
3433                 /* software strips receive CRC, so leave room for it */
3434                 min_rx_space = adapter->max_frame_size;
3435                 min_rx_space = ALIGN(min_rx_space, 1024);
3436                 min_rx_space >>= 10;
3437
3438                 /*
3439                  * If current Tx allocation is less than the min Tx FIFO size,
3440                  * and the min Tx FIFO size is less than the current Rx FIFO
3441                  * allocation, take space away from current Rx allocation
3442                  */
3443                 if ((tx_space < min_tx_space) &&
3444                     ((min_tx_space - tx_space) < pba)) {
3445                         pba -= min_tx_space - tx_space;
3446
3447                         /*
3448                          * if short on Rx space, Rx wins and must trump Tx
3449                          * adjustment or use Early Receive if available
3450                          */
3451                         if (pba < min_rx_space)
3452                                 pba = min_rx_space;
3453                 }
3454
3455                 ew32(PBA, pba);
3456         }
3457
3458         /*
3459          * flow control settings
3460          *
3461          * The high water mark must be low enough to fit one full frame
3462          * (or the size used for early receive) above it in the Rx FIFO.
3463          * Set it to the lower of:
3464          * - 90% of the Rx FIFO size, and
3465          * - the full Rx FIFO size minus one full frame
3466          */
3467         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3468                 fc->pause_time = 0xFFFF;
3469         else
3470                 fc->pause_time = E1000_FC_PAUSE_TIME;
3471         fc->send_xon = true;
3472         fc->current_mode = fc->requested_mode;
3473
3474         switch (hw->mac.type) {
3475         case e1000_ich9lan:
3476         case e1000_ich10lan:
3477                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3478                         pba = 14;
3479                         ew32(PBA, pba);
3480                         fc->high_water = 0x2800;
3481                         fc->low_water = fc->high_water - 8;
3482                         break;
3483                 }
3484                 /* fall-through */
3485         default:
3486                 hwm = min(((pba << 10) * 9 / 10),
3487                           ((pba << 10) - adapter->max_frame_size));
3488
3489                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3490                 fc->low_water = fc->high_water - 8;
3491                 break;
3492         case e1000_pchlan:
3493                 /*
3494                  * Workaround PCH LOM adapter hangs with certain network
3495                  * loads.  If hangs persist, try disabling Tx flow control.
3496                  */
3497                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3498                         fc->high_water = 0x3500;
3499                         fc->low_water  = 0x1500;
3500                 } else {
3501                         fc->high_water = 0x5000;
3502                         fc->low_water  = 0x3000;
3503                 }
3504                 fc->refresh_time = 0x1000;
3505                 break;
3506         case e1000_pch2lan:
3507         case e1000_pch_lpt:
3508                 fc->high_water = 0x05C20;
3509                 fc->low_water = 0x05048;
3510                 fc->pause_time = 0x0650;
3511                 fc->refresh_time = 0x0400;
3512                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3513                         pba = 14;
3514                         ew32(PBA, pba);
3515                 }
3516                 break;
3517         }
3518
3519         /*
3520          * Alignment of Tx data is on an arbitrary byte boundary with the
3521          * maximum size per Tx descriptor limited only to the transmit
3522          * allocation of the packet buffer minus 96 bytes with an upper
3523          * limit of 24KB due to receive synchronization limitations.
3524          */
3525         adapter->tx_fifo_limit = min_t(u32, ((er32(PBA) >> 16) << 10) - 96,
3526                                        24 << 10);
3527
3528         /*
3529          * Disable Adaptive Interrupt Moderation if 2 full packets cannot
3530          * fit in receive buffer.
3531          */
3532         if (adapter->itr_setting & 0x3) {
3533                 if ((adapter->max_frame_size * 2) > (pba << 10)) {
3534                         if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
3535                                 dev_info(&adapter->pdev->dev,
3536                                         "Interrupt Throttle Rate turned off\n");
3537                                 adapter->flags2 |= FLAG2_DISABLE_AIM;
3538                                 e1000e_write_itr(adapter, 0);
3539                         }
3540                 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
3541                         dev_info(&adapter->pdev->dev,
3542                                  "Interrupt Throttle Rate turned on\n");
3543                         adapter->flags2 &= ~FLAG2_DISABLE_AIM;
3544                         adapter->itr = 20000;
3545                         e1000e_write_itr(adapter, adapter->itr);
3546                 }
3547         }
3548
3549         /* Allow time for pending master requests to run */
3550         mac->ops.reset_hw(hw);
3551
3552         /*
3553          * For parts with AMT enabled, let the firmware know
3554          * that the network interface is in control
3555          */
3556         if (adapter->flags & FLAG_HAS_AMT)
3557                 e1000e_get_hw_control(adapter);
3558
3559         ew32(WUC, 0);
3560
3561         if (mac->ops.init_hw(hw))
3562                 e_err("Hardware Error\n");
3563
3564         e1000_update_mng_vlan(adapter);
3565
3566         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3567         ew32(VET, ETH_P_8021Q);
3568
3569         e1000e_reset_adaptive(hw);
3570
3571         if (!netif_running(adapter->netdev) &&
3572             !test_bit(__E1000_TESTING, &adapter->state)) {
3573                 e1000_power_down_phy(adapter);
3574                 return;
3575         }
3576
3577         e1000_get_phy_info(hw);
3578
3579         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3580             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3581                 u16 phy_data = 0;
3582                 /*
3583                  * speed up time to link by disabling smart power down, ignore
3584                  * the return value of this function because there is nothing
3585                  * different we would do if it failed
3586                  */
3587                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3588                 phy_data &= ~IGP02E1000_PM_SPD;
3589                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3590         }
3591 }
3592
3593 int e1000e_up(struct e1000_adapter *adapter)
3594 {
3595         struct e1000_hw *hw = &adapter->hw;
3596
3597         /* hardware has been reset, we need to reload some things */
3598         e1000_configure(adapter);
3599
3600         clear_bit(__E1000_DOWN, &adapter->state);
3601
3602         if (adapter->msix_entries)
3603                 e1000_configure_msix(adapter);
3604         e1000_irq_enable(adapter);
3605
3606         netif_start_queue(adapter->netdev);
3607
3608         /* fire a link change interrupt to start the watchdog */
3609         if (adapter->msix_entries)
3610                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3611         else
3612                 ew32(ICS, E1000_ICS_LSC);
3613
3614         return 0;
3615 }
3616
3617 static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
3618 {
3619         struct e1000_hw *hw = &adapter->hw;
3620
3621         if (!(adapter->flags2 & FLAG2_DMA_BURST))
3622                 return;
3623
3624         /* flush pending descriptor writebacks to memory */
3625         ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3626         ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3627
3628         /* execute the writes immediately */
3629         e1e_flush();
3630
3631         /*
3632          * due to rare timing issues, write to TIDV/RDTR again to ensure the
3633          * write is successful
3634          */
3635         ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3636         ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3637
3638         /* execute the writes immediately */
3639         e1e_flush();
3640 }
3641
3642 static void e1000e_update_stats(struct e1000_adapter *adapter);
3643
3644 void e1000e_down(struct e1000_adapter *adapter)
3645 {
3646         struct net_device *netdev = adapter->netdev;
3647         struct e1000_hw *hw = &adapter->hw;
3648         u32 tctl, rctl;
3649
3650         /*
3651          * signal that we're down so the interrupt handler does not
3652          * reschedule our watchdog timer
3653          */
3654         set_bit(__E1000_DOWN, &adapter->state);
3655
3656         /* disable receives in the hardware */
3657         rctl = er32(RCTL);
3658         if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3659                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3660         /* flush and sleep below */
3661
3662         netif_stop_queue(netdev);
3663
3664         /* disable transmits in the hardware */
3665         tctl = er32(TCTL);
3666         tctl &= ~E1000_TCTL_EN;
3667         ew32(TCTL, tctl);
3668
3669         /* flush both disables and wait for them to finish */
3670         e1e_flush();
3671         usleep_range(10000, 20000);
3672
3673         e1000_irq_disable(adapter);
3674
3675         del_timer_sync(&adapter->watchdog_timer);
3676         del_timer_sync(&adapter->phy_info_timer);
3677
3678         netif_carrier_off(netdev);
3679
3680         spin_lock(&adapter->stats64_lock);
3681         e1000e_update_stats(adapter);
3682         spin_unlock(&adapter->stats64_lock);
3683
3684         e1000e_flush_descriptors(adapter);
3685         e1000_clean_tx_ring(adapter->tx_ring);
3686         e1000_clean_rx_ring(adapter->rx_ring);
3687
3688         adapter->link_speed = 0;
3689         adapter->link_duplex = 0;
3690
3691         if (!pci_channel_offline(adapter->pdev))
3692                 e1000e_reset(adapter);
3693
3694         /*
3695          * TODO: for power management, we could drop the link and
3696          * pci_disable_device here.
3697          */
3698 }
3699
3700 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3701 {
3702         might_sleep();
3703         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3704                 usleep_range(1000, 2000);
3705         e1000e_down(adapter);
3706         e1000e_up(adapter);
3707         clear_bit(__E1000_RESETTING, &adapter->state);
3708 }
3709
3710 /**
3711  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3712  * @adapter: board private structure to initialize
3713  *
3714  * e1000_sw_init initializes the Adapter private data structure.
3715  * Fields are initialized based on PCI device information and
3716  * OS network device settings (MTU size).
3717  **/
3718 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3719 {
3720         struct net_device *netdev = adapter->netdev;
3721
3722         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3723         adapter->rx_ps_bsize0 = 128;
3724         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3725         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3726         adapter->tx_ring_count = E1000_DEFAULT_TXD;
3727         adapter->rx_ring_count = E1000_DEFAULT_RXD;
3728
3729         spin_lock_init(&adapter->stats64_lock);
3730
3731         e1000e_set_interrupt_capability(adapter);
3732
3733         if (e1000_alloc_queues(adapter))
3734                 return -ENOMEM;
3735
3736         /* Explicitly disable IRQ since the NIC can be in any state. */
3737         e1000_irq_disable(adapter);
3738
3739         set_bit(__E1000_DOWN, &adapter->state);
3740         return 0;
3741 }
3742
3743 /**
3744  * e1000_intr_msi_test - Interrupt Handler
3745  * @irq: interrupt number
3746  * @data: pointer to a network interface device structure
3747  **/
3748 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3749 {
3750         struct net_device *netdev = data;
3751         struct e1000_adapter *adapter = netdev_priv(netdev);
3752         struct e1000_hw *hw = &adapter->hw;
3753         u32 icr = er32(ICR);
3754
3755         e_dbg("icr is %08X\n", icr);
3756         if (icr & E1000_ICR_RXSEQ) {
3757                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3758                 wmb();
3759         }
3760
3761         return IRQ_HANDLED;
3762 }
3763
3764 /**
3765  * e1000_test_msi_interrupt - Returns 0 for successful test
3766  * @adapter: board private struct
3767  *
3768  * code flow taken from tg3.c
3769  **/
3770 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3771 {
3772         struct net_device *netdev = adapter->netdev;
3773         struct e1000_hw *hw = &adapter->hw;
3774         int err;
3775
3776         /* poll_enable hasn't been called yet, so don't need disable */
3777         /* clear any pending events */
3778         er32(ICR);
3779
3780         /* free the real vector and request a test handler */
3781         e1000_free_irq(adapter);
3782         e1000e_reset_interrupt_capability(adapter);
3783
3784         /* Assume that the test fails, if it succeeds then the test
3785          * MSI irq handler will unset this flag */
3786         adapter->flags |= FLAG_MSI_TEST_FAILED;
3787
3788         err = pci_enable_msi(adapter->pdev);
3789         if (err)
3790                 goto msi_test_failed;
3791
3792         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3793                           netdev->name, netdev);
3794         if (err) {
3795                 pci_disable_msi(adapter->pdev);
3796                 goto msi_test_failed;
3797         }
3798
3799         wmb();
3800
3801         e1000_irq_enable(adapter);
3802
3803         /* fire an unusual interrupt on the test handler */
3804         ew32(ICS, E1000_ICS_RXSEQ);
3805         e1e_flush();
3806         msleep(100);
3807
3808         e1000_irq_disable(adapter);
3809
3810         rmb();
3811
3812         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3813                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3814                 e_info("MSI interrupt test failed, using legacy interrupt.\n");
3815         } else {
3816                 e_dbg("MSI interrupt test succeeded!\n");
3817         }
3818
3819         free_irq(adapter->pdev->irq, netdev);
3820         pci_disable_msi(adapter->pdev);
3821
3822 msi_test_failed:
3823         e1000e_set_interrupt_capability(adapter);
3824         return e1000_request_irq(adapter);
3825 }
3826
3827 /**
3828  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3829  * @adapter: board private struct
3830  *
3831  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3832  **/
3833 static int e1000_test_msi(struct e1000_adapter *adapter)
3834 {
3835         int err;
3836         u16 pci_cmd;
3837
3838         if (!(adapter->flags & FLAG_MSI_ENABLED))
3839                 return 0;
3840
3841         /* disable SERR in case the MSI write causes a master abort */
3842         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3843         if (pci_cmd & PCI_COMMAND_SERR)
3844                 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3845                                       pci_cmd & ~PCI_COMMAND_SERR);
3846
3847         err = e1000_test_msi_interrupt(adapter);
3848
3849         /* re-enable SERR */
3850         if (pci_cmd & PCI_COMMAND_SERR) {
3851                 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3852                 pci_cmd |= PCI_COMMAND_SERR;
3853                 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3854         }
3855
3856         return err;
3857 }
3858
3859 /**
3860  * e1000_open - Called when a network interface is made active
3861  * @netdev: network interface device structure
3862  *
3863  * Returns 0 on success, negative value on failure
3864  *
3865  * The open entry point is called when a network interface is made
3866  * active by the system (IFF_UP).  At this point all resources needed
3867  * for transmit and receive operations are allocated, the interrupt
3868  * handler is registered with the OS, the watchdog timer is started,
3869  * and the stack is notified that the interface is ready.
