drivers/net/e1000e/netdev.c

   1 /*******************************************************************************
   2
   3   Intel PRO/1000 Linux driver
   4   Copyright(c) 1999 - 2010 Intel Corporation.
   5
   6   This program is free software; you can redistribute it and/or modify it
   7   under the terms and conditions of the GNU General Public License,
   8   version 2, as published by the Free Software Foundation.
   9
  10   This program is distributed in the hope it will be useful, but WITHOUT
  11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13   more details.
  14
  15   You should have received a copy of the GNU General Public License along with
  16   this program; if not, write to the Free Software Foundation, Inc.,
  17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18
  19   The full GNU General Public License is included in this distribution in
  20   the file called "COPYING".
  21
  22   Contact Information:
  23   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_EXTRAVERSION "-k2"
  56
  57 #define DRV_VERSION "1.2.7" DRV_EXTRAVERSION
  58 char e1000e_driver_name[] = "e1000e";
  59 const char e1000e_driver_version[] = DRV_VERSION;
  60
  61 static const struct e1000_info *e1000_info_tbl[] = {
  62         [board_82571]           = &e1000_82571_info,
  63         [board_82572]           = &e1000_82572_info,
  64         [board_82573]           = &e1000_82573_info,
  65         [board_82574]           = &e1000_82574_info,
  66         [board_82583]           = &e1000_82583_info,
  67         [board_80003es2lan]     = &e1000_es2_info,
  68         [board_ich8lan]         = &e1000_ich8_info,
  69         [board_ich9lan]         = &e1000_ich9_info,
  70         [board_ich10lan]        = &e1000_ich10_info,
  71         [board_pchlan]          = &e1000_pch_info,
  72         [board_pch2lan]         = &e1000_pch2_info,
  73 };
  74
  75 struct e1000_reg_info {
  76         u32 ofs;
  77         char *name;
  78 };
  79
  80 #define E1000_RDFH      0x02410 /* Rx Data FIFO Head - RW */
  81 #define E1000_RDFT      0x02418 /* Rx Data FIFO Tail - RW */
  82 #define E1000_RDFHS     0x02420 /* Rx Data FIFO Head Saved - RW */
  83 #define E1000_RDFTS     0x02428 /* Rx Data FIFO Tail Saved - RW */
  84 #define E1000_RDFPC     0x02430 /* Rx Data FIFO Packet Count - RW */
  85
  86 #define E1000_TDFH      0x03410 /* Tx Data FIFO Head - RW */
  87 #define E1000_TDFT      0x03418 /* Tx Data FIFO Tail - RW */
  88 #define E1000_TDFHS     0x03420 /* Tx Data FIFO Head Saved - RW */
  89 #define E1000_TDFTS     0x03428 /* Tx Data FIFO Tail Saved - RW */
  90 #define E1000_TDFPC     0x03430 /* Tx Data FIFO Packet Count - RW */
  91
  92 static const struct e1000_reg_info e1000_reg_info_tbl[] = {
  93
  94         /* General Registers */
  95         {E1000_CTRL, "CTRL"},
  96         {E1000_STATUS, "STATUS"},
  97         {E1000_CTRL_EXT, "CTRL_EXT"},
  98
  99         /* Interrupt Registers */
 100         {E1000_ICR, "ICR"},
 101
 102         /* RX Registers */
 103         {E1000_RCTL, "RCTL"},
 104         {E1000_RDLEN, "RDLEN"},
 105         {E1000_RDH, "RDH"},
 106         {E1000_RDT, "RDT"},
 107         {E1000_RDTR, "RDTR"},
 108         {E1000_RXDCTL(0), "RXDCTL"},
 109         {E1000_ERT, "ERT"},
 110         {E1000_RDBAL, "RDBAL"},
 111         {E1000_RDBAH, "RDBAH"},
 112         {E1000_RDFH, "RDFH"},
 113         {E1000_RDFT, "RDFT"},
 114         {E1000_RDFHS, "RDFHS"},
 115         {E1000_RDFTS, "RDFTS"},
 116         {E1000_RDFPC, "RDFPC"},
 117
 118         /* TX Registers */
 119         {E1000_TCTL, "TCTL"},
 120         {E1000_TDBAL, "TDBAL"},
 121         {E1000_TDBAH, "TDBAH"},
 122         {E1000_TDLEN, "TDLEN"},
 123         {E1000_TDH, "TDH"},
 124         {E1000_TDT, "TDT"},
 125         {E1000_TIDV, "TIDV"},
 126         {E1000_TXDCTL(0), "TXDCTL"},
 127         {E1000_TADV, "TADV"},
 128         {E1000_TARC(0), "TARC"},
 129         {E1000_TDFH, "TDFH"},
 130         {E1000_TDFT, "TDFT"},
 131         {E1000_TDFHS, "TDFHS"},
 132         {E1000_TDFTS, "TDFTS"},
 133         {E1000_TDFPC, "TDFPC"},
 134
 135         /* List Terminator */
 136         {}
 137 };
 138
 139 /*
 140  * e1000_regdump - register printout routine
 141  */
 142 static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
 143 {
 144         int n = 0;
 145         char rname[16];
 146         u32 regs[8];
 147
 148         switch (reginfo->ofs) {
 149         case E1000_RXDCTL(0):
 150                 for (n = 0; n < 2; n++)
 151                         regs[n] = __er32(hw, E1000_RXDCTL(n));
 152                 break;
 153         case E1000_TXDCTL(0):
 154                 for (n = 0; n < 2; n++)
 155                         regs[n] = __er32(hw, E1000_TXDCTL(n));
 156                 break;
 157         case E1000_TARC(0):
 158                 for (n = 0; n < 2; n++)
 159                         regs[n] = __er32(hw, E1000_TARC(n));
 160                 break;
 161         default:
 162                 printk(KERN_INFO "%-15s %08x\n",
 163                         reginfo->name, __er32(hw, reginfo->ofs));
 164                 return;
 165         }
 166
 167         snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
 168         printk(KERN_INFO "%-15s ", rname);
 169         for (n = 0; n < 2; n++)
 170                 printk(KERN_CONT "%08x ", regs[n]);
 171         printk(KERN_CONT "\n");
 172 }
 173
 174
 175 /*
 176  * e1000e_dump - Print registers, tx-ring and rx-ring
 177  */
 178 static void e1000e_dump(struct e1000_adapter *adapter)
 179 {
 180         struct net_device *netdev = adapter->netdev;
 181         struct e1000_hw *hw = &adapter->hw;
 182         struct e1000_reg_info *reginfo;
 183         struct e1000_ring *tx_ring = adapter->tx_ring;
 184         struct e1000_tx_desc *tx_desc;
 185         struct my_u0 { u64 a; u64 b; } *u0;
 186         struct e1000_buffer *buffer_info;
 187         struct e1000_ring *rx_ring = adapter->rx_ring;
 188         union e1000_rx_desc_packet_split *rx_desc_ps;
 189         struct e1000_rx_desc *rx_desc;
 190         struct my_u1 { u64 a; u64 b; u64 c; u64 d; } *u1;
 191         u32 staterr;
 192         int i = 0;
 193
 194         if (!netif_msg_hw(adapter))
 195                 return;
 196
 197         /* Print netdevice Info */
 198         if (netdev) {
 199                 dev_info(&adapter->pdev->dev, "Net device Info\n");
 200                 printk(KERN_INFO "Device Name     state            "
 201                         "trans_start      last_rx\n");
 202                 printk(KERN_INFO "%-15s %016lX %016lX %016lX\n",
 203                         netdev->name,
 204                         netdev->state,
 205                         netdev->trans_start,
 206                         netdev->last_rx);
 207         }
 208
 209         /* Print Registers */
 210         dev_info(&adapter->pdev->dev, "Register Dump\n");
 211         printk(KERN_INFO " Register Name   Value\n");
 212         for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
 213              reginfo->name; reginfo++) {
 214                 e1000_regdump(hw, reginfo);
 215         }
 216
 217         /* Print TX Ring Summary */
 218         if (!netdev || !netif_running(netdev))
 219                 goto exit;
 220
 221         dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
 222         printk(KERN_INFO "Queue [NTU] [NTC] [bi(ntc)->dma  ]"
 223                 " leng ntw timestamp\n");
 224         buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
 225         printk(KERN_INFO " %5d %5X %5X %016llX %04X %3X %016llX\n",
 226                 0, tx_ring->next_to_use, tx_ring->next_to_clean,
 227                 (unsigned long long)buffer_info->dma,
 228                 buffer_info->length,
 229                 buffer_info->next_to_watch,
 230                 (unsigned long long)buffer_info->time_stamp);
 231
 232         /* Print TX Rings */
 233         if (!netif_msg_tx_done(adapter))
 234                 goto rx_ring_summary;
 235
 236         dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
 237
 238         /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
 239          *
 240          * Legacy Transmit Descriptor
 241          *   +--------------------------------------------------------------+
 242          * 0 |         Buffer Address [63:0] (Reserved on Write Back)       |
 243          *   +--------------------------------------------------------------+
 244          * 8 | Special  |    CSS     | Status |  CMD    |  CSO   |  Length  |
