drivers/net/e1000e/netdev.c

   1 /*******************************************************************************
   2
   3   Intel PRO/1000 Linux driver
   4   Copyright(c) 1999 - 2009 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/tcp.h>
  40 #include <linux/ipv6.h>
  41 #include <linux/slab.h>
  42 #include <net/checksum.h>
  43 #include <net/ip6_checksum.h>
  44 #include <linux/mii.h>
  45 #include <linux/ethtool.h>
  46 #include <linux/if_vlan.h>
  47 #include <linux/cpu.h>
  48 #include <linux/smp.h>
  49 #include <linux/pm_qos_params.h>
  50 #include <linux/pm_runtime.h>
  51 #include <linux/aer.h>
  52
  53 #include "e1000.h"
  54
  55 #define DRV_VERSION "1.0.2-k2"
  56 char e1000e_driver_name[] = "e1000e";
  57 const char e1000e_driver_version[] = DRV_VERSION;
  58
  59 static const struct e1000_info *e1000_info_tbl[] = {
  60         [board_82571]           = &e1000_82571_info,
  61         [board_82572]           = &e1000_82572_info,
  62         [board_82573]           = &e1000_82573_info,
  63         [board_82574]           = &e1000_82574_info,
  64         [board_82583]           = &e1000_82583_info,
  65         [board_80003es2lan]     = &e1000_es2_info,
  66         [board_ich8lan]         = &e1000_ich8_info,
  67         [board_ich9lan]         = &e1000_ich9_info,
  68         [board_ich10lan]        = &e1000_ich10_info,
  69         [board_pchlan]          = &e1000_pch_info,
  70 };
  71
  72 struct e1000_reg_info {
  73         u32 ofs;
  74         char *name;
  75 };
  76
  77 #define E1000_RDFH      0x02410 /* Rx Data FIFO Head - RW */
  78 #define E1000_RDFT      0x02418 /* Rx Data FIFO Tail - RW */
  79 #define E1000_RDFHS     0x02420 /* Rx Data FIFO Head Saved - RW */
  80 #define E1000_RDFTS     0x02428 /* Rx Data FIFO Tail Saved - RW */
  81 #define E1000_RDFPC     0x02430 /* Rx Data FIFO Packet Count - RW */
  82
  83 #define E1000_TDFH      0x03410 /* Tx Data FIFO Head - RW */
  84 #define E1000_TDFT      0x03418 /* Tx Data FIFO Tail - RW */
  85 #define E1000_TDFHS     0x03420 /* Tx Data FIFO Head Saved - RW */
  86 #define E1000_TDFTS     0x03428 /* Tx Data FIFO Tail Saved - RW */
  87 #define E1000_TDFPC     0x03430 /* Tx Data FIFO Packet Count - RW */
  88
  89 static const struct e1000_reg_info e1000_reg_info_tbl[] = {
  90
  91         /* General Registers */
  92         {E1000_CTRL, "CTRL"},
  93         {E1000_STATUS, "STATUS"},
  94         {E1000_CTRL_EXT, "CTRL_EXT"},
  95
  96         /* Interrupt Registers */
  97         {E1000_ICR, "ICR"},
  98
  99         /* RX Registers */
 100         {E1000_RCTL, "RCTL"},
 101         {E1000_RDLEN, "RDLEN"},
 102         {E1000_RDH, "RDH"},
 103         {E1000_RDT, "RDT"},
 104         {E1000_RDTR, "RDTR"},
 105         {E1000_RXDCTL(0), "RXDCTL"},
 106         {E1000_ERT, "ERT"},
 107         {E1000_RDBAL, "RDBAL"},
 108         {E1000_RDBAH, "RDBAH"},
 109         {E1000_RDFH, "RDFH"},
 110         {E1000_RDFT, "RDFT"},
 111         {E1000_RDFHS, "RDFHS"},
 112         {E1000_RDFTS, "RDFTS"},
 113         {E1000_RDFPC, "RDFPC"},
 114
 115         /* TX Registers */
 116         {E1000_TCTL, "TCTL"},
 117         {E1000_TDBAL, "TDBAL"},
 118         {E1000_TDBAH, "TDBAH"},
 119         {E1000_TDLEN, "TDLEN"},
 120         {E1000_TDH, "TDH"},
 121         {E1000_TDT, "TDT"},
 122         {E1000_TIDV, "TIDV"},
 123         {E1000_TXDCTL(0), "TXDCTL"},
 124         {E1000_TADV, "TADV"},
 125         {E1000_TARC(0), "TARC"},
 126         {E1000_TDFH, "TDFH"},
 127         {E1000_TDFT, "TDFT"},
 128         {E1000_TDFHS, "TDFHS"},
 129         {E1000_TDFTS, "TDFTS"},
 130         {E1000_TDFPC, "TDFPC"},
 131
 132         /* List Terminator */
 133         {}
 134 };
 135
 136 /*
 137  * e1000_regdump - register printout routine
 138  */
 139 static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
 140 {
 141         int n = 0;
 142         char rname[16];
 143         u32 regs[8];
 144
 145         switch (reginfo->ofs) {
 146         case E1000_RXDCTL(0):
 147                 for (n = 0; n < 2; n++)
 148                         regs[n] = __er32(hw, E1000_RXDCTL(n));
 149                 break;
 150         case E1000_TXDCTL(0):
 151                 for (n = 0; n < 2; n++)
 152                         regs[n] = __er32(hw, E1000_TXDCTL(n));
 153                 break;
 154         case E1000_TARC(0):
 155                 for (n = 0; n < 2; n++)
 156                         regs[n] = __er32(hw, E1000_TARC(n));
 157                 break;
 158         default:
 159                 printk(KERN_INFO "%-15s %08x\n",
 160                         reginfo->name, __er32(hw, reginfo->ofs));
 161                 return;
 162         }
 163
 164         snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
 165         printk(KERN_INFO "%-15s ", rname);
 166         for (n = 0; n < 2; n++)
 167                 printk(KERN_CONT "%08x ", regs[n]);
 168         printk(KERN_CONT "\n");
 169 }
 170
 171
 172 /*
 173  * e1000e_dump - Print registers, tx-ring and rx-ring
 174  */
 175 static void e1000e_dump(struct e1000_adapter *adapter)
 176 {
 177         struct net_device *netdev = adapter->netdev;
 178         struct e1000_hw *hw = &adapter->hw;
 179         struct e1000_reg_info *reginfo;
 180         struct e1000_ring *tx_ring = adapter->tx_ring;
 181         struct e1000_tx_desc *tx_desc;
 182         struct my_u0 { u64 a; u64 b; } *u0;
 183         struct e1000_buffer *buffer_info;
 184         struct e1000_ring *rx_ring = adapter->rx_ring;
 185         union e1000_rx_desc_packet_split *rx_desc_ps;
 186         struct e1000_rx_desc *rx_desc;
 187         struct my_u1 { u64 a; u64 b; u64 c; u64 d; } *u1;
 188         u32 staterr;
 189         int i = 0;
 190
 191         if (!netif_msg_hw(adapter))
 192                 return;
 193
 194         /* Print netdevice Info */
 195         if (netdev) {
 196                 dev_info(&adapter->pdev->dev, "Net device Info\n");
 197                 printk(KERN_INFO "Device Name     state            "
 198                         "trans_start      last_rx\n");
 199                 printk(KERN_INFO "%-15s %016lX %016lX %016lX\n",
 200                         netdev->name,
 201                         netdev->state,
 202                         netdev->trans_start,
 203                         netdev->last_rx);
 204         }
 205
 206         /* Print Registers */
 207         dev_info(&adapter->pdev->dev, "Register Dump\n");
 208         printk(KERN_INFO " Register Name   Value\n");
 209         for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
 210              reginfo->name; reginfo++) {
 211                 e1000_regdump(hw, reginfo);
 212         }
 213
 214         /* Print TX Ring Summary */
 215         if (!netdev || !netif_running(netdev))
 216                 goto exit;
 217
 218         dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
 219         printk(KERN_INFO "Queue [NTU] [NTC] [bi(ntc)->dma  ]"
 220                 " leng ntw timestamp\n");
 221         buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
 222         printk(KERN_INFO " %5d %5X %5X %016llX %04X %3X %016llX\n",
 223                 0, tx_ring->next_to_use, tx_ring->next_to_clean,
 224                 (u64)buffer_info->dma,
 225                 buffer_info->length,
 226                 buffer_info->next_to_watch,
 227                 (u64)buffer_info->time_stamp);
 228
 229         /* Print TX Rings */
 230         if (!netif_msg_tx_done(adapter))
 231                 goto rx_ring_summary;
 232
 233         dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
 234
 235         /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
 236          *
 237          * Legacy Transmit Descriptor
 238          *   +--------------------------------------------------------------+
 239          * 0 |         Buffer Address [63:0] (Reserved on Write Back)       |
 240          *   +--------------------------------------------------------------+
 241          * 8 | Special  |    CSS     | Status |  CMD    |  CSO   |  Length  |
 242          *   +--------------------------------------------------------------+
 243          *   63       48 47        36 35    32 31     24 23    16 15        0
 244          *
 245          * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
 246          *   63      48 47    40 39       32 31             16 15    8 7      0
 247          *   +----------------------------------------------------------------+
 248          * 0 |  TUCSE  | TUCS0  |   TUCSS   |     IPCSE       | IPCS0 | IPCSS |
 249          *   +----------------------------------------------------------------+
 250          * 8 |   MSS   | HDRLEN | RSV | STA | TUCMD | DTYP |      PAYLEN      |
 251          *   +----------------------------------------------------------------+
 252          *   63      48 47    40 39 36 35 32 31   24 23  20 19                0
 253          *
 254          * Extended Data Descriptor (DTYP=0x1)
 255          *   +----------------------------------------------------------------+
 256          * 0 |                     Buffer Address [63:0]                      |
 257          *   +----------------------------------------------------------------+
 258          * 8 | VLAN tag |  POPTS  | Rsvd | Status | Command | DTYP |  DTALEN  |
 259          *   +----------------------------------------------------------------+
 260          *   63       48 47     40 39  36 35    32 31     24 23  20 19        0
 261          */
 262         printk(KERN_INFO "Tl[desc]     [address 63:0  ] [SpeCssSCmCsLen]"
 263                 " [bi->dma       ] leng  ntw timestamp        bi->skb "
 264                 "<-- Legacy format\n");
 265         printk(KERN_INFO "Tc[desc]     [Ce CoCsIpceCoS] [MssHlRSCm0Plen]"
 266                 " [bi->dma       ] leng  ntw timestamp        bi->skb "
 267                 "<-- Ext Context format\n");
 268         printk(KERN_INFO "Td[desc]     [address 63:0  ] [VlaPoRSCm1Dlen]"
 269                 " [bi->dma       ] leng  ntw timestamp        bi->skb "
 270                 "<-- Ext Data format\n");
 271         for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
 272                 tx_desc = E1000_TX_DESC(*tx_ring, i);
 273                 buffer_info = &tx_ring->buffer_info[i];
 274                 u0 = (struct my_u0 *)tx_desc;
 275                 printk(KERN_INFO "T%c[0x%03X]    %016llX %016llX %016llX "
 276                         "%04X  %3X %016llX %p",
 277                        (!(le64_to_cpu(u0->b) & (1<<29)) ? 'l' :
 278                         ((le64_to_cpu(u0->b) & (1<<20)) ? 'd' : 'c')), i,
 279                        le64_to_cpu(u0->a), le64_to_cpu(u0->b),
 280                        (u64)buffer_info->dma, buffer_info->length,
 281                        buffer_info->next_to_watch, (u64)buffer_info->time_stamp,
 282                        buffer_info->skb);
 283                 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
 284                         printk(KERN_CONT " NTC/U\n");
 285                 else if (i == tx_ring->next_to_use)
 286                         printk(KERN_CONT " NTU\n");
 287                 else if (i == tx_ring->next_to_clean)
 288                         printk(KERN_CONT " NTC\n");
 289                 else
 290                         printk(KERN_CONT "\n");
 291
 292                 if (netif_msg_pktdata(adapter) && buffer_info->dma != 0)
 293                         print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
 294                                         16, 1, phys_to_virt(buffer_info->dma),
 295                                         buffer_info->length, true);
 296         }
 297
 298         /* Print RX Rings Summary */
 299 rx_ring_summary:
 300         dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
 301         printk(KERN_INFO "Queue [NTU] [NTC]\n");
 302         printk(KERN_INFO " %5d %5X %5X\n", 0,
 303                 rx_ring->next_to_use, rx_ring->next_to_clean);
 304
 305         /* Print RX Rings */
 306         if (!netif_msg_rx_status(adapter))
 307                 goto exit;
 308
 309         dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
 310         switch (adapter->rx_ps_pages) {
 311         case 1:
 312         case 2:
 313         case 3:
 314                 /* [Extended] Packet Split Receive Descriptor Format
 315                  *
 316                  *    +-----------------------------------------------------+
 317                  *  0 |                Buffer Address 0 [63:0]              |
 318                  *    +-----------------------------------------------------+
 319                  *  8 |                Buffer Address 1 [63:0]              |
 320                  *    +-----------------------------------------------------+
 321                  * 16 |                Buffer Address 2 [63:0]              |
 322                  *    +-----------------------------------------------------+
 323                  * 24 |                Buffer Address 3 [63:0]              |
 324                  *    +-----------------------------------------------------+
 325                  */
 326                 printk(KERN_INFO "R  [desc]      [buffer 0 63:0 ] "
 327                         "[buffer 1 63:0 ] "
 328                        "[buffer 2 63:0 ] [buffer 3 63:0 ] [bi->dma       ] "
 329                        "[bi->skb] <-- Ext Pkt Split format\n");
 330                 /* [Extended] Receive Descriptor (Write-Back) Format
 331                  *
 332                  *   63       48 47    32 31     13 12    8 7    4 3        0
 333                  *   +------------------------------------------------------+
 334                  * 0 | Packet   | IP     |  Rsvd   | MRQ   | Rsvd | MRQ RSS |
 335                  *   | Checksum | Ident  |         | Queue |      |  Type   |
 336                  *   +------------------------------------------------------+
 337                  * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 338                  *   +------------------------------------------------------+
 339                  *   63       48 47    32 31            20 19               0
 340                  */
 341                 printk(KERN_INFO "RWB[desc]      [ck ipid mrqhsh] "
 342                         "[vl   l0 ee  es] "
 343                        "[ l3  l2  l1 hs] [reserved      ] ---------------- "
 344                        "[bi->skb] <-- Ext Rx Write-Back format\n");
 345                 for (i = 0; i < rx_ring->count; i++) {
 346                         buffer_info = &rx_ring->buffer_info[i];
 347                         rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
 348                         u1 = (struct my_u1 *)rx_desc_ps;
 349                         staterr =
 350                                 le32_to_cpu(rx_desc_ps->wb.middle.status_error);
 351                         if (staterr & E1000_RXD_STAT_DD) {
 352                                 /* Descriptor Done */
 353                                 printk(KERN_INFO "RWB[0x%03X]     %016llX "
 354                                         "%016llX %016llX %016llX "
 355                                         "---------------- %p", i,
 356                                         le64_to_cpu(u1->a),
 357                                         le64_to_cpu(u1->b),
 358                                         le64_to_cpu(u1->c),
 359                                         le64_to_cpu(u1->d),
 360                                         buffer_info->skb);
 361                         } else {
 362                                 printk(KERN_INFO "R  [0x%03X]     %016llX "
 363                                         "%016llX %016llX %016llX %016llX %p", i,
 364                                         le64_to_cpu(u1->a),
 365                                         le64_to_cpu(u1->b),
 366                                         le64_to_cpu(u1->c),
 367                                         le64_to_cpu(u1->d),
 368                                         (u64)buffer_info->dma,
 369                                         buffer_info->skb);
 370
 371                                 if (netif_msg_pktdata(adapter))
 372                                         print_hex_dump(KERN_INFO, "",
 373                                                 DUMP_PREFIX_ADDRESS, 16, 1,
 374                                                 phys_to_virt(buffer_info->dma),
 375                                                 adapter->rx_ps_bsize0, true);
 376                         }
 377
 378                         if (i == rx_ring->next_to_use)
 379                                 printk(KERN_CONT " NTU\n");
 380                         else if (i == rx_ring->next_to_clean)
 381                                 printk(KERN_CONT " NTC\n");
 382                         else
 383                                 printk(KERN_CONT "\n");
 384                 }
 385                 break;
 386         default:
 387         case 0:
 388                 /* Legacy Receive Descriptor Format
 389                  *
 390                  * +-----------------------------------------------------+
 391                  * |                Buffer Address [63:0]                |
 392                  * +-----------------------------------------------------+
 393                  * | VLAN Tag | Errors | Status 0 | Packet csum | Length |
 394                  * +-----------------------------------------------------+
 395                  * 63       48 47    40 39      32 31         16 15      0
 396                  */
 397                 printk(KERN_INFO "Rl[desc]     [address 63:0  ] "
 398                         "[vl er S cks ln] [bi->dma       ] [bi->skb] "
 399                         "<-- Legacy format\n");
 400                 for (i = 0; rx_ring->desc && (i < rx_ring->count); i++) {
 401                         rx_desc = E1000_RX_DESC(*rx_ring, i);
 402                         buffer_info = &rx_ring->buffer_info[i];
 403                         u0 = (struct my_u0 *)rx_desc;
 404                         printk(KERN_INFO "Rl[0x%03X]    %016llX %016llX "
 405                                 "%016llX %p",
 406                                 i, le64_to_cpu(u0->a), le64_to_cpu(u0->b),
 407                                 (u64)buffer_info->dma, buffer_info->skb);
 408                         if (i == rx_ring->next_to_use)
 409                                 printk(KERN_CONT " NTU\n");
 410                         else if (i == rx_ring->next_to_clean)
 411                                 printk(KERN_CONT " NTC\n");
 412                         else
 413                                 printk(KERN_CONT "\n");
 414
 415                         if (netif_msg_pktdata(adapter))
 416                                 print_hex_dump(KERN_INFO, "",
 417                                         DUMP_PREFIX_ADDRESS,
 418                                         16, 1, phys_to_virt(buffer_info->dma),
 419                                         adapter->rx_buffer_len, true);
 420                 }
 421         }
 422
 423 exit:
 424         return;
 425 }
 426
 427 /**
 428  * e1000_desc_unused - calculate if we have unused descriptors
 429  **/
 430 static int e1000_desc_unused(struct e1000_ring *ring)
 431 {
 432         if (ring->next_to_clean > ring->next_to_use)
 433                 return ring->next_to_clean - ring->next_to_use - 1;
 434
 435         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
 436 }
 437
 438 /**
 439  * e1000_receive_skb - helper function to handle Rx indications
 440  * @adapter: board private structure
 441  * @status: descriptor status field as written by hardware
 442  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
 443  * @skb: pointer to sk_buff to be indicated to stack
 444  **/
 445 static void e1000_receive_skb(struct e1000_adapter *adapter,
 446                               struct net_device *netdev,
 447                               struct sk_buff *skb,
 448                               u8 status, __le16 vlan)
 449 {
 450         skb->protocol = eth_type_trans(skb, netdev);
 451
 452         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
 453                 vlan_gro_receive(&adapter->napi, adapter->vlgrp,
 454                                  le16_to_cpu(vlan), skb);
 455         else
 456                 napi_gro_receive(&adapter->napi, skb);
 457 }
 458
 459 /**
 460  * e1000_rx_checksum - Receive Checksum Offload for 82543
 461  * @adapter:     board private structure
 462  * @status_err:  receive descriptor status and error fields
 463  * @csum:       receive descriptor csum field
 464  * @sk_buff:     socket buffer with received data
 465  **/
 466 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
 467                               u32 csum, struct sk_buff *skb)
 468 {
 469         u16 status = (u16)status_err;
 470         u8 errors = (u8)(status_err >> 24);
 471         skb->ip_summed = CHECKSUM_NONE;
 472
 473         /* Ignore Checksum bit is set */
 474         if (status & E1000_RXD_STAT_IXSM)
 475                 return;
 476         /* TCP/UDP checksum error bit is set */
 477         if (errors & E1000_RXD_ERR_TCPE) {
 478                 /* let the stack verify checksum errors */
 479                 adapter->hw_csum_err++;
 480                 return;
 481         }
 482
 483         /* TCP/UDP Checksum has not been calculated */
 484         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
 485                 return;
 486
 487         /* It must be a TCP or UDP packet with a valid checksum */
 488         if (status & E1000_RXD_STAT_TCPCS) {
 489                 /* TCP checksum is good */
 490                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 491         } else {
 492                 /*
 493                  * IP fragment with UDP payload
 494                  * Hardware complements the payload checksum, so we undo it
 495                  * and then put the value in host order for further stack use.
 496                  */
 497                 __sum16 sum = (__force __sum16)htons(csum);
 498                 skb->csum = csum_unfold(~sum);
 499                 skb->ip_summed = CHECKSUM_COMPLETE;
 500         }
 501         adapter->hw_csum_good++;
 502 }
 503
 504 /**
 505  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
 506  * @adapter: address of board private structure
 507  **/
 508 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
 509                                    int cleaned_count)
 510 {
 511         struct net_device *netdev = adapter->netdev;
 512         struct pci_dev *pdev = adapter->pdev;
 513         struct e1000_ring *rx_ring = adapter->rx_ring;
 514         struct e1000_rx_desc *rx_desc;
 515         struct e1000_buffer *buffer_info;
 516         struct sk_buff *skb;
 517         unsigned int i;
 518         unsigned int bufsz = adapter->rx_buffer_len;
 519
 520         i = rx_ring->next_to_use;
 521         buffer_info = &rx_ring->buffer_info[i];
 522
 523         while (cleaned_count--) {
 524                 skb = buffer_info->skb;
 525                 if (skb) {
 526                         skb_trim(skb, 0);
 527                         goto map_skb;
 528                 }
 529
 530                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 531                 if (!skb) {
 532                         /* Better luck next round */
 533                         adapter->alloc_rx_buff_failed++;
 534                         break;
 535                 }
 536
 537                 buffer_info->skb = skb;
 538 map_skb:
 539                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 540                                                   adapter->rx_buffer_len,
 541                                                   DMA_FROM_DEVICE);
 542                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 543                         dev_err(&pdev->dev, "RX DMA map failed\n");
 544                         adapter->rx_dma_failed++;
 545                         break;
 546                 }
 547
 548                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 549                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 550
 551                 i++;
 552                 if (i == rx_ring->count)
 553                         i = 0;
 554                 buffer_info = &rx_ring->buffer_info[i];
 555         }
 556
 557         if (rx_ring->next_to_use != i) {
 558                 rx_ring->next_to_use = i;
 559                 if (i-- == 0)
 560                         i = (rx_ring->count - 1);
 561
 562                 /*
 563                  * Force memory writes to complete before letting h/w
 564                  * know there are new descriptors to fetch.  (Only
 565                  * applicable for weak-ordered memory model archs,
 566                  * such as IA-64).
 567                  */
 568                 wmb();
 569                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 570         }
 571 }
 572
 573 /**
 574  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
 575  * @adapter: address of board private structure
 576  **/
 577 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 578                                       int cleaned_count)
 579 {
 580         struct net_device *netdev = adapter->netdev;
 581         struct pci_dev *pdev = adapter->pdev;
 582         union e1000_rx_desc_packet_split *rx_desc;
 583         struct e1000_ring *rx_ring = adapter->rx_ring;
 584         struct e1000_buffer *buffer_info;
 585         struct e1000_ps_page *ps_page;
 586         struct sk_buff *skb;
 587         unsigned int i, j;
 588
 589         i = rx_ring->next_to_use;
 590         buffer_info = &rx_ring->buffer_info[i];
 591
 592         while (cleaned_count--) {
 593                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 594
 595                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 596                         ps_page = &buffer_info->ps_pages[j];
 597                         if (j >= adapter->rx_ps_pages) {
 598                                 /* all unused desc entries get hw null ptr */
 599                                 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
 600                                 continue;
 601                         }
 602                         if (!ps_page->page) {
 603                                 ps_page->page = alloc_page(GFP_ATOMIC);
 604                                 if (!ps_page->page) {
 605                                         adapter->alloc_rx_buff_failed++;
 606                                         goto no_buffers;
 607                                 }
 608                                 ps_page->dma = dma_map_page(&pdev->dev,
 609                                                             ps_page->page,
 610                                                             0, PAGE_SIZE,
 611                                                             DMA_FROM_DEVICE);
 612                                 if (dma_mapping_error(&pdev->dev,
 613                                                       ps_page->dma)) {
 614                                         dev_err(&adapter->pdev->dev,
 615                                           "RX DMA page map failed\n");
 616                                         adapter->rx_dma_failed++;
 617                                         goto no_buffers;
 618                                 }
 619                         }
 620                         /*
 621                          * Refresh the desc even if buffer_addrs
 622                          * didn't change because each write-back
 623                          * erases this info.