3870  **/
3871 static int e1000_open(struct net_device *netdev)
3872 {
3873         struct e1000_adapter *adapter = netdev_priv(netdev);
3874         struct e1000_hw *hw = &adapter->hw;
3875         struct pci_dev *pdev = adapter->pdev;
3876         int err;
3877
3878         /* disallow open during test */
3879         if (test_bit(__E1000_TESTING, &adapter->state))
3880                 return -EBUSY;
3881
3882         pm_runtime_get_sync(&pdev->dev);
3883
3884         netif_carrier_off(netdev);
3885
3886         /* allocate transmit descriptors */
3887         err = e1000e_setup_tx_resources(adapter->tx_ring);
3888         if (err)
3889                 goto err_setup_tx;
3890
3891         /* allocate receive descriptors */
3892         err = e1000e_setup_rx_resources(adapter->rx_ring);
3893         if (err)
3894                 goto err_setup_rx;
3895
3896         /*
3897          * If AMT is enabled, let the firmware know that the network
3898          * interface is now open and reset the part to a known state.
3899          */
3900         if (adapter->flags & FLAG_HAS_AMT) {
3901                 e1000e_get_hw_control(adapter);
3902                 e1000e_reset(adapter);
3903         }
3904
3905         e1000e_power_up_phy(adapter);
3906
3907         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3908         if ((adapter->hw.mng_cookie.status &
3909              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3910                 e1000_update_mng_vlan(adapter);
3911
3912         /* DMA latency requirement to workaround jumbo issue */
3913         if (adapter->hw.mac.type == e1000_pch2lan)
3914                 pm_qos_add_request(&adapter->netdev->pm_qos_req,
3915                                    PM_QOS_CPU_DMA_LATENCY,
3916                                    PM_QOS_DEFAULT_VALUE);
3917
3918         /*
3919          * before we allocate an interrupt, we must be ready to handle it.
3920          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3921          * as soon as we call pci_request_irq, so we have to setup our
3922          * clean_rx handler before we do so.
3923          */
3924         e1000_configure(adapter);
3925
3926         err = e1000_request_irq(adapter);
3927         if (err)
3928                 goto err_req_irq;
3929
3930         /*
3931          * Work around PCIe errata with MSI interrupts causing some chipsets to
3932          * ignore e1000e MSI messages, which means we need to test our MSI
3933          * interrupt now
3934          */
3935         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3936                 err = e1000_test_msi(adapter);
3937                 if (err) {
3938                         e_err("Interrupt allocation failed\n");
3939                         goto err_req_irq;
3940                 }
3941         }
3942
3943         /* From here on the code is the same as e1000e_up() */
3944         clear_bit(__E1000_DOWN, &adapter->state);
3945
3946         napi_enable(&adapter->napi);
3947
3948         e1000_irq_enable(adapter);
3949
3950         adapter->tx_hang_recheck = false;
3951         netif_start_queue(netdev);
3952
3953         adapter->idle_check = true;
3954         pm_runtime_put(&pdev->dev);
3955
3956         /* fire a link status change interrupt to start the watchdog */
3957         if (adapter->msix_entries)
3958                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3959         else
3960                 ew32(ICS, E1000_ICS_LSC);
3961
3962         return 0;
3963
3964 err_req_irq:
3965         e1000e_release_hw_control(adapter);
3966         e1000_power_down_phy(adapter);
3967         e1000e_free_rx_resources(adapter->rx_ring);
3968 err_setup_rx:
3969         e1000e_free_tx_resources(adapter->tx_ring);
3970 err_setup_tx:
3971         e1000e_reset(adapter);
3972         pm_runtime_put_sync(&pdev->dev);
3973
3974         return err;
3975 }
3976
3977 /**
3978  * e1000_close - Disables a network interface
3979  * @netdev: network interface device structure
3980  *
3981  * Returns 0, this is not allowed to fail
3982  *
3983  * The close entry point is called when an interface is de-activated
3984  * by the OS.  The hardware is still under the drivers control, but
3985  * needs to be disabled.  A global MAC reset is issued to stop the
3986  * hardware, and all transmit and receive resources are freed.
3987  **/
3988 static int e1000_close(struct net_device *netdev)
3989 {
3990         struct e1000_adapter *adapter = netdev_priv(netdev);
3991         struct pci_dev *pdev = adapter->pdev;
3992         int count = E1000_CHECK_RESET_COUNT;
3993
3994         while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
3995                 usleep_range(10000, 20000);
3996
3997         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3998
3999         pm_runtime_get_sync(&pdev->dev);
4000
4001         napi_disable(&adapter->napi);
4002
4003         if (!test_bit(__E1000_DOWN, &adapter->state)) {
4004                 e1000e_down(adapter);
4005                 e1000_free_irq(adapter);
4006         }
4007         e1000_power_down_phy(adapter);
4008
4009         e1000e_free_tx_resources(adapter->tx_ring);
4010         e1000e_free_rx_resources(adapter->rx_ring);
4011
4012         /*
4013          * kill manageability vlan ID if supported, but not if a vlan with
4014          * the same ID is registered on the host OS (let 8021q kill it)
4015          */
4016         if (adapter->hw.mng_cookie.status &
4017             E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
4018                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
4019
4020         /*
4021          * If AMT is enabled, let the firmware know that the network
4022          * interface is now closed
4023          */
4024         if ((adapter->flags & FLAG_HAS_AMT) &&
4025             !test_bit(__E1000_TESTING, &adapter->state))
4026                 e1000e_release_hw_control(adapter);
4027
4028         if (adapter->hw.mac.type == e1000_pch2lan)
4029                 pm_qos_remove_request(&adapter->netdev->pm_qos_req);
4030
4031         pm_runtime_put_sync(&pdev->dev);
4032
4033         return 0;
4034 }
4035 /**
4036  * e1000_set_mac - Change the Ethernet Address of the NIC
4037  * @netdev: network interface device structure
4038  * @p: pointer to an address structure
4039  *
4040  * Returns 0 on success, negative on failure
4041  **/
4042 static int e1000_set_mac(struct net_device *netdev, void *p)
4043 {
4044         struct e1000_adapter *adapter = netdev_priv(netdev);
4045         struct e1000_hw *hw = &adapter->hw;
4046         struct sockaddr *addr = p;
4047
4048         if (!is_valid_ether_addr(addr->sa_data))
4049                 return -EADDRNOTAVAIL;
4050
4051         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
4052         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
4053
4054         hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
4055
4056         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
4057                 /* activate the work around */
4058                 e1000e_set_laa_state_82571(&adapter->hw, 1);
4059
4060                 /*
4061                  * Hold a copy of the LAA in RAR[14] This is done so that
4062                  * between the time RAR[0] gets clobbered  and the time it
4063                  * gets fixed (in e1000_watchdog), the actual LAA is in one
4064                  * of the RARs and no incoming packets directed to this port
4065                  * are dropped. Eventually the LAA will be in RAR[0] and
4066                  * RAR[14]
4067                  */
4068                 hw->mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr,
4069                                     adapter->hw.mac.rar_entry_count - 1);
4070         }
4071
4072         return 0;
4073 }
4074
4075 /**
4076  * e1000e_update_phy_task - work thread to update phy
4077  * @work: pointer to our work struct
4078  *
4079  * this worker thread exists because we must acquire a
4080  * semaphore to read the phy, which we could msleep while
4081  * waiting for it, and we can't msleep in a timer.
4082  **/
4083 static void e1000e_update_phy_task(struct work_struct *work)
4084 {
4085         struct e1000_adapter *adapter = container_of(work,
4086                                         struct e1000_adapter, update_phy_task);
4087
4088         if (test_bit(__E1000_DOWN, &adapter->state))
4089                 return;
4090
4091         e1000_get_phy_info(&adapter->hw);
4092 }
4093
4094 /*
4095  * Need to wait a few seconds after link up to get diagnostic information from
4096  * the phy
4097  */
4098 static void e1000_update_phy_info(unsigned long data)
4099 {
4100         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4101
4102         if (test_bit(__E1000_DOWN, &adapter->state))
4103                 return;
4104
4105         schedule_work(&adapter->update_phy_task);
4106 }
4107
4108 /**
4109  * e1000e_update_phy_stats - Update the PHY statistics counters
4110  * @adapter: board private structure
4111  *
4112  * Read/clear the upper 16-bit PHY registers and read/accumulate lower
4113  **/
4114 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
4115 {
4116         struct e1000_hw *hw = &adapter->hw;
4117         s32 ret_val;
4118         u16 phy_data;
4119
4120         ret_val = hw->phy.ops.acquire(hw);
4121         if (ret_val)
4122                 return;
4123
4124         /*
4125          * A page set is expensive so check if already on desired page.
4126          * If not, set to the page with the PHY status registers.
4127          */
4128         hw->phy.addr = 1;
4129         ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4130                                            &phy_data);
4131         if (ret_val)
4132                 goto release;
4133         if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
4134                 ret_val = hw->phy.ops.set_page(hw,
4135                                                HV_STATS_PAGE << IGP_PAGE_SHIFT);
4136                 if (ret_val)
4137                         goto release;
4138         }
4139
4140         /* Single Collision Count */
4141         hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
4142         ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
4143         if (!ret_val)
4144                 adapter->stats.scc += phy_data;
4145
4146         /* Excessive Collision Count */
4147         hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
4148         ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
4149         if (!ret_val)
4150                 adapter->stats.ecol += phy_data;
4151
4152         /* Multiple Collision Count */
4153         hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
4154         ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
4155         if (!ret_val)
4156                 adapter->stats.mcc += phy_data;
4157
4158         /* Late Collision Count */
4159         hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
4160         ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
4161         if (!ret_val)
4162                 adapter->stats.latecol += phy_data;
4163
4164         /* Collision Count - also used for adaptive IFS */
4165         hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
4166         ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
4167         if (!ret_val)
4168                 hw->mac.collision_delta = phy_data;
4169
4170         /* Defer Count */
4171         hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
4172         ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
4173         if (!ret_val)
4174                 adapter->stats.dc += phy_data;
4175
4176         /* Transmit with no CRS */
4177         hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
4178         ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
4179         if (!ret_val)
4180                 adapter->stats.tncrs += phy_data;
4181
4182 release:
4183         hw->phy.ops.release(hw);
4184 }
4185
4186 /**
4187  * e1000e_update_stats - Update the board statistics counters
4188  * @adapter: board private structure
4189  **/
4190 static void e1000e_update_stats(struct e1000_adapter *adapter)
4191 {
4192         struct net_device *netdev = adapter->netdev;
4193         struct e1000_hw *hw = &adapter->hw;
4194         struct pci_dev *pdev = adapter->pdev;
4195
4196         /*
4197          * Prevent stats update while adapter is being reset, or if the pci
4198          * connection is down.