 245          *   +--------------------------------------------------------------+
 246          *   63       48 47        36 35    32 31     24 23    16 15        0
 247          *
 248          * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
 249          *   63      48 47    40 39       32 31             16 15    8 7      0
 250          *   +----------------------------------------------------------------+
 251          * 0 |  TUCSE  | TUCS0  |   TUCSS   |     IPCSE       | IPCS0 | IPCSS |
 252          *   +----------------------------------------------------------------+
 253          * 8 |   MSS   | HDRLEN | RSV | STA | TUCMD | DTYP |      PAYLEN      |
 254          *   +----------------------------------------------------------------+
 255          *   63      48 47    40 39 36 35 32 31   24 23  20 19                0
 256          *
 257          * Extended Data Descriptor (DTYP=0x1)
 258          *   +----------------------------------------------------------------+
 259          * 0 |                     Buffer Address [63:0]                      |
 260          *   +----------------------------------------------------------------+
 261          * 8 | VLAN tag |  POPTS  | Rsvd | Status | Command | DTYP |  DTALEN  |
 262          *   +----------------------------------------------------------------+
 263          *   63       48 47     40 39  36 35    32 31     24 23  20 19        0
 264          */
 265         printk(KERN_INFO "Tl[desc]     [address 63:0  ] [SpeCssSCmCsLen]"
 266                 " [bi->dma       ] leng  ntw timestamp        bi->skb "
 267                 "<-- Legacy format\n");
 268         printk(KERN_INFO "Tc[desc]     [Ce CoCsIpceCoS] [MssHlRSCm0Plen]"
 269                 " [bi->dma       ] leng  ntw timestamp        bi->skb "
 270                 "<-- Ext Context format\n");
 271         printk(KERN_INFO "Td[desc]     [address 63:0  ] [VlaPoRSCm1Dlen]"
 272                 " [bi->dma       ] leng  ntw timestamp        bi->skb "
 273                 "<-- Ext Data format\n");
 274         for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
 275                 tx_desc = E1000_TX_DESC(*tx_ring, i);
 276                 buffer_info = &tx_ring->buffer_info[i];
 277                 u0 = (struct my_u0 *)tx_desc;
 278                 printk(KERN_INFO "T%c[0x%03X]    %016llX %016llX %016llX "
 279                         "%04X  %3X %016llX %p",
 280                        (!(le64_to_cpu(u0->b) & (1<<29)) ? 'l' :
 281                         ((le64_to_cpu(u0->b) & (1<<20)) ? 'd' : 'c')), i,
 282                        (unsigned long long)le64_to_cpu(u0->a),
 283                        (unsigned long long)le64_to_cpu(u0->b),
 284                        (unsigned long long)buffer_info->dma,
 285                        buffer_info->length, buffer_info->next_to_watch,
 286                        (unsigned long long)buffer_info->time_stamp,
 287                        buffer_info->skb);
 288                 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
 289                         printk(KERN_CONT " NTC/U\n");
 290                 else if (i == tx_ring->next_to_use)
 291                         printk(KERN_CONT " NTU\n");
 292                 else if (i == tx_ring->next_to_clean)
 293                         printk(KERN_CONT " NTC\n");
 294                 else
 295                         printk(KERN_CONT "\n");
 296
 297                 if (netif_msg_pktdata(adapter) && buffer_info->dma != 0)
 298                         print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
 299                                         16, 1, phys_to_virt(buffer_info->dma),
 300                                         buffer_info->length, true);
 301         }
 302
 303         /* Print RX Rings Summary */
 304 rx_ring_summary:
 305         dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
 306         printk(KERN_INFO "Queue [NTU] [NTC]\n");
 307         printk(KERN_INFO " %5d %5X %5X\n", 0,
 308                 rx_ring->next_to_use, rx_ring->next_to_clean);
 309
 310         /* Print RX Rings */
 311         if (!netif_msg_rx_status(adapter))
 312                 goto exit;
 313
 314         dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
 315         switch (adapter->rx_ps_pages) {
 316         case 1:
 317         case 2:
 318         case 3:
 319                 /* [Extended] Packet Split Receive Descriptor Format
 320                  *
 321                  *    +-----------------------------------------------------+
 322                  *  0 |                Buffer Address 0 [63:0]              |
 323                  *    +-----------------------------------------------------+
 324                  *  8 |                Buffer Address 1 [63:0]              |
 325                  *    +-----------------------------------------------------+
 326                  * 16 |                Buffer Address 2 [63:0]              |
 327                  *    +-----------------------------------------------------+
 328                  * 24 |                Buffer Address 3 [63:0]              |
 329                  *    +-----------------------------------------------------+
 330                  */
 331                 printk(KERN_INFO "R  [desc]      [buffer 0 63:0 ] "
 332                         "[buffer 1 63:0 ] "
 333                        "[buffer 2 63:0 ] [buffer 3 63:0 ] [bi->dma       ] "
 334                        "[bi->skb] <-- Ext Pkt Split format\n");
 335                 /* [Extended] Receive Descriptor (Write-Back) Format
 336                  *
 337                  *   63       48 47    32 31     13 12    8 7    4 3        0
 338                  *   +------------------------------------------------------+
 339                  * 0 | Packet   | IP     |  Rsvd   | MRQ   | Rsvd | MRQ RSS |
 340                  *   | Checksum | Ident  |         | Queue |      |  Type   |
 341                  *   +------------------------------------------------------+
 342                  * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 343                  *   +------------------------------------------------------+
 344                  *   63       48 47    32 31            20 19               0
 345                  */
 346                 printk(KERN_INFO "RWB[desc]      [ck ipid mrqhsh] "
 347                         "[vl   l0 ee  es] "
 348                        "[ l3  l2  l1 hs] [reserved      ] ---------------- "
 349                        "[bi->skb] <-- Ext Rx Write-Back format\n");
 350                 for (i = 0; i < rx_ring->count; i++) {
 351                         buffer_info = &rx_ring->buffer_info[i];
 352                         rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
 353                         u1 = (struct my_u1 *)rx_desc_ps;
 354                         staterr =
 355                                 le32_to_cpu(rx_desc_ps->wb.middle.status_error);
 356                         if (staterr & E1000_RXD_STAT_DD) {
 357                                 /* Descriptor Done */
 358                                 printk(KERN_INFO "RWB[0x%03X]     %016llX "
 359                                         "%016llX %016llX %016llX "
 360                                         "---------------- %p", i,
 361                                         (unsigned long long)le64_to_cpu(u1->a),
 362                                         (unsigned long long)le64_to_cpu(u1->b),
 363                                         (unsigned long long)le64_to_cpu(u1->c),
 364                                         (unsigned long long)le64_to_cpu(u1->d),
 365                                         buffer_info->skb);
 366                         } else {
 367                                 printk(KERN_INFO "R  [0x%03X]     %016llX "
 368                                         "%016llX %016llX %016llX %016llX %p", i,
 369                                         (unsigned long long)le64_to_cpu(u1->a),
 370                                         (unsigned long long)le64_to_cpu(u1->b),
 371                                         (unsigned long long)le64_to_cpu(u1->c),
 372                                         (unsigned long long)le64_to_cpu(u1->d),
 373                                         (unsigned long long)buffer_info->dma,
 374                                         buffer_info->skb);
 375
 376                                 if (netif_msg_pktdata(adapter))
 377                                         print_hex_dump(KERN_INFO, "",