 624                          */
 625                         rx_desc->read.buffer_addr[j+1] =
 626                              cpu_to_le64(ps_page->dma);
 627                 }
 628
 629                 skb = netdev_alloc_skb_ip_align(netdev,
 630                                                 adapter->rx_ps_bsize0);
 631
 632                 if (!skb) {
 633                         adapter->alloc_rx_buff_failed++;
 634                         break;
 635                 }
 636
 637                 buffer_info->skb = skb;
 638                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 639                                                   adapter->rx_ps_bsize0,
 640                                                   DMA_FROM_DEVICE);
 641                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 642                         dev_err(&pdev->dev, "RX DMA map failed\n");
 643                         adapter->rx_dma_failed++;
 644                         /* cleanup skb */
 645                         dev_kfree_skb_any(skb);
 646                         buffer_info->skb = NULL;
 647                         break;
 648                 }
 649
 650                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 651
 652                 i++;
 653                 if (i == rx_ring->count)
 654                         i = 0;
 655                 buffer_info = &rx_ring->buffer_info[i];
 656         }
 657
 658 no_buffers:
 659         if (rx_ring->next_to_use != i) {
 660                 rx_ring->next_to_use = i;
 661
 662                 if (!(i--))
 663                         i = (rx_ring->count - 1);
 664
 665                 /*
 666                  * Force memory writes to complete before letting h/w
 667                  * know there are new descriptors to fetch.  (Only
 668                  * applicable for weak-ordered memory model archs,
 669                  * such as IA-64).
 670                  */
 671                 wmb();
 672                 /*
 673                  * Hardware increments by 16 bytes, but packet split
 674                  * descriptors are 32 bytes...so we increment tail
 675                  * twice as much.
 676                  */
 677                 writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
 678         }
 679 }
 680
 681 /**
 682  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
 683  * @adapter: address of board private structure
 684  * @cleaned_count: number of buffers to allocate this pass
 685  **/
 686
 687 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
 688                                          int cleaned_count)
 689 {
 690         struct net_device *netdev = adapter->netdev;
 691         struct pci_dev *pdev = adapter->pdev;
 692         struct e1000_rx_desc *rx_desc;
 693         struct e1000_ring *rx_ring = adapter->rx_ring;
 694         struct e1000_buffer *buffer_info;
 695         struct sk_buff *skb;
 696         unsigned int i;
 697         unsigned int bufsz = 256 - 16 /* for skb_reserve */;
 698
 699         i = rx_ring->next_to_use;
 700         buffer_info = &rx_ring->buffer_info[i];
 701
 702         while (cleaned_count--) {
 703                 skb = buffer_info->skb;
 704                 if (skb) {
 705                         skb_trim(skb, 0);
 706                         goto check_page;
 707                 }
 708
 709                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 710                 if (unlikely(!skb)) {
 711                         /* Better luck next round */
 712                         adapter->alloc_rx_buff_failed++;
 713                         break;
 714                 }
 715
 716                 buffer_info->skb = skb;
 717 check_page:
 718                 /* allocate a new page if necessary */
 719                 if (!buffer_info->page) {
 720                         buffer_info->page = alloc_page(GFP_ATOMIC);
 721                         if (unlikely(!buffer_info->page)) {
 722                                 adapter->alloc_rx_buff_failed++;
 723                                 break;
 724                         }
 725                 }
 726
 727                 if (!buffer_info->dma)
 728                         buffer_info->dma = dma_map_page(&pdev->dev,
 729                                                         buffer_info->page, 0,
 730                                                         PAGE_SIZE,
 731                                                         DMA_FROM_DEVICE);
 732
 733                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 734                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 735
 736                 if (unlikely(++i == rx_ring->count))
 737                         i = 0;
 738                 buffer_info = &rx_ring->buffer_info[i];
 739         }
 740
 741         if (likely(rx_ring->next_to_use != i)) {
 742                 rx_ring->next_to_use = i;
 743                 if (unlikely(i-- == 0))
 744                         i = (rx_ring->count - 1);
 745
 746                 /* Force memory writes to complete before letting h/w
 747                  * know there are new descriptors to fetch.  (Only
 748                  * applicable for weak-ordered memory model archs,
 749                  * such as IA-64). */
 750                 wmb();
 751                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 752         }
 753 }
 754
 755 /**
 756  * e1000_clean_rx_irq - Send received data up the network stack; legacy
 757  * @adapter: board private structure
 758  *
 759  * the return value indicates whether actual cleaning was done, there
 760  * is no guarantee that everything was cleaned
 761  **/
 762 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
 763                                int *work_done, int work_to_do)
 764 {
 765         struct net_device *netdev = adapter->netdev;
 766         struct pci_dev *pdev = adapter->pdev;
 767         struct e1000_hw *hw = &adapter->hw;
 768         struct e1000_ring *rx_ring = adapter->rx_ring;
 769         struct e1000_rx_desc *rx_desc, *next_rxd;
 770         struct e1000_buffer *buffer_info, *next_buffer;
 771         u32 length;
 772         unsigned int i;
 773         int cleaned_count = 0;
 774         bool cleaned = 0;
 775         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 776
 777         i = rx_ring->next_to_clean;
 778         rx_desc = E1000_RX_DESC(*rx_ring, i);
 779         buffer_info = &rx_ring->buffer_info[i];
 780
 781         while (rx_desc->status & E1000_RXD_STAT_DD) {
 782                 struct sk_buff *skb;
 783                 u8 status;
 784
 785                 if (*work_done >= work_to_do)
 786                         break;
 787                 (*work_done)++;
 788
 789                 status = rx_desc->status;
 790                 skb = buffer_info->skb;
 791                 buffer_info->skb = NULL;
 792
 793                 prefetch(skb->data - NET_IP_ALIGN);
 794
 795                 i++;
 796                 if (i == rx_ring->count)
 797                         i = 0;
 798                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 799                 prefetch(next_rxd);
 800
 801                 next_buffer = &rx_ring->buffer_info[i];
 802
 803                 cleaned = 1;
 804                 cleaned_count++;
 805                 dma_unmap_single(&pdev->dev,
 806                                  buffer_info->dma,
 807                                  adapter->rx_buffer_len,
 808                                  DMA_FROM_DEVICE);
 809                 buffer_info->dma = 0;
 810
 811                 length = le16_to_cpu(rx_desc->length);
 812
 813                 /*
 814                  * !EOP means multiple descriptors were used to store a single
 815                  * packet, if that's the case we need to toss it.  In fact, we
 816                  * need to toss every packet with the EOP bit clear and the
 817                  * next frame that _does_ have the EOP bit set, as it is by
 818                  * definition only a frame fragment
 819                  */
 820                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
 821                         adapter->flags2 |= FLAG2_IS_DISCARDING;
 822
 823                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
 824                         /* All receives must fit into a single buffer */
 825                         e_dbg("Receive packet consumed multiple buffers\n");
 826                         /* recycle */
 827                         buffer_info->skb = skb;
 828                         if (status & E1000_RXD_STAT_EOP)
 829                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
 830                         goto next_desc;
 831                 }
 832
 833                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
 834                         /* recycle */
 835                         buffer_info->skb = skb;
 836                         goto next_desc;
 837                 }
 838
 839                 /* adjust length to remove Ethernet CRC */
 840                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 841                         length -= 4;
 842
 843                 total_rx_bytes += length;
 844                 total_rx_packets++;
 845
 846                 /*
 847                  * code added for copybreak, this should improve
 848                  * performance for small packets with large amounts
 849                  * of reassembly being done in the stack
 850                  */
 851                 if (length < copybreak) {
 852                         struct sk_buff *new_skb =
 853                             netdev_alloc_skb_ip_align(netdev, length);
 854                         if (new_skb) {
 855                                 skb_copy_to_linear_data_offset(new_skb,
 856                                                                -NET_IP_ALIGN,
 857                                                                (skb->data -
 858                                                                 NET_IP_ALIGN),
 859                                                                (length +
 860                                                                 NET_IP_ALIGN));
 861                                 /* save the skb in buffer_info as good */
 862                                 buffer_info->skb = skb;
 863                                 skb = new_skb;
 864                         }
 865                         /* else just continue with the old one */
 866                 }
 867                 /* end copybreak code */
 868                 skb_put(skb, length);
 869
 870                 /* Receive Checksum Offload */
 871                 e1000_rx_checksum(adapter,
 872                                   (u32)(status) |
 873                                   ((u32)(rx_desc->errors) << 24),
 874                                   le16_to_cpu(rx_desc->csum), skb);
 875
 876                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
 877
 878 next_desc:
 879                 rx_desc->status = 0;
 880
 881                 /* return some buffers to hardware, one at a time is too slow */
 882                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 883                         adapter->alloc_rx_buf(adapter, cleaned_count);
 884                         cleaned_count = 0;
 885                 }
 886
 887                 /* use prefetched values */
 888                 rx_desc = next_rxd;
 889                 buffer_info = next_buffer;
 890         }
 891         rx_ring->next_to_clean = i;
 892
 893         cleaned_count = e1000_desc_unused(rx_ring);
 894         if (cleaned_count)
 895                 adapter->alloc_rx_buf(adapter, cleaned_count);
 896
 897         adapter->total_rx_bytes += total_rx_bytes;
 898         adapter->total_rx_packets += total_rx_packets;
 899         netdev->stats.rx_bytes += total_rx_bytes;
 900         netdev->stats.rx_packets += total_rx_packets;
 901         return cleaned;
 902 }
 903
 904 static void e1000_put_txbuf(struct e1000_adapter *adapter,
 905                              struct e1000_buffer *buffer_info)
 906 {
 907         if (buffer_info->dma) {
 908                 if (buffer_info->mapped_as_page)
 909                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
 910                                        buffer_info->length, DMA_TO_DEVICE);
 911                 else
 912                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
 913                                          buffer_info->length, DMA_TO_DEVICE);
 914                 buffer_info->dma = 0;
 915         }
 916         if (buffer_info->skb) {
 917                 dev_kfree_skb_any(buffer_info->skb);
 918                 buffer_info->skb = NULL;
 919         }
 920         buffer_info->time_stamp = 0;
 921 }
 922
 923 static void e1000_print_hw_hang(struct work_struct *work)
 924 {
 925         struct e1000_adapter *adapter = container_of(work,
 926                                                      struct e1000_adapter,
 927                                                      print_hang_task);
 928         struct e1000_ring *tx_ring = adapter->tx_ring;
 929         unsigned int i = tx_ring->next_to_clean;
 930         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 931         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
 932         struct e1000_hw *hw = &adapter->hw;
 933         u16 phy_status, phy_1000t_status, phy_ext_status;
 934         u16 pci_status;
 935
 936         e1e_rphy(hw, PHY_STATUS, &phy_status);
 937         e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
 938         e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
 939
 940         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
 941
 942         /* detected Hardware unit hang */
 943         e_err("Detected Hardware Unit Hang:\n"
 944               "  TDH                  <%x>\n"
 945               "  TDT                  <%x>\n"
 946               "  next_to_use          <%x>\n"
 947               "  next_to_clean        <%x>\n"
 948               "buffer_info[next_to_clean]:\n"
 949               "  time_stamp           <%lx>\n"
 950               "  next_to_watch        <%x>\n"
 951               "  jiffies              <%lx>\n"
 952               "  next_to_watch.status <%x>\n"
 953               "MAC Status             <%x>\n"
 954               "PHY Status             <%x>\n"
 955               "PHY 1000BASE-T Status  <%x>\n"
 956               "PHY Extended Status    <%x>\n"
 957               "PCI Status             <%x>\n",
 958               readl(adapter->hw.hw_addr + tx_ring->head),
 959               readl(adapter->hw.hw_addr + tx_ring->tail),
 960               tx_ring->next_to_use,
 961               tx_ring->next_to_clean,
 962               tx_ring->buffer_info[eop].time_stamp,
 963               eop,
 964               jiffies,
 965               eop_desc->upper.fields.status,
 966               er32(STATUS),
 967               phy_status,
 968               phy_1000t_status,
 969               phy_ext_status,
 970               pci_status);
 971 }
 972
 973 /**
 974  * e1000_clean_tx_irq - Reclaim resources after transmit completes
 975  * @adapter: board private structure
 976  *
 977  * the return value indicates whether actual cleaning was done, there
 978  * is no guarantee that everything was cleaned
 979  **/
 980 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
 981 {
 982         struct net_device *netdev = adapter->netdev;
 983         struct e1000_hw *hw = &adapter->hw;
 984         struct e1000_ring *tx_ring = adapter->tx_ring;
 985         struct e1000_tx_desc *tx_desc, *eop_desc;
 986         struct e1000_buffer *buffer_info;
 987         unsigned int i, eop;
 988         unsigned int count = 0;
 989         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
 990
 991         i = tx_ring->next_to_clean;
 992         eop = tx_ring->buffer_info[i].next_to_watch;
 993         eop_desc = E1000_TX_DESC(*tx_ring, eop);
 994
 995         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
 996                (count < tx_ring->count)) {
 997                 bool cleaned = false;
 998                 for (; !cleaned; count++) {
 999                         tx_desc = E1000_TX_DESC(*tx_ring, i);
1000                         buffer_info = &tx_ring->buffer_info[i];
1001                         cleaned = (i == eop);
1002
1003                         if (cleaned) {
1004                                 struct sk_buff *skb = buffer_info->skb;
1005                                 unsigned int segs, bytecount;
1006                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
1007                                 /* multiply data chunks by size of headers */
1008                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
1009                                             skb->len;
1010                                 total_tx_packets += segs;
1011                                 total_tx_bytes += bytecount;
1012                         }
1013
1014                         e1000_put_txbuf(adapter, buffer_info);
1015                         tx_desc->upper.data = 0;
1016
1017                         i++;
1018                         if (i == tx_ring->count)
1019                                 i = 0;
1020                 }
1021
1022                 if (i == tx_ring->next_to_use)
1023                         break;
1024                 eop = tx_ring->buffer_info[i].next_to_watch;
1025                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1026         }
1027
1028         tx_ring->next_to_clean = i;
1029
1030 #define TX_WAKE_THRESHOLD 32
1031         if (count && netif_carrier_ok(netdev) &&
1032             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1033                 /* Make sure that anybody stopping the queue after this
1034                  * sees the new next_to_clean.
1035                  */
1036                 smp_mb();
1037
1038                 if (netif_queue_stopped(netdev) &&
1039                     !(test_bit(__E1000_DOWN, &adapter->state))) {
1040                         netif_wake_queue(netdev);
1041                         ++adapter->restart_queue;
1042                 }
1043         }
1044
1045         if (adapter->detect_tx_hung) {
1046                 /*
1047                  * Detect a transmit hang in hardware, this serializes the
1048                  * check with the clearing of time_stamp and movement of i
1049                  */
1050                 adapter->detect_tx_hung = 0;
1051                 if (tx_ring->buffer_info[i].time_stamp &&
1052                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1053                                + (adapter->tx_timeout_factor * HZ)) &&
1054                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
1055                         schedule_work(&adapter->print_hang_task);
1056                         netif_stop_queue(netdev);
1057                 }
1058         }
1059         adapter->total_tx_bytes += total_tx_bytes;
1060         adapter->total_tx_packets += total_tx_packets;
1061         netdev->stats.tx_bytes += total_tx_bytes;
1062         netdev->stats.tx_packets += total_tx_packets;
1063         return (count < tx_ring->count);
1064 }
1065
1066 /**
1067  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1068  * @adapter: board private structure
1069  *
1070  * the return value indicates whether actual cleaning was done, there
1071  * is no guarantee that everything was cleaned
1072  **/
1073 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
1074                                   int *work_done, int work_to_do)
1075 {
1076         struct e1000_hw *hw = &adapter->hw;
1077         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1078         struct net_device *netdev = adapter->netdev;
1079         struct pci_dev *pdev = adapter->pdev;
1080         struct e1000_ring *rx_ring = adapter->rx_ring;
1081         struct e1000_buffer *buffer_info, *next_buffer;
1082         struct e1000_ps_page *ps_page;
1083         struct sk_buff *skb;
1084         unsigned int i, j;
1085         u32 length, staterr;
1086         int cleaned_count = 0;
1087         bool cleaned = 0;
1088         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1089
1090         i = rx_ring->next_to_clean;
1091         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1092         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1093         buffer_info = &rx_ring->buffer_info[i];
1094
1095         while (staterr & E1000_RXD_STAT_DD) {
1096                 if (*work_done >= work_to_do)
1097                         break;
1098                 (*work_done)++;
1099                 skb = buffer_info->skb;
1100
1101                 /* in the packet split case this is header only */
1102                 prefetch(skb->data - NET_IP_ALIGN);
1103
1104                 i++;
1105                 if (i == rx_ring->count)
1106                         i = 0;
1107                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1108                 prefetch(next_rxd);
1109
1110                 next_buffer = &rx_ring->buffer_info[i];
1111
1112                 cleaned = 1;
1113                 cleaned_count++;
1114                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1115                                  adapter->rx_ps_bsize0,
1116                                  DMA_FROM_DEVICE);
1117                 buffer_info->dma = 0;
1118
1119                 /* see !EOP comment in other rx routine */
1120                 if (!(staterr & E1000_RXD_STAT_EOP))
1121                         adapter->flags2 |= FLAG2_IS_DISCARDING;
1122
1123                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1124                         e_dbg("Packet Split buffers didn't pick up the full "
1125                               "packet\n");
1126                         dev_kfree_skb_irq(skb);
1127                         if (staterr & E1000_RXD_STAT_EOP)
1128                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1129                         goto next_desc;
1130                 }
1131
1132                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
1133                         dev_kfree_skb_irq(skb);
1134                         goto next_desc;
1135                 }
1136
1137                 length = le16_to_cpu(rx_desc->wb.middle.length0);
1138
1139                 if (!length) {
1140                         e_dbg("Last part of the packet spanning multiple "
1141                               "descriptors\n");
1142                         dev_kfree_skb_irq(skb);
1143                         goto next_desc;
1144                 }
1145
1146                 /* Good Receive */
1147                 skb_put(skb, length);
1148
1149                 {
1150                 /*
1151                  * this looks ugly, but it seems compiler issues make it
1152                  * more efficient than reusing j
1153                  */
1154                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1155
1156                 /*
1157                  * page alloc/put takes too long and effects small packet
1158                  * throughput, so unsplit small packets and save the alloc/put
1159                  * only valid in softirq (napi) context to call kmap_*
1160                  */
1161                 if (l1 && (l1 <= copybreak) &&
1162                     ((length + l1) <= adapter->rx_ps_bsize0)) {
1163                         u8 *vaddr;
1164
1165                         ps_page = &buffer_info->ps_pages[0];
1166
1167                         /*
1168                          * there is no documentation about how to call
1169                          * kmap_atomic, so we can't hold the mapping
1170                          * very long
1171                          */
1172                         dma_sync_single_for_cpu(&pdev->dev, ps_page->dma,
1173                                                 PAGE_SIZE, DMA_FROM_DEVICE);
1174                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
1175                         memcpy(skb_tail_pointer(skb), vaddr, l1);
1176                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
1177                         dma_sync_single_for_device(&pdev->dev, ps_page->dma,
1178                                                    PAGE_SIZE, DMA_FROM_DEVICE);
1179
1180                         /* remove the CRC */
1181                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1182                                 l1 -= 4;
1183
1184                         skb_put(skb, l1);
1185                         goto copydone;
1186                 } /* if */
1187                 }
1188
1189                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1190                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1191                         if (!length)
1192                                 break;
1193
1194                         ps_page = &buffer_info->ps_pages[j];
1195                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1196                                        DMA_FROM_DEVICE);
1197                         ps_page->dma = 0;
1198                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1199                         ps_page->page = NULL;
1200                         skb->len += length;
1201                         skb->data_len += length;
1202                         skb->truesize += length;
1203                 }
1204
1205                 /* strip the ethernet crc, problem is we're using pages now so
1206                  * this whole operation can get a little cpu intensive
1207                  */
1208                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1209                         pskb_trim(skb, skb->len - 4);
1210
1211 copydone:
1212                 total_rx_bytes += skb->len;
1213                 total_rx_packets++;
1214
1215                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
1216                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
1217
1218                 if (rx_desc->wb.upper.header_status &
1219                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1220                         adapter->rx_hdr_split++;
1221
1222                 e1000_receive_skb(adapter, netdev, skb,
1223                                   staterr, rx_desc->wb.middle.vlan);
1224
1225 next_desc:
1226                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1227                 buffer_info->skb = NULL;
1228
1229                 /* return some buffers to hardware, one at a time is too slow */
1230                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1231                         adapter->alloc_rx_buf(adapter, cleaned_count);
1232                         cleaned_count = 0;
1233                 }
1234
1235                 /* use prefetched values */
1236                 rx_desc = next_rxd;
1237                 buffer_info = next_buffer;
1238
1239                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1240         }
1241         rx_ring->next_to_clean = i;
1242
1243         cleaned_count = e1000_desc_unused(rx_ring);
1244         if (cleaned_count)
1245                 adapter->alloc_rx_buf(adapter, cleaned_count);
1246
1247         adapter->total_rx_bytes += total_rx_bytes;
1248         adapter->total_rx_packets += total_rx_packets;
1249         netdev->stats.rx_bytes += total_rx_bytes;
1250         netdev->stats.rx_packets += total_rx_packets;
1251         return cleaned;
1252 }
1253
1254 /**
1255  * e1000_consume_page - helper function
1256  **/
1257 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1258                                u16 length)
1259 {
1260         bi->page = NULL;
1261         skb->len += length;
1262         skb->data_len += length;
1263         skb->truesize += length;
1264 }
1265
1266 /**
1267  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1268  * @adapter: board private structure
1269  *
1270  * the return value indicates whether actual cleaning was done, there
1271  * is no guarantee that everything was cleaned
1272  **/
1273
1274 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
1275                                      int *work_done, int work_to_do)
1276 {
1277         struct net_device *netdev = adapter->netdev;
1278         struct pci_dev *pdev = adapter->pdev;
1279         struct e1000_ring *rx_ring = adapter->rx_ring;
1280         struct e1000_rx_desc *rx_desc, *next_rxd;
1281         struct e1000_buffer *buffer_info, *next_buffer;
1282         u32 length;
1283         unsigned int i;
1284         int cleaned_count = 0;
1285         bool cleaned = false;
1286         unsigned int total_rx_bytes=0, total_rx_packets=0;
1287
1288         i = rx_ring->next_to_clean;
1289         rx_desc = E1000_RX_DESC(*rx_ring, i);
1290         buffer_info = &rx_ring->buffer_info[i];
1291
1292         while (rx_desc->status & E1000_RXD_STAT_DD) {
1293                 struct sk_buff *skb;
1294                 u8 status;
1295
1296                 if (*work_done >= work_to_do)
1297                         break;
1298                 (*work_done)++;
1299
1300                 status = rx_desc->status;
1301                 skb = buffer_info->skb;
1302                 buffer_info->skb = NULL;
1303
1304                 ++i;
1305                 if (i == rx_ring->count)
1306                         i = 0;
1307                 next_rxd = E1000_RX_DESC(*rx_ring, i);
1308                 prefetch(next_rxd);
1309
1310                 next_buffer = &rx_ring->buffer_info[i];
1311
1312                 cleaned = true;
1313                 cleaned_count++;