4199          */
4200         if (adapter->link_speed == 0)
4201                 return;
4202         if (pci_channel_offline(pdev))
4203                 return;
4204
4205         adapter->stats.crcerrs += er32(CRCERRS);
4206         adapter->stats.gprc += er32(GPRC);
4207         adapter->stats.gorc += er32(GORCL);
4208         er32(GORCH); /* Clear gorc */
4209         adapter->stats.bprc += er32(BPRC);
4210         adapter->stats.mprc += er32(MPRC);
4211         adapter->stats.roc += er32(ROC);
4212
4213         adapter->stats.mpc += er32(MPC);
4214
4215         /* Half-duplex statistics */
4216         if (adapter->link_duplex == HALF_DUPLEX) {
4217                 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
4218                         e1000e_update_phy_stats(adapter);
4219                 } else {
4220                         adapter->stats.scc += er32(SCC);
4221                         adapter->stats.ecol += er32(ECOL);
4222                         adapter->stats.mcc += er32(MCC);
4223                         adapter->stats.latecol += er32(LATECOL);
4224                         adapter->stats.dc += er32(DC);
4225
4226                         hw->mac.collision_delta = er32(COLC);
4227
4228                         if ((hw->mac.type != e1000_82574) &&
4229                             (hw->mac.type != e1000_82583))
4230                                 adapter->stats.tncrs += er32(TNCRS);
4231                 }
4232                 adapter->stats.colc += hw->mac.collision_delta;
4233         }
4234
4235         adapter->stats.xonrxc += er32(XONRXC);
4236         adapter->stats.xontxc += er32(XONTXC);
4237         adapter->stats.xoffrxc += er32(XOFFRXC);
4238         adapter->stats.xofftxc += er32(XOFFTXC);
4239         adapter->stats.gptc += er32(GPTC);
4240         adapter->stats.gotc += er32(GOTCL);
4241         er32(GOTCH); /* Clear gotc */
4242         adapter->stats.rnbc += er32(RNBC);
4243         adapter->stats.ruc += er32(RUC);
4244
4245         adapter->stats.mptc += er32(MPTC);
4246         adapter->stats.bptc += er32(BPTC);
4247
4248         /* used for adaptive IFS */
4249
4250         hw->mac.tx_packet_delta = er32(TPT);
4251         adapter->stats.tpt += hw->mac.tx_packet_delta;
4252
4253         adapter->stats.algnerrc += er32(ALGNERRC);
4254         adapter->stats.rxerrc += er32(RXERRC);
4255         adapter->stats.cexterr += er32(CEXTERR);
4256         adapter->stats.tsctc += er32(TSCTC);
4257         adapter->stats.tsctfc += er32(TSCTFC);
4258
4259         /* Fill out the OS statistics structure */
4260         netdev->stats.multicast = adapter->stats.mprc;
4261         netdev->stats.collisions = adapter->stats.colc;
4262
4263         /* Rx Errors */
4264
4265         /*
4266          * RLEC on some newer hardware can be incorrect so build
4267          * our own version based on RUC and ROC
4268          */
4269         netdev->stats.rx_errors = adapter->stats.rxerrc +
4270                 adapter->stats.crcerrs + adapter->stats.algnerrc +
4271                 adapter->stats.ruc + adapter->stats.roc +
4272                 adapter->stats.cexterr;
4273         netdev->stats.rx_length_errors = adapter->stats.ruc +
4274                                               adapter->stats.roc;
4275         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4276         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4277         netdev->stats.rx_missed_errors = adapter->stats.mpc;
4278
4279         /* Tx Errors */
4280         netdev->stats.tx_errors = adapter->stats.ecol +
4281                                        adapter->stats.latecol;
4282         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4283         netdev->stats.tx_window_errors = adapter->stats.latecol;
4284         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4285
4286         /* Tx Dropped needs to be maintained elsewhere */
4287
4288         /* Management Stats */
4289         adapter->stats.mgptc += er32(MGTPTC);
4290         adapter->stats.mgprc += er32(MGTPRC);
4291         adapter->stats.mgpdc += er32(MGTPDC);
4292 }
4293
4294 /**
4295  * e1000_phy_read_status - Update the PHY register status snapshot
4296  * @adapter: board private structure
4297  **/
4298 static void e1000_phy_read_status(struct e1000_adapter *adapter)
4299 {
4300         struct e1000_hw *hw = &adapter->hw;
4301         struct e1000_phy_regs *phy = &adapter->phy_regs;
4302
4303         if ((er32(STATUS) & E1000_STATUS_LU) &&
4304             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
4305                 int ret_val;
4306
4307                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
4308                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
4309                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
4310                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
4311                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
4312                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
4313                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
4314                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
4315                 if (ret_val)
4316                         e_warn("Error reading PHY register\n");
4317         } else {
4318                 /*
4319                  * Do not read PHY registers if link is not up
4320                  * Set values to typical power-on defaults
4321                  */
4322                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4323                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4324                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4325                              BMSR_ERCAP);
4326                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4327                                   ADVERTISE_ALL | ADVERTISE_CSMA);
4328                 phy->lpa = 0;
4329                 phy->expansion = EXPANSION_ENABLENPAGE;
4330                 phy->ctrl1000 = ADVERTISE_1000FULL;
4331                 phy->stat1000 = 0;
4332                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4333         }
4334 }
4335
4336 static void e1000_print_link_info(struct e1000_adapter *adapter)
4337 {
4338         struct e1000_hw *hw = &adapter->hw;
4339         u32 ctrl = er32(CTRL);
4340
4341         /* Link status message must follow this format for user tools */
4342         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
4343                 adapter->netdev->name,
4344                 adapter->link_speed,
4345                 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half",
4346                 (ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE) ? "Rx/Tx" :
4347                 (ctrl & E1000_CTRL_RFCE) ? "Rx" :
4348                 (ctrl & E1000_CTRL_TFCE) ? "Tx" : "None");
4349 }
4350
4351 static bool e1000e_has_link(struct e1000_adapter *adapter)
4352 {
4353         struct e1000_hw *hw = &adapter->hw;
4354         bool link_active = false;
4355         s32 ret_val = 0;
4356
4357         /*
4358          * get_link_status is set on LSC (link status) interrupt or
4359          * Rx sequence error interrupt.  get_link_status will stay
4360          * false until the check_for_link establishes link
4361          * for copper adapters ONLY
4362          */
4363         switch (hw->phy.media_type) {
4364         case e1000_media_type_copper:
4365                 if (hw->mac.get_link_status) {
4366                         ret_val = hw->mac.ops.check_for_link(hw);
4367                         link_active = !hw->mac.get_link_status;
4368                 } else {
4369                         link_active = true;
4370                 }
4371                 break;
4372         case e1000_media_type_fiber:
4373                 ret_val = hw->mac.ops.check_for_link(hw);
4374                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4375                 break;
4376         case e1000_media_type_internal_serdes:
4377                 ret_val = hw->mac.ops.check_for_link(hw);
4378                 link_active = adapter->hw.mac.serdes_has_link;
4379                 break;
4380         default:
4381         case e1000_media_type_unknown:
4382                 break;
4383         }
4384
4385         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4386             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4387                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4388                 e_info("Gigabit has been disabled, downgrading speed\n");
4389         }
4390
4391         return link_active;
4392 }
4393
4394 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4395 {
4396         /* make sure the receive unit is started */
4397         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4398             (adapter->flags & FLAG_RX_RESTART_NOW)) {
4399                 struct e1000_hw *hw = &adapter->hw;
4400                 u32 rctl = er32(RCTL);
4401                 ew32(RCTL, rctl | E1000_RCTL_EN);
4402                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
4403         }
4404 }
4405
4406 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
4407 {
4408         struct e1000_hw *hw = &adapter->hw;
4409
4410         /*
4411          * With 82574 controllers, PHY needs to be checked periodically
4412          * for hung state and reset, if two calls return true
4413          */
4414         if (e1000_check_phy_82574(hw))
4415                 adapter->phy_hang_count++;
4416         else
4417                 adapter->phy_hang_count = 0;
4418
4419         if (adapter->phy_hang_count > 1) {
4420                 adapter->phy_hang_count = 0;
4421                 schedule_work(&adapter->reset_task);
4422         }
4423 }
4424
4425 /**
4426  * e1000_watchdog - Timer Call-back
4427  * @data: pointer to adapter cast into an unsigned long
4428  **/
4429 static void e1000_watchdog(unsigned long data)
4430 {
4431         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4432
4433         /* Do the rest outside of interrupt context */
4434         schedule_work(&adapter->watchdog_task);
4435
4436         /* TODO: make this use queue_delayed_work() */
4437 }
4438
4439 static void e1000_watchdog_task(struct work_struct *work)
4440 {
4441         struct e1000_adapter *adapter = container_of(work,
4442                                         struct e1000_adapter, watchdog_task);
4443         struct net_device *netdev = adapter->netdev;
4444         struct e1000_mac_info *mac = &adapter->hw.mac;
4445         struct e1000_phy_info *phy = &adapter->hw.phy;
4446         struct e1000_ring *tx_ring = adapter->tx_ring;
4447         struct e1000_hw *hw = &adapter->hw;
4448         u32 link, tctl;
4449
4450         if (test_bit(__E1000_DOWN, &adapter->state))
4451                 return;
4452
4453         link = e1000e_has_link(adapter);
4454         if ((netif_carrier_ok(netdev)) && link) {
4455                 /* Cancel scheduled suspend requests. */
4456                 pm_runtime_resume(netdev->dev.parent);
4457
4458                 e1000e_enable_receives(adapter);
4459                 goto link_up;
4460         }
4461
4462         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4463             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4464                 e1000_update_mng_vlan(adapter);
4465
4466         if (link) {
4467                 if (!netif_carrier_ok(netdev)) {
4468                         bool txb2b = true;
4469
4470                         /* Cancel scheduled suspend requests. */
4471                         pm_runtime_resume(netdev->dev.parent);
4472
4473                         /* update snapshot of PHY registers on LSC */
4474                         e1000_phy_read_status(adapter);
4475                         mac->ops.get_link_up_info(&adapter->hw,
4476                                                    &adapter->link_speed,
4477                                                    &adapter->link_duplex);
4478                         e1000_print_link_info(adapter);
4479                         /*
4480                          * On supported PHYs, check for duplex mismatch only
4481                          * if link has autonegotiated at 10/100 half
4482                          */
4483                         if ((hw->phy.type == e1000_phy_igp_3 ||
4484                              hw->phy.type == e1000_phy_bm) &&
4485                             (hw->mac.autoneg == true) &&
4486                             (adapter->link_speed == SPEED_10 ||
4487                              adapter->link_speed == SPEED_100) &&
4488                             (adapter->link_duplex == HALF_DUPLEX)) {
4489                                 u16 autoneg_exp;
4490
4491                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4492
4493                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4494                                         e_info("Autonegotiated half duplex but link partner cannot autoneg.  Try forcing full duplex if link gets many collisions.\n");
4495                         }
4496
4497                         /* adjust timeout factor according to speed/duplex */
4498                         adapter->tx_timeout_factor = 1;
4499                         switch (adapter->link_speed) {
4500                         case SPEED_10:
4501                                 txb2b = false;
4502                                 adapter->tx_timeout_factor = 16;
4503                                 break;
4504                         case SPEED_100:
4505                                 txb2b = false;
4506                                 adapter->tx_timeout_factor = 10;
4507                                 break;
4508                         }
4509
4510                         /*
4511                          * workaround: re-program speed mode bit after
4512                          * link-up event
4513                          */
4514                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4515                             !txb2b) {
4516                                 u32 tarc0;
4517                                 tarc0 = er32(TARC(0));
4518                                 tarc0 &= ~SPEED_MODE_BIT;
4519                                 ew32(TARC(0), tarc0);
4520                         }
4521
4522                         /*
4523                          * disable TSO for pcie and 10/100 speeds, to avoid
4524                          * some hardware issues
4525                          */
4526                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
4527                                 switch (adapter->link_speed) {
4528                                 case SPEED_10:
4529                                 case SPEED_100:
4530                                         e_info("10/100 speed: disabling TSO\n");
4531                                         netdev->features &= ~NETIF_F_TSO;
4532                                         netdev->features &= ~NETIF_F_TSO6;
4533                                         break;
4534                                 case SPEED_1000:
4535                                         netdev->features |= NETIF_F_TSO;
4536                                         netdev->features |= NETIF_F_TSO6;
4537                                         break;
4538                                 default:
4539                                         /* oops */
4540                                         break;
4541                                 }
4542                         }
4543
4544                         /*
4545                          * enable transmits in the hardware, need to do this
4546                          * after setting TARC(0)
4547                          */
4548                         tctl = er32(TCTL);
4549                         tctl |= E1000_TCTL_EN;
4550                         ew32(TCTL, tctl);
4551
4552                         /*
4553                          * Perform any post-link-up configuration before
4554                          * reporting link up.
4555                          */
4556                         if (phy->ops.cfg_on_link_up)
4557                                 phy->ops.cfg_on_link_up(hw);
4558
4559                         netif_carrier_on(netdev);
4560
4561                         if (!test_bit(__E1000_DOWN, &adapter->state))
4562                                 mod_timer(&adapter->phy_info_timer,
4563                                           round_jiffies(jiffies + 2 * HZ));
4564                 }
4565         } else {
4566                 if (netif_carrier_ok(netdev)) {
4567                         adapter->link_speed = 0;
4568                         adapter->link_duplex = 0;
4569                         /* Link status message must follow this format */
4570                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4571                                adapter->netdev->name);
4572                         netif_carrier_off(netdev);
4573                         if (!test_bit(__E1000_DOWN, &adapter->state))
4574                                 mod_timer(&adapter->phy_info_timer,
4575                                           round_jiffies(jiffies + 2 * HZ));
4576
4577                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4578                                 schedule_work(&adapter->reset_task);
4579                         else
4580                                 pm_schedule_suspend(netdev->dev.parent,
4581                                                         LINK_TIMEOUT);
4582                 }
4583         }
4584
4585 link_up:
4586         spin_lock(&adapter->stats64_lock);
4587         e1000e_update_stats(adapter);
4588
4589         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4590         adapter->tpt_old = adapter->stats.tpt;
4591         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4592         adapter->colc_old = adapter->stats.colc;
4593
4594         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4595         adapter->gorc_old = adapter->stats.gorc;
4596         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4597         adapter->gotc_old = adapter->stats.gotc;
4598         spin_unlock(&adapter->stats64_lock);
4599
4600         e1000e_update_adaptive(&adapter->hw);
4601
4602         if (!netif_carrier_ok(netdev) &&
4603             (e1000_desc_unused(tx_ring) + 1 < tx_ring->count)) {
4604                 /*
4605                  * We've lost link, so the controller stops DMA,
4606                  * but we've got queued Tx work that's never going
4607                  * to get done, so reset controller to flush Tx.
4608                  * (Do the reset outside of interrupt context).
4609                  */
4610                 schedule_work(&adapter->reset_task);
4611                 /* return immediately since reset is imminent */
4612                 return;
4613         }
4614
4615         /* Simple mode for Interrupt Throttle Rate (ITR) */
4616         if (adapter->itr_setting == 4) {
4617                 /*
4618                  * Symmetric Tx/Rx gets a reduced ITR=2000;
4619                  * Total asymmetrical Tx or Rx gets ITR=8000;
4620                  * everyone else is between 2000-8000.
4621                  */
4622                 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4623                 u32 dif = (adapter->gotc > adapter->gorc ?