 378                                                 DUMP_PREFIX_ADDRESS, 16, 1,
 379                                                 phys_to_virt(buffer_info->dma),
 380                                                 adapter->rx_ps_bsize0, true);
 381                         }
 382
 383                         if (i == rx_ring->next_to_use)
 384                                 printk(KERN_CONT " NTU\n");
 385                         else if (i == rx_ring->next_to_clean)
 386                                 printk(KERN_CONT " NTC\n");
 387                         else
 388                                 printk(KERN_CONT "\n");
 389                 }
 390                 break;
 391         default:
 392         case 0:
 393                 /* Legacy Receive Descriptor Format
 394                  *
 395                  * +-----------------------------------------------------+
 396                  * |                Buffer Address [63:0]                |
 397                  * +-----------------------------------------------------+
 398                  * | VLAN Tag | Errors | Status 0 | Packet csum | Length |
 399                  * +-----------------------------------------------------+
 400                  * 63       48 47    40 39      32 31         16 15      0
 401                  */
 402                 printk(KERN_INFO "Rl[desc]     [address 63:0  ] "
 403                         "[vl er S cks ln] [bi->dma       ] [bi->skb] "
 404                         "<-- Legacy format\n");
 405                 for (i = 0; rx_ring->desc && (i < rx_ring->count); i++) {
 406                         rx_desc = E1000_RX_DESC(*rx_ring, i);
 407                         buffer_info = &rx_ring->buffer_info[i];
 408                         u0 = (struct my_u0 *)rx_desc;
 409                         printk(KERN_INFO "Rl[0x%03X]    %016llX %016llX "
 410                                 "%016llX %p", i,
 411                                 (unsigned long long)le64_to_cpu(u0->a),
 412                                 (unsigned long long)le64_to_cpu(u0->b),
 413                                 (unsigned long long)buffer_info->dma,
 414                                 buffer_info->skb);
 415                         if (i == rx_ring->next_to_use)
 416                                 printk(KERN_CONT " NTU\n");
 417                         else if (i == rx_ring->next_to_clean)
 418                                 printk(KERN_CONT " NTC\n");
 419                         else
 420                                 printk(KERN_CONT "\n");
 421
 422                         if (netif_msg_pktdata(adapter))
 423                                 print_hex_dump(KERN_INFO, "",
 424                                         DUMP_PREFIX_ADDRESS,
 425                                         16, 1, phys_to_virt(buffer_info->dma),
 426                                         adapter->rx_buffer_len, true);
 427                 }
 428         }
 429
 430 exit:
 431         return;
 432 }
 433
 434 /**
 435  * e1000_desc_unused - calculate if we have unused descriptors
 436  **/
 437 static int e1000_desc_unused(struct e1000_ring *ring)
 438 {
 439         if (ring->next_to_clean > ring->next_to_use)
 440                 return ring->next_to_clean - ring->next_to_use - 1;
 441
 442         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
 443 }
 444
 445 /**
 446  * e1000_receive_skb - helper function to handle Rx indications
 447  * @adapter: board private structure
 448  * @status: descriptor status field as written by hardware
 449  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
 450  * @skb: pointer to sk_buff to be indicated to stack
 451  **/
 452 static void e1000_receive_skb(struct e1000_adapter *adapter,
 453                               struct net_device *netdev,
 454                               struct sk_buff *skb,
 455                               u8 status, __le16 vlan)
 456 {
 457         skb->protocol = eth_type_trans(skb, netdev);
 458
 459         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
 460                 vlan_gro_receive(&adapter->napi, adapter->vlgrp,
 461                                  le16_to_cpu(vlan), skb);
 462         else
 463                 napi_gro_receive(&adapter->napi, skb);
 464 }
 465
 466 /**
 467  * e1000_rx_checksum - Receive Checksum Offload for 82543
 468  * @adapter:     board private structure
 469  * @status_err:  receive descriptor status and error fields
 470  * @csum:       receive descriptor csum field
 471  * @sk_buff:     socket buffer with received data
 472  **/
 473 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
 474                               u32 csum, struct sk_buff *skb)
 475 {
 476         u16 status = (u16)status_err;
 477         u8 errors = (u8)(status_err >> 24);
 478         skb->ip_summed = CHECKSUM_NONE;
 479
 480         /* Ignore Checksum bit is set */
 481         if (status & E1000_RXD_STAT_IXSM)
 482                 return;
 483         /* TCP/UDP checksum error bit is set */
 484         if (errors & E1000_RXD_ERR_TCPE) {
 485                 /* let the stack verify checksum errors */
 486                 adapter->hw_csum_err++;
 487                 return;
 488         }
 489
 490         /* TCP/UDP Checksum has not been calculated */
 491         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
 492                 return;
 493
 494         /* It must be a TCP or UDP packet with a valid checksum */
 495         if (status & E1000_RXD_STAT_TCPCS) {
 496                 /* TCP checksum is good */
 497                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 498         } else {
 499                 /*
 500                  * IP fragment with UDP payload
 501                  * Hardware complements the payload checksum, so we undo it
 502                  * and then put the value in host order for further stack use.
 503                  */
 504                 __sum16 sum = (__force __sum16)htons(csum);
 505                 skb->csum = csum_unfold(~sum);
 506                 skb->ip_summed = CHECKSUM_COMPLETE;
 507         }
 508         adapter->hw_csum_good++;
 509 }
 510
 511 /**
 512  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
 513  * @adapter: address of board private structure
 514  **/
 515 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
 516                                    int cleaned_count)
 517 {
 518         struct net_device *netdev = adapter->netdev;
 519         struct pci_dev *pdev = adapter->pdev;
 520         struct e1000_ring *rx_ring = adapter->rx_ring;
 521         struct e1000_rx_desc *rx_desc;
 522         struct e1000_buffer *buffer_info;
 523         struct sk_buff *skb;
 524         unsigned int i;
 525         unsigned int bufsz = adapter->rx_buffer_len;
 526
 527         i = rx_ring->next_to_use;
 528         buffer_info = &rx_ring->buffer_info[i];
 529
 530         while (cleaned_count--) {
 531                 skb = buffer_info->skb;
 532                 if (skb) {
 533                         skb_trim(skb, 0);
 534                         goto map_skb;
 535                 }
 536
 537                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 538                 if (!skb) {
 539                         /* Better luck next round */
 540                         adapter->alloc_rx_buff_failed++;
 541                         break;
 542                 }
 543
 544                 buffer_info->skb = skb;
 545 map_skb:
 546                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 547                                                   adapter->rx_buffer_len,
 548                                                   DMA_FROM_DEVICE);
 549                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 550                         dev_err(&pdev->dev, "RX DMA map failed\n");
 551                         adapter->rx_dma_failed++;
 552                         break;
 553                 }
 554
 555                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 556                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 557
 558                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 559                         /*
 560                          * Force memory writes to complete before letting h/w
 561                          * know there are new descriptors to fetch.  (Only
 562                          * applicable for weak-ordered memory model archs,
 563                          * such as IA-64).