1314                 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1315                                DMA_FROM_DEVICE);
1316                 buffer_info->dma = 0;
1317
1318                 length = le16_to_cpu(rx_desc->length);
1319
1320                 /* errors is only valid for DD + EOP descriptors */
1321                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
1322                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
1323                                 /* recycle both page and skb */
1324                                 buffer_info->skb = skb;
1325                                 /* an error means any chain goes out the window
1326                                  * too */
1327                                 if (rx_ring->rx_skb_top)
1328                                         dev_kfree_skb(rx_ring->rx_skb_top);
1329                                 rx_ring->rx_skb_top = NULL;
1330                                 goto next_desc;
1331                 }
1332
1333 #define rxtop rx_ring->rx_skb_top
1334                 if (!(status & E1000_RXD_STAT_EOP)) {
1335                         /* this descriptor is only the beginning (or middle) */
1336                         if (!rxtop) {
1337                                 /* this is the beginning of a chain */
1338                                 rxtop = skb;
1339                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1340                                                    0, length);
1341                         } else {
1342                                 /* this is the middle of a chain */
1343                                 skb_fill_page_desc(rxtop,
1344                                     skb_shinfo(rxtop)->nr_frags,
1345                                     buffer_info->page, 0, length);
1346                                 /* re-use the skb, only consumed the page */
1347                                 buffer_info->skb = skb;
1348                         }
1349                         e1000_consume_page(buffer_info, rxtop, length);
1350                         goto next_desc;
1351                 } else {
1352                         if (rxtop) {
1353                                 /* end of the chain */
1354                                 skb_fill_page_desc(rxtop,
1355                                     skb_shinfo(rxtop)->nr_frags,
1356                                     buffer_info->page, 0, length);
1357                                 /* re-use the current skb, we only consumed the
1358                                  * page */
1359                                 buffer_info->skb = skb;
1360                                 skb = rxtop;
1361                                 rxtop = NULL;
1362                                 e1000_consume_page(buffer_info, skb, length);
1363                         } else {
1364                                 /* no chain, got EOP, this buf is the packet
1365                                  * copybreak to save the put_page/alloc_page */
1366                                 if (length <= copybreak &&
1367                                     skb_tailroom(skb) >= length) {
1368                                         u8 *vaddr;
1369                                         vaddr = kmap_atomic(buffer_info->page,
1370                                                            KM_SKB_DATA_SOFTIRQ);
1371                                         memcpy(skb_tail_pointer(skb), vaddr,
1372                                                length);
1373                                         kunmap_atomic(vaddr,
1374                                                       KM_SKB_DATA_SOFTIRQ);
1375                                         /* re-use the page, so don't erase
1376                                          * buffer_info->page */
1377                                         skb_put(skb, length);
1378                                 } else {
1379                                         skb_fill_page_desc(skb, 0,
1380                                                            buffer_info->page, 0,
1381                                                            length);
1382                                         e1000_consume_page(buffer_info, skb,
1383                                                            length);
1384                                 }
1385                         }
1386                 }
1387
1388                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1389                 e1000_rx_checksum(adapter,
1390                                   (u32)(status) |
1391                                   ((u32)(rx_desc->errors) << 24),
1392                                   le16_to_cpu(rx_desc->csum), skb);
1393
1394                 /* probably a little skewed due to removing CRC */
1395                 total_rx_bytes += skb->len;
1396                 total_rx_packets++;
1397
1398                 /* eth type trans needs skb->data to point to something */
1399                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1400                         e_err("pskb_may_pull failed.\n");
1401                         dev_kfree_skb(skb);
1402                         goto next_desc;
1403                 }
1404
1405                 e1000_receive_skb(adapter, netdev, skb, status,
1406                                   rx_desc->special);
1407
1408 next_desc:
1409                 rx_desc->status = 0;
1410
1411                 /* return some buffers to hardware, one at a time is too slow */
1412                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1413                         adapter->alloc_rx_buf(adapter, cleaned_count);
1414                         cleaned_count = 0;
1415                 }
1416
1417                 /* use prefetched values */
1418                 rx_desc = next_rxd;
1419                 buffer_info = next_buffer;
1420         }
1421         rx_ring->next_to_clean = i;
1422
1423         cleaned_count = e1000_desc_unused(rx_ring);
1424         if (cleaned_count)
1425                 adapter->alloc_rx_buf(adapter, cleaned_count);
1426
1427         adapter->total_rx_bytes += total_rx_bytes;
1428         adapter->total_rx_packets += total_rx_packets;
1429         netdev->stats.rx_bytes += total_rx_bytes;
1430         netdev->stats.rx_packets += total_rx_packets;
1431         return cleaned;
1432 }
1433
1434 /**
1435  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1436  * @adapter: board private structure
1437  **/
1438 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1439 {
1440         struct e1000_ring *rx_ring = adapter->rx_ring;
1441         struct e1000_buffer *buffer_info;
1442         struct e1000_ps_page *ps_page;
1443         struct pci_dev *pdev = adapter->pdev;
1444         unsigned int i, j;
1445
1446         /* Free all the Rx ring sk_buffs */
1447         for (i = 0; i < rx_ring->count; i++) {
1448                 buffer_info = &rx_ring->buffer_info[i];
1449                 if (buffer_info->dma) {
1450                         if (adapter->clean_rx == e1000_clean_rx_irq)
1451                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1452                                                  adapter->rx_buffer_len,
1453                                                  DMA_FROM_DEVICE);
1454                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1455                                 dma_unmap_page(&pdev->dev, buffer_info->dma,
1456                                                PAGE_SIZE,
1457                                                DMA_FROM_DEVICE);
1458                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1459                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1460                                                  adapter->rx_ps_bsize0,
1461                                                  DMA_FROM_DEVICE);
1462                         buffer_info->dma = 0;
1463                 }
1464
1465                 if (buffer_info->page) {
1466                         put_page(buffer_info->page);
1467                         buffer_info->page = NULL;
1468                 }
1469
1470                 if (buffer_info->skb) {
1471                         dev_kfree_skb(buffer_info->skb);
1472                         buffer_info->skb = NULL;
1473                 }
1474
1475                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1476                         ps_page = &buffer_info->ps_pages[j];
1477                         if (!ps_page->page)
1478                                 break;
1479                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1480                                        DMA_FROM_DEVICE);
1481                         ps_page->dma = 0;
1482                         put_page(ps_page->page);
1483                         ps_page->page = NULL;
1484                 }
1485         }
1486
1487         /* there also may be some cached data from a chained receive */
1488         if (rx_ring->rx_skb_top) {
1489                 dev_kfree_skb(rx_ring->rx_skb_top);
1490                 rx_ring->rx_skb_top = NULL;
1491         }
1492
1493         /* Zero out the descriptor ring */
1494         memset(rx_ring->desc, 0, rx_ring->size);
1495
1496         rx_ring->next_to_clean = 0;
1497         rx_ring->next_to_use = 0;
1498         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1499
1500         writel(0, adapter->hw.hw_addr + rx_ring->head);
1501         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1502 }
1503
1504 static void e1000e_downshift_workaround(struct work_struct *work)
1505 {
1506         struct e1000_adapter *adapter = container_of(work,
1507                                         struct e1000_adapter, downshift_task);
1508
1509         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1510 }
1511
1512 /**
1513  * e1000_intr_msi - Interrupt Handler
1514  * @irq: interrupt number
1515  * @data: pointer to a network interface device structure
1516  **/
1517 static irqreturn_t e1000_intr_msi(int irq, void *data)
1518 {
1519         struct net_device *netdev = data;
1520         struct e1000_adapter *adapter = netdev_priv(netdev);
1521         struct e1000_hw *hw = &adapter->hw;
1522         u32 icr = er32(ICR);
1523
1524         /*
1525          * read ICR disables interrupts using IAM
1526          */
1527
1528         if (icr & E1000_ICR_LSC) {
1529                 hw->mac.get_link_status = 1;
1530                 /*
1531                  * ICH8 workaround-- Call gig speed drop workaround on cable
1532                  * disconnect (LSC) before accessing any PHY registers
1533                  */
1534                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1535                     (!(er32(STATUS) & E1000_STATUS_LU)))
1536                         schedule_work(&adapter->downshift_task);
1537
1538                 /*
1539                  * 80003ES2LAN workaround-- For packet buffer work-around on
1540                  * link down event; disable receives here in the ISR and reset
1541                  * adapter in watchdog
1542                  */
1543                 if (netif_carrier_ok(netdev) &&
1544                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1545                         /* disable receives */
1546                         u32 rctl = er32(RCTL);
1547                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1548                         adapter->flags |= FLAG_RX_RESTART_NOW;
1549                 }
1550                 /* guard against interrupt when we're going down */
1551                 if (!test_bit(__E1000_DOWN, &adapter->state))
1552                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1553         }
1554
1555         if (napi_schedule_prep(&adapter->napi)) {
1556                 adapter->total_tx_bytes = 0;
1557                 adapter->total_tx_packets = 0;
1558                 adapter->total_rx_bytes = 0;
1559                 adapter->total_rx_packets = 0;
1560                 __napi_schedule(&adapter->napi);
1561         }
1562
1563         return IRQ_HANDLED;
1564 }
1565
1566 /**
1567  * e1000_intr - Interrupt Handler
1568  * @irq: interrupt number
1569  * @data: pointer to a network interface device structure
1570  **/
1571 static irqreturn_t e1000_intr(int irq, void *data)
1572 {
1573         struct net_device *netdev = data;
1574         struct e1000_adapter *adapter = netdev_priv(netdev);
1575         struct e1000_hw *hw = &adapter->hw;
1576         u32 rctl, icr = er32(ICR);
1577
1578         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1579                 return IRQ_NONE;  /* Not our interrupt */
1580
1581         /*
1582          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1583          * not set, then the adapter didn't send an interrupt
1584          */
1585         if (!(icr & E1000_ICR_INT_ASSERTED))
1586                 return IRQ_NONE;
1587
1588         /*
1589          * Interrupt Auto-Mask...upon reading ICR,
1590          * interrupts are masked.  No need for the
1591          * IMC write
1592          */
1593
1594         if (icr & E1000_ICR_LSC) {
1595                 hw->mac.get_link_status = 1;
1596                 /*
1597                  * ICH8 workaround-- Call gig speed drop workaround on cable
1598                  * disconnect (LSC) before accessing any PHY registers
1599                  */
1600                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1601                     (!(er32(STATUS) & E1000_STATUS_LU)))
1602                         schedule_work(&adapter->downshift_task);
1603
1604                 /*
1605                  * 80003ES2LAN workaround--
1606                  * For packet buffer work-around on link down event;
1607                  * disable receives here in the ISR and
1608                  * reset adapter in watchdog
1609                  */
1610                 if (netif_carrier_ok(netdev) &&
1611                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1612                         /* disable receives */
1613                         rctl = er32(RCTL);
1614                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1615                         adapter->flags |= FLAG_RX_RESTART_NOW;
1616                 }
1617                 /* guard against interrupt when we're going down */
1618                 if (!test_bit(__E1000_DOWN, &adapter->state))
1619                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1620         }
1621
1622         if (napi_schedule_prep(&adapter->napi)) {
1623                 adapter->total_tx_bytes = 0;
1624                 adapter->total_tx_packets = 0;
1625                 adapter->total_rx_bytes = 0;
1626                 adapter->total_rx_packets = 0;
1627                 __napi_schedule(&adapter->napi);
1628         }
1629
1630         return IRQ_HANDLED;
1631 }
1632
1633 static irqreturn_t e1000_msix_other(int irq, void *data)
1634 {
1635         struct net_device *netdev = data;
1636         struct e1000_adapter *adapter = netdev_priv(netdev);
1637         struct e1000_hw *hw = &adapter->hw;
1638         u32 icr = er32(ICR);
1639
1640         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1641                 if (!test_bit(__E1000_DOWN, &adapter->state))
1642                         ew32(IMS, E1000_IMS_OTHER);
1643                 return IRQ_NONE;
1644         }
1645
1646         if (icr & adapter->eiac_mask)
1647                 ew32(ICS, (icr & adapter->eiac_mask));
1648
1649         if (icr & E1000_ICR_OTHER) {
1650                 if (!(icr & E1000_ICR_LSC))
1651                         goto no_link_interrupt;
1652                 hw->mac.get_link_status = 1;
1653                 /* guard against interrupt when we're going down */
1654                 if (!test_bit(__E1000_DOWN, &adapter->state))
1655                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1656         }
1657
1658 no_link_interrupt:
1659         if (!test_bit(__E1000_DOWN, &adapter->state))
1660                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1661
1662         return IRQ_HANDLED;
1663 }
1664
1665
1666 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1667 {
1668         struct net_device *netdev = data;
1669         struct e1000_adapter *adapter = netdev_priv(netdev);
1670         struct e1000_hw *hw = &adapter->hw;
1671         struct e1000_ring *tx_ring = adapter->tx_ring;
1672
1673
1674         adapter->total_tx_bytes = 0;
1675         adapter->total_tx_packets = 0;
1676
1677         if (!e1000_clean_tx_irq(adapter))
1678                 /* Ring was not completely cleaned, so fire another interrupt */
1679                 ew32(ICS, tx_ring->ims_val);
1680
1681         return IRQ_HANDLED;
1682 }
1683
1684 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1685 {
1686         struct net_device *netdev = data;
1687         struct e1000_adapter *adapter = netdev_priv(netdev);
1688
1689         /* Write the ITR value calculated at the end of the
1690          * previous interrupt.
1691          */
1692         if (adapter->rx_ring->set_itr) {
1693                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1694                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1695                 adapter->rx_ring->set_itr = 0;
1696         }
1697
1698         if (napi_schedule_prep(&adapter->napi)) {
1699                 adapter->total_rx_bytes = 0;
1700                 adapter->total_rx_packets = 0;
1701                 __napi_schedule(&adapter->napi);
1702         }
1703         return IRQ_HANDLED;
1704 }
1705
1706 /**
1707  * e1000_configure_msix - Configure MSI-X hardware
1708  *
1709  * e1000_configure_msix sets up the hardware to properly
1710  * generate MSI-X interrupts.
1711  **/
1712 static void e1000_configure_msix(struct e1000_adapter *adapter)
1713 {
1714         struct e1000_hw *hw = &adapter->hw;
1715         struct e1000_ring *rx_ring = adapter->rx_ring;
1716         struct e1000_ring *tx_ring = adapter->tx_ring;
1717         int vector = 0;
1718         u32 ctrl_ext, ivar = 0;
1719
1720         adapter->eiac_mask = 0;
1721
1722         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1723         if (hw->mac.type == e1000_82574) {
1724                 u32 rfctl = er32(RFCTL);
1725                 rfctl |= E1000_RFCTL_ACK_DIS;
1726                 ew32(RFCTL, rfctl);
1727         }
1728
1729 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1730         /* Configure Rx vector */
1731         rx_ring->ims_val = E1000_IMS_RXQ0;
1732         adapter->eiac_mask |= rx_ring->ims_val;
1733         if (rx_ring->itr_val)
1734                 writel(1000000000 / (rx_ring->itr_val * 256),
1735                        hw->hw_addr + rx_ring->itr_register);
1736         else
1737                 writel(1, hw->hw_addr + rx_ring->itr_register);
1738         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1739
1740         /* Configure Tx vector */
1741         tx_ring->ims_val = E1000_IMS_TXQ0;
1742         vector++;
1743         if (tx_ring->itr_val)
1744                 writel(1000000000 / (tx_ring->itr_val * 256),
1745                        hw->hw_addr + tx_ring->itr_register);
1746         else
1747                 writel(1, hw->hw_addr + tx_ring->itr_register);
1748         adapter->eiac_mask |= tx_ring->ims_val;
1749         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1750
1751         /* set vector for Other Causes, e.g. link changes */
1752         vector++;
1753         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1754         if (rx_ring->itr_val)
1755                 writel(1000000000 / (rx_ring->itr_val * 256),
1756                        hw->hw_addr + E1000_EITR_82574(vector));
1757         else
1758                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1759
1760         /* Cause Tx interrupts on every write back */
1761         ivar |= (1 << 31);
1762
1763         ew32(IVAR, ivar);
1764
1765         /* enable MSI-X PBA support */
1766         ctrl_ext = er32(CTRL_EXT);
1767         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1768
1769         /* Auto-Mask Other interrupts upon ICR read */
1770 #define E1000_EIAC_MASK_82574   0x01F00000
1771         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1772         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1773         ew32(CTRL_EXT, ctrl_ext);
1774         e1e_flush();
1775 }
1776
1777 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1778 {
1779         if (adapter->msix_entries) {
1780                 pci_disable_msix(adapter->pdev);
1781                 kfree(adapter->msix_entries);
1782                 adapter->msix_entries = NULL;
1783         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1784                 pci_disable_msi(adapter->pdev);
1785                 adapter->flags &= ~FLAG_MSI_ENABLED;
1786         }
1787
1788         return;
1789 }
1790
1791 /**
1792  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1793  *
1794  * Attempt to configure interrupts using the best available
1795  * capabilities of the hardware and kernel.
1796  **/
1797 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1798 {
1799         int err;
1800         int numvecs, i;
1801
1802
1803         switch (adapter->int_mode) {
1804         case E1000E_INT_MODE_MSIX:
1805                 if (adapter->flags & FLAG_HAS_MSIX) {
1806                         numvecs = 3; /* RxQ0, TxQ0 and other */
1807                         adapter->msix_entries = kcalloc(numvecs,
1808                                                       sizeof(struct msix_entry),
1809                                                       GFP_KERNEL);
1810                         if (adapter->msix_entries) {
1811                                 for (i = 0; i < numvecs; i++)
1812                                         adapter->msix_entries[i].entry = i;
1813
1814                                 err = pci_enable_msix(adapter->pdev,
1815                                                       adapter->msix_entries,
1816                                                       numvecs);
1817                                 if (err == 0)
1818                                         return;
1819                         }
1820                         /* MSI-X failed, so fall through and try MSI */
1821                         e_err("Failed to initialize MSI-X interrupts.  "
1822                               "Falling back to MSI interrupts.\n");
1823                         e1000e_reset_interrupt_capability(adapter);
1824                 }
1825                 adapter->int_mode = E1000E_INT_MODE_MSI;
1826                 /* Fall through */
1827         case E1000E_INT_MODE_MSI:
1828                 if (!pci_enable_msi(adapter->pdev)) {
1829                         adapter->flags |= FLAG_MSI_ENABLED;
1830                 } else {
1831                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1832                         e_err("Failed to initialize MSI interrupts.  Falling "
1833                               "back to legacy interrupts.\n");
1834                 }
1835                 /* Fall through */
1836         case E1000E_INT_MODE_LEGACY:
1837                 /* Don't do anything; this is the system default */
1838                 break;
1839         }
1840
1841         return;
1842 }
1843
1844 /**
1845  * e1000_request_msix - Initialize MSI-X interrupts
1846  *
1847  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1848  * kernel.
1849  **/
1850 static int e1000_request_msix(struct e1000_adapter *adapter)
1851 {
1852         struct net_device *netdev = adapter->netdev;
1853         int err = 0, vector = 0;
1854
1855         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1856                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1857         else
1858                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1859         err = request_irq(adapter->msix_entries[vector].vector,
1860                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1861                           netdev);
1862         if (err)
1863                 goto out;
1864         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1865         adapter->rx_ring->itr_val = adapter->itr;
1866         vector++;
1867
1868         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1869                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1870         else
1871                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1872         err = request_irq(adapter->msix_entries[vector].vector,
1873                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1874                           netdev);
1875         if (err)
1876                 goto out;
1877         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1878         adapter->tx_ring->itr_val = adapter->itr;
1879         vector++;
1880
1881         err = request_irq(adapter->msix_entries[vector].vector,
1882                           e1000_msix_other, 0, netdev->name, netdev);
1883         if (err)
1884                 goto out;
1885
1886         e1000_configure_msix(adapter);
1887         return 0;
1888 out:
1889         return err;
1890 }
1891
1892 /**
1893  * e1000_request_irq - initialize interrupts
1894  *
1895  * Attempts to configure interrupts using the best available
1896  * capabilities of the hardware and kernel.
1897  **/
1898 static int e1000_request_irq(struct e1000_adapter *adapter)
1899 {
1900         struct net_device *netdev = adapter->netdev;
1901         int err;
1902
1903         if (adapter->msix_entries) {
1904                 err = e1000_request_msix(adapter);
1905                 if (!err)
1906                         return err;
1907                 /* fall back to MSI */
1908                 e1000e_reset_interrupt_capability(adapter);
1909                 adapter->int_mode = E1000E_INT_MODE_MSI;
1910                 e1000e_set_interrupt_capability(adapter);
1911         }
1912         if (adapter->flags & FLAG_MSI_ENABLED) {
1913                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1914                                   netdev->name, netdev);
1915                 if (!err)
1916                         return err;
1917
1918                 /* fall back to legacy interrupt */
1919                 e1000e_reset_interrupt_capability(adapter);
1920                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1921         }
1922
1923         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1924                           netdev->name, netdev);
1925         if (err)
1926                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1927
1928         return err;
1929 }
1930
1931 static void e1000_free_irq(struct e1000_adapter *adapter)
1932 {
1933         struct net_device *netdev = adapter->netdev;
1934
1935         if (adapter->msix_entries) {
1936                 int vector = 0;
1937
1938                 free_irq(adapter->msix_entries[vector].vector, netdev);
1939                 vector++;
1940
1941                 free_irq(adapter->msix_entries[vector].vector, netdev);
1942                 vector++;
1943
1944                 /* Other Causes interrupt vector */
1945                 free_irq(adapter->msix_entries[vector].vector, netdev);
1946                 return;
1947         }
1948
1949         free_irq(adapter->pdev->irq, netdev);
1950 }
1951
1952 /**
1953  * e1000_irq_disable - Mask off interrupt generation on the NIC
1954  **/
1955 static void e1000_irq_disable(struct e1000_adapter *adapter)
1956 {
1957         struct e1000_hw *hw = &adapter->hw;
1958
1959         ew32(IMC, ~0);
1960         if (adapter->msix_entries)
1961                 ew32(EIAC_82574, 0);
1962         e1e_flush();
1963         synchronize_irq(adapter->pdev->irq);
1964 }
1965
1966 /**
1967  * e1000_irq_enable - Enable default interrupt generation settings
1968  **/
1969 static void e1000_irq_enable(struct e1000_adapter *adapter)
1970 {
1971         struct e1000_hw *hw = &adapter->hw;
1972
1973         if (adapter->msix_entries) {
1974                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1975                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1976         } else {
1977                 ew32(IMS, IMS_ENABLE_MASK);
1978         }
1979         e1e_flush();
1980 }
1981
1982 /**
1983  * e1000_get_hw_control - get control of the h/w from f/w
1984  * @adapter: address of board private structure
1985  *
1986  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1987  * For ASF and Pass Through versions of f/w this means that
1988  * the driver is loaded. For AMT version (only with 82573)
1989  * of the f/w this means that the network i/f is open.
1990  **/
1991 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1992 {
1993         struct e1000_hw *hw = &adapter->hw;
1994         u32 ctrl_ext;
1995         u32 swsm;
1996
1997         /* Let firmware know the driver has taken over */
1998         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1999                 swsm = er32(SWSM);
2000                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2001         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2002                 ctrl_ext = er32(CTRL_EXT);
2003                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2004         }
2005 }
2006
2007 /**
2008  * e1000_release_hw_control - release control of the h/w to f/w
2009  * @adapter: address of board private structure
2010  *
2011  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2012  * For ASF and Pass Through versions of f/w this means that the
2013  * driver is no longer loaded. For AMT version (only with 82573) i
2014  * of the f/w this means that the network i/f is closed.