4624                             adapter->gotc - adapter->gorc :
4625                             adapter->gorc - adapter->gotc) / 10000;
4626                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4627
4628                 e1000e_write_itr(adapter, itr);
4629         }
4630
4631         /* Cause software interrupt to ensure Rx ring is cleaned */
4632         if (adapter->msix_entries)
4633                 ew32(ICS, adapter->rx_ring->ims_val);
4634         else
4635                 ew32(ICS, E1000_ICS_RXDMT0);
4636
4637         /* flush pending descriptors to memory before detecting Tx hang */
4638         e1000e_flush_descriptors(adapter);
4639
4640         /* Force detection of hung controller every watchdog period */
4641         adapter->detect_tx_hung = true;
4642
4643         /*
4644          * With 82571 controllers, LAA may be overwritten due to controller
4645          * reset from the other port. Set the appropriate LAA in RAR[0]
4646          */
4647         if (e1000e_get_laa_state_82571(hw))
4648                 hw->mac.ops.rar_set(hw, adapter->hw.mac.addr, 0);
4649
4650         if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
4651                 e1000e_check_82574_phy_workaround(adapter);
4652
4653         /* Reset the timer */
4654         if (!test_bit(__E1000_DOWN, &adapter->state))
4655                 mod_timer(&adapter->watchdog_timer,
4656                           round_jiffies(jiffies + 2 * HZ));
4657 }
4658
4659 #define E1000_TX_FLAGS_CSUM             0x00000001
4660 #define E1000_TX_FLAGS_VLAN             0x00000002
4661 #define E1000_TX_FLAGS_TSO              0x00000004
4662 #define E1000_TX_FLAGS_IPV4             0x00000008
4663 #define E1000_TX_FLAGS_NO_FCS           0x00000010
4664 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
4665 #define E1000_TX_FLAGS_VLAN_SHIFT       16
4666
4667 static int e1000_tso(struct e1000_ring *tx_ring, struct sk_buff *skb)
4668 {
4669         struct e1000_context_desc *context_desc;
4670         struct e1000_buffer *buffer_info;
4671         unsigned int i;
4672         u32 cmd_length = 0;
4673         u16 ipcse = 0, tucse, mss;
4674         u8 ipcss, ipcso, tucss, tucso, hdr_len;
4675
4676         if (!skb_is_gso(skb))
4677                 return 0;
4678
4679         if (skb_header_cloned(skb)) {
4680                 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4681
4682                 if (err)
4683                         return err;
4684         }
4685
4686         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4687         mss = skb_shinfo(skb)->gso_size;
4688         if (skb->protocol == htons(ETH_P_IP)) {
4689                 struct iphdr *iph = ip_hdr(skb);
4690                 iph->tot_len = 0;
4691                 iph->check = 0;
4692                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4693                                                          0, IPPROTO_TCP, 0);
4694                 cmd_length = E1000_TXD_CMD_IP;
4695                 ipcse = skb_transport_offset(skb) - 1;
4696         } else if (skb_is_gso_v6(skb)) {
4697                 ipv6_hdr(skb)->payload_len = 0;
4698                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4699                                                        &ipv6_hdr(skb)->daddr,
4700                                                        0, IPPROTO_TCP, 0);
4701                 ipcse = 0;
4702         }
4703         ipcss = skb_network_offset(skb);
4704         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4705         tucss = skb_transport_offset(skb);
4706         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4707         tucse = 0;
4708
4709         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4710                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4711
4712         i = tx_ring->next_to_use;
4713         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4714         buffer_info = &tx_ring->buffer_info[i];
4715
4716         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4717         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4718         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4719         context_desc->upper_setup.tcp_fields.tucss = tucss;
4720         context_desc->upper_setup.tcp_fields.tucso = tucso;
4721         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4722         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4723         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4724         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4725
4726         buffer_info->time_stamp = jiffies;
4727         buffer_info->next_to_watch = i;
4728
4729         i++;
4730         if (i == tx_ring->count)
4731                 i = 0;
4732         tx_ring->next_to_use = i;
4733
4734         return 1;
4735 }
4736
4737 static bool e1000_tx_csum(struct e1000_ring *tx_ring, struct sk_buff *skb)
4738 {
4739         struct e1000_adapter *adapter = tx_ring->adapter;
4740         struct e1000_context_desc *context_desc;
4741         struct e1000_buffer *buffer_info;
4742         unsigned int i;
4743         u8 css;
4744         u32 cmd_len = E1000_TXD_CMD_DEXT;
4745         __be16 protocol;
4746
4747         if (skb->ip_summed != CHECKSUM_PARTIAL)
4748                 return 0;
4749
4750         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4751                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4752         else
4753                 protocol = skb->protocol;
4754
4755         switch (protocol) {
4756         case cpu_to_be16(ETH_P_IP):
4757                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4758                         cmd_len |= E1000_TXD_CMD_TCP;
4759                 break;
4760         case cpu_to_be16(ETH_P_IPV6):
4761                 /* XXX not handling all IPV6 headers */
4762                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4763                         cmd_len |= E1000_TXD_CMD_TCP;
4764                 break;
4765         default:
4766                 if (unlikely(net_ratelimit()))
4767                         e_warn("checksum_partial proto=%x!\n",
4768                                be16_to_cpu(protocol));
4769                 break;
4770         }
4771
4772         css = skb_checksum_start_offset(skb);
4773
4774         i = tx_ring->next_to_use;
4775         buffer_info = &tx_ring->buffer_info[i];
4776         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4777
4778         context_desc->lower_setup.ip_config = 0;
4779         context_desc->upper_setup.tcp_fields.tucss = css;
4780         context_desc->upper_setup.tcp_fields.tucso =
4781                                 css + skb->csum_offset;
4782         context_desc->upper_setup.tcp_fields.tucse = 0;
4783         context_desc->tcp_seg_setup.data = 0;
4784         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4785
4786         buffer_info->time_stamp = jiffies;
4787         buffer_info->next_to_watch = i;
4788
4789         i++;
4790         if (i == tx_ring->count)
4791                 i = 0;
4792         tx_ring->next_to_use = i;
4793
4794         return 1;
4795 }
4796
4797 static int e1000_tx_map(struct e1000_ring *tx_ring, struct sk_buff *skb,
4798                         unsigned int first, unsigned int max_per_txd,
4799                         unsigned int nr_frags)
4800 {
4801         struct e1000_adapter *adapter = tx_ring->adapter;
4802         struct pci_dev *pdev = adapter->pdev;
4803         struct e1000_buffer *buffer_info;
4804         unsigned int len = skb_headlen(skb);
4805         unsigned int offset = 0, size, count = 0, i;
4806         unsigned int f, bytecount, segs;
4807
4808         i = tx_ring->next_to_use;
4809
4810         while (len) {
4811                 buffer_info = &tx_ring->buffer_info[i];
4812                 size = min(len, max_per_txd);
4813
4814                 buffer_info->length = size;
4815                 buffer_info->time_stamp = jiffies;
4816                 buffer_info->next_to_watch = i;
4817                 buffer_info->dma = dma_map_single(&pdev->dev,
4818                                                   skb->data + offset,
4819                                                   size, DMA_TO_DEVICE);
4820                 buffer_info->mapped_as_page = false;
4821                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4822                         goto dma_error;
4823
4824                 len -= size;
4825                 offset += size;
4826                 count++;
4827
4828                 if (len) {
4829                         i++;
4830                         if (i == tx_ring->count)
4831                                 i = 0;
4832                 }
4833         }
4834
4835         for (f = 0; f < nr_frags; f++) {
4836                 const struct skb_frag_struct *frag;
4837
4838                 frag = &skb_shinfo(skb)->frags[f];
4839                 len = skb_frag_size(frag);
4840                 offset = 0;
4841
4842                 while (len) {
4843                         i++;
4844                         if (i == tx_ring->count)
4845                                 i = 0;
4846
4847                         buffer_info = &tx_ring->buffer_info[i];
4848                         size = min(len, max_per_txd);
4849
4850                         buffer_info->length = size;
4851                         buffer_info->time_stamp = jiffies;
4852                         buffer_info->next_to_watch = i;
4853                         buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
4854                                                 offset, size, DMA_TO_DEVICE);
4855                         buffer_info->mapped_as_page = true;
4856                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4857                                 goto dma_error;
4858
4859                         len -= size;
4860                         offset += size;
4861                         count++;
4862                 }
4863         }
4864
4865         segs = skb_shinfo(skb)->gso_segs ? : 1;
4866         /* multiply data chunks by size of headers */
4867         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4868
4869         tx_ring->buffer_info[i].skb = skb;
4870         tx_ring->buffer_info[i].segs = segs;
4871         tx_ring->buffer_info[i].bytecount = bytecount;
4872         tx_ring->buffer_info[first].next_to_watch = i;
4873
4874         return count;
4875
4876 dma_error:
4877         dev_err(&pdev->dev, "Tx DMA map failed\n");
4878         buffer_info->dma = 0;
4879         if (count)
4880                 count--;
4881
4882         while (count--) {
4883                 if (i == 0)
4884                         i += tx_ring->count;
4885                 i--;
4886                 buffer_info = &tx_ring->buffer_info[i];
4887                 e1000_put_txbuf(tx_ring, buffer_info);
4888         }
4889
4890         return 0;
4891 }
4892
4893 static void e1000_tx_queue(struct e1000_ring *tx_ring, int tx_flags, int count)
4894 {
4895         struct e1000_adapter *adapter = tx_ring->adapter;
4896         struct e1000_tx_desc *tx_desc = NULL;
4897         struct e1000_buffer *buffer_info;
4898         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4899         unsigned int i;
4900
4901         if (tx_flags & E1000_TX_FLAGS_TSO) {
4902                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4903                              E1000_TXD_CMD_TSE;
4904                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4905
4906                 if (tx_flags & E1000_TX_FLAGS_IPV4)
4907                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4908         }
4909
4910         if (tx_flags & E1000_TX_FLAGS_CSUM) {
4911                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4912                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4913         }
4914
4915         if (tx_flags & E1000_TX_FLAGS_VLAN) {
4916                 txd_lower |= E1000_TXD_CMD_VLE;
4917                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4918         }
4919
4920         if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
4921                 txd_lower &= ~(E1000_TXD_CMD_IFCS);
4922
4923         i = tx_ring->next_to_use;
4924
4925         do {
4926                 buffer_info = &tx_ring->buffer_info[i];
4927                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4928                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4929                 tx_desc->lower.data =
4930                         cpu_to_le32(txd_lower | buffer_info->length);
4931                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4932
4933                 i++;
4934                 if (i == tx_ring->count)
4935                         i = 0;
4936         } while (--count > 0);
4937
4938         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4939
4940         /* txd_cmd re-enables FCS, so we'll re-disable it here as desired. */
4941         if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
4942                 tx_desc->lower.data &= ~(cpu_to_le32(E1000_TXD_CMD_IFCS));
4943
4944         /*
4945          * Force memory writes to complete before letting h/w
4946          * know there are new descriptors to fetch.  (Only
4947          * applicable for weak-ordered memory model archs,
4948          * such as IA-64).
4949          */
4950         wmb();
4951
4952         tx_ring->next_to_use = i;
4953
4954         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
4955                 e1000e_update_tdt_wa(tx_ring, i);
4956         else
4957                 writel(i, tx_ring->tail);
4958
4959         /*
4960          * we need this if more than one processor can write to our tail
4961          * at a time, it synchronizes IO on IA64/Altix systems
4962          */
4963         mmiowb();
4964 }
4965
4966 #define MINIMUM_DHCP_PACKET_SIZE 282
4967 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4968                                     struct sk_buff *skb)
4969 {
4970         struct e1000_hw *hw =  &adapter->hw;
4971         u16 length, offset;
4972
4973         if (vlan_tx_tag_present(skb)) {
4974                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4975                     (adapter->hw.mng_cookie.status &
4976                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4977                         return 0;
4978         }
4979
4980         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4981                 return 0;
4982
4983         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4984                 return 0;
4985
4986         {
4987                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4988                 struct udphdr *udp;
4989
4990                 if (ip->protocol != IPPROTO_UDP)
4991                         return 0;
4992
4993                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4994                 if (ntohs(udp->dest) != 67)
4995                         return 0;
4996
4997                 offset = (u8 *)udp + 8 - skb->data;
4998                 length = skb->len - offset;
4999                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
5000         }
5001
5002         return 0;
5003 }
5004
5005 static int __e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5006 {
5007         struct e1000_adapter *adapter = tx_ring->adapter;
5008
5009         netif_stop_queue(adapter->netdev);
5010         /*
5011          * Herbert's original patch had:
5012          *  smp_mb__after_netif_stop_queue();
5013          * but since that doesn't exist yet, just open code it.
5014          */
5015         smp_mb();
5016
5017         /*
5018          * We need to check again in a case another CPU has just
5019          * made room available.