 564                          */
 565                         wmb();
 566                         writel(i, adapter->hw.hw_addr + rx_ring->tail);
 567                 }
 568                 i++;
 569                 if (i == rx_ring->count)
 570                         i = 0;
 571                 buffer_info = &rx_ring->buffer_info[i];
 572         }
 573
 574         rx_ring->next_to_use = i;
 575 }
 576
 577 /**
 578  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
 579  * @adapter: address of board private structure
 580  **/
 581 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 582                                       int cleaned_count)
 583 {
 584         struct net_device *netdev = adapter->netdev;
 585         struct pci_dev *pdev = adapter->pdev;
 586         union e1000_rx_desc_packet_split *rx_desc;
 587         struct e1000_ring *rx_ring = adapter->rx_ring;
 588         struct e1000_buffer *buffer_info;
 589         struct e1000_ps_page *ps_page;
 590         struct sk_buff *skb;
 591         unsigned int i, j;
 592
 593         i = rx_ring->next_to_use;
 594         buffer_info = &rx_ring->buffer_info[i];
 595
 596         while (cleaned_count--) {
 597                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 598
 599                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 600                         ps_page = &buffer_info->ps_pages[j];
 601                         if (j >= adapter->rx_ps_pages) {
 602                                 /* all unused desc entries get hw null ptr */
 603                                 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
 604                                 continue;
 605                         }
 606                         if (!ps_page->page) {
 607                                 ps_page->page = alloc_page(GFP_ATOMIC);
 608                                 if (!ps_page->page) {
 609                                         adapter->alloc_rx_buff_failed++;
 610                                         goto no_buffers;
 611                                 }
 612                                 ps_page->dma = dma_map_page(&pdev->dev,
 613                                                             ps_page->page,
 614                                                             0, PAGE_SIZE,
 615                                                             DMA_FROM_DEVICE);
 616                                 if (dma_mapping_error(&pdev->dev,
 617                                                       ps_page->dma)) {
 618                                         dev_err(&adapter->pdev->dev,
 619                                           "RX DMA page map failed\n");
 620                                         adapter->rx_dma_failed++;
 621                                         goto no_buffers;
 622                                 }
 623                         }
 624                         /*
 625                          * Refresh the desc even if buffer_addrs
 626                          * didn't change because each write-back
 627                          * erases this info.
 628                          */
 629                         rx_desc->read.buffer_addr[j+1] =
 630                              cpu_to_le64(ps_page->dma);
 631                 }
 632
 633                 skb = netdev_alloc_skb_ip_align(netdev,
 634                                                 adapter->rx_ps_bsize0);
 635
 636                 if (!skb) {
 637                         adapter->alloc_rx_buff_failed++;
 638                         break;
 639                 }
 640
 641                 buffer_info->skb = skb;
 642                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 643                                                   adapter->rx_ps_bsize0,
 644                                                   DMA_FROM_DEVICE);
 645                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 646                         dev_err(&pdev->dev, "RX DMA map failed\n");
 647                         adapter->rx_dma_failed++;
 648                         /* cleanup skb */
 649                         dev_kfree_skb_any(skb);
 650                         buffer_info->skb = NULL;
 651                         break;
 652                 }
 653
 654                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 655
 656                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 657                         /*
 658                          * Force memory writes to complete before letting h/w
 659                          * know there are new descriptors to fetch.  (Only
 660                          * applicable for weak-ordered memory model archs,
 661                          * such as IA-64).
 662                          */
 663                         wmb();
 664                         writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
 665                 }
 666
 667                 i++;
 668                 if (i == rx_ring->count)
 669                         i = 0;
 670                 buffer_info = &rx_ring->buffer_info[i];
 671         }
 672
 673 no_buffers:
 674         rx_ring->next_to_use = i;
 675 }
 676
 677 /**
 678  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
 679  * @adapter: address of board private structure
 680  * @cleaned_count: number of buffers to allocate this pass
 681  **/
 682
 683 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
 684                                          int cleaned_count)
 685 {
 686         struct net_device *netdev = adapter->netdev;
 687         struct pci_dev *pdev = adapter->pdev;
 688         struct e1000_rx_desc *rx_desc;
 689         struct e1000_ring *rx_ring = adapter->rx_ring;
 690         struct e1000_buffer *buffer_info;
 691         struct sk_buff *skb;
 692         unsigned int i;
 693         unsigned int bufsz = 256 - 16 /* for skb_reserve */;
 694
 695         i = rx_ring->next_to_use;
 696         buffer_info = &rx_ring->buffer_info[i];
 697
 698         while (cleaned_count--) {
 699                 skb = buffer_info->skb;
 700                 if (skb) {
 701                         skb_trim(skb, 0);
 702                         goto check_page;
 703                 }
 704
 705                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 706                 if (unlikely(!skb)) {
 707                         /* Better luck next round */
 708                         adapter->alloc_rx_buff_failed++;
 709                         break;
 710                 }
 711
 712                 buffer_info->skb = skb;
 713 check_page:
 714                 /* allocate a new page if necessary */
 715                 if (!buffer_info->page) {
 716                         buffer_info->page = alloc_page(GFP_ATOMIC);
 717                         if (unlikely(!buffer_info->page)) {
 718                                 adapter->alloc_rx_buff_failed++;
 719                                 break;
 720                         }
 721                 }
 722
 723                 if (!buffer_info->dma)
 724                         buffer_info->dma = dma_map_page(&pdev->dev,
 725                                                         buffer_info->page, 0,
 726                                                         PAGE_SIZE,
 727                                                         DMA_FROM_DEVICE);
 728
 729                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 730                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 731
 732                 if (unlikely(++i == rx_ring->count))
 733                         i = 0;
 734                 buffer_info = &rx_ring->buffer_info[i];
 735         }
 736
 737         if (likely(rx_ring->next_to_use != i)) {
 738                 rx_ring->next_to_use = i;
 739                 if (unlikely(i-- == 0))
 740                         i = (rx_ring->count - 1);
 741
 742                 /* Force memory writes to complete before letting h/w
 743                  * know there are new descriptors to fetch.  (Only
 744                  * applicable for weak-ordered memory model archs,
 745                  * such as IA-64). */
 746                 wmb();
 747                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 748         }
 749 }
 750
 751 /**
 752  * e1000_clean_rx_irq - Send received data up the network stack; legacy
 753  * @adapter: board private structure
 754  *
 755  * the return value indicates whether actual cleaning was done, there
 756  * is no guarantee that everything was cleaned
 757  **/
 758 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
 759                                int *work_done, int work_to_do)
 760 {
 761         struct net_device *netdev = adapter->netdev;
 762         struct pci_dev *pdev = adapter->pdev;
 763         struct e1000_hw *hw = &adapter->hw;
 764         struct e1000_ring *rx_ring = adapter->rx_ring;
 765         struct e1000_rx_desc *rx_desc, *next_rxd;
 766         struct e1000_buffer *buffer_info, *next_buffer;
 767         u32 length;
 768         unsigned int i;
 769         int cleaned_count = 0;
 770         bool cleaned = 0;
 771         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 772
 773         i = rx_ring->next_to_clean;
 774         rx_desc = E1000_RX_DESC(*rx_ring, i);
 775         buffer_info = &rx_ring->buffer_info[i];
 776
 777         while (rx_desc->status & E1000_RXD_STAT_DD) {
 778                 struct sk_buff *skb;
 779                 u8 status;
 780
 781                 if (*work_done >= work_to_do)
 782                         break;
 783                 (*work_done)++;
 784
 785                 status = rx_desc->status;
 786                 skb = buffer_info->skb;
 787                 buffer_info->skb = NULL;
 788
 789                 prefetch(skb->data - NET_IP_ALIGN);
 790
 791                 i++;
 792                 if (i == rx_ring->count)
 793                         i = 0;
 794                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 795                 prefetch(next_rxd);
 796
 797                 next_buffer = &rx_ring->buffer_info[i];
 798
 799                 cleaned = 1;
 800                 cleaned_count++;