2015  *
2016  **/
2017 static void e1000_release_hw_control(struct e1000_adapter *adapter)
2018 {
2019         struct e1000_hw *hw = &adapter->hw;
2020         u32 ctrl_ext;
2021         u32 swsm;
2022
2023         /* Let firmware taken over control of h/w */
2024         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2025                 swsm = er32(SWSM);
2026                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2027         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2028                 ctrl_ext = er32(CTRL_EXT);
2029                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2030         }
2031 }
2032
2033 /**
2034  * @e1000_alloc_ring - allocate memory for a ring structure
2035  **/
2036 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2037                                 struct e1000_ring *ring)
2038 {
2039         struct pci_dev *pdev = adapter->pdev;
2040
2041         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2042                                         GFP_KERNEL);
2043         if (!ring->desc)
2044                 return -ENOMEM;
2045
2046         return 0;
2047 }
2048
2049 /**
2050  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2051  * @adapter: board private structure
2052  *
2053  * Return 0 on success, negative on failure
2054  **/
2055 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
2056 {
2057         struct e1000_ring *tx_ring = adapter->tx_ring;
2058         int err = -ENOMEM, size;
2059
2060         size = sizeof(struct e1000_buffer) * tx_ring->count;
2061         tx_ring->buffer_info = vmalloc(size);
2062         if (!tx_ring->buffer_info)
2063                 goto err;
2064         memset(tx_ring->buffer_info, 0, size);
2065
2066         /* round up to nearest 4K */
2067         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2068         tx_ring->size = ALIGN(tx_ring->size, 4096);
2069
2070         err = e1000_alloc_ring_dma(adapter, tx_ring);
2071         if (err)
2072                 goto err;
2073
2074         tx_ring->next_to_use = 0;
2075         tx_ring->next_to_clean = 0;
2076
2077         return 0;
2078 err:
2079         vfree(tx_ring->buffer_info);
2080         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2081         return err;
2082 }
2083
2084 /**
2085  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2086  * @adapter: board private structure
2087  *
2088  * Returns 0 on success, negative on failure
2089  **/
2090 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
2091 {
2092         struct e1000_ring *rx_ring = adapter->rx_ring;
2093         struct e1000_buffer *buffer_info;
2094         int i, size, desc_len, err = -ENOMEM;
2095
2096         size = sizeof(struct e1000_buffer) * rx_ring->count;
2097         rx_ring->buffer_info = vmalloc(size);
2098         if (!rx_ring->buffer_info)
2099                 goto err;
2100         memset(rx_ring->buffer_info, 0, size);
2101
2102         for (i = 0; i < rx_ring->count; i++) {
2103                 buffer_info = &rx_ring->buffer_info[i];
2104                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2105                                                 sizeof(struct e1000_ps_page),
2106                                                 GFP_KERNEL);
2107                 if (!buffer_info->ps_pages)
2108                         goto err_pages;
2109         }
2110
2111         desc_len = sizeof(union e1000_rx_desc_packet_split);
2112
2113         /* Round up to nearest 4K */
2114         rx_ring->size = rx_ring->count * desc_len;
2115         rx_ring->size = ALIGN(rx_ring->size, 4096);
2116
2117         err = e1000_alloc_ring_dma(adapter, rx_ring);
2118         if (err)
2119                 goto err_pages;
2120
2121         rx_ring->next_to_clean = 0;
2122         rx_ring->next_to_use = 0;
2123         rx_ring->rx_skb_top = NULL;
2124
2125         return 0;
2126
2127 err_pages:
2128         for (i = 0; i < rx_ring->count; i++) {
2129                 buffer_info = &rx_ring->buffer_info[i];
2130                 kfree(buffer_info->ps_pages);
2131         }
2132 err:
2133         vfree(rx_ring->buffer_info);
2134         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2135         return err;
2136 }
2137
2138 /**
2139  * e1000_clean_tx_ring - Free Tx Buffers
2140  * @adapter: board private structure
2141  **/
2142 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
2143 {
2144         struct e1000_ring *tx_ring = adapter->tx_ring;
2145         struct e1000_buffer *buffer_info;
2146         unsigned long size;
2147         unsigned int i;
2148
2149         for (i = 0; i < tx_ring->count; i++) {
2150                 buffer_info = &tx_ring->buffer_info[i];
2151                 e1000_put_txbuf(adapter, buffer_info);
2152         }
2153
2154         size = sizeof(struct e1000_buffer) * tx_ring->count;
2155         memset(tx_ring->buffer_info, 0, size);
2156
2157         memset(tx_ring->desc, 0, tx_ring->size);
2158
2159         tx_ring->next_to_use = 0;
2160         tx_ring->next_to_clean = 0;
2161
2162         writel(0, adapter->hw.hw_addr + tx_ring->head);
2163         writel(0, adapter->hw.hw_addr + tx_ring->tail);
2164 }
2165
2166 /**
2167  * e1000e_free_tx_resources - Free Tx Resources per Queue
2168  * @adapter: board private structure
2169  *
2170  * Free all transmit software resources
2171  **/
2172 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
2173 {
2174         struct pci_dev *pdev = adapter->pdev;
2175         struct e1000_ring *tx_ring = adapter->tx_ring;
2176
2177         e1000_clean_tx_ring(adapter);
2178
2179         vfree(tx_ring->buffer_info);
2180         tx_ring->buffer_info = NULL;
2181
2182         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2183                           tx_ring->dma);
2184         tx_ring->desc = NULL;
2185 }
2186
2187 /**
2188  * e1000e_free_rx_resources - Free Rx Resources
2189  * @adapter: board private structure
2190  *
2191  * Free all receive software resources
2192  **/
2193
2194 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
2195 {
2196         struct pci_dev *pdev = adapter->pdev;
2197         struct e1000_ring *rx_ring = adapter->rx_ring;
2198         int i;
2199
2200         e1000_clean_rx_ring(adapter);
2201
2202         for (i = 0; i < rx_ring->count; i++) {
2203                 kfree(rx_ring->buffer_info[i].ps_pages);
2204         }
2205
2206         vfree(rx_ring->buffer_info);
2207         rx_ring->buffer_info = NULL;
2208
2209         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2210                           rx_ring->dma);
2211         rx_ring->desc = NULL;
2212 }
2213
2214 /**
2215  * e1000_update_itr - update the dynamic ITR value based on statistics
2216  * @adapter: pointer to adapter
2217  * @itr_setting: current adapter->itr
2218  * @packets: the number of packets during this measurement interval
2219  * @bytes: the number of bytes during this measurement interval
2220  *
2221  *      Stores a new ITR value based on packets and byte
2222  *      counts during the last interrupt.  The advantage of per interrupt
2223  *      computation is faster updates and more accurate ITR for the current
2224  *      traffic pattern.  Constants in this function were computed
2225  *      based on theoretical maximum wire speed and thresholds were set based
2226  *      on testing data as well as attempting to minimize response time
2227  *      while increasing bulk throughput.  This functionality is controlled
2228  *      by the InterruptThrottleRate module parameter.
2229  **/
2230 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2231                                      u16 itr_setting, int packets,
2232                                      int bytes)
2233 {
2234         unsigned int retval = itr_setting;
2235
2236         if (packets == 0)
2237                 goto update_itr_done;
2238
2239         switch (itr_setting) {
2240         case lowest_latency:
2241                 /* handle TSO and jumbo frames */
2242                 if (bytes/packets > 8000)
2243                         retval = bulk_latency;
2244                 else if ((packets < 5) && (bytes > 512)) {
2245                         retval = low_latency;
2246                 }
2247                 break;
2248         case low_latency:  /* 50 usec aka 20000 ints/s */
2249                 if (bytes > 10000) {
2250                         /* this if handles the TSO accounting */
2251                         if (bytes/packets > 8000) {
2252                                 retval = bulk_latency;
2253                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
2254                                 retval = bulk_latency;
2255                         } else if ((packets > 35)) {
2256                                 retval = lowest_latency;
2257                         }
2258                 } else if (bytes/packets > 2000) {
2259                         retval = bulk_latency;
2260                 } else if (packets <= 2 && bytes < 512) {
2261                         retval = lowest_latency;
2262                 }
2263                 break;
2264         case bulk_latency: /* 250 usec aka 4000 ints/s */
2265                 if (bytes > 25000) {
2266                         if (packets > 35) {
2267                                 retval = low_latency;
2268                         }
2269                 } else if (bytes < 6000) {
2270                         retval = low_latency;
2271                 }
2272                 break;
2273         }
2274
2275 update_itr_done:
2276         return retval;
2277 }
2278
2279 static void e1000_set_itr(struct e1000_adapter *adapter)
2280 {
2281         struct e1000_hw *hw = &adapter->hw;
2282         u16 current_itr;
2283         u32 new_itr = adapter->itr;
2284
2285         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2286         if (adapter->link_speed != SPEED_1000) {
2287                 current_itr = 0;
2288                 new_itr = 4000;
2289                 goto set_itr_now;
2290         }
2291
2292         adapter->tx_itr = e1000_update_itr(adapter,
2293                                     adapter->tx_itr,
2294                                     adapter->total_tx_packets,
2295                                     adapter->total_tx_bytes);
2296         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2297         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2298                 adapter->tx_itr = low_latency;
2299
2300         adapter->rx_itr = e1000_update_itr(adapter,
2301                                     adapter->rx_itr,
2302                                     adapter->total_rx_packets,
2303                                     adapter->total_rx_bytes);
2304         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2305         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2306                 adapter->rx_itr = low_latency;
2307
2308         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2309
2310         switch (current_itr) {
2311         /* counts and packets in update_itr are dependent on these numbers */
2312         case lowest_latency:
2313                 new_itr = 70000;
2314                 break;
2315         case low_latency:
2316                 new_itr = 20000; /* aka hwitr = ~200 */
2317                 break;
2318         case bulk_latency:
2319                 new_itr = 4000;
2320                 break;
2321         default:
2322                 break;
2323         }
2324
2325 set_itr_now:
2326         if (new_itr != adapter->itr) {
2327                 /*
2328                  * this attempts to bias the interrupt rate towards Bulk
2329                  * by adding intermediate steps when interrupt rate is
2330                  * increasing
2331                  */
2332                 new_itr = new_itr > adapter->itr ?
2333                              min(adapter->itr + (new_itr >> 2), new_itr) :
2334                              new_itr;
2335                 adapter->itr = new_itr;
2336                 adapter->rx_ring->itr_val = new_itr;
2337                 if (adapter->msix_entries)
2338                         adapter->rx_ring->set_itr = 1;
2339                 else
2340                         ew32(ITR, 1000000000 / (new_itr * 256));
2341         }
2342 }
2343
2344 /**
2345  * e1000_alloc_queues - Allocate memory for all rings
2346  * @adapter: board private structure to initialize
2347  **/
2348 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2349 {
2350         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2351         if (!adapter->tx_ring)
2352                 goto err;
2353
2354         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2355         if (!adapter->rx_ring)
2356                 goto err;
2357
2358         return 0;
2359 err:
2360         e_err("Unable to allocate memory for queues\n");
2361         kfree(adapter->rx_ring);
2362         kfree(adapter->tx_ring);
2363         return -ENOMEM;
2364 }
2365
2366 /**
2367  * e1000_clean - NAPI Rx polling callback
2368  * @napi: struct associated with this polling callback
2369  * @budget: amount of packets driver is allowed to process this poll
2370  **/
2371 static int e1000_clean(struct napi_struct *napi, int budget)
2372 {
2373         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2374         struct e1000_hw *hw = &adapter->hw;
2375         struct net_device *poll_dev = adapter->netdev;
2376         int tx_cleaned = 1, work_done = 0;
2377
2378         adapter = netdev_priv(poll_dev);
2379
2380         if (adapter->msix_entries &&
2381             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2382                 goto clean_rx;
2383
2384         tx_cleaned = e1000_clean_tx_irq(adapter);
2385
2386 clean_rx:
2387         adapter->clean_rx(adapter, &work_done, budget);
2388
2389         if (!tx_cleaned)
2390                 work_done = budget;
2391
2392         /* If budget not fully consumed, exit the polling mode */
2393         if (work_done < budget) {
2394                 if (adapter->itr_setting & 3)
2395                         e1000_set_itr(adapter);
2396                 napi_complete(napi);
2397                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2398                         if (adapter->msix_entries)
2399                                 ew32(IMS, adapter->rx_ring->ims_val);
2400                         else
2401                                 e1000_irq_enable(adapter);
2402                 }
2403         }
2404
2405         return work_done;
2406 }
2407
2408 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2409 {
2410         struct e1000_adapter *adapter = netdev_priv(netdev);
2411         struct e1000_hw *hw = &adapter->hw;
2412         u32 vfta, index;
2413
2414         /* don't update vlan cookie if already programmed */
2415         if ((adapter->hw.mng_cookie.status &
2416              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2417             (vid == adapter->mng_vlan_id))
2418                 return;
2419
2420         /* add VID to filter table */
2421         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2422                 index = (vid >> 5) & 0x7F;
2423                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2424                 vfta |= (1 << (vid & 0x1F));
2425                 hw->mac.ops.write_vfta(hw, index, vfta);
2426         }
2427 }
2428
2429 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2430 {
2431         struct e1000_adapter *adapter = netdev_priv(netdev);
2432         struct e1000_hw *hw = &adapter->hw;
2433         u32 vfta, index;
2434
2435         if (!test_bit(__E1000_DOWN, &adapter->state))
2436                 e1000_irq_disable(adapter);
2437         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2438
2439         if (!test_bit(__E1000_DOWN, &adapter->state))
2440                 e1000_irq_enable(adapter);
2441
2442         if ((adapter->hw.mng_cookie.status &
2443              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2444             (vid == adapter->mng_vlan_id)) {
2445                 /* release control to f/w */
2446                 e1000_release_hw_control(adapter);
2447                 return;
2448         }
2449
2450         /* remove VID from filter table */
2451         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2452                 index = (vid >> 5) & 0x7F;
2453                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2454                 vfta &= ~(1 << (vid & 0x1F));
2455                 hw->mac.ops.write_vfta(hw, index, vfta);
2456         }
2457 }
2458
2459 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2460 {
2461         struct net_device *netdev = adapter->netdev;
2462         u16 vid = adapter->hw.mng_cookie.vlan_id;
2463         u16 old_vid = adapter->mng_vlan_id;
2464
2465         if (!adapter->vlgrp)
2466                 return;
2467
2468         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2469                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2470                 if (adapter->hw.mng_cookie.status &
2471                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2472                         e1000_vlan_rx_add_vid(netdev, vid);
2473                         adapter->mng_vlan_id = vid;
2474                 }
2475
2476                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2477                                 (vid != old_vid) &&
2478                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2479                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2480         } else {
2481                 adapter->mng_vlan_id = vid;
2482         }
2483 }
2484
2485
2486 static void e1000_vlan_rx_register(struct net_device *netdev,
2487                                    struct vlan_group *grp)
2488 {
2489         struct e1000_adapter *adapter = netdev_priv(netdev);
2490         struct e1000_hw *hw = &adapter->hw;
2491         u32 ctrl, rctl;
2492
2493         if (!test_bit(__E1000_DOWN, &adapter->state))
2494                 e1000_irq_disable(adapter);
2495         adapter->vlgrp = grp;
2496
2497         if (grp) {
2498                 /* enable VLAN tag insert/strip */
2499                 ctrl = er32(CTRL);
2500                 ctrl |= E1000_CTRL_VME;
2501                 ew32(CTRL, ctrl);
2502
2503                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2504                         /* enable VLAN receive filtering */
2505                         rctl = er32(RCTL);
2506                         rctl &= ~E1000_RCTL_CFIEN;
2507                         ew32(RCTL, rctl);
2508                         e1000_update_mng_vlan(adapter);
2509                 }
2510         } else {
2511                 /* disable VLAN tag insert/strip */
2512                 ctrl = er32(CTRL);
2513                 ctrl &= ~E1000_CTRL_VME;
2514                 ew32(CTRL, ctrl);
2515
2516                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2517                         if (adapter->mng_vlan_id !=
2518                             (u16)E1000_MNG_VLAN_NONE) {
2519                                 e1000_vlan_rx_kill_vid(netdev,
2520                                                        adapter->mng_vlan_id);
2521                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2522                         }
2523                 }
2524         }
2525
2526         if (!test_bit(__E1000_DOWN, &adapter->state))
2527                 e1000_irq_enable(adapter);
2528 }
2529
2530 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2531 {
2532         u16 vid;
2533
2534         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2535
2536         if (!adapter->vlgrp)
2537                 return;
2538
2539         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2540                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2541                         continue;
2542                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2543         }
2544 }
2545
2546 static void e1000_init_manageability(struct e1000_adapter *adapter)
2547 {
2548         struct e1000_hw *hw = &adapter->hw;
2549         u32 manc, manc2h;
2550
2551         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2552                 return;
2553
2554         manc = er32(MANC);
2555
2556         /*
2557          * enable receiving management packets to the host. this will probably
2558          * generate destination unreachable messages from the host OS, but
2559          * the packets will be handled on SMBUS
2560          */
2561         manc |= E1000_MANC_EN_MNG2HOST;
2562         manc2h = er32(MANC2H);
2563 #define E1000_MNG2HOST_PORT_623 (1 << 5)
2564 #define E1000_MNG2HOST_PORT_664 (1 << 6)
2565         manc2h |= E1000_MNG2HOST_PORT_623;
2566         manc2h |= E1000_MNG2HOST_PORT_664;
2567         ew32(MANC2H, manc2h);
2568         ew32(MANC, manc);
2569 }
2570
2571 /**
2572  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2573  * @adapter: board private structure
2574  *
2575  * Configure the Tx unit of the MAC after a reset.
2576  **/
2577 static void e1000_configure_tx(struct e1000_adapter *adapter)
2578 {
2579         struct e1000_hw *hw = &adapter->hw;
2580         struct e1000_ring *tx_ring = adapter->tx_ring;
2581         u64 tdba;
2582         u32 tdlen, tctl, tipg, tarc;
2583         u32 ipgr1, ipgr2;
2584
2585         /* Setup the HW Tx Head and Tail descriptor pointers */
2586         tdba = tx_ring->dma;
2587         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2588         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2589         ew32(TDBAH, (tdba >> 32));
2590         ew32(TDLEN, tdlen);
2591         ew32(TDH, 0);
2592         ew32(TDT, 0);
2593         tx_ring->head = E1000_TDH;
2594         tx_ring->tail = E1000_TDT;
2595
2596         /* Set the default values for the Tx Inter Packet Gap timer */
2597         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2598         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2599         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2600
2601         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2602                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2603
2604         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2605         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2606         ew32(TIPG, tipg);
2607
2608         /* Set the Tx Interrupt Delay register */
2609         ew32(TIDV, adapter->tx_int_delay);
2610         /* Tx irq moderation */
2611         ew32(TADV, adapter->tx_abs_int_delay);
2612
2613         /* Program the Transmit Control Register */
2614         tctl = er32(TCTL);
2615         tctl &= ~E1000_TCTL_CT;
2616         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2617                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2618
2619         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2620                 tarc = er32(TARC(0));
2621                 /*
2622                  * set the speed mode bit, we'll clear it if we're not at
2623                  * gigabit link later
2624                  */
2625 #define SPEED_MODE_BIT (1 << 21)
2626                 tarc |= SPEED_MODE_BIT;
2627                 ew32(TARC(0), tarc);
2628         }
2629
2630         /* errata: program both queues to unweighted RR */
2631         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2632                 tarc = er32(TARC(0));
2633                 tarc |= 1;
2634                 ew32(TARC(0), tarc);
2635                 tarc = er32(TARC(1));
2636                 tarc |= 1;
2637                 ew32(TARC(1), tarc);
2638         }
2639
2640         /* Setup Transmit Descriptor Settings for eop descriptor */
2641         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2642
2643         /* only set IDE if we are delaying interrupts using the timers */
2644         if (adapter->tx_int_delay)
2645                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2646
2647         /* enable Report Status bit */
2648         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2649
2650         ew32(TCTL, tctl);
2651
2652         e1000e_config_collision_dist(hw);
2653 }
2654
2655 /**
2656  * e1000_setup_rctl - configure the receive control registers
2657  * @adapter: Board private structure
2658  **/
2659 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2660                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2661 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2662 {
2663         struct e1000_hw *hw = &adapter->hw;
2664         u32 rctl, rfctl;
2665         u32 psrctl = 0;
2666         u32 pages = 0;
2667
2668         /* Program MC offset vector base */
2669         rctl = er32(RCTL);
2670         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2671         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2672                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2673                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2674
2675         /* Do not Store bad packets */
2676         rctl &= ~E1000_RCTL_SBP;
2677
2678         /* Enable Long Packet receive */
2679         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2680                 rctl &= ~E1000_RCTL_LPE;
2681         else
2682                 rctl |= E1000_RCTL_LPE;
2683
2684         /* Some systems expect that the CRC is included in SMBUS traffic. The
2685          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2686          * host memory when this is enabled
2687          */
2688         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2689                 rctl |= E1000_RCTL_SECRC;
2690
2691         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2692         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2693                 u16 phy_data;
2694
2695                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2696                 phy_data &= 0xfff8;
2697                 phy_data |= (1 << 2);
2698                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2699
2700                 e1e_rphy(hw, 22, &phy_data);
2701                 phy_data &= 0x0fff;
2702                 phy_data |= (1 << 14);
2703                 e1e_wphy(hw, 0x10, 0x2823);
2704                 e1e_wphy(hw, 0x11, 0x0003);
2705                 e1e_wphy(hw, 22, phy_data);
2706         }
2707
2708         /* Setup buffer sizes */
2709         rctl &= ~E1000_RCTL_SZ_4096;
2710         rctl |= E1000_RCTL_BSEX;
2711         switch (adapter->rx_buffer_len) {
2712         case 2048:
2713         default:
2714                 rctl |= E1000_RCTL_SZ_2048;
2715                 rctl &= ~E1000_RCTL_BSEX;
2716                 break;
2717         case 4096:
2718                 rctl |= E1000_RCTL_SZ_4096;
2719                 break;
2720         case 8192:
2721                 rctl |= E1000_RCTL_SZ_8192;
2722                 break;
2723         case 16384:
2724                 rctl |= E1000_RCTL_SZ_16384;
2725                 break;
2726         }
2727
2728         /*
2729          * 82571 and greater support packet-split where the protocol
2730          * header is placed in skb->data and the packet data is
2731          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2732          * In the case of a non-split, skb->data is linearly filled,
2733          * followed by the page buffers.  Therefore, skb->data is
2734          * sized to hold the largest protocol header.
2735          *
2736          * allocations using alloc_page take too long for regular MTU
2737          * so only enable packet split for jumbo frames
2738          *
2739          * Using pages when the page size is greater than 16k wastes
2740          * a lot of memory, since we allocate 3 pages at all times
2741          * per packet.
2742          */
2743         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2744         if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2745             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2746                 adapter->rx_ps_pages = pages;
2747         else
2748                 adapter->rx_ps_pages = 0;
2749
2750         if (adapter->rx_ps_pages) {
2751                 /* Configure extra packet-split registers */
2752                 rfctl = er32(RFCTL);
2753                 rfctl |= E1000_RFCTL_EXTEN;
2754                 /*
2755                  * disable packet split support for IPv6 extension headers,
2756                  * because some malformed IPv6 headers can hang the Rx
2757                  */
2758                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2759                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2760
2761                 ew32(RFCTL, rfctl);
2762
2763                 /* Enable Packet split descriptors */
2764                 rctl |= E1000_RCTL_DTYP_PS;
2765
2766                 psrctl |= adapter->rx_ps_bsize0 >>
2767                         E1000_PSRCTL_BSIZE0_SHIFT;
2768
2769                 switch (adapter->rx_ps_pages) {
2770                 case 3:
2771                         psrctl |= PAGE_SIZE <<
2772                                 E1000_PSRCTL_BSIZE3_SHIFT;
2773                 case 2:
2774                         psrctl |= PAGE_SIZE <<
2775                                 E1000_PSRCTL_BSIZE2_SHIFT;
2776                 case 1:
2777                         psrctl |= PAGE_SIZE >>
2778                                 E1000_PSRCTL_BSIZE1_SHIFT;
2779                         break;
2780                 }
2781
2782                 ew32(PSRCTL, psrctl);
2783         }
2784
2785         ew32(RCTL, rctl);
2786         /* just started the receive unit, no need to restart */
2787         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2788 }
2789
2790 /**
2791  * e1000_configure_rx - Configure Receive Unit after Reset
2792  * @adapter: board private structure
2793  *
2794  * Configure the Rx unit of the MAC after a reset.
2795  **/
2796 static void e1000_configure_rx(struct e1000_adapter *adapter)
2797 {
2798         struct e1000_hw *hw = &adapter->hw;
2799         struct e1000_ring *rx_ring = adapter->rx_ring;
2800         u64 rdba;
2801         u32 rdlen, rctl, rxcsum, ctrl_ext;
2802
2803         if (adapter->rx_ps_pages) {
2804                 /* this is a 32 byte descriptor */
2805                 rdlen = rx_ring->count *
2806                         sizeof(union e1000_rx_desc_packet_split);
2807                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2808                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2809         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2810                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2811                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2812                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2813         } else {
2814                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2815                 adapter->clean_rx = e1000_clean_rx_irq;
2816                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2817         }
2818
2819         /* disable receives while setting up the descriptors */
2820         rctl = er32(RCTL);
2821         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2822         e1e_flush();
2823         msleep(10);
2824
2825         /* set the Receive Delay Timer Register */
2826         ew32(RDTR, adapter->rx_int_delay);
2827
2828         /* irq moderation */
2829         ew32(RADV, adapter->rx_abs_int_delay);
2830         if (adapter->itr_setting != 0)
2831                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2832
2833         ctrl_ext = er32(CTRL_EXT);
2834         /* Auto-Mask interrupts upon ICR access */
2835         ctrl_ext |= E1000_CTRL_EXT_IAME;
2836         ew32(IAM, 0xffffffff);
2837         ew32(CTRL_EXT, ctrl_ext);
2838         e1e_flush();
2839
2840         /*
2841          * Setup the HW Rx Head and Tail Descriptor Pointers and
2842          * the Base and Length of the Rx Descriptor Ring
2843          */
2844         rdba = rx_ring->dma;
2845         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2846         ew32(RDBAH, (rdba >> 32));
2847         ew32(RDLEN, rdlen);
2848         ew32(RDH, 0);
2849         ew32(RDT, 0);
2850         rx_ring->head = E1000_RDH;
2851         rx_ring->tail = E1000_RDT;
2852
2853         /* Enable Receive Checksum Offload for TCP and UDP */
2854         rxcsum = er32(RXCSUM);
2855         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2856                 rxcsum |= E1000_RXCSUM_TUOFL;
2857
2858                 /*
2859                  * IPv4 payload checksum for UDP fragments must be
2860                  * used in conjunction with packet-split.
2861                  */
2862                 if (adapter->rx_ps_pages)
2863                         rxcsum |= E1000_RXCSUM_IPPCSE;
2864         } else {
2865                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2866                 /* no need to clear IPPCSE as it defaults to 0 */
2867         }
2868         ew32(RXCSUM, rxcsum);
2869
2870         /*
2871          * Enable early receives on supported devices, only takes effect when
2872          * packet size is equal or larger than the specified value (in 8 byte
2873          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2874          */
2875         if (adapter->flags & FLAG_HAS_ERT) {
2876                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2877                         u32 rxdctl = er32(RXDCTL(0));
2878                         ew32(RXDCTL(0), rxdctl | 0x3);
2879                         ew32(ERT, E1000_ERT_2048 | (1 << 13));
2880                         /*
2881                          * With jumbo frames and early-receive enabled,
2882                          * excessive C-state transition latencies result in
2883                          * dropped transactions.