5020          */
5021         if (e1000_desc_unused(tx_ring) < size)
5022                 return -EBUSY;
5023
5024         /* A reprieve! */
5025         netif_start_queue(adapter->netdev);
5026         ++adapter->restart_queue;
5027         return 0;
5028 }
5029
5030 static int e1000_maybe_stop_tx(struct e1000_ring *tx_ring, int size)
5031 {
5032         BUG_ON(size > tx_ring->count);
5033
5034         if (e1000_desc_unused(tx_ring) >= size)
5035                 return 0;
5036         return __e1000_maybe_stop_tx(tx_ring, size);
5037 }
5038
5039 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
5040                                     struct net_device *netdev)
5041 {
5042         struct e1000_adapter *adapter = netdev_priv(netdev);
5043         struct e1000_ring *tx_ring = adapter->tx_ring;
5044         unsigned int first;
5045         unsigned int tx_flags = 0;
5046         unsigned int len = skb_headlen(skb);
5047         unsigned int nr_frags;
5048         unsigned int mss;
5049         int count = 0;
5050         int tso;
5051         unsigned int f;
5052
5053         if (test_bit(__E1000_DOWN, &adapter->state)) {
5054                 dev_kfree_skb_any(skb);
5055                 return NETDEV_TX_OK;
5056         }
5057
5058         if (skb->len <= 0) {
5059                 dev_kfree_skb_any(skb);
5060                 return NETDEV_TX_OK;
5061         }
5062
5063         mss = skb_shinfo(skb)->gso_size;
5064         if (mss) {
5065                 u8 hdr_len;
5066
5067                 /*
5068                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
5069                  * points to just header, pull a few bytes of payload from
5070                  * frags into skb->data
5071                  */
5072                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
5073                 /*
5074                  * we do this workaround for ES2LAN, but it is un-necessary,
5075                  * avoiding it could save a lot of cycles
5076                  */
5077                 if (skb->data_len && (hdr_len == len)) {
5078                         unsigned int pull_size;
5079
5080                         pull_size = min_t(unsigned int, 4, skb->data_len);
5081                         if (!__pskb_pull_tail(skb, pull_size)) {
5082                                 e_err("__pskb_pull_tail failed.\n");
5083                                 dev_kfree_skb_any(skb);
5084                                 return NETDEV_TX_OK;
5085                         }
5086                         len = skb_headlen(skb);
5087                 }
5088         }
5089
5090         /* reserve a descriptor for the offload context */
5091         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
5092                 count++;
5093         count++;
5094
5095         count += DIV_ROUND_UP(len, adapter->tx_fifo_limit);
5096
5097         nr_frags = skb_shinfo(skb)->nr_frags;
5098         for (f = 0; f < nr_frags; f++)
5099                 count += DIV_ROUND_UP(skb_frag_size(&skb_shinfo(skb)->frags[f]),
5100                                       adapter->tx_fifo_limit);
5101
5102         if (adapter->hw.mac.tx_pkt_filtering)
5103                 e1000_transfer_dhcp_info(adapter, skb);
5104
5105         /*
5106          * need: count + 2 desc gap to keep tail from touching
5107          * head, otherwise try next time
5108          */
5109         if (e1000_maybe_stop_tx(tx_ring, count + 2))
5110                 return NETDEV_TX_BUSY;
5111
5112         if (vlan_tx_tag_present(skb)) {
5113                 tx_flags |= E1000_TX_FLAGS_VLAN;
5114                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
5115         }
5116
5117         first = tx_ring->next_to_use;
5118
5119         tso = e1000_tso(tx_ring, skb);
5120         if (tso < 0) {
5121                 dev_kfree_skb_any(skb);
5122                 return NETDEV_TX_OK;
5123         }
5124
5125         if (tso)
5126                 tx_flags |= E1000_TX_FLAGS_TSO;
5127         else if (e1000_tx_csum(tx_ring, skb))
5128                 tx_flags |= E1000_TX_FLAGS_CSUM;
5129
5130         /*
5131          * Old method was to assume IPv4 packet by default if TSO was enabled.
5132          * 82571 hardware supports TSO capabilities for IPv6 as well...
5133          * no longer assume, we must.
5134          */
5135         if (skb->protocol == htons(ETH_P_IP))
5136                 tx_flags |= E1000_TX_FLAGS_IPV4;
5137
5138         if (unlikely(skb->no_fcs))
5139                 tx_flags |= E1000_TX_FLAGS_NO_FCS;
5140
5141         /* if count is 0 then mapping error has occurred */
5142         count = e1000_tx_map(tx_ring, skb, first, adapter->tx_fifo_limit,
5143                              nr_frags);
5144         if (count) {
5145                 skb_tx_timestamp(skb);
5146
5147                 netdev_sent_queue(netdev, skb->len);
5148                 e1000_tx_queue(tx_ring, tx_flags, count);
5149                 /* Make sure there is space in the ring for the next send. */
5150                 e1000_maybe_stop_tx(tx_ring,
5151                                     (MAX_SKB_FRAGS *
5152                                      DIV_ROUND_UP(PAGE_SIZE,
5153                                                   adapter->tx_fifo_limit) + 2));
5154         } else {
5155                 dev_kfree_skb_any(skb);
5156                 tx_ring->buffer_info[first].time_stamp = 0;
5157                 tx_ring->next_to_use = first;
5158         }
5159
5160         return NETDEV_TX_OK;
5161 }
5162
5163 /**
5164  * e1000_tx_timeout - Respond to a Tx Hang
5165  * @netdev: network interface device structure
5166  **/
5167 static void e1000_tx_timeout(struct net_device *netdev)
5168 {
5169         struct e1000_adapter *adapter = netdev_priv(netdev);
5170
5171         /* Do the reset outside of interrupt context */
5172         adapter->tx_timeout_count++;
5173         schedule_work(&adapter->reset_task);
5174 }
5175
5176 static void e1000_reset_task(struct work_struct *work)
5177 {
5178         struct e1000_adapter *adapter;
5179         adapter = container_of(work, struct e1000_adapter, reset_task);
5180
5181         /* don't run the task if already down */
5182         if (test_bit(__E1000_DOWN, &adapter->state))
5183                 return;
5184
5185         if (!((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
5186               (adapter->flags & FLAG_RX_RESTART_NOW))) {
5187                 e1000e_dump(adapter);
5188                 e_err("Reset adapter\n");
5189         }
5190         e1000e_reinit_locked(adapter);
5191 }
5192
5193 /**
5194  * e1000_get_stats64 - Get System Network Statistics
5195  * @netdev: network interface device structure
5196  * @stats: rtnl_link_stats64 pointer
5197  *
5198  * Returns the address of the device statistics structure.
5199  **/
5200 struct rtnl_link_stats64 *e1000e_get_stats64(struct net_device *netdev,
5201                                              struct rtnl_link_stats64 *stats)
5202 {
5203         struct e1000_adapter *adapter = netdev_priv(netdev);
5204
5205         memset(stats, 0, sizeof(struct rtnl_link_stats64));
5206         spin_lock(&adapter->stats64_lock);
5207         e1000e_update_stats(adapter);
5208         /* Fill out the OS statistics structure */
5209         stats->rx_bytes = adapter->stats.gorc;
5210         stats->rx_packets = adapter->stats.gprc;
5211         stats->tx_bytes = adapter->stats.gotc;
5212         stats->tx_packets = adapter->stats.gptc;
5213         stats->multicast = adapter->stats.mprc;
5214         stats->collisions = adapter->stats.colc;
5215
5216         /* Rx Errors */
5217
5218         /*
5219          * RLEC on some newer hardware can be incorrect so build
5220          * our own version based on RUC and ROC
5221          */
5222         stats->rx_errors = adapter->stats.rxerrc +
5223                 adapter->stats.crcerrs + adapter->stats.algnerrc +
5224                 adapter->stats.ruc + adapter->stats.roc +
5225                 adapter->stats.cexterr;
5226         stats->rx_length_errors = adapter->stats.ruc +
5227                                               adapter->stats.roc;
5228         stats->rx_crc_errors = adapter->stats.crcerrs;
5229         stats->rx_frame_errors = adapter->stats.algnerrc;
5230         stats->rx_missed_errors = adapter->stats.mpc;
5231
5232         /* Tx Errors */
5233         stats->tx_errors = adapter->stats.ecol +
5234                                        adapter->stats.latecol;
5235         stats->tx_aborted_errors = adapter->stats.ecol;
5236         stats->tx_window_errors = adapter->stats.latecol;
5237         stats->tx_carrier_errors = adapter->stats.tncrs;
5238
5239         /* Tx Dropped needs to be maintained elsewhere */
5240
5241         spin_unlock(&adapter->stats64_lock);
5242         return stats;
5243 }
5244
5245 /**
5246  * e1000_change_mtu - Change the Maximum Transfer Unit
5247  * @netdev: network interface device structure
5248  * @new_mtu: new value for maximum frame size
5249  *
5250  * Returns 0 on success, negative on failure
5251  **/
5252 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
5253 {
5254         struct e1000_adapter *adapter = netdev_priv(netdev);
5255         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5256
5257         /* Jumbo frame support */
5258         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
5259             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
5260                 e_err("Jumbo Frames not supported.\n");
5261                 return -EINVAL;
5262         }
5263
5264         /* Supported frame sizes */
5265         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
5266             (max_frame > adapter->max_hw_frame_size)) {
5267                 e_err("Unsupported MTU setting\n");
5268                 return -EINVAL;
5269         }
5270
5271         /* Jumbo frame workaround on 82579 and newer requires CRC be stripped */
5272         if ((adapter->hw.mac.type >= e1000_pch2lan) &&
5273             !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
5274             (new_mtu > ETH_DATA_LEN)) {
5275                 e_err("Jumbo Frames not supported on this device when CRC stripping is disabled.\n");
5276                 return -EINVAL;
5277         }
5278
5279         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
5280                 usleep_range(1000, 2000);
5281         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
5282         adapter->max_frame_size = max_frame;
5283         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5284         netdev->mtu = new_mtu;
5285         if (netif_running(netdev))
5286                 e1000e_down(adapter);
5287
5288         /*
5289          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
5290          * means we reserve 2 more, this pushes us to allocate from the next
5291          * larger slab size.
5292          * i.e. RXBUFFER_2048 --> size-4096 slab
5293          * However with the new *_jumbo_rx* routines, jumbo receives will use
5294          * fragmented skbs
5295          */
5296
5297         if (max_frame <= 2048)
5298                 adapter->rx_buffer_len = 2048;
5299         else
5300                 adapter->rx_buffer_len = 4096;
5301
5302         /* adjust allocation if LPE protects us, and we aren't using SBP */
5303         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
5304              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
5305                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
5306                                          + ETH_FCS_LEN;
5307
5308         if (netif_running(netdev))
5309                 e1000e_up(adapter);
5310         else
5311                 e1000e_reset(adapter);
5312
5313         clear_bit(__E1000_RESETTING, &adapter->state);
5314
5315         return 0;
5316 }
5317
5318 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
5319                            int cmd)
5320 {
5321         struct e1000_adapter *adapter = netdev_priv(netdev);
5322         struct mii_ioctl_data *data = if_mii(ifr);
5323
5324         if (adapter->hw.phy.media_type != e1000_media_type_copper)
5325                 return -EOPNOTSUPP;
5326
5327         switch (cmd) {
5328         case SIOCGMIIPHY:
5329                 data->phy_id = adapter->hw.phy.addr;
5330                 break;
5331         case SIOCGMIIREG:
5332                 e1000_phy_read_status(adapter);
5333
5334                 switch (data->reg_num & 0x1F) {
5335                 case MII_BMCR:
5336                         data->val_out = adapter->phy_regs.bmcr;
5337                         break;
5338                 case MII_BMSR:
5339                         data->val_out = adapter->phy_regs.bmsr;
5340                         break;
5341                 case MII_PHYSID1:
5342                         data->val_out = (adapter->hw.phy.id >> 16);
5343                         break;
5344                 case MII_PHYSID2:
5345                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
5346                         break;
5347                 case MII_ADVERTISE:
5348                         data->val_out = adapter->phy_regs.advertise;
5349                         break;
5350                 case MII_LPA:
5351                         data->val_out = adapter->phy_regs.lpa;
5352                         break;
5353                 case MII_EXPANSION:
5354                         data->val_out = adapter->phy_regs.expansion;
5355                         break;
5356                 case MII_CTRL1000:
5357                         data->val_out = adapter->phy_regs.ctrl1000;
5358                         break;
5359                 case MII_STAT1000:
5360                         data->val_out = adapter->phy_regs.stat1000;
5361                         break;
5362                 case MII_ESTATUS:
5363                         data->val_out = adapter->phy_regs.estatus;
5364                         break;
5365                 default:
5366                         return -EIO;
5367                 }
5368                 break;
5369         case SIOCSMIIREG:
5370         default:
5371                 return -EOPNOTSUPP;
5372         }
5373         return 0;
5374 }
5375
5376 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5377 {
5378         switch (cmd) {
5379         case SIOCGMIIPHY:
5380         case SIOCGMIIREG:
5381         case SIOCSMIIREG:
5382                 return e1000_mii_ioctl(netdev, ifr, cmd);
5383         default:
5384                 return -EOPNOTSUPP;
5385         }
5386 }
5387
5388 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
5389 {
5390         struct e1000_hw *hw = &adapter->hw;
5391         u32 i, mac_reg;
5392         u16 phy_reg, wuc_enable;
5393         int retval = 0;
5394
5395         /* copy MAC RARs to PHY RARs */
5396         e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5397
5398         retval = hw->phy.ops.acquire(hw);
5399         if (retval) {
5400                 e_err("Could not acquire PHY\n");
5401                 return retval;
5402         }
5403
5404         /* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
5405         retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5406         if (retval)
5407                 goto release;
5408
5409         /* copy MAC MTA to PHY MTA - only needed for pchlan */
5410         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5411                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5412                 hw->phy.ops.write_reg_page(hw, BM_MTA(i),
5413                                            (u16)(mac_reg & 0xFFFF));
5414                 hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
5415                                            (u16)((mac_reg >> 16) & 0xFFFF));
5416         }
5417
5418         /* configure PHY Rx Control register */
5419         hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
5420         mac_reg = er32(RCTL);