 801                 dma_unmap_single(&pdev->dev,
 802                                  buffer_info->dma,
 803                                  adapter->rx_buffer_len,
 804                                  DMA_FROM_DEVICE);
 805                 buffer_info->dma = 0;
 806
 807                 length = le16_to_cpu(rx_desc->length);
 808
 809                 /*
 810                  * !EOP means multiple descriptors were used to store a single
 811                  * packet, if that's the case we need to toss it.  In fact, we
 812                  * need to toss every packet with the EOP bit clear and the
 813                  * next frame that _does_ have the EOP bit set, as it is by
 814                  * definition only a frame fragment
 815                  */
 816                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
 817                         adapter->flags2 |= FLAG2_IS_DISCARDING;
 818
 819                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
 820                         /* All receives must fit into a single buffer */
 821                         e_dbg("Receive packet consumed multiple buffers\n");
 822                         /* recycle */
 823                         buffer_info->skb = skb;
 824                         if (status & E1000_RXD_STAT_EOP)
 825                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
 826                         goto next_desc;
 827                 }
 828
 829                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
 830                         /* recycle */
 831                         buffer_info->skb = skb;
 832                         goto next_desc;
 833                 }
 834
 835                 /* adjust length to remove Ethernet CRC */
 836                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 837                         length -= 4;
 838
 839                 total_rx_bytes += length;
 840                 total_rx_packets++;
 841
 842                 /*
 843                  * code added for copybreak, this should improve
 844                  * performance for small packets with large amounts
 845                  * of reassembly being done in the stack
 846                  */
 847                 if (length < copybreak) {
 848                         struct sk_buff *new_skb =
 849                             netdev_alloc_skb_ip_align(netdev, length);
 850                         if (new_skb) {
 851                                 skb_copy_to_linear_data_offset(new_skb,
 852                                                                -NET_IP_ALIGN,
 853                                                                (skb->data -
 854                                                                 NET_IP_ALIGN),
 855                                                                (length +
 856                                                                 NET_IP_ALIGN));
 857                                 /* save the skb in buffer_info as good */
 858                                 buffer_info->skb = skb;
 859                                 skb = new_skb;
 860                         }
 861                         /* else just continue with the old one */
 862                 }
 863                 /* end copybreak code */
 864                 skb_put(skb, length);
 865
 866                 /* Receive Checksum Offload */
 867                 e1000_rx_checksum(adapter,
 868                                   (u32)(status) |
 869                                   ((u32)(rx_desc->errors) << 24),
 870                                   le16_to_cpu(rx_desc->csum), skb);
 871
 872                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
 873
 874 next_desc:
 875                 rx_desc->status = 0;
 876
 877                 /* return some buffers to hardware, one at a time is too slow */
 878                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 879                         adapter->alloc_rx_buf(adapter, cleaned_count);
 880                         cleaned_count = 0;
 881                 }
 882
 883                 /* use prefetched values */
 884                 rx_desc = next_rxd;
 885                 buffer_info = next_buffer;
 886         }
 887         rx_ring->next_to_clean = i;
 888
 889         cleaned_count = e1000_desc_unused(rx_ring);
 890         if (cleaned_count)
 891                 adapter->alloc_rx_buf(adapter, cleaned_count);
 892
 893         adapter->total_rx_bytes += total_rx_bytes;
 894         adapter->total_rx_packets += total_rx_packets;
 895         netdev->stats.rx_bytes += total_rx_bytes;
 896         netdev->stats.rx_packets += total_rx_packets;
 897         return cleaned;
 898 }
 899
 900 static void e1000_put_txbuf(struct e1000_adapter *adapter,
 901                              struct e1000_buffer *buffer_info)
 902 {
 903         if (buffer_info->dma) {
 904                 if (buffer_info->mapped_as_page)
 905                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
 906                                        buffer_info->length, DMA_TO_DEVICE);
 907                 else
 908                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
 909                                          buffer_info->length, DMA_TO_DEVICE);
 910                 buffer_info->dma = 0;
 911         }
 912         if (buffer_info->skb) {
 913                 dev_kfree_skb_any(buffer_info->skb);
 914                 buffer_info->skb = NULL;
 915         }
 916         buffer_info->time_stamp = 0;
 917 }
 918
 919 static void e1000_print_hw_hang(struct work_struct *work)
 920 {
 921         struct e1000_adapter *adapter = container_of(work,
 922                                                      struct e1000_adapter,
 923                                                      print_hang_task);
 924         struct e1000_ring *tx_ring = adapter->tx_ring;
 925         unsigned int i = tx_ring->next_to_clean;
 926         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 927         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
 928         struct e1000_hw *hw = &adapter->hw;
 929         u16 phy_status, phy_1000t_status, phy_ext_status;
 930         u16 pci_status;
 931
 932         e1e_rphy(hw, PHY_STATUS, &phy_status);
 933         e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
 934         e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
 935
 936         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
 937
 938         /* detected Hardware unit hang */
 939         e_err("Detected Hardware Unit Hang:\n"
 940               "  TDH                  <%x>\n"
 941               "  TDT                  <%x>\n"
 942               "  next_to_use          <%x>\n"
 943               "  next_to_clean        <%x>\n"
 944               "buffer_info[next_to_clean]:\n"
 945               "  time_stamp           <%lx>\n"
 946               "  next_to_watch        <%x>\n"
 947               "  jiffies              <%lx>\n"
 948               "  next_to_watch.status <%x>\n"
 949               "MAC Status             <%x>\n"
 950               "PHY Status             <%x>\n"
 951               "PHY 1000BASE-T Status  <%x>\n"
 952               "PHY Extended Status    <%x>\n"
 953               "PCI Status             <%x>\n",
 954               readl(adapter->hw.hw_addr + tx_ring->head),
 955               readl(adapter->hw.hw_addr + tx_ring->tail),
 956               tx_ring->next_to_use,
 957               tx_ring->next_to_clean,
 958               tx_ring->buffer_info[eop].time_stamp,
 959               eop,
 960               jiffies,
 961               eop_desc->upper.fields.status,
 962               er32(STATUS),
 963               phy_status,
 964               phy_1000t_status,
 965               phy_ext_status,
 966               pci_status);
 967 }
 968
 969 /**
 970  * e1000_clean_tx_irq - Reclaim resources after transmit completes
 971  * @adapter: board private structure
 972  *
 973  * the return value indicates whether actual cleaning was done, there
 974  * is no guarantee that everything was cleaned
 975  **/
 976 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
 977 {
 978         struct net_device *netdev = adapter->netdev;
 979         struct e1000_hw *hw = &adapter->hw;
 980         struct e1000_ring *tx_ring = adapter->tx_ring;
 981         struct e1000_tx_desc *tx_desc, *eop_desc;
 982         struct e1000_buffer *buffer_info;
 983         unsigned int i, eop;
 984         unsigned int count = 0;
 985         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
 986
 987         i = tx_ring->next_to_clean;
 988         eop = tx_ring->buffer_info[i].next_to_watch;
 989         eop_desc = E1000_TX_DESC(*tx_ring, eop);
 990
 991         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
 992                (count < tx_ring->count)) {
 993                 bool cleaned = false;
 994                 for (; !cleaned; count++) {
 995                         tx_desc = E1000_TX_DESC(*tx_ring, i);
 996                         buffer_info = &tx_ring->buffer_info[i];
 997                         cleaned = (i == eop);
 998
 999                         if (cleaned) {
1000                                 total_tx_packets += buffer_info->segs;
1001                                 total_tx_bytes += buffer_info->bytecount;
1002                         }
1003
1004                         e1000_put_txbuf(adapter, buffer_info);
1005                         tx_desc->upper.data = 0;
1006
1007                         i++;
1008                         if (i == tx_ring->count)
1009                                 i = 0;
1010                 }
1011
1012                 if (i == tx_ring->next_to_use)
1013                         break;
1014                 eop = tx_ring->buffer_info[i].next_to_watch;
1015                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1016         }
1017
1018         tx_ring->next_to_clean = i;
1019
1020 #define TX_WAKE_THRESHOLD 32
1021         if (count && netif_carrier_ok(netdev) &&
1022             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1023                 /* Make sure that anybody stopping the queue after this
1024                  * sees the new next_to_clean.