2884                          */
2885                         pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2886                                                   adapter->netdev->name, 55);
2887                 } else {
2888                         pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2889                                                   adapter->netdev->name,
2890                                                   PM_QOS_DEFAULT_VALUE);
2891                 }
2892         }
2893
2894         /* Enable Receives */
2895         ew32(RCTL, rctl);
2896 }
2897
2898 /**
2899  *  e1000_update_mc_addr_list - Update Multicast addresses
2900  *  @hw: pointer to the HW structure
2901  *  @mc_addr_list: array of multicast addresses to program
2902  *  @mc_addr_count: number of multicast addresses to program
2903  *
2904  *  Updates the Multicast Table Array.
2905  *  The caller must have a packed mc_addr_list of multicast addresses.
2906  **/
2907 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2908                                       u32 mc_addr_count)
2909 {
2910         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
2911 }
2912
2913 /**
2914  * e1000_set_multi - Multicast and Promiscuous mode set
2915  * @netdev: network interface device structure
2916  *
2917  * The set_multi entry point is called whenever the multicast address
2918  * list or the network interface flags are updated.  This routine is
2919  * responsible for configuring the hardware for proper multicast,
2920  * promiscuous mode, and all-multi behavior.
2921  **/
2922 static void e1000_set_multi(struct net_device *netdev)
2923 {
2924         struct e1000_adapter *adapter = netdev_priv(netdev);
2925         struct e1000_hw *hw = &adapter->hw;
2926         struct netdev_hw_addr *ha;
2927         u8  *mta_list;
2928         u32 rctl;
2929         int i;
2930
2931         /* Check for Promiscuous and All Multicast modes */
2932
2933         rctl = er32(RCTL);
2934
2935         if (netdev->flags & IFF_PROMISC) {
2936                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2937                 rctl &= ~E1000_RCTL_VFE;
2938         } else {
2939                 if (netdev->flags & IFF_ALLMULTI) {
2940                         rctl |= E1000_RCTL_MPE;
2941                         rctl &= ~E1000_RCTL_UPE;
2942                 } else {
2943                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2944                 }
2945                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2946                         rctl |= E1000_RCTL_VFE;
2947         }
2948
2949         ew32(RCTL, rctl);
2950
2951         if (!netdev_mc_empty(netdev)) {
2952                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
2953                 if (!mta_list)
2954                         return;
2955
2956                 /* prepare a packed array of only addresses. */
2957                 i = 0;
2958                 netdev_for_each_mc_addr(ha, netdev)
2959                         memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
2960
2961                 e1000_update_mc_addr_list(hw, mta_list, i);
2962                 kfree(mta_list);
2963         } else {
2964                 /*
2965                  * if we're called from probe, we might not have
2966                  * anything to do here, so clear out the list
2967                  */
2968                 e1000_update_mc_addr_list(hw, NULL, 0);
2969         }
2970 }
2971
2972 /**
2973  * e1000_configure - configure the hardware for Rx and Tx
2974  * @adapter: private board structure
2975  **/
2976 static void e1000_configure(struct e1000_adapter *adapter)
2977 {
2978         e1000_set_multi(adapter->netdev);
2979
2980         e1000_restore_vlan(adapter);
2981         e1000_init_manageability(adapter);
2982
2983         e1000_configure_tx(adapter);
2984         e1000_setup_rctl(adapter);
2985         e1000_configure_rx(adapter);
2986         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
2987 }
2988
2989 /**
2990  * e1000e_power_up_phy - restore link in case the phy was powered down
2991  * @adapter: address of board private structure
2992  *
2993  * The phy may be powered down to save power and turn off link when the
2994  * driver is unloaded and wake on lan is not enabled (among others)
2995  * *** this routine MUST be followed by a call to e1000e_reset ***
2996  **/
2997 void e1000e_power_up_phy(struct e1000_adapter *adapter)
2998 {
2999         if (adapter->hw.phy.ops.power_up)
3000                 adapter->hw.phy.ops.power_up(&adapter->hw);
3001
3002         adapter->hw.mac.ops.setup_link(&adapter->hw);
3003 }
3004
3005 /**
3006  * e1000_power_down_phy - Power down the PHY
3007  *
3008  * Power down the PHY so no link is implied when interface is down.
3009  * The PHY cannot be powered down if management or WoL is active.
3010  */
3011 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3012 {
3013         /* WoL is enabled */
3014         if (adapter->wol)
3015                 return;
3016
3017         if (adapter->hw.phy.ops.power_down)
3018                 adapter->hw.phy.ops.power_down(&adapter->hw);
3019 }
3020
3021 /**
3022  * e1000e_reset - bring the hardware into a known good state
3023  *
3024  * This function boots the hardware and enables some settings that
3025  * require a configuration cycle of the hardware - those cannot be
3026  * set/changed during runtime. After reset the device needs to be
3027  * properly configured for Rx, Tx etc.
3028  */
3029 void e1000e_reset(struct e1000_adapter *adapter)
3030 {
3031         struct e1000_mac_info *mac = &adapter->hw.mac;
3032         struct e1000_fc_info *fc = &adapter->hw.fc;
3033         struct e1000_hw *hw = &adapter->hw;
3034         u32 tx_space, min_tx_space, min_rx_space;
3035         u32 pba = adapter->pba;
3036         u16 hwm;
3037
3038         /* reset Packet Buffer Allocation to default */
3039         ew32(PBA, pba);
3040
3041         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3042                 /*
3043                  * To maintain wire speed transmits, the Tx FIFO should be
3044                  * large enough to accommodate two full transmit packets,
3045                  * rounded up to the next 1KB and expressed in KB.  Likewise,
3046                  * the Rx FIFO should be large enough to accommodate at least
3047                  * one full receive packet and is similarly rounded up and
3048                  * expressed in KB.
3049                  */
3050                 pba = er32(PBA);
3051                 /* upper 16 bits has Tx packet buffer allocation size in KB */
3052                 tx_space = pba >> 16;
3053                 /* lower 16 bits has Rx packet buffer allocation size in KB */
3054                 pba &= 0xffff;
3055                 /*
3056                  * the Tx fifo also stores 16 bytes of information about the tx
3057                  * but don't include ethernet FCS because hardware appends it
3058                  */
3059                 min_tx_space = (adapter->max_frame_size +
3060                                 sizeof(struct e1000_tx_desc) -
3061                                 ETH_FCS_LEN) * 2;
3062                 min_tx_space = ALIGN(min_tx_space, 1024);
3063                 min_tx_space >>= 10;
3064                 /* software strips receive CRC, so leave room for it */
3065                 min_rx_space = adapter->max_frame_size;
3066                 min_rx_space = ALIGN(min_rx_space, 1024);
3067                 min_rx_space >>= 10;
3068
3069                 /*
3070                  * If current Tx allocation is less than the min Tx FIFO size,
3071                  * and the min Tx FIFO size is less than the current Rx FIFO
3072                  * allocation, take space away from current Rx allocation
3073                  */
3074                 if ((tx_space < min_tx_space) &&
3075                     ((min_tx_space - tx_space) < pba)) {
3076                         pba -= min_tx_space - tx_space;
3077
3078                         /*
3079                          * if short on Rx space, Rx wins and must trump tx
3080                          * adjustment or use Early Receive if available
3081                          */
3082                         if ((pba < min_rx_space) &&
3083                             (!(adapter->flags & FLAG_HAS_ERT)))
3084                                 /* ERT enabled in e1000_configure_rx */
3085                                 pba = min_rx_space;
3086                 }
3087
3088                 ew32(PBA, pba);
3089         }
3090
3091
3092         /*
3093          * flow control settings
3094          *
3095          * The high water mark must be low enough to fit one full frame
3096          * (or the size used for early receive) above it in the Rx FIFO.
3097          * Set it to the lower of:
3098          * - 90% of the Rx FIFO size, and
3099          * - the full Rx FIFO size minus the early receive size (for parts
3100          *   with ERT support assuming ERT set to E1000_ERT_2048), or
3101          * - the full Rx FIFO size minus one full frame
3102          */
3103         if (hw->mac.type == e1000_pchlan) {
3104                 /*
3105                  * Workaround PCH LOM adapter hangs with certain network
3106                  * loads.  If hangs persist, try disabling Tx flow control.
3107                  */
3108                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3109                         fc->high_water = 0x3500;
3110                         fc->low_water  = 0x1500;
3111                 } else {
3112                         fc->high_water = 0x5000;
3113                         fc->low_water  = 0x3000;
3114                 }
3115         } else {
3116                 if ((adapter->flags & FLAG_HAS_ERT) &&
3117                     (adapter->netdev->mtu > ETH_DATA_LEN))
3118                         hwm = min(((pba << 10) * 9 / 10),
3119                                   ((pba << 10) - (E1000_ERT_2048 << 3)));
3120                 else
3121                         hwm = min(((pba << 10) * 9 / 10),
3122                                   ((pba << 10) - adapter->max_frame_size));
3123
3124                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3125                 fc->low_water = fc->high_water - 8;
3126         }
3127
3128         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3129                 fc->pause_time = 0xFFFF;
3130         else
3131                 fc->pause_time = E1000_FC_PAUSE_TIME;
3132         fc->send_xon = 1;
3133         fc->current_mode = fc->requested_mode;
3134
3135         /* Allow time for pending master requests to run */
3136         mac->ops.reset_hw(hw);
3137
3138         /*
3139          * For parts with AMT enabled, let the firmware know
3140          * that the network interface is in control
3141          */
3142         if (adapter->flags & FLAG_HAS_AMT)
3143                 e1000_get_hw_control(adapter);
3144
3145         ew32(WUC, 0);
3146         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP)
3147                 e1e_wphy(&adapter->hw, BM_WUC, 0);
3148
3149         if (mac->ops.init_hw(hw))
3150                 e_err("Hardware Error\n");
3151
3152         /* additional part of the flow-control workaround above */
3153         if (hw->mac.type == e1000_pchlan)
3154                 ew32(FCRTV_PCH, 0x1000);
3155
3156         e1000_update_mng_vlan(adapter);
3157
3158         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3159         ew32(VET, ETH_P_8021Q);
3160
3161         e1000e_reset_adaptive(hw);
3162         e1000_get_phy_info(hw);
3163
3164         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3165             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3166                 u16 phy_data = 0;
3167                 /*
3168                  * speed up time to link by disabling smart power down, ignore
3169                  * the return value of this function because there is nothing
3170                  * different we would do if it failed
3171                  */
3172                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3173                 phy_data &= ~IGP02E1000_PM_SPD;
3174                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3175         }
3176 }
3177
3178 int e1000e_up(struct e1000_adapter *adapter)
3179 {
3180         struct e1000_hw *hw = &adapter->hw;
3181
3182         /* DMA latency requirement to workaround early-receive/jumbo issue */
3183         if (adapter->flags & FLAG_HAS_ERT)
3184                 pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY,
3185                                        adapter->netdev->name,
3186                                        PM_QOS_DEFAULT_VALUE);
3187
3188         /* hardware has been reset, we need to reload some things */
3189         e1000_configure(adapter);
3190
3191         clear_bit(__E1000_DOWN, &adapter->state);
3192
3193         napi_enable(&adapter->napi);
3194         if (adapter->msix_entries)
3195                 e1000_configure_msix(adapter);
3196         e1000_irq_enable(adapter);
3197
3198         netif_wake_queue(adapter->netdev);
3199
3200         /* fire a link change interrupt to start the watchdog */
3201         ew32(ICS, E1000_ICS_LSC);
3202         return 0;
3203 }
3204
3205 void e1000e_down(struct e1000_adapter *adapter)
3206 {
3207         struct net_device *netdev = adapter->netdev;
3208         struct e1000_hw *hw = &adapter->hw;
3209         u32 tctl, rctl;
3210
3211         /*
3212          * signal that we're down so the interrupt handler does not
3213          * reschedule our watchdog timer
3214          */
3215         set_bit(__E1000_DOWN, &adapter->state);
3216
3217         /* disable receives in the hardware */
3218         rctl = er32(RCTL);
3219         ew32(RCTL, rctl & ~E1000_RCTL_EN);
3220         /* flush and sleep below */
3221
3222         netif_stop_queue(netdev);
3223
3224         /* disable transmits in the hardware */
3225         tctl = er32(TCTL);
3226         tctl &= ~E1000_TCTL_EN;
3227         ew32(TCTL, tctl);
3228         /* flush both disables and wait for them to finish */
3229         e1e_flush();
3230         msleep(10);
3231
3232         napi_disable(&adapter->napi);
3233         e1000_irq_disable(adapter);
3234
3235         del_timer_sync(&adapter->watchdog_timer);
3236         del_timer_sync(&adapter->phy_info_timer);
3237
3238         netif_carrier_off(netdev);
3239         adapter->link_speed = 0;
3240         adapter->link_duplex = 0;
3241
3242         if (!pci_channel_offline(adapter->pdev))
3243                 e1000e_reset(adapter);
3244         e1000_clean_tx_ring(adapter);
3245         e1000_clean_rx_ring(adapter);
3246
3247         if (adapter->flags & FLAG_HAS_ERT)
3248                 pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY,
3249                                           adapter->netdev->name);
3250
3251         /*
3252          * TODO: for power management, we could drop the link and
3253          * pci_disable_device here.
3254          */
3255 }
3256
3257 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3258 {
3259         might_sleep();
3260         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3261                 msleep(1);
3262         e1000e_down(adapter);
3263         e1000e_up(adapter);
3264         clear_bit(__E1000_RESETTING, &adapter->state);
3265 }
3266
3267 /**
3268  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3269  * @adapter: board private structure to initialize
3270  *
3271  * e1000_sw_init initializes the Adapter private data structure.
3272  * Fields are initialized based on PCI device information and
3273  * OS network device settings (MTU size).
3274  **/
3275 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3276 {
3277         struct net_device *netdev = adapter->netdev;
3278
3279         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3280         adapter->rx_ps_bsize0 = 128;
3281         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3282         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3283
3284         e1000e_set_interrupt_capability(adapter);
3285
3286         if (e1000_alloc_queues(adapter))
3287                 return -ENOMEM;
3288
3289         /* Explicitly disable IRQ since the NIC can be in any state. */
3290         e1000_irq_disable(adapter);
3291
3292         set_bit(__E1000_DOWN, &adapter->state);
3293         return 0;
3294 }
3295
3296 /**
3297  * e1000_intr_msi_test - Interrupt Handler
3298  * @irq: interrupt number
3299  * @data: pointer to a network interface device structure
3300  **/
3301 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3302 {
3303         struct net_device *netdev = data;
3304         struct e1000_adapter *adapter = netdev_priv(netdev);
3305         struct e1000_hw *hw = &adapter->hw;
3306         u32 icr = er32(ICR);
3307
3308         e_dbg("icr is %08X\n", icr);
3309         if (icr & E1000_ICR_RXSEQ) {
3310                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3311                 wmb();
3312         }
3313
3314         return IRQ_HANDLED;
3315 }
3316
3317 /**
3318  * e1000_test_msi_interrupt - Returns 0 for successful test
3319  * @adapter: board private struct
3320  *
3321  * code flow taken from tg3.c
3322  **/
3323 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3324 {
3325         struct net_device *netdev = adapter->netdev;
3326         struct e1000_hw *hw = &adapter->hw;
3327         int err;
3328
3329         /* poll_enable hasn't been called yet, so don't need disable */
3330         /* clear any pending events */
3331         er32(ICR);
3332
3333         /* free the real vector and request a test handler */
3334         e1000_free_irq(adapter);
3335         e1000e_reset_interrupt_capability(adapter);
3336
3337         /* Assume that the test fails, if it succeeds then the test
3338          * MSI irq handler will unset this flag */
3339         adapter->flags |= FLAG_MSI_TEST_FAILED;
3340
3341         err = pci_enable_msi(adapter->pdev);
3342         if (err)
3343                 goto msi_test_failed;
3344
3345         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3346                           netdev->name, netdev);
3347         if (err) {
3348                 pci_disable_msi(adapter->pdev);
3349                 goto msi_test_failed;
3350         }
3351
3352         wmb();
3353
3354         e1000_irq_enable(adapter);
3355
3356         /* fire an unusual interrupt on the test handler */
3357         ew32(ICS, E1000_ICS_RXSEQ);
3358         e1e_flush();
3359         msleep(50);
3360
3361         e1000_irq_disable(adapter);
3362
3363         rmb();
3364
3365         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3366                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3367                 err = -EIO;
3368                 e_info("MSI interrupt test failed!\n");
3369         }
3370
3371         free_irq(adapter->pdev->irq, netdev);
3372         pci_disable_msi(adapter->pdev);
3373
3374         if (err == -EIO)
3375                 goto msi_test_failed;
3376
3377         /* okay so the test worked, restore settings */
3378         e_dbg("MSI interrupt test succeeded!\n");
3379 msi_test_failed:
3380         e1000e_set_interrupt_capability(adapter);
3381         e1000_request_irq(adapter);
3382         return err;
3383 }
3384
3385 /**
3386  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3387  * @adapter: board private struct
3388  *
3389  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3390  **/
3391 static int e1000_test_msi(struct e1000_adapter *adapter)
3392 {
3393         int err;
3394         u16 pci_cmd;
3395
3396         if (!(adapter->flags & FLAG_MSI_ENABLED))
3397                 return 0;
3398
3399         /* disable SERR in case the MSI write causes a master abort */
3400         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3401         pci_write_config_word(adapter->pdev, PCI_COMMAND,
3402                               pci_cmd & ~PCI_COMMAND_SERR);
3403
3404         err = e1000_test_msi_interrupt(adapter);
3405
3406         /* restore previous setting of command word */
3407         pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3408
3409         /* success ! */
3410         if (!err)
3411                 return 0;
3412
3413         /* EIO means MSI test failed */
3414         if (err != -EIO)
3415                 return err;
3416
3417         /* back to INTx mode */
3418         e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3419
3420         e1000_free_irq(adapter);
3421
3422         err = e1000_request_irq(adapter);
3423
3424         return err;
3425 }
3426
3427 /**
3428  * e1000_open - Called when a network interface is made active
3429  * @netdev: network interface device structure
3430  *
3431  * Returns 0 on success, negative value on failure
3432  *
3433  * The open entry point is called when a network interface is made
3434  * active by the system (IFF_UP).  At this point all resources needed
3435  * for transmit and receive operations are allocated, the interrupt
3436  * handler is registered with the OS, the watchdog timer is started,
3437  * and the stack is notified that the interface is ready.
3438  **/
3439 static int e1000_open(struct net_device *netdev)
3440 {
3441         struct e1000_adapter *adapter = netdev_priv(netdev);
3442         struct e1000_hw *hw = &adapter->hw;
3443         struct pci_dev *pdev = adapter->pdev;
3444         int err;
3445
3446         /* disallow open during test */
3447         if (test_bit(__E1000_TESTING, &adapter->state))
3448                 return -EBUSY;
3449
3450         pm_runtime_get_sync(&pdev->dev);
3451
3452         netif_carrier_off(netdev);
3453
3454         /* allocate transmit descriptors */
3455         err = e1000e_setup_tx_resources(adapter);
3456         if (err)
3457                 goto err_setup_tx;
3458
3459         /* allocate receive descriptors */
3460         err = e1000e_setup_rx_resources(adapter);
3461         if (err)
3462                 goto err_setup_rx;
3463
3464         e1000e_power_up_phy(adapter);
3465
3466         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3467         if ((adapter->hw.mng_cookie.status &
3468              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3469                 e1000_update_mng_vlan(adapter);
3470
3471         /*
3472          * If AMT is enabled, let the firmware know that the network
3473          * interface is now open
3474          */
3475         if (adapter->flags & FLAG_HAS_AMT)
3476                 e1000_get_hw_control(adapter);
3477
3478         /*
3479          * before we allocate an interrupt, we must be ready to handle it.
3480          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3481          * as soon as we call pci_request_irq, so we have to setup our
3482          * clean_rx handler before we do so.
3483          */
3484         e1000_configure(adapter);
3485
3486         err = e1000_request_irq(adapter);
3487         if (err)
3488                 goto err_req_irq;
3489
3490         /*
3491          * Work around PCIe errata with MSI interrupts causing some chipsets to
3492          * ignore e1000e MSI messages, which means we need to test our MSI
3493          * interrupt now
3494          */
3495         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3496                 err = e1000_test_msi(adapter);
3497                 if (err) {
3498                         e_err("Interrupt allocation failed\n");
3499                         goto err_req_irq;
3500                 }
3501         }
3502
3503         /* From here on the code is the same as e1000e_up() */
3504         clear_bit(__E1000_DOWN, &adapter->state);
3505
3506         napi_enable(&adapter->napi);
3507
3508         e1000_irq_enable(adapter);
3509
3510         netif_start_queue(netdev);
3511
3512         adapter->idle_check = true;
3513         pm_runtime_put(&pdev->dev);
3514
3515         /* fire a link status change interrupt to start the watchdog */
3516         ew32(ICS, E1000_ICS_LSC);
3517
3518         return 0;
3519
3520 err_req_irq:
3521         e1000_release_hw_control(adapter);
3522         e1000_power_down_phy(adapter);
3523         e1000e_free_rx_resources(adapter);
3524 err_setup_rx:
3525         e1000e_free_tx_resources(adapter);
3526 err_setup_tx:
3527         e1000e_reset(adapter);
3528         pm_runtime_put_sync(&pdev->dev);
3529
3530         return err;
3531 }
3532
3533 /**
3534  * e1000_close - Disables a network interface
3535  * @netdev: network interface device structure
3536  *
3537  * Returns 0, this is not allowed to fail
3538  *
3539  * The close entry point is called when an interface is de-activated
3540  * by the OS.  The hardware is still under the drivers control, but
3541  * needs to be disabled.  A global MAC reset is issued to stop the
3542  * hardware, and all transmit and receive resources are freed.
3543  **/
3544 static int e1000_close(struct net_device *netdev)
3545 {
3546         struct e1000_adapter *adapter = netdev_priv(netdev);
3547         struct pci_dev *pdev = adapter->pdev;
3548
3549         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3550
3551         pm_runtime_get_sync(&pdev->dev);
3552
3553         if (!test_bit(__E1000_DOWN, &adapter->state)) {
3554                 e1000e_down(adapter);
3555                 e1000_free_irq(adapter);
3556         }
3557         e1000_power_down_phy(adapter);
3558
3559         e1000e_free_tx_resources(adapter);
3560         e1000e_free_rx_resources(adapter);
3561
3562         /*
3563          * kill manageability vlan ID if supported, but not if a vlan with
3564          * the same ID is registered on the host OS (let 8021q kill it)
3565          */
3566         if ((adapter->hw.mng_cookie.status &
3567                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3568              !(adapter->vlgrp &&
3569                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3570                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3571
3572         /*
3573          * If AMT is enabled, let the firmware know that the network
3574          * interface is now closed
3575          */
3576         if (adapter->flags & FLAG_HAS_AMT)
3577                 e1000_release_hw_control(adapter);
3578
3579         pm_runtime_put_sync(&pdev->dev);
3580
3581         return 0;
3582 }
3583 /**
3584  * e1000_set_mac - Change the Ethernet Address of the NIC
3585  * @netdev: network interface device structure
3586  * @p: pointer to an address structure
3587  *
3588  * Returns 0 on success, negative on failure
3589  **/
3590 static int e1000_set_mac(struct net_device *netdev, void *p)
3591 {
3592         struct e1000_adapter *adapter = netdev_priv(netdev);
3593         struct sockaddr *addr = p;
3594
3595         if (!is_valid_ether_addr(addr->sa_data))
3596                 return -EADDRNOTAVAIL;
3597
3598         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3599         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3600
3601         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3602
3603         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3604                 /* activate the work around */
3605                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3606
3607                 /*
3608                  * Hold a copy of the LAA in RAR[14] This is done so that
3609                  * between the time RAR[0] gets clobbered  and the time it
3610                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3611                  * of the RARs and no incoming packets directed to this port
3612                  * are dropped. Eventually the LAA will be in RAR[0] and
3613                  * RAR[14]
3614                  */
3615                 e1000e_rar_set(&adapter->hw,
3616                               adapter->hw.mac.addr,
3617                               adapter->hw.mac.rar_entry_count - 1);
3618         }
3619
3620         return 0;
3621 }
3622
3623 /**
3624  * e1000e_update_phy_task - work thread to update phy
3625  * @work: pointer to our work struct
3626  *
3627  * this worker thread exists because we must acquire a
3628  * semaphore to read the phy, which we could msleep while
3629  * waiting for it, and we can't msleep in a timer.
3630  **/
3631 static void e1000e_update_phy_task(struct work_struct *work)
3632 {
3633         struct e1000_adapter *adapter = container_of(work,
3634                                         struct e1000_adapter, update_phy_task);
3635         e1000_get_phy_info(&adapter->hw);
3636 }
3637
3638 /*
3639  * Need to wait a few seconds after link up to get diagnostic information from
3640  * the phy
3641  */
3642 static void e1000_update_phy_info(unsigned long data)
3643 {
3644         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3645         schedule_work(&adapter->update_phy_task);
3646 }
3647
3648 /**
3649  * e1000e_update_stats - Update the board statistics counters
3650  * @adapter: board private structure
3651  **/
3652 void e1000e_update_stats(struct e1000_adapter *adapter)
3653 {
3654         struct net_device *netdev = adapter->netdev;
3655         struct e1000_hw *hw = &adapter->hw;
3656         struct pci_dev *pdev = adapter->pdev;
3657         u16 phy_data;
3658
3659         /*
3660          * Prevent stats update while adapter is being reset, or if the pci
3661          * connection is down.