5421         if (mac_reg & E1000_RCTL_UPE)
5422                 phy_reg |= BM_RCTL_UPE;
5423         if (mac_reg & E1000_RCTL_MPE)
5424                 phy_reg |= BM_RCTL_MPE;
5425         phy_reg &= ~(BM_RCTL_MO_MASK);
5426         if (mac_reg & E1000_RCTL_MO_3)
5427                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5428                                 << BM_RCTL_MO_SHIFT);
5429         if (mac_reg & E1000_RCTL_BAM)
5430                 phy_reg |= BM_RCTL_BAM;
5431         if (mac_reg & E1000_RCTL_PMCF)
5432                 phy_reg |= BM_RCTL_PMCF;
5433         mac_reg = er32(CTRL);
5434         if (mac_reg & E1000_CTRL_RFCE)
5435                 phy_reg |= BM_RCTL_RFCE;
5436         hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
5437
5438         /* enable PHY wakeup in MAC register */
5439         ew32(WUFC, wufc);
5440         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
5441
5442         /* configure and enable PHY wakeup in PHY registers */
5443         hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
5444         hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
5445
5446         /* activate PHY wakeup */
5447         wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5448         retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5449         if (retval)
5450                 e_err("Could not set PHY Host Wakeup bit\n");
5451 release:
5452         hw->phy.ops.release(hw);
5453
5454         return retval;
5455 }
5456
5457 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5458                             bool runtime)
5459 {
5460         struct net_device *netdev = pci_get_drvdata(pdev);
5461         struct e1000_adapter *adapter = netdev_priv(netdev);
5462         struct e1000_hw *hw = &adapter->hw;
5463         u32 ctrl, ctrl_ext, rctl, status;
5464         /* Runtime suspend should only enable wakeup for link changes */
5465         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5466         int retval = 0;
5467
5468         netif_device_detach(netdev);
5469
5470         if (netif_running(netdev)) {
5471                 int count = E1000_CHECK_RESET_COUNT;
5472
5473                 while (test_bit(__E1000_RESETTING, &adapter->state) && count--)
5474                         usleep_range(10000, 20000);
5475
5476                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5477                 e1000e_down(adapter);
5478                 e1000_free_irq(adapter);
5479         }
5480         e1000e_reset_interrupt_capability(adapter);
5481
5482         retval = pci_save_state(pdev);
5483         if (retval)
5484                 return retval;
5485
5486         status = er32(STATUS);
5487         if (status & E1000_STATUS_LU)
5488                 wufc &= ~E1000_WUFC_LNKC;
5489
5490         if (wufc) {
5491                 e1000_setup_rctl(adapter);
5492                 e1000e_set_rx_mode(netdev);
5493
5494                 /* turn on all-multi mode if wake on multicast is enabled */
5495                 if (wufc & E1000_WUFC_MC) {
5496                         rctl = er32(RCTL);
5497                         rctl |= E1000_RCTL_MPE;
5498                         ew32(RCTL, rctl);
5499                 }
5500
5501                 ctrl = er32(CTRL);
5502                 /* advertise wake from D3Cold */
5503                 #define E1000_CTRL_ADVD3WUC 0x00100000
5504                 /* phy power management enable */
5505                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5506                 ctrl |= E1000_CTRL_ADVD3WUC;
5507                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5508                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5509                 ew32(CTRL, ctrl);
5510
5511                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5512                     adapter->hw.phy.media_type ==
5513                     e1000_media_type_internal_serdes) {
5514                         /* keep the laser running in D3 */
5515                         ctrl_ext = er32(CTRL_EXT);
5516                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5517                         ew32(CTRL_EXT, ctrl_ext);
5518                 }
5519
5520                 if (adapter->flags & FLAG_IS_ICH)
5521                         e1000_suspend_workarounds_ich8lan(&adapter->hw);
5522
5523                 /* Allow time for pending master requests to run */
5524                 e1000e_disable_pcie_master(&adapter->hw);
5525
5526                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5527                         /* enable wakeup by the PHY */
5528                         retval = e1000_init_phy_wakeup(adapter, wufc);
5529                         if (retval)
5530                                 return retval;
5531                 } else {
5532                         /* enable wakeup by the MAC */
5533                         ew32(WUFC, wufc);
5534                         ew32(WUC, E1000_WUC_PME_EN);
5535                 }
5536         } else {
5537                 ew32(WUC, 0);
5538                 ew32(WUFC, 0);
5539         }
5540
5541         *enable_wake = !!wufc;
5542
5543         /* make sure adapter isn't asleep if manageability is enabled */
5544         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5545             (hw->mac.ops.check_mng_mode(hw)))
5546                 *enable_wake = true;
5547
5548         if (adapter->hw.phy.type == e1000_phy_igp_3)
5549                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5550
5551         /*
5552          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5553          * would have already happened in close and is redundant.
5554          */
5555         e1000e_release_hw_control(adapter);
5556
5557         pci_disable_device(pdev);
5558
5559         return 0;
5560 }
5561
5562 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5563 {
5564         if (sleep && wake) {
5565                 pci_prepare_to_sleep(pdev);
5566                 return;
5567         }
5568
5569         pci_wake_from_d3(pdev, wake);
5570         pci_set_power_state(pdev, PCI_D3hot);
5571 }
5572
5573 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5574                                     bool wake)
5575 {
5576         struct net_device *netdev = pci_get_drvdata(pdev);
5577         struct e1000_adapter *adapter = netdev_priv(netdev);
5578
5579         /*
5580          * The pci-e switch on some quad port adapters will report a
5581          * correctable error when the MAC transitions from D0 to D3.  To
5582          * prevent this we need to mask off the correctable errors on the
5583          * downstream port of the pci-e switch.
5584          */
5585         if (adapter->flags & FLAG_IS_QUAD_PORT) {
5586                 struct pci_dev *us_dev = pdev->bus->self;
5587                 u16 devctl;
5588
5589                 pcie_capability_read_word(us_dev, PCI_EXP_DEVCTL, &devctl);
5590                 pcie_capability_write_word(us_dev, PCI_EXP_DEVCTL,
5591                                            (devctl & ~PCI_EXP_DEVCTL_CERE));
5592
5593                 e1000_power_off(pdev, sleep, wake);
5594
5595                 pcie_capability_write_word(us_dev, PCI_EXP_DEVCTL, devctl);
5596         } else {
5597                 e1000_power_off(pdev, sleep, wake);
5598         }
5599 }
5600
5601 #ifdef CONFIG_PCIEASPM
5602 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5603 {
5604         pci_disable_link_state_locked(pdev, state);
5605 }
5606 #else
5607 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5608 {
5609         /*
5610          * Both device and parent should have the same ASPM setting.
5611          * Disable ASPM in downstream component first and then upstream.
5612          */
5613         pcie_capability_clear_word(pdev, PCI_EXP_LNKCTL, state);
5614
5615         if (pdev->bus->self)
5616                 pcie_capability_clear_word(pdev->bus->self, PCI_EXP_LNKCTL,
5617                                            state);
5618 }
5619 #endif
5620 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5621 {
5622         dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5623                  (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5624                  (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5625
5626         __e1000e_disable_aspm(pdev, state);
5627 }
5628
5629 #ifdef CONFIG_PM
5630 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5631 {
5632         return !!adapter->tx_ring->buffer_info;
5633 }
5634
5635 static int __e1000_resume(struct pci_dev *pdev)
5636 {
5637         struct net_device *netdev = pci_get_drvdata(pdev);
5638         struct e1000_adapter *adapter = netdev_priv(netdev);
5639         struct e1000_hw *hw = &adapter->hw;
5640         u16 aspm_disable_flag = 0;
5641         u32 err;
5642
5643         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5644                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5645         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5646                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5647         if (aspm_disable_flag)
5648                 e1000e_disable_aspm(pdev, aspm_disable_flag);
5649
5650         pci_set_power_state(pdev, PCI_D0);
5651         pci_restore_state(pdev);
5652         pci_save_state(pdev);
5653
5654         e1000e_set_interrupt_capability(adapter);
5655         if (netif_running(netdev)) {
5656                 err = e1000_request_irq(adapter);
5657                 if (err)
5658                         return err;
5659         }
5660
5661         if (hw->mac.type >= e1000_pch2lan)
5662                 e1000_resume_workarounds_pchlan(&adapter->hw);
5663
5664         e1000e_power_up_phy(adapter);
5665
5666         /* report the system wakeup cause from S3/S4 */
5667         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5668                 u16 phy_data;
5669
5670                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5671                 if (phy_data) {
5672                         e_info("PHY Wakeup cause - %s\n",
5673                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5674                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5675                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5676                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5677                                 phy_data & E1000_WUS_LNKC ?
5678                                 "Link Status Change" : "other");
5679                 }
5680                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5681         } else {
5682                 u32 wus = er32(WUS);
5683                 if (wus) {
5684                         e_info("MAC Wakeup cause - %s\n",
5685                                 wus & E1000_WUS_EX ? "Unicast Packet" :
5686                                 wus & E1000_WUS_MC ? "Multicast Packet" :
5687                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
5688                                 wus & E1000_WUS_MAG ? "Magic Packet" :
5689                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
5690                                 "other");
5691                 }
5692                 ew32(WUS, ~0);
5693         }
5694
5695         e1000e_reset(adapter);
5696
5697         e1000_init_manageability_pt(adapter);
5698
5699         if (netif_running(netdev))
5700                 e1000e_up(adapter);
5701
5702         netif_device_attach(netdev);
5703
5704         /*
5705          * If the controller has AMT, do not set DRV_LOAD until the interface
5706          * is up.  For all other cases, let the f/w know that the h/w is now
5707          * under the control of the driver.
5708          */
5709         if (!(adapter->flags & FLAG_HAS_AMT))
5710                 e1000e_get_hw_control(adapter);
5711
5712         return 0;
5713 }
5714
5715 #ifdef CONFIG_PM_SLEEP
5716 static int e1000_suspend(struct device *dev)
5717 {
5718         struct pci_dev *pdev = to_pci_dev(dev);
5719         int retval;
5720         bool wake;
5721
5722         retval = __e1000_shutdown(pdev, &wake, false);
5723         if (!retval)
5724                 e1000_complete_shutdown(pdev, true, wake);
5725
5726         return retval;
5727 }
5728
5729 static int e1000_resume(struct device *dev)
5730 {
5731         struct pci_dev *pdev = to_pci_dev(dev);
5732         struct net_device *netdev = pci_get_drvdata(pdev);
5733         struct e1000_adapter *adapter = netdev_priv(netdev);
5734
5735         if (e1000e_pm_ready(adapter))
5736                 adapter->idle_check = true;
5737
5738         return __e1000_resume(pdev);
5739 }
5740 #endif /* CONFIG_PM_SLEEP */
5741
5742 #ifdef CONFIG_PM_RUNTIME
5743 static int e1000_runtime_suspend(struct device *dev)
5744 {
5745         struct pci_dev *pdev = to_pci_dev(dev);
5746         struct net_device *netdev = pci_get_drvdata(pdev);
5747         struct e1000_adapter *adapter = netdev_priv(netdev);
5748
5749         if (e1000e_pm_ready(adapter)) {
5750                 bool wake;
5751
5752                 __e1000_shutdown(pdev, &wake, true);
5753         }
5754
5755         return 0;
5756 }
5757
5758 static int e1000_idle(struct device *dev)
5759 {
5760         struct pci_dev *pdev = to_pci_dev(dev);
5761         struct net_device *netdev = pci_get_drvdata(pdev);
5762         struct e1000_adapter *adapter = netdev_priv(netdev);
5763
5764         if (!e1000e_pm_ready(adapter))
5765                 return 0;
5766
5767         if (adapter->idle_check) {
5768                 adapter->idle_check = false;
5769                 if (!e1000e_has_link(adapter))
5770                         pm_schedule_suspend(dev, MSEC_PER_SEC);
5771         }
5772
5773         return -EBUSY;
5774 }
5775
5776 static int e1000_runtime_resume(struct device *dev)
5777 {
5778         struct pci_dev *pdev = to_pci_dev(dev);
5779         struct net_device *netdev = pci_get_drvdata(pdev);
5780         struct e1000_adapter *adapter = netdev_priv(netdev);
5781
5782         if (!e1000e_pm_ready(adapter))
5783                 return 0;
5784
5785         adapter->idle_check = !dev->power.runtime_auto;
5786         return __e1000_resume(pdev);
5787 }
5788 #endif /* CONFIG_PM_RUNTIME */
5789 #endif /* CONFIG_PM */
5790
5791 static void e1000_shutdown(struct pci_dev *pdev)
5792 {
5793         bool wake = false;
5794
5795         __e1000_shutdown(pdev, &wake, false);
5796
5797         if (system_state == SYSTEM_POWER_OFF)
5798                 e1000_complete_shutdown(pdev, false, wake);
5799 }
5800
5801 #ifdef CONFIG_NET_POLL_CONTROLLER
5802
5803 static irqreturn_t e1000_intr_msix(int irq, void *data)
5804 {
5805         struct net_device *netdev = data;
5806         struct e1000_adapter *adapter = netdev_priv(netdev);
5807
5808         if (adapter->msix_entries) {
5809                 int vector, msix_irq;
5810
5811                 vector = 0;
5812                 msix_irq = adapter->msix_entries[vector].vector;
5813                 disable_irq(msix_irq);
5814                 e1000_intr_msix_rx(msix_irq, netdev);
5815                 enable_irq(msix_irq);
5816
5817                 vector++;
5818                 msix_irq = adapter->msix_entries[vector].vector;
5819                 disable_irq(msix_irq);
5820                 e1000_intr_msix_tx(msix_irq, netdev);
5821                 enable_irq(msix_irq);
5822
5823                 vector++;
5824                 msix_irq = adapter->msix_entries[vector].vector;
5825                 disable_irq(msix_irq);
5826                 e1000_msix_other(msix_irq, netdev);
5827                 enable_irq(msix_irq);
5828         }
5829
5830         return IRQ_HANDLED;
5831 }
5832
5833 /*
5834  * Polling 'interrupt' - used by things like netconsole to send skbs
5835  * without having to re-enable interrupts. It's not called while
5836  * the interrupt routine is executing.