1025                  */
1026                 smp_mb();
1027
1028                 if (netif_queue_stopped(netdev) &&
1029                     !(test_bit(__E1000_DOWN, &adapter->state))) {
1030                         netif_wake_queue(netdev);
1031                         ++adapter->restart_queue;
1032                 }
1033         }
1034
1035         if (adapter->detect_tx_hung) {
1036                 /*
1037                  * Detect a transmit hang in hardware, this serializes the
1038                  * check with the clearing of time_stamp and movement of i
1039                  */
1040                 adapter->detect_tx_hung = 0;
1041                 if (tx_ring->buffer_info[i].time_stamp &&
1042                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1043                                + (adapter->tx_timeout_factor * HZ)) &&
1044                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
1045                         schedule_work(&adapter->print_hang_task);
1046                         netif_stop_queue(netdev);
1047                 }
1048         }
1049         adapter->total_tx_bytes += total_tx_bytes;
1050         adapter->total_tx_packets += total_tx_packets;
1051         netdev->stats.tx_bytes += total_tx_bytes;
1052         netdev->stats.tx_packets += total_tx_packets;
1053         return (count < tx_ring->count);
1054 }
1055
1056 /**
1057  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1058  * @adapter: board private structure
1059  *
1060  * the return value indicates whether actual cleaning was done, there
1061  * is no guarantee that everything was cleaned
1062  **/
1063 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
1064                                   int *work_done, int work_to_do)
1065 {
1066         struct e1000_hw *hw = &adapter->hw;
1067         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1068         struct net_device *netdev = adapter->netdev;
1069         struct pci_dev *pdev = adapter->pdev;
1070         struct e1000_ring *rx_ring = adapter->rx_ring;
1071         struct e1000_buffer *buffer_info, *next_buffer;
1072         struct e1000_ps_page *ps_page;
1073         struct sk_buff *skb;
1074         unsigned int i, j;
1075         u32 length, staterr;
1076         int cleaned_count = 0;
1077         bool cleaned = 0;
1078         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1079
1080         i = rx_ring->next_to_clean;
1081         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1082         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1083         buffer_info = &rx_ring->buffer_info[i];
1084
1085         while (staterr & E1000_RXD_STAT_DD) {
1086                 if (*work_done >= work_to_do)
1087                         break;
1088                 (*work_done)++;
1089                 skb = buffer_info->skb;
1090
1091                 /* in the packet split case this is header only */
1092                 prefetch(skb->data - NET_IP_ALIGN);
1093
1094                 i++;
1095                 if (i == rx_ring->count)
1096                         i = 0;
1097                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1098                 prefetch(next_rxd);
1099
1100                 next_buffer = &rx_ring->buffer_info[i];
1101
1102                 cleaned = 1;
1103                 cleaned_count++;
1104                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1105                                  adapter->rx_ps_bsize0,
1106                                  DMA_FROM_DEVICE);
1107                 buffer_info->dma = 0;
1108
1109                 /* see !EOP comment in other rx routine */
1110                 if (!(staterr & E1000_RXD_STAT_EOP))
1111                         adapter->flags2 |= FLAG2_IS_DISCARDING;
1112
1113                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1114                         e_dbg("Packet Split buffers didn't pick up the full "
1115                               "packet\n");
1116                         dev_kfree_skb_irq(skb);
1117                         if (staterr & E1000_RXD_STAT_EOP)
1118                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1119                         goto next_desc;
1120                 }
1121
1122                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
1123                         dev_kfree_skb_irq(skb);
1124                         goto next_desc;
1125                 }
1126
1127                 length = le16_to_cpu(rx_desc->wb.middle.length0);
1128
1129                 if (!length) {
1130                         e_dbg("Last part of the packet spanning multiple "
1131                               "descriptors\n");
1132                         dev_kfree_skb_irq(skb);
1133                         goto next_desc;
1134                 }
1135
1136                 /* Good Receive */
1137                 skb_put(skb, length);
1138
1139                 {
1140                 /*
1141                  * this looks ugly, but it seems compiler issues make it
1142                  * more efficient than reusing j
1143                  */
1144                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1145
1146                 /*
1147                  * page alloc/put takes too long and effects small packet
1148                  * throughput, so unsplit small packets and save the alloc/put
1149                  * only valid in softirq (napi) context to call kmap_*
1150                  */
1151                 if (l1 && (l1 <= copybreak) &&
1152                     ((length + l1) <= adapter->rx_ps_bsize0)) {
1153                         u8 *vaddr;
1154
1155                         ps_page = &buffer_info->ps_pages[0];
1156
1157                         /*
1158                          * there is no documentation about how to call
1159                          * kmap_atomic, so we can't hold the mapping
1160                          * very long
1161                          */
1162                         dma_sync_single_for_cpu(&pdev->dev, ps_page->dma,
1163                                                 PAGE_SIZE, DMA_FROM_DEVICE);
1164                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
1165                         memcpy(skb_tail_pointer(skb), vaddr, l1);
1166                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
1167                         dma_sync_single_for_device(&pdev->dev, ps_page->dma,
1168                                                    PAGE_SIZE, DMA_FROM_DEVICE);
1169
1170                         /* remove the CRC */
1171                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1172                                 l1 -= 4;
1173
1174                         skb_put(skb, l1);
1175                         goto copydone;
1176                 } /* if */
1177                 }
1178
1179                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1180                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1181                         if (!length)
1182                                 break;
1183
1184                         ps_page = &buffer_info->ps_pages[j];
1185                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1186                                        DMA_FROM_DEVICE);
1187                         ps_page->dma = 0;
1188                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1189                         ps_page->page = NULL;
1190                         skb->len += length;
1191                         skb->data_len += length;
1192                         skb->truesize += length;
1193                 }
1194
1195                 /* strip the ethernet crc, problem is we're using pages now so
1196                  * this whole operation can get a little cpu intensive
1197                  */
1198                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1199                         pskb_trim(skb, skb->len - 4);
1200
1201 copydone:
1202                 total_rx_bytes += skb->len;
1203                 total_rx_packets++;
1204
1205                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
1206                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
1207
1208                 if (rx_desc->wb.upper.header_status &
1209                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1210                         adapter->rx_hdr_split++;
1211
1212                 e1000_receive_skb(adapter, netdev, skb,
1213                                   staterr, rx_desc->wb.middle.vlan);
1214
1215 next_desc:
1216                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1217                 buffer_info->skb = NULL;
1218
1219                 /* return some buffers to hardware, one at a time is too slow */
1220                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1221                         adapter->alloc_rx_buf(adapter, cleaned_count);
1222                         cleaned_count = 0;
1223                 }
1224
1225                 /* use prefetched values */
1226                 rx_desc = next_rxd;
1227                 buffer_info = next_buffer;
1228
1229                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1230         }
1231         rx_ring->next_to_clean = i;
1232
1233         cleaned_count = e1000_desc_unused(rx_ring);
1234         if (cleaned_count)
1235                 adapter->alloc_rx_buf(adapter, cleaned_count);
1236
1237         adapter->total_rx_bytes += total_rx_bytes;
1238         adapter->total_rx_packets += total_rx_packets;
1239         netdev->stats.rx_bytes += total_rx_bytes;
1240         netdev->stats.rx_packets += total_rx_packets;
1241         return cleaned;
1242 }
1243
1244 /**
1245  * e1000_consume_page - helper function
1246  **/
1247 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1248                                u16 length)
1249 {
1250         bi->page = NULL;
1251         skb->len += length;
1252         skb->data_len += length;
1253         skb->truesize += length;
1254 }
1255
1256 /**
1257  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1258  * @adapter: board private structure
1259  *
1260  * the return value indicates whether actual cleaning was done, there
1261  * is no guarantee that everything was cleaned
1262  **/