3662          */
3663         if (adapter->link_speed == 0)
3664                 return;
3665         if (pci_channel_offline(pdev))
3666                 return;
3667
3668         adapter->stats.crcerrs += er32(CRCERRS);
3669         adapter->stats.gprc += er32(GPRC);
3670         adapter->stats.gorc += er32(GORCL);
3671         er32(GORCH); /* Clear gorc */
3672         adapter->stats.bprc += er32(BPRC);
3673         adapter->stats.mprc += er32(MPRC);
3674         adapter->stats.roc += er32(ROC);
3675
3676         adapter->stats.mpc += er32(MPC);
3677         if ((hw->phy.type == e1000_phy_82578) ||
3678             (hw->phy.type == e1000_phy_82577)) {
3679                 e1e_rphy(hw, HV_SCC_UPPER, &phy_data);
3680                 if (!e1e_rphy(hw, HV_SCC_LOWER, &phy_data))
3681                         adapter->stats.scc += phy_data;
3682
3683                 e1e_rphy(hw, HV_ECOL_UPPER, &phy_data);
3684                 if (!e1e_rphy(hw, HV_ECOL_LOWER, &phy_data))
3685                         adapter->stats.ecol += phy_data;
3686
3687                 e1e_rphy(hw, HV_MCC_UPPER, &phy_data);
3688                 if (!e1e_rphy(hw, HV_MCC_LOWER, &phy_data))
3689                         adapter->stats.mcc += phy_data;
3690
3691                 e1e_rphy(hw, HV_LATECOL_UPPER, &phy_data);
3692                 if (!e1e_rphy(hw, HV_LATECOL_LOWER, &phy_data))
3693                         adapter->stats.latecol += phy_data;
3694
3695                 e1e_rphy(hw, HV_DC_UPPER, &phy_data);
3696                 if (!e1e_rphy(hw, HV_DC_LOWER, &phy_data))
3697                         adapter->stats.dc += phy_data;
3698         } else {
3699                 adapter->stats.scc += er32(SCC);
3700                 adapter->stats.ecol += er32(ECOL);
3701                 adapter->stats.mcc += er32(MCC);
3702                 adapter->stats.latecol += er32(LATECOL);
3703                 adapter->stats.dc += er32(DC);
3704         }
3705         adapter->stats.xonrxc += er32(XONRXC);
3706         adapter->stats.xontxc += er32(XONTXC);
3707         adapter->stats.xoffrxc += er32(XOFFRXC);
3708         adapter->stats.xofftxc += er32(XOFFTXC);
3709         adapter->stats.gptc += er32(GPTC);
3710         adapter->stats.gotc += er32(GOTCL);
3711         er32(GOTCH); /* Clear gotc */
3712         adapter->stats.rnbc += er32(RNBC);
3713         adapter->stats.ruc += er32(RUC);
3714
3715         adapter->stats.mptc += er32(MPTC);
3716         adapter->stats.bptc += er32(BPTC);
3717
3718         /* used for adaptive IFS */
3719
3720         hw->mac.tx_packet_delta = er32(TPT);
3721         adapter->stats.tpt += hw->mac.tx_packet_delta;
3722         if ((hw->phy.type == e1000_phy_82578) ||
3723             (hw->phy.type == e1000_phy_82577)) {
3724                 e1e_rphy(hw, HV_COLC_UPPER, &phy_data);
3725                 if (!e1e_rphy(hw, HV_COLC_LOWER, &phy_data))
3726                         hw->mac.collision_delta = phy_data;
3727         } else {
3728                 hw->mac.collision_delta = er32(COLC);
3729         }
3730         adapter->stats.colc += hw->mac.collision_delta;
3731
3732         adapter->stats.algnerrc += er32(ALGNERRC);
3733         adapter->stats.rxerrc += er32(RXERRC);
3734         if ((hw->phy.type == e1000_phy_82578) ||
3735             (hw->phy.type == e1000_phy_82577)) {
3736                 e1e_rphy(hw, HV_TNCRS_UPPER, &phy_data);
3737                 if (!e1e_rphy(hw, HV_TNCRS_LOWER, &phy_data))
3738                         adapter->stats.tncrs += phy_data;
3739         } else {
3740                 if ((hw->mac.type != e1000_82574) &&
3741                     (hw->mac.type != e1000_82583))
3742                         adapter->stats.tncrs += er32(TNCRS);
3743         }
3744         adapter->stats.cexterr += er32(CEXTERR);
3745         adapter->stats.tsctc += er32(TSCTC);
3746         adapter->stats.tsctfc += er32(TSCTFC);
3747
3748         /* Fill out the OS statistics structure */
3749         netdev->stats.multicast = adapter->stats.mprc;
3750         netdev->stats.collisions = adapter->stats.colc;
3751
3752         /* Rx Errors */
3753
3754         /*
3755          * RLEC on some newer hardware can be incorrect so build
3756          * our own version based on RUC and ROC
3757          */
3758         netdev->stats.rx_errors = adapter->stats.rxerrc +
3759                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3760                 adapter->stats.ruc + adapter->stats.roc +
3761                 adapter->stats.cexterr;
3762         netdev->stats.rx_length_errors = adapter->stats.ruc +
3763                                               adapter->stats.roc;
3764         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3765         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3766         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3767
3768         /* Tx Errors */
3769         netdev->stats.tx_errors = adapter->stats.ecol +
3770                                        adapter->stats.latecol;
3771         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3772         netdev->stats.tx_window_errors = adapter->stats.latecol;
3773         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3774
3775         /* Tx Dropped needs to be maintained elsewhere */
3776
3777         /* Management Stats */
3778         adapter->stats.mgptc += er32(MGTPTC);
3779         adapter->stats.mgprc += er32(MGTPRC);
3780         adapter->stats.mgpdc += er32(MGTPDC);
3781 }
3782
3783 /**
3784  * e1000_phy_read_status - Update the PHY register status snapshot
3785  * @adapter: board private structure
3786  **/
3787 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3788 {
3789         struct e1000_hw *hw = &adapter->hw;
3790         struct e1000_phy_regs *phy = &adapter->phy_regs;
3791         int ret_val;
3792
3793         if ((er32(STATUS) & E1000_STATUS_LU) &&
3794             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3795                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3796                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3797                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3798                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3799                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3800                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3801                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3802                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3803                 if (ret_val)
3804                         e_warn("Error reading PHY register\n");
3805         } else {
3806                 /*
3807                  * Do not read PHY registers if link is not up
3808                  * Set values to typical power-on defaults
3809                  */
3810                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3811                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3812                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3813                              BMSR_ERCAP);
3814                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3815                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3816                 phy->lpa = 0;
3817                 phy->expansion = EXPANSION_ENABLENPAGE;
3818                 phy->ctrl1000 = ADVERTISE_1000FULL;
3819                 phy->stat1000 = 0;
3820                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3821         }
3822 }
3823
3824 static void e1000_print_link_info(struct e1000_adapter *adapter)
3825 {
3826         struct e1000_hw *hw = &adapter->hw;
3827         u32 ctrl = er32(CTRL);
3828
3829         /* Link status message must follow this format for user tools */
3830         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
3831                "Flow Control: %s\n",
3832                adapter->netdev->name,
3833                adapter->link_speed,
3834                (adapter->link_duplex == FULL_DUPLEX) ?
3835                                 "Full Duplex" : "Half Duplex",
3836                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3837                                 "RX/TX" :
3838                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3839                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3840 }
3841
3842 bool e1000e_has_link(struct e1000_adapter *adapter)
3843 {
3844         struct e1000_hw *hw = &adapter->hw;
3845         bool link_active = 0;
3846         s32 ret_val = 0;
3847
3848         /*
3849          * get_link_status is set on LSC (link status) interrupt or
3850          * Rx sequence error interrupt.  get_link_status will stay
3851          * false until the check_for_link establishes link
3852          * for copper adapters ONLY
3853          */
3854         switch (hw->phy.media_type) {
3855         case e1000_media_type_copper:
3856                 if (hw->mac.get_link_status) {
3857                         ret_val = hw->mac.ops.check_for_link(hw);
3858                         link_active = !hw->mac.get_link_status;
3859                 } else {
3860                         link_active = 1;
3861                 }
3862                 break;
3863         case e1000_media_type_fiber:
3864                 ret_val = hw->mac.ops.check_for_link(hw);
3865                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3866                 break;
3867         case e1000_media_type_internal_serdes:
3868                 ret_val = hw->mac.ops.check_for_link(hw);
3869                 link_active = adapter->hw.mac.serdes_has_link;
3870                 break;
3871         default:
3872         case e1000_media_type_unknown:
3873                 break;
3874         }
3875
3876         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
3877             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3878                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3879                 e_info("Gigabit has been disabled, downgrading speed\n");
3880         }
3881
3882         return link_active;
3883 }
3884
3885 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3886 {
3887         /* make sure the receive unit is started */
3888         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3889             (adapter->flags & FLAG_RX_RESTART_NOW)) {
3890                 struct e1000_hw *hw = &adapter->hw;
3891                 u32 rctl = er32(RCTL);
3892                 ew32(RCTL, rctl | E1000_RCTL_EN);
3893                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3894         }
3895 }
3896
3897 /**
3898  * e1000_watchdog - Timer Call-back
3899  * @data: pointer to adapter cast into an unsigned long
3900  **/
3901 static void e1000_watchdog(unsigned long data)
3902 {
3903         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3904
3905         /* Do the rest outside of interrupt context */
3906         schedule_work(&adapter->watchdog_task);
3907
3908         /* TODO: make this use queue_delayed_work() */
3909 }
3910
3911 static void e1000_watchdog_task(struct work_struct *work)
3912 {
3913         struct e1000_adapter *adapter = container_of(work,
3914                                         struct e1000_adapter, watchdog_task);
3915         struct net_device *netdev = adapter->netdev;
3916         struct e1000_mac_info *mac = &adapter->hw.mac;
3917         struct e1000_phy_info *phy = &adapter->hw.phy;
3918         struct e1000_ring *tx_ring = adapter->tx_ring;
3919         struct e1000_hw *hw = &adapter->hw;
3920         u32 link, tctl;
3921         int tx_pending = 0;
3922
3923         link = e1000e_has_link(adapter);
3924         if ((netif_carrier_ok(netdev)) && link) {
3925                 /* Cancel scheduled suspend requests. */
3926                 pm_runtime_resume(netdev->dev.parent);
3927
3928                 e1000e_enable_receives(adapter);
3929                 goto link_up;
3930         }
3931
3932         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3933             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3934                 e1000_update_mng_vlan(adapter);
3935
3936         if (link) {
3937                 if (!netif_carrier_ok(netdev)) {
3938                         bool txb2b = 1;
3939
3940                         /* Cancel scheduled suspend requests. */
3941                         pm_runtime_resume(netdev->dev.parent);
3942
3943                         /* update snapshot of PHY registers on LSC */
3944                         e1000_phy_read_status(adapter);
3945                         mac->ops.get_link_up_info(&adapter->hw,
3946                                                    &adapter->link_speed,
3947                                                    &adapter->link_duplex);
3948                         e1000_print_link_info(adapter);
3949                         /*
3950                          * On supported PHYs, check for duplex mismatch only
3951                          * if link has autonegotiated at 10/100 half
3952                          */
3953                         if ((hw->phy.type == e1000_phy_igp_3 ||
3954                              hw->phy.type == e1000_phy_bm) &&
3955                             (hw->mac.autoneg == true) &&
3956                             (adapter->link_speed == SPEED_10 ||
3957                              adapter->link_speed == SPEED_100) &&
3958                             (adapter->link_duplex == HALF_DUPLEX)) {
3959                                 u16 autoneg_exp;
3960
3961                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
3962
3963                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
3964                                         e_info("Autonegotiated half duplex but"
3965                                                " link partner cannot autoneg. "
3966                                                " Try forcing full duplex if "
3967                                                "link gets many collisions.\n");
3968                         }
3969
3970                         /* adjust timeout factor according to speed/duplex */
3971                         adapter->tx_timeout_factor = 1;
3972                         switch (adapter->link_speed) {
3973                         case SPEED_10:
3974                                 txb2b = 0;
3975                                 adapter->tx_timeout_factor = 16;
3976                                 break;
3977                         case SPEED_100:
3978                                 txb2b = 0;
3979                                 adapter->tx_timeout_factor = 10;
3980                                 break;
3981                         }
3982
3983                         /*
3984                          * workaround: re-program speed mode bit after
3985                          * link-up event
3986                          */
3987                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
3988                             !txb2b) {
3989                                 u32 tarc0;
3990                                 tarc0 = er32(TARC(0));
3991                                 tarc0 &= ~SPEED_MODE_BIT;
3992                                 ew32(TARC(0), tarc0);
3993                         }
3994
3995                         /*
3996                          * disable TSO for pcie and 10/100 speeds, to avoid
3997                          * some hardware issues
3998                          */
3999                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
4000                                 switch (adapter->link_speed) {
4001                                 case SPEED_10:
4002                                 case SPEED_100:
4003                                         e_info("10/100 speed: disabling TSO\n");
4004                                         netdev->features &= ~NETIF_F_TSO;
4005                                         netdev->features &= ~NETIF_F_TSO6;
4006                                         break;
4007                                 case SPEED_1000:
4008                                         netdev->features |= NETIF_F_TSO;
4009                                         netdev->features |= NETIF_F_TSO6;
4010                                         break;
4011                                 default:
4012                                         /* oops */
4013                                         break;
4014                                 }
4015                         }
4016
4017                         /*
4018                          * enable transmits in the hardware, need to do this
4019                          * after setting TARC(0)
4020                          */
4021                         tctl = er32(TCTL);
4022                         tctl |= E1000_TCTL_EN;
4023                         ew32(TCTL, tctl);
4024
4025                         /*
4026                          * Perform any post-link-up configuration before
4027                          * reporting link up.
4028                          */
4029                         if (phy->ops.cfg_on_link_up)
4030                                 phy->ops.cfg_on_link_up(hw);
4031
4032                         netif_carrier_on(netdev);
4033
4034                         if (!test_bit(__E1000_DOWN, &adapter->state))
4035                                 mod_timer(&adapter->phy_info_timer,
4036                                           round_jiffies(jiffies + 2 * HZ));
4037                 }
4038         } else {
4039                 if (netif_carrier_ok(netdev)) {
4040                         adapter->link_speed = 0;
4041                         adapter->link_duplex = 0;
4042                         /* Link status message must follow this format */
4043                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4044                                adapter->netdev->name);
4045                         netif_carrier_off(netdev);
4046                         if (!test_bit(__E1000_DOWN, &adapter->state))
4047                                 mod_timer(&adapter->phy_info_timer,
4048                                           round_jiffies(jiffies + 2 * HZ));
4049
4050                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4051                                 schedule_work(&adapter->reset_task);
4052                         else
4053                                 pm_schedule_suspend(netdev->dev.parent,
4054                                                         LINK_TIMEOUT);
4055                 }
4056         }
4057
4058 link_up:
4059         e1000e_update_stats(adapter);
4060
4061         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4062         adapter->tpt_old = adapter->stats.tpt;
4063         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4064         adapter->colc_old = adapter->stats.colc;
4065
4066         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4067         adapter->gorc_old = adapter->stats.gorc;
4068         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4069         adapter->gotc_old = adapter->stats.gotc;
4070
4071         e1000e_update_adaptive(&adapter->hw);
4072
4073         if (!netif_carrier_ok(netdev)) {
4074                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
4075                                tx_ring->count);
4076                 if (tx_pending) {
4077                         /*
4078                          * We've lost link, so the controller stops DMA,
4079                          * but we've got queued Tx work that's never going
4080                          * to get done, so reset controller to flush Tx.
4081                          * (Do the reset outside of interrupt context).
4082                          */
4083                         adapter->tx_timeout_count++;
4084                         schedule_work(&adapter->reset_task);
4085                         /* return immediately since reset is imminent */
4086                         return;
4087                 }
4088         }
4089
4090         /* Cause software interrupt to ensure Rx ring is cleaned */
4091         if (adapter->msix_entries)
4092                 ew32(ICS, adapter->rx_ring->ims_val);
4093         else
4094                 ew32(ICS, E1000_ICS_RXDMT0);
4095
4096         /* Force detection of hung controller every watchdog period */
4097         adapter->detect_tx_hung = 1;
4098
4099         /*
4100          * With 82571 controllers, LAA may be overwritten due to controller
4101          * reset from the other port. Set the appropriate LAA in RAR[0]
4102          */
4103         if (e1000e_get_laa_state_82571(hw))
4104                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4105
4106         /* Reset the timer */
4107         if (!test_bit(__E1000_DOWN, &adapter->state))
4108                 mod_timer(&adapter->watchdog_timer,
4109                           round_jiffies(jiffies + 2 * HZ));
4110 }
4111
4112 #define E1000_TX_FLAGS_CSUM             0x00000001
4113 #define E1000_TX_FLAGS_VLAN             0x00000002
4114 #define E1000_TX_FLAGS_TSO              0x00000004
4115 #define E1000_TX_FLAGS_IPV4             0x00000008
4116 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
4117 #define E1000_TX_FLAGS_VLAN_SHIFT       16
4118
4119 static int e1000_tso(struct e1000_adapter *adapter,
4120                      struct sk_buff *skb)
4121 {
4122         struct e1000_ring *tx_ring = adapter->tx_ring;
4123         struct e1000_context_desc *context_desc;
4124         struct e1000_buffer *buffer_info;
4125         unsigned int i;
4126         u32 cmd_length = 0;
4127         u16 ipcse = 0, tucse, mss;
4128         u8 ipcss, ipcso, tucss, tucso, hdr_len;
4129         int err;
4130
4131         if (!skb_is_gso(skb))
4132                 return 0;
4133
4134         if (skb_header_cloned(skb)) {
4135                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4136                 if (err)
4137                         return err;
4138         }
4139
4140         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4141         mss = skb_shinfo(skb)->gso_size;
4142         if (skb->protocol == htons(ETH_P_IP)) {
4143                 struct iphdr *iph = ip_hdr(skb);
4144                 iph->tot_len = 0;
4145                 iph->check = 0;
4146                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4147                                                          0, IPPROTO_TCP, 0);
4148                 cmd_length = E1000_TXD_CMD_IP;
4149                 ipcse = skb_transport_offset(skb) - 1;
4150         } else if (skb_is_gso_v6(skb)) {
4151                 ipv6_hdr(skb)->payload_len = 0;
4152                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4153                                                        &ipv6_hdr(skb)->daddr,
4154                                                        0, IPPROTO_TCP, 0);
4155                 ipcse = 0;
4156         }
4157         ipcss = skb_network_offset(skb);
4158         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4159         tucss = skb_transport_offset(skb);
4160         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4161         tucse = 0;
4162
4163         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4164                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4165
4166         i = tx_ring->next_to_use;
4167         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4168         buffer_info = &tx_ring->buffer_info[i];
4169
4170         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4171         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4172         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4173         context_desc->upper_setup.tcp_fields.tucss = tucss;
4174         context_desc->upper_setup.tcp_fields.tucso = tucso;
4175         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4176         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4177         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4178         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4179
4180         buffer_info->time_stamp = jiffies;
4181         buffer_info->next_to_watch = i;
4182
4183         i++;
4184         if (i == tx_ring->count)
4185                 i = 0;
4186         tx_ring->next_to_use = i;
4187
4188         return 1;
4189 }
4190
4191 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
4192 {
4193         struct e1000_ring *tx_ring = adapter->tx_ring;
4194         struct e1000_context_desc *context_desc;
4195         struct e1000_buffer *buffer_info;
4196         unsigned int i;
4197         u8 css;
4198         u32 cmd_len = E1000_TXD_CMD_DEXT;
4199         __be16 protocol;
4200
4201         if (skb->ip_summed != CHECKSUM_PARTIAL)
4202                 return 0;
4203
4204         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4205                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4206         else
4207                 protocol = skb->protocol;
4208
4209         switch (protocol) {
4210         case cpu_to_be16(ETH_P_IP):
4211                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4212                         cmd_len |= E1000_TXD_CMD_TCP;
4213                 break;
4214         case cpu_to_be16(ETH_P_IPV6):
4215                 /* XXX not handling all IPV6 headers */
4216                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4217                         cmd_len |= E1000_TXD_CMD_TCP;
4218                 break;
4219         default:
4220                 if (unlikely(net_ratelimit()))
4221                         e_warn("checksum_partial proto=%x!\n",
4222                                be16_to_cpu(protocol));
4223                 break;
4224         }
4225
4226         css = skb_transport_offset(skb);
4227
4228         i = tx_ring->next_to_use;
4229         buffer_info = &tx_ring->buffer_info[i];
4230         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4231
4232         context_desc->lower_setup.ip_config = 0;
4233         context_desc->upper_setup.tcp_fields.tucss = css;
4234         context_desc->upper_setup.tcp_fields.tucso =
4235                                 css + skb->csum_offset;
4236         context_desc->upper_setup.tcp_fields.tucse = 0;
4237         context_desc->tcp_seg_setup.data = 0;
4238         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4239
4240         buffer_info->time_stamp = jiffies;
4241         buffer_info->next_to_watch = i;
4242
4243         i++;
4244         if (i == tx_ring->count)
4245                 i = 0;
4246         tx_ring->next_to_use = i;
4247
4248         return 1;
4249 }
4250
4251 #define E1000_MAX_PER_TXD       8192
4252 #define E1000_MAX_TXD_PWR       12
4253
4254 static int e1000_tx_map(struct e1000_adapter *adapter,
4255                         struct sk_buff *skb, unsigned int first,
4256                         unsigned int max_per_txd, unsigned int nr_frags,
4257                         unsigned int mss)
4258 {
4259         struct e1000_ring *tx_ring = adapter->tx_ring;
4260         struct pci_dev *pdev = adapter->pdev;
4261         struct e1000_buffer *buffer_info;
4262         unsigned int len = skb_headlen(skb);
4263         unsigned int offset = 0, size, count = 0, i;
4264         unsigned int f;
4265
4266         i = tx_ring->next_to_use;
4267
4268         while (len) {
4269                 buffer_info = &tx_ring->buffer_info[i];
4270                 size = min(len, max_per_txd);
4271
4272                 buffer_info->length = size;
4273                 buffer_info->time_stamp = jiffies;
4274                 buffer_info->next_to_watch = i;
4275                 buffer_info->dma = dma_map_single(&pdev->dev,
4276                                                   skb->data + offset,
4277                                                   size, DMA_TO_DEVICE);
4278                 buffer_info->mapped_as_page = false;
4279                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4280                         goto dma_error;
4281
4282                 len -= size;
4283                 offset += size;
4284                 count++;
4285
4286                 if (len) {
4287                         i++;
4288                         if (i == tx_ring->count)
4289                                 i = 0;
4290                 }
4291         }
4292
4293         for (f = 0; f < nr_frags; f++) {
4294                 struct skb_frag_struct *frag;
4295
4296                 frag = &skb_shinfo(skb)->frags[f];
4297                 len = frag->size;
4298                 offset = frag->page_offset;
4299
4300                 while (len) {
4301                         i++;
4302                         if (i == tx_ring->count)
4303                                 i = 0;
4304
4305                         buffer_info = &tx_ring->buffer_info[i];
4306                         size = min(len, max_per_txd);
4307
4308                         buffer_info->length = size;
4309                         buffer_info->time_stamp = jiffies;
4310                         buffer_info->next_to_watch = i;
4311                         buffer_info->dma = dma_map_page(&pdev->dev, frag->page,
4312                                                         offset, size,
4313                                                         DMA_TO_DEVICE);
4314                         buffer_info->mapped_as_page = true;
4315                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4316                                 goto dma_error;
4317
4318                         len -= size;
4319                         offset += size;
4320                         count++;
4321                 }
4322         }
4323
4324         tx_ring->buffer_info[i].skb = skb;
4325         tx_ring->buffer_info[first].next_to_watch = i;
4326
4327         return count;
4328
4329 dma_error:
4330         dev_err(&pdev->dev, "TX DMA map failed\n");
4331         buffer_info->dma = 0;
4332         if (count)
4333                 count--;
4334
4335         while (count--) {
4336                 if (i==0)
4337                         i += tx_ring->count;
4338                 i--;
4339                 buffer_info = &tx_ring->buffer_info[i];
4340                 e1000_put_txbuf(adapter, buffer_info);;
4341         }
4342
4343         return 0;
4344 }
4345
4346 static void e1000_tx_queue(struct e1000_adapter *adapter,
4347                            int tx_flags, int count)
4348 {
4349         struct e1000_ring *tx_ring = adapter->tx_ring;
4350         struct e1000_tx_desc *tx_desc = NULL;
4351         struct e1000_buffer *buffer_info;
4352         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4353         unsigned int i;
4354
4355         if (tx_flags & E1000_TX_FLAGS_TSO) {
4356                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4357                              E1000_TXD_CMD_TSE;
4358                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4359
4360                 if (tx_flags & E1000_TX_FLAGS_IPV4)
4361                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4362         }
4363
4364         if (tx_flags & E1000_TX_FLAGS_CSUM) {
4365                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4366                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4367         }
4368
4369         if (tx_flags & E1000_TX_FLAGS_VLAN) {
4370                 txd_lower |= E1000_TXD_CMD_VLE;
4371                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4372         }
4373
4374         i = tx_ring->next_to_use;
4375
4376         while (count--) {
4377                 buffer_info = &tx_ring->buffer_info[i];
4378                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4379                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4380                 tx_desc->lower.data =
4381                         cpu_to_le32(txd_lower | buffer_info->length);
4382                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4383
4384                 i++;
4385                 if (i == tx_ring->count)
4386                         i = 0;
4387         }
4388
4389         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4390
4391         /*
4392          * Force memory writes to complete before letting h/w
4393          * know there are new descriptors to fetch.  (Only
4394          * applicable for weak-ordered memory model archs,
4395          * such as IA-64).