5837  */
5838 static void e1000_netpoll(struct net_device *netdev)
5839 {
5840         struct e1000_adapter *adapter = netdev_priv(netdev);
5841
5842         switch (adapter->int_mode) {
5843         case E1000E_INT_MODE_MSIX:
5844                 e1000_intr_msix(adapter->pdev->irq, netdev);
5845                 break;
5846         case E1000E_INT_MODE_MSI:
5847                 disable_irq(adapter->pdev->irq);
5848                 e1000_intr_msi(adapter->pdev->irq, netdev);
5849                 enable_irq(adapter->pdev->irq);
5850                 break;
5851         default: /* E1000E_INT_MODE_LEGACY */
5852                 disable_irq(adapter->pdev->irq);
5853                 e1000_intr(adapter->pdev->irq, netdev);
5854                 enable_irq(adapter->pdev->irq);
5855                 break;
5856         }
5857 }
5858 #endif
5859
5860 /**
5861  * e1000_io_error_detected - called when PCI error is detected
5862  * @pdev: Pointer to PCI device
5863  * @state: The current pci connection state
5864  *
5865  * This function is called after a PCI bus error affecting
5866  * this device has been detected.
5867  */
5868 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5869                                                 pci_channel_state_t state)
5870 {
5871         struct net_device *netdev = pci_get_drvdata(pdev);
5872         struct e1000_adapter *adapter = netdev_priv(netdev);
5873
5874         netif_device_detach(netdev);
5875
5876         if (state == pci_channel_io_perm_failure)
5877                 return PCI_ERS_RESULT_DISCONNECT;
5878
5879         if (netif_running(netdev))
5880                 e1000e_down(adapter);
5881         pci_disable_device(pdev);
5882
5883         /* Request a slot slot reset. */
5884         return PCI_ERS_RESULT_NEED_RESET;
5885 }
5886
5887 /**
5888  * e1000_io_slot_reset - called after the pci bus has been reset.
5889  * @pdev: Pointer to PCI device
5890  *
5891  * Restart the card from scratch, as if from a cold-boot. Implementation
5892  * resembles the first-half of the e1000_resume routine.
5893  */
5894 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5895 {
5896         struct net_device *netdev = pci_get_drvdata(pdev);
5897         struct e1000_adapter *adapter = netdev_priv(netdev);
5898         struct e1000_hw *hw = &adapter->hw;
5899         u16 aspm_disable_flag = 0;
5900         int err;
5901         pci_ers_result_t result;
5902
5903         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5904                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5905         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5906                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5907         if (aspm_disable_flag)
5908                 e1000e_disable_aspm(pdev, aspm_disable_flag);
5909
5910         err = pci_enable_device_mem(pdev);
5911         if (err) {
5912                 dev_err(&pdev->dev,
5913                         "Cannot re-enable PCI device after reset.\n");
5914                 result = PCI_ERS_RESULT_DISCONNECT;
5915         } else {
5916                 pci_set_master(pdev);
5917                 pdev->state_saved = true;
5918                 pci_restore_state(pdev);
5919
5920                 pci_enable_wake(pdev, PCI_D3hot, 0);
5921                 pci_enable_wake(pdev, PCI_D3cold, 0);
5922
5923                 e1000e_reset(adapter);
5924                 ew32(WUS, ~0);
5925                 result = PCI_ERS_RESULT_RECOVERED;
5926         }
5927
5928         pci_cleanup_aer_uncorrect_error_status(pdev);
5929
5930         return result;
5931 }
5932
5933 /**
5934  * e1000_io_resume - called when traffic can start flowing again.
5935  * @pdev: Pointer to PCI device
5936  *
5937  * This callback is called when the error recovery driver tells us that
5938  * its OK to resume normal operation. Implementation resembles the
5939  * second-half of the e1000_resume routine.
5940  */
5941 static void e1000_io_resume(struct pci_dev *pdev)
5942 {
5943         struct net_device *netdev = pci_get_drvdata(pdev);
5944         struct e1000_adapter *adapter = netdev_priv(netdev);
5945
5946         e1000_init_manageability_pt(adapter);
5947
5948         if (netif_running(netdev)) {
5949                 if (e1000e_up(adapter)) {
5950                         dev_err(&pdev->dev,
5951                                 "can't bring device back up after reset\n");
5952                         return;
5953                 }
5954         }
5955
5956         netif_device_attach(netdev);
5957
5958         /*
5959          * If the controller has AMT, do not set DRV_LOAD until the interface
5960          * is up.  For all other cases, let the f/w know that the h/w is now
5961          * under the control of the driver.
5962          */
5963         if (!(adapter->flags & FLAG_HAS_AMT))
5964                 e1000e_get_hw_control(adapter);
5965
5966 }
5967
5968 static void e1000_print_device_info(struct e1000_adapter *adapter)
5969 {
5970         struct e1000_hw *hw = &adapter->hw;
5971         struct net_device *netdev = adapter->netdev;
5972         u32 ret_val;
5973         u8 pba_str[E1000_PBANUM_LENGTH];
5974
5975         /* print bus type/speed/width info */
5976         e_info("(PCI Express:2.5GT/s:%s) %pM\n",
5977                /* bus width */
5978                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5979                 "Width x1"),
5980                /* MAC address */
5981                netdev->dev_addr);
5982         e_info("Intel(R) PRO/%s Network Connection\n",
5983                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5984         ret_val = e1000_read_pba_string_generic(hw, pba_str,
5985                                                 E1000_PBANUM_LENGTH);
5986         if (ret_val)
5987                 strlcpy((char *)pba_str, "Unknown", sizeof(pba_str));
5988         e_info("MAC: %d, PHY: %d, PBA No: %s\n",
5989                hw->mac.type, hw->phy.type, pba_str);
5990 }
5991
5992 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5993 {
5994         struct e1000_hw *hw = &adapter->hw;
5995         int ret_val;
5996         u16 buf = 0;
5997
5998         if (hw->mac.type != e1000_82573)
5999                 return;
6000
6001         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
6002         le16_to_cpus(&buf);
6003         if (!ret_val && (!(buf & (1 << 0)))) {
6004                 /* Deep Smart Power Down (DSPD) */
6005                 dev_warn(&adapter->pdev->dev,
6006                          "Warning: detected DSPD enabled in EEPROM\n");
6007         }
6008 }
6009
6010 static int e1000_set_features(struct net_device *netdev,
6011                               netdev_features_t features)
6012 {
6013         struct e1000_adapter *adapter = netdev_priv(netdev);
6014         netdev_features_t changed = features ^ netdev->features;
6015
6016         if (changed & (NETIF_F_TSO | NETIF_F_TSO6))
6017                 adapter->flags |= FLAG_TSO_FORCE;
6018
6019         if (!(changed & (NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX |
6020                          NETIF_F_RXCSUM | NETIF_F_RXHASH | NETIF_F_RXFCS |
6021                          NETIF_F_RXALL)))
6022                 return 0;
6023
6024         if (changed & NETIF_F_RXFCS) {
6025                 if (features & NETIF_F_RXFCS) {
6026                         adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6027                 } else {
6028                         /* We need to take it back to defaults, which might mean
6029                          * stripping is still disabled at the adapter level.
6030                          */
6031                         if (adapter->flags2 & FLAG2_DFLT_CRC_STRIPPING)
6032                                 adapter->flags2 |= FLAG2_CRC_STRIPPING;
6033                         else
6034                                 adapter->flags2 &= ~FLAG2_CRC_STRIPPING;
6035                 }
6036         }
6037
6038         netdev->features = features;
6039
6040         if (netif_running(netdev))
6041                 e1000e_reinit_locked(adapter);
6042         else
6043                 e1000e_reset(adapter);
6044
6045         return 0;
6046 }
6047
6048 static const struct net_device_ops e1000e_netdev_ops = {
6049         .ndo_open               = e1000_open,
6050         .ndo_stop               = e1000_close,
6051         .ndo_start_xmit         = e1000_xmit_frame,
6052         .ndo_get_stats64        = e1000e_get_stats64,
6053         .ndo_set_rx_mode        = e1000e_set_rx_mode,
6054         .ndo_set_mac_address    = e1000_set_mac,
6055         .ndo_change_mtu         = e1000_change_mtu,
6056         .ndo_do_ioctl           = e1000_ioctl,
6057         .ndo_tx_timeout         = e1000_tx_timeout,
6058         .ndo_validate_addr      = eth_validate_addr,
6059
6060         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
6061         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
6062 #ifdef CONFIG_NET_POLL_CONTROLLER
6063         .ndo_poll_controller    = e1000_netpoll,
6064 #endif
6065         .ndo_set_features = e1000_set_features,
6066 };
6067
6068 /**
6069  * e1000_probe - Device Initialization Routine
6070  * @pdev: PCI device information struct
6071  * @ent: entry in e1000_pci_tbl
6072  *
6073  * Returns 0 on success, negative on failure
6074  *
6075  * e1000_probe initializes an adapter identified by a pci_dev structure.
6076  * The OS initialization, configuring of the adapter private structure,
6077  * and a hardware reset occur.
6078  **/
6079 static int __devinit e1000_probe(struct pci_dev *pdev,
6080                                  const struct pci_device_id *ent)
6081 {
6082         struct net_device *netdev;
6083         struct e1000_adapter *adapter;
6084         struct e1000_hw *hw;
6085         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
6086         resource_size_t mmio_start, mmio_len;
6087         resource_size_t flash_start, flash_len;
6088         static int cards_found;
6089         u16 aspm_disable_flag = 0;
6090         int i, err, pci_using_dac;
6091         u16 eeprom_data = 0;
6092         u16 eeprom_apme_mask = E1000_EEPROM_APME;
6093
6094         if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
6095                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
6096         if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
6097                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
6098         if (aspm_disable_flag)
6099                 e1000e_disable_aspm(pdev, aspm_disable_flag);
6100
6101         err = pci_enable_device_mem(pdev);
6102         if (err)
6103                 return err;
6104
6105         pci_using_dac = 0;
6106         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
6107         if (!err) {
6108                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
6109                 if (!err)
6110                         pci_using_dac = 1;
6111         } else {
6112                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
6113                 if (err) {
6114                         err = dma_set_coherent_mask(&pdev->dev,
6115                                                     DMA_BIT_MASK(32));
6116                         if (err) {
6117                                 dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
6118                                 goto err_dma;
6119                         }
6120                 }
6121         }
6122
6123         err = pci_request_selected_regions_exclusive(pdev,
6124                                           pci_select_bars(pdev, IORESOURCE_MEM),
6125                                           e1000e_driver_name);
6126         if (err)
6127                 goto err_pci_reg;
6128
6129         /* AER (Advanced Error Reporting) hooks */
6130         pci_enable_pcie_error_reporting(pdev);
6131
6132         pci_set_master(pdev);
6133         /* PCI config space info */
6134         err = pci_save_state(pdev);
6135         if (err)
6136                 goto err_alloc_etherdev;
6137
6138         err = -ENOMEM;
6139         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
6140         if (!netdev)
6141                 goto err_alloc_etherdev;
6142
6143         SET_NETDEV_DEV(netdev, &pdev->dev);
6144
6145         netdev->irq = pdev->irq;
6146
6147         pci_set_drvdata(pdev, netdev);
6148         adapter = netdev_priv(netdev);
6149         hw = &adapter->hw;
6150         adapter->netdev = netdev;
6151         adapter->pdev = pdev;
6152         adapter->ei = ei;
6153         adapter->pba = ei->pba;
6154         adapter->flags = ei->flags;
6155         adapter->flags2 = ei->flags2;
6156         adapter->hw.adapter = adapter;
6157         adapter->hw.mac.type = ei->mac;
6158         adapter->max_hw_frame_size = ei->max_hw_frame_size;
6159         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
6160
6161         mmio_start = pci_resource_start(pdev, 0);
6162         mmio_len = pci_resource_len(pdev, 0);
6163
6164         err = -EIO;
6165         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
6166         if (!adapter->hw.hw_addr)
6167                 goto err_ioremap;
6168
6169         if ((adapter->flags & FLAG_HAS_FLASH) &&
6170             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
6171                 flash_start = pci_resource_start(pdev, 1);
6172                 flash_len = pci_resource_len(pdev, 1);
6173                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
6174                 if (!adapter->hw.flash_address)
6175                         goto err_flashmap;
6176         }
6177
6178         /* construct the net_device struct */
6179         netdev->netdev_ops              = &e1000e_netdev_ops;
6180         e1000e_set_ethtool_ops(netdev);
6181         netdev->watchdog_timeo          = 5 * HZ;
6182         netif_napi_add(netdev, &adapter->napi, e1000e_poll, 64);
6183         strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
6184
6185         netdev->mem_start = mmio_start;
6186         netdev->mem_end = mmio_start + mmio_len;
6187
6188         adapter->bd_number = cards_found++;
6189
6190         e1000e_check_options(adapter);
6191
6192         /* setup adapter struct */
6193         err = e1000_sw_init(adapter);
6194         if (err)
6195                 goto err_sw_init;
6196
6197         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
6198         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
6199         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
6200
6201         err = ei->get_variants(adapter);
6202         if (err)
6203                 goto err_hw_init;
6204
6205         if ((adapter->flags & FLAG_IS_ICH) &&
6206             (adapter->flags & FLAG_READ_ONLY_NVM))
6207                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
6208
6209         hw->mac.ops.get_bus_info(&adapter->hw);
6210
6211         adapter->hw.phy.autoneg_wait_to_complete = 0;
6212
6213         /* Copper options */
6214         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
6215                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
6216                 adapter->hw.phy.disable_polarity_correction = 0;
6217                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
6218         }
6219
6220         if (hw->phy.ops.check_reset_block && hw->phy.ops.check_reset_block(hw))
6221                 dev_info(&pdev->dev,
6222                          "PHY reset is blocked due to SOL/IDER session.\n");
6223
6224         /* Set initial default active device features */
6225         netdev->features = (NETIF_F_SG |
6226                             NETIF_F_HW_VLAN_RX |
6227                             NETIF_F_HW_VLAN_TX |
6228                             NETIF_F_TSO |
6229                             NETIF_F_TSO6 |
6230                             NETIF_F_RXHASH |
6231                             NETIF_F_RXCSUM |
6232                             NETIF_F_HW_CSUM);
6233
6234         /* Set user-changeable features (subset of all device features) */
6235         netdev->hw_features = netdev->features;