1263
1264 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
1265                                      int *work_done, int work_to_do)
1266 {
1267         struct net_device *netdev = adapter->netdev;
1268         struct pci_dev *pdev = adapter->pdev;
1269         struct e1000_ring *rx_ring = adapter->rx_ring;
1270         struct e1000_rx_desc *rx_desc, *next_rxd;
1271         struct e1000_buffer *buffer_info, *next_buffer;
1272         u32 length;
1273         unsigned int i;
1274         int cleaned_count = 0;
1275         bool cleaned = false;
1276         unsigned int total_rx_bytes=0, total_rx_packets=0;
1277
1278         i = rx_ring->next_to_clean;
1279         rx_desc = E1000_RX_DESC(*rx_ring, i);
1280         buffer_info = &rx_ring->buffer_info[i];
1281
1282         while (rx_desc->status & E1000_RXD_STAT_DD) {
1283                 struct sk_buff *skb;
1284                 u8 status;
1285
1286                 if (*work_done >= work_to_do)
1287                         break;
1288                 (*work_done)++;
1289
1290                 status = rx_desc->status;
1291                 skb = buffer_info->skb;
1292                 buffer_info->skb = NULL;
1293
1294                 ++i;
1295                 if (i == rx_ring->count)
1296                         i = 0;
1297                 next_rxd = E1000_RX_DESC(*rx_ring, i);
1298                 prefetch(next_rxd);
1299
1300                 next_buffer = &rx_ring->buffer_info[i];
1301
1302                 cleaned = true;
1303                 cleaned_count++;
1304                 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1305                                DMA_FROM_DEVICE);
1306                 buffer_info->dma = 0;
1307
1308                 length = le16_to_cpu(rx_desc->length);
1309
1310                 /* errors is only valid for DD + EOP descriptors */
1311                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
1312                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
1313                                 /* recycle both page and skb */
1314                                 buffer_info->skb = skb;
1315                                 /* an error means any chain goes out the window
1316                                  * too */
1317                                 if (rx_ring->rx_skb_top)
1318                                         dev_kfree_skb(rx_ring->rx_skb_top);
1319                                 rx_ring->rx_skb_top = NULL;
1320                                 goto next_desc;
1321                 }
1322
1323 #define rxtop rx_ring->rx_skb_top
1324                 if (!(status & E1000_RXD_STAT_EOP)) {
1325                         /* this descriptor is only the beginning (or middle) */
1326                         if (!rxtop) {
1327                                 /* this is the beginning of a chain */
1328                                 rxtop = skb;
1329                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1330                                                    0, length);
1331                         } else {
1332                                 /* this is the middle of a chain */
1333                                 skb_fill_page_desc(rxtop,
1334                                     skb_shinfo(rxtop)->nr_frags,
1335                                     buffer_info->page, 0, length);
1336                                 /* re-use the skb, only consumed the page */
1337                                 buffer_info->skb = skb;
1338                         }
1339                         e1000_consume_page(buffer_info, rxtop, length);
1340                         goto next_desc;
1341                 } else {
1342                         if (rxtop) {
1343                                 /* end of the chain */
1344                                 skb_fill_page_desc(rxtop,
1345                                     skb_shinfo(rxtop)->nr_frags,
1346                                     buffer_info->page, 0, length);
1347                                 /* re-use the current skb, we only consumed the
1348                                  * page */
1349                                 buffer_info->skb = skb;
1350                                 skb = rxtop;
1351                                 rxtop = NULL;
1352                                 e1000_consume_page(buffer_info, skb, length);
1353                         } else {
1354                                 /* no chain, got EOP, this buf is the packet
1355                                  * copybreak to save the put_page/alloc_page */
1356                                 if (length <= copybreak &&
1357                                     skb_tailroom(skb) >= length) {
1358                                         u8 *vaddr;
1359                                         vaddr = kmap_atomic(buffer_info->page,
1360                                                            KM_SKB_DATA_SOFTIRQ);
1361                                         memcpy(skb_tail_pointer(skb), vaddr,
1362                                                length);
1363                                         kunmap_atomic(vaddr,
1364                                                       KM_SKB_DATA_SOFTIRQ);
1365                                         /* re-use the page, so don't erase
1366                                          * buffer_info->page */
1367                                         skb_put(skb, length);
1368                                 } else {
1369                                         skb_fill_page_desc(skb, 0,
1370                                                            buffer_info->page, 0,
1371                                                            length);
1372                                         e1000_consume_page(buffer_info, skb,
1373                                                            length);
1374                                 }
1375                         }
1376                 }
1377
1378                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1379                 e1000_rx_checksum(adapter,
1380                                   (u32)(status) |
1381                                   ((u32)(rx_desc->errors) << 24),
1382                                   le16_to_cpu(rx_desc->csum), skb);
1383
1384                 /* probably a little skewed due to removing CRC */
1385                 total_rx_bytes += skb->len;
1386                 total_rx_packets++;
1387
1388                 /* eth type trans needs skb->data to point to something */
1389                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1390                         e_err("pskb_may_pull failed.\n");
1391                         dev_kfree_skb(skb);
1392                         goto next_desc;
1393                 }
1394
1395                 e1000_receive_skb(adapter, netdev, skb, status,
1396                                   rx_desc->special);
1397
1398 next_desc:
1399                 rx_desc->status = 0;
1400
1401                 /* return some buffers to hardware, one at a time is too slow */
1402                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1403                         adapter->alloc_rx_buf(adapter, cleaned_count);
1404                         cleaned_count = 0;
1405                 }
1406
1407                 /* use prefetched values */
1408                 rx_desc = next_rxd;
1409                 buffer_info = next_buffer;
1410         }
1411         rx_ring->next_to_clean = i;
1412
1413         cleaned_count = e1000_desc_unused(rx_ring);
1414         if (cleaned_count)
1415                 adapter->alloc_rx_buf(adapter, cleaned_count);
1416
1417         adapter->total_rx_bytes += total_rx_bytes;
1418         adapter->total_rx_packets += total_rx_packets;
1419         netdev->stats.rx_bytes += total_rx_bytes;
1420         netdev->stats.rx_packets += total_rx_packets;
1421         return cleaned;
1422 }
1423
1424 /**
1425  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1426  * @adapter: board private structure
1427  **/
1428 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1429 {
1430         struct e1000_ring *rx_ring = adapter->rx_ring;
1431         struct e1000_buffer *buffer_info;
1432         struct e1000_ps_page *ps_page;
1433         struct pci_dev *pdev = adapter->pdev;
1434         unsigned int i, j;
1435
1436         /* Free all the Rx ring sk_buffs */
1437         for (i = 0; i < rx_ring->count; i++) {
1438                 buffer_info = &rx_ring->buffer_info[i];
1439                 if (buffer_info->dma) {
1440                         if (adapter->clean_rx == e1000_clean_rx_irq)
1441                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1442                                                  adapter->rx_buffer_len,
1443                                                  DMA_FROM_DEVICE);
1444                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1445                                 dma_unmap_page(&pdev->dev, buffer_info->dma,
1446                                                PAGE_SIZE,
1447                                                DMA_FROM_DEVICE);
1448                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1449                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1450                                                  adapter->rx_ps_bsize0,
1451                                                  DMA_FROM_DEVICE);
1452                         buffer_info->dma = 0;
1453                 }
1454
1455                 if (buffer_info->page) {
1456                         put_page(buffer_info->page);
1457                         buffer_info->page = NULL;
1458                 }
1459
1460                 if (buffer_info->skb) {
1461                         dev_kfree_skb(buffer_info->skb);
1462                         buffer_info->skb = NULL;
1463                 }
1464
1465                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1466                         ps_page = &buffer_info->ps_pages[j];
1467                         if (!ps_page->page)
1468                                 break;
1469                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1470                                        DMA_FROM_DEVICE);
1471                         ps_page->dma = 0;