4396          */
4397         wmb();
4398
4399         tx_ring->next_to_use = i;
4400         writel(i, adapter->hw.hw_addr + tx_ring->tail);
4401         /*
4402          * we need this if more than one processor can write to our tail
4403          * at a time, it synchronizes IO on IA64/Altix systems
4404          */
4405         mmiowb();
4406 }
4407
4408 #define MINIMUM_DHCP_PACKET_SIZE 282
4409 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4410                                     struct sk_buff *skb)
4411 {
4412         struct e1000_hw *hw =  &adapter->hw;
4413         u16 length, offset;
4414
4415         if (vlan_tx_tag_present(skb)) {
4416                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4417                     (adapter->hw.mng_cookie.status &
4418                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4419                         return 0;
4420         }
4421
4422         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4423                 return 0;
4424
4425         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4426                 return 0;
4427
4428         {
4429                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4430                 struct udphdr *udp;
4431
4432                 if (ip->protocol != IPPROTO_UDP)
4433                         return 0;
4434
4435                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4436                 if (ntohs(udp->dest) != 67)
4437                         return 0;
4438
4439                 offset = (u8 *)udp + 8 - skb->data;
4440                 length = skb->len - offset;
4441                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4442         }
4443
4444         return 0;
4445 }
4446
4447 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4448 {
4449         struct e1000_adapter *adapter = netdev_priv(netdev);
4450
4451         netif_stop_queue(netdev);
4452         /*
4453          * Herbert's original patch had:
4454          *  smp_mb__after_netif_stop_queue();
4455          * but since that doesn't exist yet, just open code it.
4456          */
4457         smp_mb();
4458
4459         /*
4460          * We need to check again in a case another CPU has just
4461          * made room available.
4462          */
4463         if (e1000_desc_unused(adapter->tx_ring) < size)
4464                 return -EBUSY;
4465
4466         /* A reprieve! */
4467         netif_start_queue(netdev);
4468         ++adapter->restart_queue;
4469         return 0;
4470 }
4471
4472 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4473 {
4474         struct e1000_adapter *adapter = netdev_priv(netdev);
4475
4476         if (e1000_desc_unused(adapter->tx_ring) >= size)
4477                 return 0;
4478         return __e1000_maybe_stop_tx(netdev, size);
4479 }
4480
4481 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4482 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4483                                     struct net_device *netdev)
4484 {
4485         struct e1000_adapter *adapter = netdev_priv(netdev);
4486         struct e1000_ring *tx_ring = adapter->tx_ring;
4487         unsigned int first;
4488         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4489         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4490         unsigned int tx_flags = 0;
4491         unsigned int len = skb_headlen(skb);
4492         unsigned int nr_frags;
4493         unsigned int mss;
4494         int count = 0;
4495         int tso;
4496         unsigned int f;
4497
4498         if (test_bit(__E1000_DOWN, &adapter->state)) {
4499                 dev_kfree_skb_any(skb);
4500                 return NETDEV_TX_OK;
4501         }
4502
4503         if (skb->len <= 0) {
4504                 dev_kfree_skb_any(skb);
4505                 return NETDEV_TX_OK;
4506         }
4507
4508         mss = skb_shinfo(skb)->gso_size;
4509         /*
4510          * The controller does a simple calculation to
4511          * make sure there is enough room in the FIFO before
4512          * initiating the DMA for each buffer.  The calc is:
4513          * 4 = ceil(buffer len/mss).  To make sure we don't
4514          * overrun the FIFO, adjust the max buffer len if mss
4515          * drops.
4516          */
4517         if (mss) {
4518                 u8 hdr_len;
4519                 max_per_txd = min(mss << 2, max_per_txd);
4520                 max_txd_pwr = fls(max_per_txd) - 1;
4521
4522                 /*
4523                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4524                  * points to just header, pull a few bytes of payload from
4525                  * frags into skb->data
4526                  */
4527                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4528                 /*
4529                  * we do this workaround for ES2LAN, but it is un-necessary,
4530                  * avoiding it could save a lot of cycles
4531                  */
4532                 if (skb->data_len && (hdr_len == len)) {
4533                         unsigned int pull_size;
4534
4535                         pull_size = min((unsigned int)4, skb->data_len);
4536                         if (!__pskb_pull_tail(skb, pull_size)) {
4537                                 e_err("__pskb_pull_tail failed.\n");
4538                                 dev_kfree_skb_any(skb);
4539                                 return NETDEV_TX_OK;
4540                         }
4541                         len = skb_headlen(skb);
4542                 }
4543         }
4544
4545         /* reserve a descriptor for the offload context */
4546         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4547                 count++;
4548         count++;
4549
4550         count += TXD_USE_COUNT(len, max_txd_pwr);
4551
4552         nr_frags = skb_shinfo(skb)->nr_frags;
4553         for (f = 0; f < nr_frags; f++)
4554                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4555                                        max_txd_pwr);
4556
4557         if (adapter->hw.mac.tx_pkt_filtering)
4558                 e1000_transfer_dhcp_info(adapter, skb);
4559
4560         /*
4561          * need: count + 2 desc gap to keep tail from touching
4562          * head, otherwise try next time
4563          */
4564         if (e1000_maybe_stop_tx(netdev, count + 2))
4565                 return NETDEV_TX_BUSY;
4566
4567         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4568                 tx_flags |= E1000_TX_FLAGS_VLAN;
4569                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4570         }
4571
4572         first = tx_ring->next_to_use;
4573
4574         tso = e1000_tso(adapter, skb);
4575         if (tso < 0) {
4576                 dev_kfree_skb_any(skb);
4577                 return NETDEV_TX_OK;
4578         }
4579
4580         if (tso)
4581                 tx_flags |= E1000_TX_FLAGS_TSO;
4582         else if (e1000_tx_csum(adapter, skb))
4583                 tx_flags |= E1000_TX_FLAGS_CSUM;
4584
4585         /*
4586          * Old method was to assume IPv4 packet by default if TSO was enabled.
4587          * 82571 hardware supports TSO capabilities for IPv6 as well...
4588          * no longer assume, we must.
4589          */
4590         if (skb->protocol == htons(ETH_P_IP))
4591                 tx_flags |= E1000_TX_FLAGS_IPV4;
4592
4593         /* if count is 0 then mapping error has occured */
4594         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4595         if (count) {
4596                 e1000_tx_queue(adapter, tx_flags, count);
4597                 /* Make sure there is space in the ring for the next send. */
4598                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4599
4600         } else {
4601                 dev_kfree_skb_any(skb);
4602                 tx_ring->buffer_info[first].time_stamp = 0;
4603                 tx_ring->next_to_use = first;
4604         }
4605
4606         return NETDEV_TX_OK;
4607 }
4608
4609 /**
4610  * e1000_tx_timeout - Respond to a Tx Hang
4611  * @netdev: network interface device structure
4612  **/
4613 static void e1000_tx_timeout(struct net_device *netdev)
4614 {
4615         struct e1000_adapter *adapter = netdev_priv(netdev);
4616
4617         /* Do the reset outside of interrupt context */
4618         adapter->tx_timeout_count++;
4619         schedule_work(&adapter->reset_task);
4620 }
4621
4622 static void e1000_reset_task(struct work_struct *work)
4623 {
4624         struct e1000_adapter *adapter;
4625         adapter = container_of(work, struct e1000_adapter, reset_task);
4626
4627         e1000e_dump(adapter);
4628         e_err("Reset adapter\n");
4629         e1000e_reinit_locked(adapter);
4630 }
4631
4632 /**
4633  * e1000_get_stats - Get System Network Statistics
4634  * @netdev: network interface device structure
4635  *
4636  * Returns the address of the device statistics structure.
4637  * The statistics are actually updated from the timer callback.
4638  **/
4639 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4640 {
4641         /* only return the current stats */
4642         return &netdev->stats;
4643 }
4644
4645 /**
4646  * e1000_change_mtu - Change the Maximum Transfer Unit
4647  * @netdev: network interface device structure
4648  * @new_mtu: new value for maximum frame size
4649  *
4650  * Returns 0 on success, negative on failure
4651  **/
4652 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4653 {
4654         struct e1000_adapter *adapter = netdev_priv(netdev);
4655         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4656
4657         /* Jumbo frame support */
4658         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
4659             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4660                 e_err("Jumbo Frames not supported.\n");
4661                 return -EINVAL;
4662         }
4663
4664         /* Supported frame sizes */
4665         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4666             (max_frame > adapter->max_hw_frame_size)) {
4667                 e_err("Unsupported MTU setting\n");
4668                 return -EINVAL;
4669         }
4670
4671         /* 82573 Errata 17 */
4672         if (((adapter->hw.mac.type == e1000_82573) ||
4673              (adapter->hw.mac.type == e1000_82574)) &&
4674             (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
4675                 adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
4676                 e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
4677         }
4678
4679         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4680                 msleep(1);
4681         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
4682         adapter->max_frame_size = max_frame;
4683         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4684         netdev->mtu = new_mtu;
4685         if (netif_running(netdev))
4686                 e1000e_down(adapter);
4687
4688         /*
4689          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4690          * means we reserve 2 more, this pushes us to allocate from the next
4691          * larger slab size.
4692          * i.e. RXBUFFER_2048 --> size-4096 slab
4693          * However with the new *_jumbo_rx* routines, jumbo receives will use
4694          * fragmented skbs
4695          */
4696
4697         if (max_frame <= 2048)
4698                 adapter->rx_buffer_len = 2048;
4699         else
4700                 adapter->rx_buffer_len = 4096;
4701
4702         /* adjust allocation if LPE protects us, and we aren't using SBP */
4703         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4704              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4705                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4706                                          + ETH_FCS_LEN;
4707
4708         if (netif_running(netdev))
4709                 e1000e_up(adapter);
4710         else
4711                 e1000e_reset(adapter);
4712
4713         clear_bit(__E1000_RESETTING, &adapter->state);
4714
4715         return 0;
4716 }
4717
4718 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4719                            int cmd)
4720 {
4721         struct e1000_adapter *adapter = netdev_priv(netdev);
4722         struct mii_ioctl_data *data = if_mii(ifr);
4723
4724         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4725                 return -EOPNOTSUPP;
4726
4727         switch (cmd) {
4728         case SIOCGMIIPHY:
4729                 data->phy_id = adapter->hw.phy.addr;
4730                 break;
4731         case SIOCGMIIREG:
4732                 e1000_phy_read_status(adapter);
4733
4734                 switch (data->reg_num & 0x1F) {
4735                 case MII_BMCR:
4736                         data->val_out = adapter->phy_regs.bmcr;
4737                         break;
4738                 case MII_BMSR:
4739                         data->val_out = adapter->phy_regs.bmsr;
4740                         break;
4741                 case MII_PHYSID1:
4742                         data->val_out = (adapter->hw.phy.id >> 16);
4743                         break;
4744                 case MII_PHYSID2:
4745                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
4746                         break;
4747                 case MII_ADVERTISE:
4748                         data->val_out = adapter->phy_regs.advertise;
4749                         break;
4750                 case MII_LPA:
4751                         data->val_out = adapter->phy_regs.lpa;
4752                         break;
4753                 case MII_EXPANSION:
4754                         data->val_out = adapter->phy_regs.expansion;
4755                         break;
4756                 case MII_CTRL1000:
4757                         data->val_out = adapter->phy_regs.ctrl1000;
4758                         break;
4759                 case MII_STAT1000:
4760                         data->val_out = adapter->phy_regs.stat1000;
4761                         break;
4762                 case MII_ESTATUS:
4763                         data->val_out = adapter->phy_regs.estatus;
4764                         break;
4765                 default:
4766                         return -EIO;
4767                 }
4768                 break;
4769         case SIOCSMIIREG:
4770         default:
4771                 return -EOPNOTSUPP;
4772         }
4773         return 0;
4774 }
4775
4776 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4777 {
4778         switch (cmd) {
4779         case SIOCGMIIPHY:
4780         case SIOCGMIIREG:
4781         case SIOCSMIIREG:
4782                 return e1000_mii_ioctl(netdev, ifr, cmd);
4783         default:
4784                 return -EOPNOTSUPP;
4785         }
4786 }
4787
4788 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
4789 {
4790         struct e1000_hw *hw = &adapter->hw;
4791         u32 i, mac_reg;
4792         u16 phy_reg;
4793         int retval = 0;
4794
4795         /* copy MAC RARs to PHY RARs */
4796         for (i = 0; i < adapter->hw.mac.rar_entry_count; i++) {
4797                 mac_reg = er32(RAL(i));
4798                 e1e_wphy(hw, BM_RAR_L(i), (u16)(mac_reg & 0xFFFF));
4799                 e1e_wphy(hw, BM_RAR_M(i), (u16)((mac_reg >> 16) & 0xFFFF));
4800                 mac_reg = er32(RAH(i));
4801                 e1e_wphy(hw, BM_RAR_H(i), (u16)(mac_reg & 0xFFFF));
4802                 e1e_wphy(hw, BM_RAR_CTRL(i), (u16)((mac_reg >> 16) & 0xFFFF));
4803         }
4804
4805         /* copy MAC MTA to PHY MTA */
4806         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
4807                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
4808                 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
4809                 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
4810         }
4811
4812         /* configure PHY Rx Control register */
4813         e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
4814         mac_reg = er32(RCTL);
4815         if (mac_reg & E1000_RCTL_UPE)
4816                 phy_reg |= BM_RCTL_UPE;
4817         if (mac_reg & E1000_RCTL_MPE)
4818                 phy_reg |= BM_RCTL_MPE;
4819         phy_reg &= ~(BM_RCTL_MO_MASK);
4820         if (mac_reg & E1000_RCTL_MO_3)
4821                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
4822                                 << BM_RCTL_MO_SHIFT);
4823         if (mac_reg & E1000_RCTL_BAM)
4824                 phy_reg |= BM_RCTL_BAM;
4825         if (mac_reg & E1000_RCTL_PMCF)
4826                 phy_reg |= BM_RCTL_PMCF;
4827         mac_reg = er32(CTRL);
4828         if (mac_reg & E1000_CTRL_RFCE)
4829                 phy_reg |= BM_RCTL_RFCE;
4830         e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
4831
4832         /* enable PHY wakeup in MAC register */
4833         ew32(WUFC, wufc);
4834         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
4835
4836         /* configure and enable PHY wakeup in PHY registers */
4837         e1e_wphy(&adapter->hw, BM_WUFC, wufc);
4838         e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
4839
4840         /* activate PHY wakeup */
4841         retval = hw->phy.ops.acquire(hw);
4842         if (retval) {
4843                 e_err("Could not acquire PHY\n");
4844                 return retval;
4845         }
4846         e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4847                                  (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
4848         retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
4849         if (retval) {
4850                 e_err("Could not read PHY page 769\n");
4851                 goto out;
4852         }
4853         phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
4854         retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
4855         if (retval)
4856                 e_err("Could not set PHY Host Wakeup bit\n");
4857 out:
4858         hw->phy.ops.release(hw);
4859
4860         return retval;
4861 }
4862
4863 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
4864                             bool runtime)
4865 {
4866         struct net_device *netdev = pci_get_drvdata(pdev);
4867         struct e1000_adapter *adapter = netdev_priv(netdev);
4868         struct e1000_hw *hw = &adapter->hw;
4869         u32 ctrl, ctrl_ext, rctl, status;
4870         /* Runtime suspend should only enable wakeup for link changes */
4871         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
4872         int retval = 0;
4873
4874         netif_device_detach(netdev);
4875
4876         if (netif_running(netdev)) {
4877                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4878                 e1000e_down(adapter);
4879                 e1000_free_irq(adapter);
4880         }
4881         e1000e_reset_interrupt_capability(adapter);
4882
4883         retval = pci_save_state(pdev);
4884         if (retval)
4885                 return retval;
4886
4887         status = er32(STATUS);
4888         if (status & E1000_STATUS_LU)
4889                 wufc &= ~E1000_WUFC_LNKC;
4890
4891         if (wufc) {
4892                 e1000_setup_rctl(adapter);
4893                 e1000_set_multi(netdev);
4894
4895                 /* turn on all-multi mode if wake on multicast is enabled */
4896                 if (wufc & E1000_WUFC_MC) {
4897                         rctl = er32(RCTL);
4898                         rctl |= E1000_RCTL_MPE;
4899                         ew32(RCTL, rctl);
4900                 }
4901
4902                 ctrl = er32(CTRL);
4903                 /* advertise wake from D3Cold */
4904                 #define E1000_CTRL_ADVD3WUC 0x00100000
4905                 /* phy power management enable */
4906                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4907                 ctrl |= E1000_CTRL_ADVD3WUC;
4908                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
4909                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
4910                 ew32(CTRL, ctrl);
4911
4912                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
4913                     adapter->hw.phy.media_type ==
4914                     e1000_media_type_internal_serdes) {
4915                         /* keep the laser running in D3 */
4916                         ctrl_ext = er32(CTRL_EXT);
4917                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
4918                         ew32(CTRL_EXT, ctrl_ext);
4919                 }
4920
4921                 if (adapter->flags & FLAG_IS_ICH)
4922                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
4923
4924                 /* Allow time for pending master requests to run */
4925                 e1000e_disable_pcie_master(&adapter->hw);
4926
4927                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
4928                         /* enable wakeup by the PHY */
4929                         retval = e1000_init_phy_wakeup(adapter, wufc);
4930                         if (retval)
4931                                 return retval;
4932                 } else {
4933                         /* enable wakeup by the MAC */
4934                         ew32(WUFC, wufc);
4935                         ew32(WUC, E1000_WUC_PME_EN);
4936                 }
4937         } else {
4938                 ew32(WUC, 0);
4939                 ew32(WUFC, 0);
4940         }
4941
4942         *enable_wake = !!wufc;
4943
4944         /* make sure adapter isn't asleep if manageability is enabled */
4945         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
4946             (hw->mac.ops.check_mng_mode(hw)))
4947                 *enable_wake = true;
4948
4949         if (adapter->hw.phy.type == e1000_phy_igp_3)
4950                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
4951
4952         /*
4953          * Release control of h/w to f/w.  If f/w is AMT enabled, this
4954          * would have already happened in close and is redundant.
4955          */
4956         e1000_release_hw_control(adapter);
4957
4958         pci_disable_device(pdev);
4959
4960         return 0;
4961 }
4962
4963 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
4964 {
4965         if (sleep && wake) {
4966                 pci_prepare_to_sleep(pdev);
4967                 return;
4968         }
4969
4970         pci_wake_from_d3(pdev, wake);
4971         pci_set_power_state(pdev, PCI_D3hot);
4972 }
4973
4974 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
4975                                     bool wake)
4976 {
4977         struct net_device *netdev = pci_get_drvdata(pdev);
4978         struct e1000_adapter *adapter = netdev_priv(netdev);
4979
4980         /*
4981          * The pci-e switch on some quad port adapters will report a
4982          * correctable error when the MAC transitions from D0 to D3.  To
4983          * prevent this we need to mask off the correctable errors on the
4984          * downstream port of the pci-e switch.
4985          */
4986         if (adapter->flags & FLAG_IS_QUAD_PORT) {
4987                 struct pci_dev *us_dev = pdev->bus->self;
4988                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
4989                 u16 devctl;
4990
4991                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
4992                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
4993                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
4994
4995                 e1000_power_off(pdev, sleep, wake);
4996
4997                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
4998         } else {
4999                 e1000_power_off(pdev, sleep, wake);
5000         }
5001 }
5002
5003 #ifdef CONFIG_PCIEASPM
5004 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5005 {
5006         pci_disable_link_state(pdev, state);
5007 }
5008 #else
5009 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5010 {
5011         int pos;
5012         u16 reg16;
5013
5014         /*
5015          * Both device and parent should have the same ASPM setting.
5016          * Disable ASPM in downstream component first and then upstream.
5017          */
5018         pos = pci_pcie_cap(pdev);
5019         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5020         reg16 &= ~state;
5021         pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5022
5023         if (!pdev->bus->self)
5024                 return;
5025
5026         pos = pci_pcie_cap(pdev->bus->self);
5027         pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5028         reg16 &= ~state;
5029         pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5030 }
5031 #endif
5032 void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5033 {
5034         dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5035                  (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5036                  (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5037
5038         __e1000e_disable_aspm(pdev, state);
5039 }
5040
5041 #ifdef CONFIG_PM_OPS
5042 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5043 {
5044         return !!adapter->tx_ring->buffer_info;
5045 }
5046
5047 static int __e1000_resume(struct pci_dev *pdev)
5048 {
5049         struct net_device *netdev = pci_get_drvdata(pdev);
5050         struct e1000_adapter *adapter = netdev_priv(netdev);
5051         struct e1000_hw *hw = &adapter->hw;
5052         u32 err;
5053
5054         pci_set_power_state(pdev, PCI_D0);
5055         pci_restore_state(pdev);
5056         pci_save_state(pdev);
5057         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5058                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5059
5060         e1000e_set_interrupt_capability(adapter);
5061         if (netif_running(netdev)) {
5062                 err = e1000_request_irq(adapter);
5063                 if (err)
5064                         return err;
5065         }
5066
5067         e1000e_power_up_phy(adapter);
5068
5069         /* report the system wakeup cause from S3/S4 */
5070         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5071                 u16 phy_data;
5072
5073                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5074                 if (phy_data) {
5075                         e_info("PHY Wakeup cause - %s\n",
5076                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5077                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5078                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5079                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5080                                 phy_data & E1000_WUS_LNKC ? "Link Status "
5081                                 " Change" : "other");
5082                 }
5083                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5084         } else {
5085                 u32 wus = er32(WUS);
5086                 if (wus) {
5087                         e_info("MAC Wakeup cause - %s\n",
5088                                 wus & E1000_WUS_EX ? "Unicast Packet" :
5089                                 wus & E1000_WUS_MC ? "Multicast Packet" :
5090                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
5091                                 wus & E1000_WUS_MAG ? "Magic Packet" :
5092                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
5093                                 "other");
5094                 }
5095                 ew32(WUS, ~0);
5096         }
5097
5098         e1000e_reset(adapter);
5099
5100         e1000_init_manageability(adapter);
5101
5102         if (netif_running(netdev))
5103                 e1000e_up(adapter);
5104
5105         netif_device_attach(netdev);
5106
5107         /*
5108          * If the controller has AMT, do not set DRV_LOAD until the interface
5109          * is up.  For all other cases, let the f/w know that the h/w is now
5110          * under the control of the driver.