6236         netdev->hw_features |= NETIF_F_RXFCS;
6237         netdev->priv_flags |= IFF_SUPP_NOFCS;
6238         netdev->hw_features |= NETIF_F_RXALL;
6239
6240         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
6241                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
6242
6243         netdev->vlan_features |= (NETIF_F_SG |
6244                                   NETIF_F_TSO |
6245                                   NETIF_F_TSO6 |
6246                                   NETIF_F_HW_CSUM);
6247
6248         netdev->priv_flags |= IFF_UNICAST_FLT;
6249
6250         if (pci_using_dac) {
6251                 netdev->features |= NETIF_F_HIGHDMA;
6252                 netdev->vlan_features |= NETIF_F_HIGHDMA;
6253         }
6254
6255         if (e1000e_enable_mng_pass_thru(&adapter->hw))
6256                 adapter->flags |= FLAG_MNG_PT_ENABLED;
6257
6258         /*
6259          * before reading the NVM, reset the controller to
6260          * put the device in a known good starting state
6261          */
6262         adapter->hw.mac.ops.reset_hw(&adapter->hw);
6263
6264         /*
6265          * systems with ASPM and others may see the checksum fail on the first
6266          * attempt. Let's give it a few tries
6267          */
6268         for (i = 0;; i++) {
6269                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
6270                         break;
6271                 if (i == 2) {
6272                         dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
6273                         err = -EIO;
6274                         goto err_eeprom;
6275                 }
6276         }
6277
6278         e1000_eeprom_checks(adapter);
6279
6280         /* copy the MAC address */
6281         if (e1000e_read_mac_addr(&adapter->hw))
6282                 dev_err(&pdev->dev,
6283                         "NVM Read Error while reading MAC address\n");
6284
6285         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
6286         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
6287
6288         if (!is_valid_ether_addr(netdev->perm_addr)) {
6289                 dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
6290                         netdev->perm_addr);
6291                 err = -EIO;
6292                 goto err_eeprom;
6293         }
6294
6295         init_timer(&adapter->watchdog_timer);
6296         adapter->watchdog_timer.function = e1000_watchdog;
6297         adapter->watchdog_timer.data = (unsigned long) adapter;
6298
6299         init_timer(&adapter->phy_info_timer);
6300         adapter->phy_info_timer.function = e1000_update_phy_info;
6301         adapter->phy_info_timer.data = (unsigned long) adapter;
6302
6303         INIT_WORK(&adapter->reset_task, e1000_reset_task);
6304         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
6305         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
6306         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
6307         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
6308
6309         /* Initialize link parameters. User can change them with ethtool */
6310         adapter->hw.mac.autoneg = 1;
6311         adapter->fc_autoneg = true;
6312         adapter->hw.fc.requested_mode = e1000_fc_default;
6313         adapter->hw.fc.current_mode = e1000_fc_default;
6314         adapter->hw.phy.autoneg_advertised = 0x2f;
6315
6316         /* ring size defaults */
6317         adapter->rx_ring->count = E1000_DEFAULT_RXD;
6318         adapter->tx_ring->count = E1000_DEFAULT_TXD;
6319
6320         /*
6321          * Initial Wake on LAN setting - If APM wake is enabled in
6322          * the EEPROM, enable the ACPI Magic Packet filter
6323          */
6324         if (adapter->flags & FLAG_APME_IN_WUC) {
6325                 /* APME bit in EEPROM is mapped to WUC.APME */
6326                 eeprom_data = er32(WUC);
6327                 eeprom_apme_mask = E1000_WUC_APME;
6328                 if ((hw->mac.type > e1000_ich10lan) &&
6329                     (eeprom_data & E1000_WUC_PHY_WAKE))
6330                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
6331         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
6332                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
6333                     (adapter->hw.bus.func == 1))
6334                         e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_B,
6335                                        1, &eeprom_data);
6336                 else
6337                         e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_A,
6338                                        1, &eeprom_data);
6339         }
6340
6341         /* fetch WoL from EEPROM */
6342         if (eeprom_data & eeprom_apme_mask)
6343                 adapter->eeprom_wol |= E1000_WUFC_MAG;
6344
6345         /*
6346          * now that we have the eeprom settings, apply the special cases
6347          * where the eeprom may be wrong or the board simply won't support
6348          * wake on lan on a particular port
6349          */
6350         if (!(adapter->flags & FLAG_HAS_WOL))
6351                 adapter->eeprom_wol = 0;
6352
6353         /* initialize the wol settings based on the eeprom settings */
6354         adapter->wol = adapter->eeprom_wol;
6355         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6356
6357         /* save off EEPROM version number */
6358         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
6359
6360         /* reset the hardware with the new settings */
6361         e1000e_reset(adapter);
6362
6363         /*
6364          * If the controller has AMT, do not set DRV_LOAD until the interface
6365          * is up.  For all other cases, let the f/w know that the h/w is now
6366          * under the control of the driver.
6367          */
6368         if (!(adapter->flags & FLAG_HAS_AMT))
6369                 e1000e_get_hw_control(adapter);
6370
6371         strlcpy(netdev->name, "eth%d", sizeof(netdev->name));
6372         err = register_netdev(netdev);
6373         if (err)
6374                 goto err_register;
6375
6376         /* carrier off reporting is important to ethtool even BEFORE open */
6377         netif_carrier_off(netdev);
6378
6379         e1000_print_device_info(adapter);
6380
6381         if (pci_dev_run_wake(pdev))
6382                 pm_runtime_put_noidle(&pdev->dev);
6383
6384         return 0;
6385
6386 err_register:
6387         if (!(adapter->flags & FLAG_HAS_AMT))
6388                 e1000e_release_hw_control(adapter);
6389 err_eeprom:
6390         if (hw->phy.ops.check_reset_block && !hw->phy.ops.check_reset_block(hw))
6391                 e1000_phy_hw_reset(&adapter->hw);
6392 err_hw_init:
6393         kfree(adapter->tx_ring);
6394         kfree(adapter->rx_ring);
6395 err_sw_init:
6396         if (adapter->hw.flash_address)
6397                 iounmap(adapter->hw.flash_address);
6398         e1000e_reset_interrupt_capability(adapter);
6399 err_flashmap:
6400         iounmap(adapter->hw.hw_addr);
6401 err_ioremap:
6402         free_netdev(netdev);
6403 err_alloc_etherdev:
6404         pci_release_selected_regions(pdev,
6405                                      pci_select_bars(pdev, IORESOURCE_MEM));
6406 err_pci_reg:
6407 err_dma:
6408         pci_disable_device(pdev);
6409         return err;
6410 }
6411
6412 /**
6413  * e1000_remove - Device Removal Routine
6414  * @pdev: PCI device information struct
6415  *
6416  * e1000_remove is called by the PCI subsystem to alert the driver
6417  * that it should release a PCI device.  The could be caused by a
6418  * Hot-Plug event, or because the driver is going to be removed from
6419  * memory.
6420  **/
6421 static void __devexit e1000_remove(struct pci_dev *pdev)
6422 {
6423         struct net_device *netdev = pci_get_drvdata(pdev);
6424         struct e1000_adapter *adapter = netdev_priv(netdev);
6425         bool down = test_bit(__E1000_DOWN, &adapter->state);
6426
6427         /*
6428          * The timers may be rescheduled, so explicitly disable them
6429          * from being rescheduled.
6430          */
6431         if (!down)
6432                 set_bit(__E1000_DOWN, &adapter->state);
6433         del_timer_sync(&adapter->watchdog_timer);
6434         del_timer_sync(&adapter->phy_info_timer);
6435
6436         cancel_work_sync(&adapter->reset_task);
6437         cancel_work_sync(&adapter->watchdog_task);
6438         cancel_work_sync(&adapter->downshift_task);
6439         cancel_work_sync(&adapter->update_phy_task);
6440         cancel_work_sync(&adapter->print_hang_task);
6441
6442         if (!(netdev->flags & IFF_UP))
6443                 e1000_power_down_phy(adapter);
6444
6445         /* Don't lie to e1000_close() down the road. */
6446         if (!down)
6447                 clear_bit(__E1000_DOWN, &adapter->state);
6448         unregister_netdev(netdev);
6449
6450         if (pci_dev_run_wake(pdev))
6451                 pm_runtime_get_noresume(&pdev->dev);
6452
6453         /*
6454          * Release control of h/w to f/w.  If f/w is AMT enabled, this
6455          * would have already happened in close and is redundant.
6456          */
6457         e1000e_release_hw_control(adapter);
6458
6459         e1000e_reset_interrupt_capability(adapter);
6460         kfree(adapter->tx_ring);
6461         kfree(adapter->rx_ring);
6462
6463         iounmap(adapter->hw.hw_addr);
6464         if (adapter->hw.flash_address)
6465                 iounmap(adapter->hw.flash_address);
6466         pci_release_selected_regions(pdev,
6467                                      pci_select_bars(pdev, IORESOURCE_MEM));
6468
6469         free_netdev(netdev);
6470
6471         /* AER disable */
6472         pci_disable_pcie_error_reporting(pdev);
6473
6474         pci_disable_device(pdev);
6475 }
6476
6477 /* PCI Error Recovery (ERS) */
6478 static const struct pci_error_handlers e1000_err_handler = {
6479         .error_detected = e1000_io_error_detected,
6480         .slot_reset = e1000_io_slot_reset,
6481         .resume = e1000_io_resume,
6482 };
6483
6484 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
6485         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
6486         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
6487         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
6488         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
6489         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
6490         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
6491         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
6492         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
6493         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
6494
6495         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
6496         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
6497         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
6498         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
6499
6500         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
6501         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
6502         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
6503
6504         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
6505         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
6506         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
6507
6508         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
6509           board_80003es2lan },
6510         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
6511           board_80003es2lan },
6512         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
6513           board_80003es2lan },
6514         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
6515           board_80003es2lan },
6516
6517         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
6518         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
6519         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
6520         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
6521         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
6522         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
6523         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
6524         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
6525
6526         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
6527         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
6528         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
6529         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
6530         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
6531         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
6532         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
6533         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
6534         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
6535
6536         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
6537         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
6538         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
6539
6540         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
6541         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
6542         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
6543
6544         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
6545         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
6546         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
6547         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
6548
6549         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
6550         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
6551
6552         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_LM), board_pch_lpt },
6553         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_LPT_I217_V), board_pch_lpt },
6554
6555         { 0, 0, 0, 0, 0, 0, 0 } /* terminate list */
6556 };
6557 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
6558
6559 #ifdef CONFIG_PM
6560 static const struct dev_pm_ops e1000_pm_ops = {
6561         SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
6562         SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
6563                                 e1000_runtime_resume, e1000_idle)
6564 };
6565 #endif
6566
6567 /* PCI Device API Driver */
6568 static struct pci_driver e1000_driver = {
6569         .name     = e1000e_driver_name,
6570         .id_table = e1000_pci_tbl,
6571         .probe    = e1000_probe,
6572         .remove   = __devexit_p(e1000_remove),
6573 #ifdef CONFIG_PM
6574         .driver   = {
6575                 .pm = &e1000_pm_ops,
6576         },
6577 #endif
6578         .shutdown = e1000_shutdown,
6579         .err_handler = &e1000_err_handler
6580 };
6581
6582 /**
6583  * e1000_init_module - Driver Registration Routine
6584  *
6585  * e1000_init_module is the first routine called when the driver is
6586  * loaded. All it does is register with the PCI subsystem.
6587  **/
6588 static int __init e1000_init_module(void)
6589 {
6590         int ret;
6591         pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
6592                 e1000e_driver_version);
6593         pr_info("Copyright(c) 1999 - 2012 Intel Corporation.\n");
6594         ret = pci_register_driver(&e1000_driver);
6595
6596         return ret;
6597 }
6598 module_init(e1000_init_module);
6599
6600 /**
6601  * e1000_exit_module - Driver Exit Cleanup Routine
6602  *
6603  * e1000_exit_module is called just before the driver is removed
6604  * from memory.
6605  **/
6606 static void __exit e1000_exit_module(void)
6607 {
6608         pci_unregister_driver(&e1000_driver);
6609 }
6610 module_exit(e1000_exit_module);
6611
6612
6613 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
6614 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
6615 MODULE_LICENSE("GPL");
6616 MODULE_VERSION(DRV_VERSION);
6617
6618 /* netdev.c */