5111          */
5112         if (!(adapter->flags & FLAG_HAS_AMT))
5113                 e1000_get_hw_control(adapter);
5114
5115         return 0;
5116 }
5117
5118 #ifdef CONFIG_PM_SLEEP
5119 static int e1000_suspend(struct device *dev)
5120 {
5121         struct pci_dev *pdev = to_pci_dev(dev);
5122         int retval;
5123         bool wake;
5124
5125         retval = __e1000_shutdown(pdev, &wake, false);
5126         if (!retval)
5127                 e1000_complete_shutdown(pdev, true, wake);
5128
5129         return retval;
5130 }
5131
5132 static int e1000_resume(struct device *dev)
5133 {
5134         struct pci_dev *pdev = to_pci_dev(dev);
5135         struct net_device *netdev = pci_get_drvdata(pdev);
5136         struct e1000_adapter *adapter = netdev_priv(netdev);
5137
5138         if (e1000e_pm_ready(adapter))
5139                 adapter->idle_check = true;
5140
5141         return __e1000_resume(pdev);
5142 }
5143 #endif /* CONFIG_PM_SLEEP */
5144
5145 #ifdef CONFIG_PM_RUNTIME
5146 static int e1000_runtime_suspend(struct device *dev)
5147 {
5148         struct pci_dev *pdev = to_pci_dev(dev);
5149         struct net_device *netdev = pci_get_drvdata(pdev);
5150         struct e1000_adapter *adapter = netdev_priv(netdev);
5151
5152         if (e1000e_pm_ready(adapter)) {
5153                 bool wake;
5154
5155                 __e1000_shutdown(pdev, &wake, true);
5156         }
5157
5158         return 0;
5159 }
5160
5161 static int e1000_idle(struct device *dev)
5162 {
5163         struct pci_dev *pdev = to_pci_dev(dev);
5164         struct net_device *netdev = pci_get_drvdata(pdev);
5165         struct e1000_adapter *adapter = netdev_priv(netdev);
5166
5167         if (!e1000e_pm_ready(adapter))
5168                 return 0;
5169
5170         if (adapter->idle_check) {
5171                 adapter->idle_check = false;
5172                 if (!e1000e_has_link(adapter))
5173                         pm_schedule_suspend(dev, MSEC_PER_SEC);
5174         }
5175
5176         return -EBUSY;
5177 }
5178
5179 static int e1000_runtime_resume(struct device *dev)
5180 {
5181         struct pci_dev *pdev = to_pci_dev(dev);
5182         struct net_device *netdev = pci_get_drvdata(pdev);
5183         struct e1000_adapter *adapter = netdev_priv(netdev);
5184
5185         if (!e1000e_pm_ready(adapter))
5186                 return 0;
5187
5188         adapter->idle_check = !dev->power.runtime_auto;
5189         return __e1000_resume(pdev);
5190 }
5191 #endif /* CONFIG_PM_RUNTIME */
5192 #endif /* CONFIG_PM_OPS */
5193
5194 static void e1000_shutdown(struct pci_dev *pdev)
5195 {
5196         bool wake = false;
5197
5198         __e1000_shutdown(pdev, &wake, false);
5199
5200         if (system_state == SYSTEM_POWER_OFF)
5201                 e1000_complete_shutdown(pdev, false, wake);
5202 }
5203
5204 #ifdef CONFIG_NET_POLL_CONTROLLER
5205 /*
5206  * Polling 'interrupt' - used by things like netconsole to send skbs
5207  * without having to re-enable interrupts. It's not called while
5208  * the interrupt routine is executing.
5209  */
5210 static void e1000_netpoll(struct net_device *netdev)
5211 {
5212         struct e1000_adapter *adapter = netdev_priv(netdev);
5213
5214         disable_irq(adapter->pdev->irq);
5215         e1000_intr(adapter->pdev->irq, netdev);
5216
5217         enable_irq(adapter->pdev->irq);
5218 }
5219 #endif
5220
5221 /**
5222  * e1000_io_error_detected - called when PCI error is detected
5223  * @pdev: Pointer to PCI device
5224  * @state: The current pci connection state
5225  *
5226  * This function is called after a PCI bus error affecting
5227  * this device has been detected.
5228  */
5229 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5230                                                 pci_channel_state_t state)
5231 {
5232         struct net_device *netdev = pci_get_drvdata(pdev);
5233         struct e1000_adapter *adapter = netdev_priv(netdev);
5234
5235         netif_device_detach(netdev);
5236
5237         if (state == pci_channel_io_perm_failure)
5238                 return PCI_ERS_RESULT_DISCONNECT;
5239
5240         if (netif_running(netdev))
5241                 e1000e_down(adapter);
5242         pci_disable_device(pdev);
5243
5244         /* Request a slot slot reset. */
5245         return PCI_ERS_RESULT_NEED_RESET;
5246 }
5247
5248 /**
5249  * e1000_io_slot_reset - called after the pci bus has been reset.
5250  * @pdev: Pointer to PCI device
5251  *
5252  * Restart the card from scratch, as if from a cold-boot. Implementation
5253  * resembles the first-half of the e1000_resume routine.
5254  */
5255 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5256 {
5257         struct net_device *netdev = pci_get_drvdata(pdev);
5258         struct e1000_adapter *adapter = netdev_priv(netdev);
5259         struct e1000_hw *hw = &adapter->hw;
5260         int err;
5261         pci_ers_result_t result;
5262
5263         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5264                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5265         err = pci_enable_device_mem(pdev);
5266         if (err) {
5267                 dev_err(&pdev->dev,
5268                         "Cannot re-enable PCI device after reset.\n");
5269                 result = PCI_ERS_RESULT_DISCONNECT;
5270         } else {
5271                 pci_set_master(pdev);
5272                 pdev->state_saved = true;
5273                 pci_restore_state(pdev);
5274
5275                 pci_enable_wake(pdev, PCI_D3hot, 0);
5276                 pci_enable_wake(pdev, PCI_D3cold, 0);
5277
5278                 e1000e_reset(adapter);
5279                 ew32(WUS, ~0);
5280                 result = PCI_ERS_RESULT_RECOVERED;
5281         }
5282
5283         pci_cleanup_aer_uncorrect_error_status(pdev);
5284
5285         return result;
5286 }
5287
5288 /**
5289  * e1000_io_resume - called when traffic can start flowing again.
5290  * @pdev: Pointer to PCI device
5291  *
5292  * This callback is called when the error recovery driver tells us that
5293  * its OK to resume normal operation. Implementation resembles the
5294  * second-half of the e1000_resume routine.
5295  */
5296 static void e1000_io_resume(struct pci_dev *pdev)
5297 {
5298         struct net_device *netdev = pci_get_drvdata(pdev);
5299         struct e1000_adapter *adapter = netdev_priv(netdev);
5300
5301         e1000_init_manageability(adapter);
5302
5303         if (netif_running(netdev)) {
5304                 if (e1000e_up(adapter)) {
5305                         dev_err(&pdev->dev,
5306                                 "can't bring device back up after reset\n");
5307                         return;
5308                 }
5309         }
5310
5311         netif_device_attach(netdev);
5312
5313         /*
5314          * If the controller has AMT, do not set DRV_LOAD until the interface
5315          * is up.  For all other cases, let the f/w know that the h/w is now
5316          * under the control of the driver.
5317          */
5318         if (!(adapter->flags & FLAG_HAS_AMT))
5319                 e1000_get_hw_control(adapter);
5320
5321 }
5322
5323 static void e1000_print_device_info(struct e1000_adapter *adapter)
5324 {
5325         struct e1000_hw *hw = &adapter->hw;
5326         struct net_device *netdev = adapter->netdev;
5327         u32 pba_num;
5328
5329         /* print bus type/speed/width info */
5330         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
5331                /* bus width */
5332                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5333                 "Width x1"),
5334                /* MAC address */
5335                netdev->dev_addr);
5336         e_info("Intel(R) PRO/%s Network Connection\n",
5337                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5338         e1000e_read_pba_num(hw, &pba_num);
5339         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
5340                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
5341 }
5342
5343 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5344 {
5345         struct e1000_hw *hw = &adapter->hw;
5346         int ret_val;
5347         u16 buf = 0;
5348
5349         if (hw->mac.type != e1000_82573)
5350                 return;
5351
5352         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
5353         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
5354                 /* Deep Smart Power Down (DSPD) */
5355                 dev_warn(&adapter->pdev->dev,
5356                          "Warning: detected DSPD enabled in EEPROM\n");
5357         }
5358 }
5359
5360 static const struct net_device_ops e1000e_netdev_ops = {
5361         .ndo_open               = e1000_open,
5362         .ndo_stop               = e1000_close,
5363         .ndo_start_xmit         = e1000_xmit_frame,
5364         .ndo_get_stats          = e1000_get_stats,
5365         .ndo_set_multicast_list = e1000_set_multi,
5366         .ndo_set_mac_address    = e1000_set_mac,
5367         .ndo_change_mtu         = e1000_change_mtu,
5368         .ndo_do_ioctl           = e1000_ioctl,
5369         .ndo_tx_timeout         = e1000_tx_timeout,
5370         .ndo_validate_addr      = eth_validate_addr,
5371
5372         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
5373         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
5374         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
5375 #ifdef CONFIG_NET_POLL_CONTROLLER
5376         .ndo_poll_controller    = e1000_netpoll,
5377 #endif
5378 };
5379
5380 /**
5381  * e1000_probe - Device Initialization Routine
5382  * @pdev: PCI device information struct
5383  * @ent: entry in e1000_pci_tbl
5384  *
5385  * Returns 0 on success, negative on failure
5386  *
5387  * e1000_probe initializes an adapter identified by a pci_dev structure.
5388  * The OS initialization, configuring of the adapter private structure,
5389  * and a hardware reset occur.
5390  **/
5391 static int __devinit e1000_probe(struct pci_dev *pdev,
5392                                  const struct pci_device_id *ent)
5393 {
5394         struct net_device *netdev;
5395         struct e1000_adapter *adapter;
5396         struct e1000_hw *hw;
5397         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
5398         resource_size_t mmio_start, mmio_len;
5399         resource_size_t flash_start, flash_len;
5400
5401         static int cards_found;
5402         int i, err, pci_using_dac;
5403         u16 eeprom_data = 0;
5404         u16 eeprom_apme_mask = E1000_EEPROM_APME;
5405
5406         if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
5407                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5408
5409         err = pci_enable_device_mem(pdev);
5410         if (err)
5411                 return err;
5412
5413         pci_using_dac = 0;
5414         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
5415         if (!err) {
5416                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
5417                 if (!err)
5418                         pci_using_dac = 1;
5419         } else {
5420                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
5421                 if (err) {
5422                         err = dma_set_coherent_mask(&pdev->dev,
5423                                                     DMA_BIT_MASK(32));
5424                         if (err) {
5425                                 dev_err(&pdev->dev, "No usable DMA "
5426                                         "configuration, aborting\n");
5427                                 goto err_dma;
5428                         }
5429                 }
5430         }
5431
5432         err = pci_request_selected_regions_exclusive(pdev,
5433                                           pci_select_bars(pdev, IORESOURCE_MEM),
5434                                           e1000e_driver_name);
5435         if (err)
5436                 goto err_pci_reg;
5437
5438         /* AER (Advanced Error Reporting) hooks */
5439         pci_enable_pcie_error_reporting(pdev);
5440
5441         pci_set_master(pdev);
5442         /* PCI config space info */
5443         err = pci_save_state(pdev);
5444         if (err)
5445                 goto err_alloc_etherdev;
5446
5447         err = -ENOMEM;
5448         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
5449         if (!netdev)
5450                 goto err_alloc_etherdev;
5451
5452         SET_NETDEV_DEV(netdev, &pdev->dev);
5453
5454         pci_set_drvdata(pdev, netdev);
5455         adapter = netdev_priv(netdev);
5456         hw = &adapter->hw;
5457         adapter->netdev = netdev;
5458         adapter->pdev = pdev;
5459         adapter->ei = ei;
5460         adapter->pba = ei->pba;
5461         adapter->flags = ei->flags;
5462         adapter->flags2 = ei->flags2;
5463         adapter->hw.adapter = adapter;
5464         adapter->hw.mac.type = ei->mac;
5465         adapter->max_hw_frame_size = ei->max_hw_frame_size;
5466         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5467
5468         mmio_start = pci_resource_start(pdev, 0);
5469         mmio_len = pci_resource_len(pdev, 0);
5470
5471         err = -EIO;
5472         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5473         if (!adapter->hw.hw_addr)
5474                 goto err_ioremap;
5475
5476         if ((adapter->flags & FLAG_HAS_FLASH) &&
5477             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5478                 flash_start = pci_resource_start(pdev, 1);
5479                 flash_len = pci_resource_len(pdev, 1);
5480                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5481                 if (!adapter->hw.flash_address)
5482                         goto err_flashmap;
5483         }
5484
5485         /* construct the net_device struct */
5486         netdev->netdev_ops              = &e1000e_netdev_ops;
5487         e1000e_set_ethtool_ops(netdev);
5488         netdev->watchdog_timeo          = 5 * HZ;
5489         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5490         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5491
5492         netdev->mem_start = mmio_start;
5493         netdev->mem_end = mmio_start + mmio_len;
5494
5495         adapter->bd_number = cards_found++;
5496
5497         e1000e_check_options(adapter);
5498
5499         /* setup adapter struct */
5500         err = e1000_sw_init(adapter);
5501         if (err)
5502                 goto err_sw_init;
5503
5504         err = -EIO;
5505
5506         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5507         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5508         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5509
5510         err = ei->get_variants(adapter);
5511         if (err)
5512                 goto err_hw_init;
5513
5514         if ((adapter->flags & FLAG_IS_ICH) &&
5515             (adapter->flags & FLAG_READ_ONLY_NVM))
5516                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5517
5518         hw->mac.ops.get_bus_info(&adapter->hw);
5519
5520         adapter->hw.phy.autoneg_wait_to_complete = 0;
5521
5522         /* Copper options */
5523         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5524                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5525                 adapter->hw.phy.disable_polarity_correction = 0;
5526                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5527         }
5528
5529         if (e1000_check_reset_block(&adapter->hw))
5530                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5531
5532         netdev->features = NETIF_F_SG |
5533                            NETIF_F_HW_CSUM |
5534                            NETIF_F_HW_VLAN_TX |
5535                            NETIF_F_HW_VLAN_RX;
5536
5537         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5538                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5539
5540         netdev->features |= NETIF_F_TSO;
5541         netdev->features |= NETIF_F_TSO6;
5542
5543         netdev->vlan_features |= NETIF_F_TSO;
5544         netdev->vlan_features |= NETIF_F_TSO6;
5545         netdev->vlan_features |= NETIF_F_HW_CSUM;
5546         netdev->vlan_features |= NETIF_F_SG;
5547
5548         if (pci_using_dac)
5549                 netdev->features |= NETIF_F_HIGHDMA;
5550
5551         if (e1000e_enable_mng_pass_thru(&adapter->hw))
5552                 adapter->flags |= FLAG_MNG_PT_ENABLED;
5553
5554         /*
5555          * before reading the NVM, reset the controller to
5556          * put the device in a known good starting state
5557          */
5558         adapter->hw.mac.ops.reset_hw(&adapter->hw);
5559
5560         /*
5561          * systems with ASPM and others may see the checksum fail on the first
5562          * attempt. Let's give it a few tries
5563          */
5564         for (i = 0;; i++) {
5565                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5566                         break;
5567                 if (i == 2) {
5568                         e_err("The NVM Checksum Is Not Valid\n");
5569                         err = -EIO;
5570                         goto err_eeprom;
5571                 }
5572         }
5573
5574         e1000_eeprom_checks(adapter);
5575
5576         /* copy the MAC address */
5577         if (e1000e_read_mac_addr(&adapter->hw))
5578                 e_err("NVM Read Error while reading MAC address\n");
5579
5580         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5581         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5582
5583         if (!is_valid_ether_addr(netdev->perm_addr)) {
5584                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5585                 err = -EIO;
5586                 goto err_eeprom;
5587         }
5588
5589         init_timer(&adapter->watchdog_timer);
5590         adapter->watchdog_timer.function = &e1000_watchdog;
5591         adapter->watchdog_timer.data = (unsigned long) adapter;
5592
5593         init_timer(&adapter->phy_info_timer);
5594         adapter->phy_info_timer.function = &e1000_update_phy_info;
5595         adapter->phy_info_timer.data = (unsigned long) adapter;
5596
5597         INIT_WORK(&adapter->reset_task, e1000_reset_task);
5598         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
5599         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
5600         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
5601         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
5602
5603         /* Initialize link parameters. User can change them with ethtool */
5604         adapter->hw.mac.autoneg = 1;
5605         adapter->fc_autoneg = 1;
5606         adapter->hw.fc.requested_mode = e1000_fc_default;
5607         adapter->hw.fc.current_mode = e1000_fc_default;
5608         adapter->hw.phy.autoneg_advertised = 0x2f;
5609
5610         /* ring size defaults */
5611         adapter->rx_ring->count = 256;
5612         adapter->tx_ring->count = 256;
5613
5614         /*
5615          * Initial Wake on LAN setting - If APM wake is enabled in
5616          * the EEPROM, enable the ACPI Magic Packet filter
5617          */
5618         if (adapter->flags & FLAG_APME_IN_WUC) {
5619                 /* APME bit in EEPROM is mapped to WUC.APME */
5620                 eeprom_data = er32(WUC);
5621                 eeprom_apme_mask = E1000_WUC_APME;
5622                 if (eeprom_data & E1000_WUC_PHY_WAKE)
5623                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
5624         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5625                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5626                     (adapter->hw.bus.func == 1))
5627                         e1000_read_nvm(&adapter->hw,
5628                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5629                 else
5630                         e1000_read_nvm(&adapter->hw,
5631                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5632         }
5633
5634         /* fetch WoL from EEPROM */
5635         if (eeprom_data & eeprom_apme_mask)
5636                 adapter->eeprom_wol |= E1000_WUFC_MAG;
5637
5638         /*
5639          * now that we have the eeprom settings, apply the special cases
5640          * where the eeprom may be wrong or the board simply won't support
5641          * wake on lan on a particular port
5642          */
5643         if (!(adapter->flags & FLAG_HAS_WOL))
5644                 adapter->eeprom_wol = 0;
5645
5646         /* initialize the wol settings based on the eeprom settings */
5647         adapter->wol = adapter->eeprom_wol;
5648         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5649
5650         /* save off EEPROM version number */
5651         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
5652
5653         /* reset the hardware with the new settings */
5654         e1000e_reset(adapter);
5655
5656         /*
5657          * If the controller has AMT, do not set DRV_LOAD until the interface
5658          * is up.  For all other cases, let the f/w know that the h/w is now
5659          * under the control of the driver.
5660          */
5661         if (!(adapter->flags & FLAG_HAS_AMT))
5662                 e1000_get_hw_control(adapter);
5663
5664         strcpy(netdev->name, "eth%d");
5665         err = register_netdev(netdev);
5666         if (err)
5667                 goto err_register;
5668
5669         /* carrier off reporting is important to ethtool even BEFORE open */
5670         netif_carrier_off(netdev);
5671
5672         e1000_print_device_info(adapter);
5673
5674         if (pci_dev_run_wake(pdev)) {
5675                 pm_runtime_set_active(&pdev->dev);
5676                 pm_runtime_enable(&pdev->dev);
5677         }
5678         pm_schedule_suspend(&pdev->dev, MSEC_PER_SEC);
5679
5680         return 0;
5681
5682 err_register:
5683         if (!(adapter->flags & FLAG_HAS_AMT))
5684                 e1000_release_hw_control(adapter);
5685 err_eeprom:
5686         if (!e1000_check_reset_block(&adapter->hw))
5687                 e1000_phy_hw_reset(&adapter->hw);
5688 err_hw_init:
5689
5690         kfree(adapter->tx_ring);
5691         kfree(adapter->rx_ring);
5692 err_sw_init:
5693         if (adapter->hw.flash_address)
5694                 iounmap(adapter->hw.flash_address);
5695         e1000e_reset_interrupt_capability(adapter);
5696 err_flashmap:
5697         iounmap(adapter->hw.hw_addr);
5698 err_ioremap:
5699         free_netdev(netdev);
5700 err_alloc_etherdev:
5701         pci_release_selected_regions(pdev,
5702                                      pci_select_bars(pdev, IORESOURCE_MEM));
5703 err_pci_reg:
5704 err_dma:
5705         pci_disable_device(pdev);
5706         return err;
5707 }
5708
5709 /**
5710  * e1000_remove - Device Removal Routine
5711  * @pdev: PCI device information struct
5712  *
5713  * e1000_remove is called by the PCI subsystem to alert the driver
5714  * that it should release a PCI device.  The could be caused by a
5715  * Hot-Plug event, or because the driver is going to be removed from
5716  * memory.
5717  **/
5718 static void __devexit e1000_remove(struct pci_dev *pdev)
5719 {
5720         struct net_device *netdev = pci_get_drvdata(pdev);
5721         struct e1000_adapter *adapter = netdev_priv(netdev);
5722         bool down = test_bit(__E1000_DOWN, &adapter->state);
5723
5724         pm_runtime_get_sync(&pdev->dev);
5725
5726         /*
5727          * flush_scheduled work may reschedule our watchdog task, so
5728          * explicitly disable watchdog tasks from being rescheduled
5729          */
5730         if (!down)
5731                 set_bit(__E1000_DOWN, &adapter->state);
5732         del_timer_sync(&adapter->watchdog_timer);
5733         del_timer_sync(&adapter->phy_info_timer);
5734
5735         cancel_work_sync(&adapter->reset_task);
5736         cancel_work_sync(&adapter->watchdog_task);
5737         cancel_work_sync(&adapter->downshift_task);
5738         cancel_work_sync(&adapter->update_phy_task);
5739         cancel_work_sync(&adapter->print_hang_task);
5740         flush_scheduled_work();
5741
5742         if (!(netdev->flags & IFF_UP))
5743                 e1000_power_down_phy(adapter);
5744
5745         /* Don't lie to e1000_close() down the road. */
5746         if (!down)
5747                 clear_bit(__E1000_DOWN, &adapter->state);
5748         unregister_netdev(netdev);
5749
5750         if (pci_dev_run_wake(pdev)) {
5751                 pm_runtime_disable(&pdev->dev);
5752                 pm_runtime_set_suspended(&pdev->dev);
5753         }
5754         pm_runtime_put_noidle(&pdev->dev);
5755
5756         /*
5757          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5758          * would have already happened in close and is redundant.
5759          */
5760         e1000_release_hw_control(adapter);
5761
5762         e1000e_reset_interrupt_capability(adapter);
5763         kfree(adapter->tx_ring);
5764         kfree(adapter->rx_ring);
5765
5766         iounmap(adapter->hw.hw_addr);
5767         if (adapter->hw.flash_address)
5768                 iounmap(adapter->hw.flash_address);
5769         pci_release_selected_regions(pdev,
5770                                      pci_select_bars(pdev, IORESOURCE_MEM));
5771
5772         free_netdev(netdev);
5773
5774         /* AER disable */
5775         pci_disable_pcie_error_reporting(pdev);
5776
5777         pci_disable_device(pdev);
5778 }
5779
5780 /* PCI Error Recovery (ERS) */
5781 static struct pci_error_handlers e1000_err_handler = {
5782         .error_detected = e1000_io_error_detected,
5783         .slot_reset = e1000_io_slot_reset,
5784         .resume = e1000_io_resume,
5785 };
5786
5787 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
5788         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
5789         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
5790         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
5791         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
5792         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
5793         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
5794         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
5795         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
5796         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
5797
5798         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
5799         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
5800         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
5801         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
5802
5803         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
5804         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
5805         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
5806
5807         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
5808         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
5809         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
5810
5811         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
5812           board_80003es2lan },
5813         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
5814           board_80003es2lan },
5815         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
5816           board_80003es2lan },
5817         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
5818           board_80003es2lan },
5819
5820         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
5821         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
5822         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
5823         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
5824         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
5825         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
5826         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
5827         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
5828
5829         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
5830         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
5831         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
5832         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
5833         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
5834         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
5835         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
5836         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
5837         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
5838
5839         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
5840         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
5841         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
5842
5843         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
5844         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
5845
5846         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
5847         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
5848         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
5849         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
5850
5851         { }     /* terminate list */
5852 };
5853 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
5854
5855 #ifdef CONFIG_PM_OPS
5856 static const struct dev_pm_ops e1000_pm_ops = {
5857         SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
5858         SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
5859                                 e1000_runtime_resume, e1000_idle)
5860 };
5861 #endif
5862
5863 /* PCI Device API Driver */
5864 static struct pci_driver e1000_driver = {
5865         .name     = e1000e_driver_name,
5866         .id_table = e1000_pci_tbl,
5867         .probe    = e1000_probe,
5868         .remove   = __devexit_p(e1000_remove),
5869 #ifdef CONFIG_PM_OPS
5870         .driver.pm = &e1000_pm_ops,
5871 #endif
5872         .shutdown = e1000_shutdown,
5873         .err_handler = &e1000_err_handler
5874 };
5875
5876 /**
5877  * e1000_init_module - Driver Registration Routine
5878  *
5879  * e1000_init_module is the first routine called when the driver is
5880  * loaded. All it does is register with the PCI subsystem.
5881  **/
5882 static int __init e1000_init_module(void)
5883 {
5884         int ret;
5885         pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
5886                 e1000e_driver_version);
5887         pr_info("Copyright (c) 1999 - 2009 Intel Corporation.\n");
5888         ret = pci_register_driver(&e1000_driver);
5889
5890         return ret;
5891 }
5892 module_init(e1000_init_module);
5893
5894 /**
5895  * e1000_exit_module - Driver Exit Cleanup Routine
5896  *
5897  * e1000_exit_module is called just before the driver is removed
5898  * from memory.
5899  **/
5900 static void __exit e1000_exit_module(void)
5901 {
5902         pci_unregister_driver(&e1000_driver);
5903 }
5904 module_exit(e1000_exit_module);
5905
5906
5907 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
5908 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
5909 MODULE_LICENSE("GPL");
5910 MODULE_VERSION(DRV_VERSION);
5911
5912 /* e1000_main.c */