1472                         put_page(ps_page->page);
1473                         ps_page->page = NULL;
1474                 }
1475         }
1476
1477         /* there also may be some cached data from a chained receive */
1478         if (rx_ring->rx_skb_top) {
1479                 dev_kfree_skb(rx_ring->rx_skb_top);
1480                 rx_ring->rx_skb_top = NULL;
1481         }
1482
1483         /* Zero out the descriptor ring */
1484         memset(rx_ring->desc, 0, rx_ring->size);
1485
1486         rx_ring->next_to_clean = 0;
1487         rx_ring->next_to_use = 0;
1488         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1489
1490         writel(0, adapter->hw.hw_addr + rx_ring->head);
1491         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1492 }
1493
1494 static void e1000e_downshift_workaround(struct work_struct *work)
1495 {
1496         struct e1000_adapter *adapter = container_of(work,
1497                                         struct e1000_adapter, downshift_task);
1498
1499         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1500 }
1501
1502 /**
1503  * e1000_intr_msi - Interrupt Handler
1504  * @irq: interrupt number
1505  * @data: pointer to a network interface device structure
1506  **/
1507 static irqreturn_t e1000_intr_msi(int irq, void *data)
1508 {
1509         struct net_device *netdev = data;
1510         struct e1000_adapter *adapter = netdev_priv(netdev);
1511         struct e1000_hw *hw = &adapter->hw;
1512         u32 icr = er32(ICR);
1513
1514         /*
1515          * read ICR disables interrupts using IAM
1516          */
1517
1518         if (icr & E1000_ICR_LSC) {
1519                 hw->mac.get_link_status = 1;
1520                 /*
1521                  * ICH8 workaround-- Call gig speed drop workaround on cable
1522                  * disconnect (LSC) before accessing any PHY registers
1523                  */
1524                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1525                     (!(er32(STATUS) & E1000_STATUS_LU)))
1526                         schedule_work(&adapter->downshift_task);
1527
1528                 /*
1529                  * 80003ES2LAN workaround-- For packet buffer work-around on
1530                  * link down event; disable receives here in the ISR and reset
1531                  * adapter in watchdog
1532                  */
1533                 if (netif_carrier_ok(netdev) &&
1534                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1535                         /* disable receives */
1536                         u32 rctl = er32(RCTL);
1537                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1538                         adapter->flags |= FLAG_RX_RESTART_NOW;
1539                 }
1540                 /* guard against interrupt when we're going down */
1541                 if (!test_bit(__E1000_DOWN, &adapter->state))
1542                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1543         }
1544
1545         if (napi_schedule_prep(&adapter->napi)) {
1546                 adapter->total_tx_bytes = 0;
1547                 adapter->total_tx_packets = 0;
1548                 adapter->total_rx_bytes = 0;
1549                 adapter->total_rx_packets = 0;
1550                 __napi_schedule(&adapter->napi);
1551         }
1552
1553         return IRQ_HANDLED;
1554 }
1555
1556 /**
1557  * e1000_intr - Interrupt Handler
1558  * @irq: interrupt number
1559  * @data: pointer to a network interface device structure
1560  **/
1561 static irqreturn_t e1000_intr(int irq, void *data)
1562 {
1563         struct net_device *netdev = data;
1564         struct e1000_adapter *adapter = netdev_priv(netdev);
1565         struct e1000_hw *hw = &adapter->hw;
1566         u32 rctl, icr = er32(ICR);
1567
1568         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1569                 return IRQ_NONE;  /* Not our interrupt */
1570
1571         /*
1572          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1573          * not set, then the adapter didn't send an interrupt
1574          */
1575         if (!(icr & E1000_ICR_INT_ASSERTED))
1576                 return IRQ_NONE;
1577
1578         /*
1579          * Interrupt Auto-Mask...upon reading ICR,
1580          * interrupts are masked.  No need for the
1581          * IMC write
1582          */
1583
1584         if (icr & E1000_ICR_LSC) {
1585                 hw->mac.get_link_status = 1;
1586                 /*
1587                  * ICH8 workaround-- Call gig speed drop workaround on cable
1588                  * disconnect (LSC) before accessing any PHY registers
1589                  */
1590                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1591                     (!(er32(STATUS) & E1000_STATUS_LU)))
1592                         schedule_work(&adapter->downshift_task);
1593
1594                 /*
1595                  * 80003ES2LAN workaround--
1596                  * For packet buffer work-around on link down event;
1597                  * disable receives here in the ISR and
1598                  * reset adapter in watchdog
1599                  */
1600                 if (netif_carrier_ok(netdev) &&
1601                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1602                         /* disable receives */
1603                         rctl = er32(RCTL);
1604                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1605                         adapter->flags |= FLAG_RX_RESTART_NOW;
1606                 }
1607                 /* guard against interrupt when we're going down */
1608                 if (!test_bit(__E1000_DOWN, &adapter->state))
1609                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1610         }
1611
1612         if (napi_schedule_prep(&adapter->napi)) {
1613                 adapter->total_tx_bytes = 0;
1614                 adapter->total_tx_packets = 0;
1615                 adapter->total_rx_bytes = 0;
1616                 adapter->total_rx_packets = 0;
1617                 __napi_schedule(&adapter->napi);
1618         }
1619
1620         return IRQ_HANDLED;
1621 }
1622
1623 static irqreturn_t e1000_msix_other(int irq, void *data)
1624 {
1625         struct net_device *netdev = data;
1626         struct e1000_adapter *adapter = netdev_priv(netdev);
1627         struct e1000_hw *hw = &adapter->hw;
1628         u32 icr = er32(ICR);
1629
1630         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1631                 if (!test_bit(__E1000_DOWN, &adapter->state))
1632                         ew32(IMS, E1000_IMS_OTHER);
1633                 return IRQ_NONE;
1634         }
1635
1636         if (icr & adapter->eiac_mask)
1637                 ew32(ICS, (icr & adapter->eiac_mask));
1638
1639         if (icr & E1000_ICR_OTHER) {
1640                 if (!(icr & E1000_ICR_LSC))
1641                         goto no_link_interrupt;
1642                 hw->mac.get_link_status = 1;
1643                 /* guard against interrupt when we're going down */
1644                 if (!test_bit(__E1000_DOWN, &adapter->state))
1645                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1646         }
1647
1648 no_link_interrupt:
1649         if (!test_bit(__E1000_DOWN, &adapter->state))
1650                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1651
1652         return IRQ_HANDLED;
1653 }
1654
1655
1656 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1657 {
1658         struct net_device *netdev = data;
1659         struct e1000_adapter *adapter = netdev_priv(netdev);
1660         struct e1000_hw *hw = &adapter->hw;
1661         struct e1000_ring *tx_ring = adapter->tx_ring;
1662
1663
1664         adapter->total_tx_bytes = 0;
1665         adapter->total_tx_packets = 0;
1666
1667         if (!e1000_clean_tx_irq(adapter))
1668                 /* Ring was not completely cleaned, so fire another interrupt */
1669                 ew32(ICS, tx_ring->ims_val);
1670
1671         return IRQ_HANDLED;
1672 }
1673
1674 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1675 {
1676         struct net_device *netdev = data;
1677         struct e1000_adapter *adapter = netdev_priv(netdev);
1678
1679         /* Write the ITR value calculated at the end of the
1680          * previous interrupt.
1681          */
1682         if (adapter->rx_ring->set_itr) {
1683                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1684                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1685                 adapter->rx_ring->set_itr = 0;
1686         }
1687
1688         if (napi_schedule_prep(&adapter->napi)) {
1689                 adapter->total_rx_bytes = 0;
1690                 adapter->total_rx_packets = 0;
1691                 __napi_schedule(&adapter->napi);
1692         }
1693         return IRQ_HANDLED;
1694 }
1695
1696 /**
1697  * e1000_configure_msix - Configure MSI-X hardware
1698  *
1699  * e1000_configure_msix sets up the hardware to properly
1700  * generate MSI-X interrupts.
1701  **/
1702 static void e1000_configure_msix(struct e1000_adapter *adapter)
1703 {
1704         struct e1000_hw *hw = &adapter->hw;
1705         struct e1000_ring *rx_ring = adapter->rx_ring;
1706         struct e1000_ring *tx_ring = adapter->tx_ring;
1707         int vector = 0;
1708         u32 ctrl_ext, ivar = 0;
1709
1710         adapter->eiac_mask = 0;
1711
1712         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1713         if (hw->mac.type == e1000_82574) {
1714                 u32 rfctl = er32(RFCTL);
1715                 rfctl |= E1000_RFCTL_ACK_DIS;
1716                 ew32(RFCTL, rfctl);
1717         }
1718
1719 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1720         /* Configure Rx vector */
1721         rx_ring->ims_val = E1000_IMS_RXQ0;
1722         adapter->eiac_mask |= rx_ring->ims_val;
1723         if (rx_ring->itr_val)
1724                 writel(1000000000 / (rx_ring->itr_val * 256),
1725                        hw->hw_addr + rx_ring->itr_register);
1726         else
1727                 writel(1, hw->hw_addr + rx_ring->itr_register);
1728         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1729
1730         /* Configure Tx vector */
1731         tx_ring->ims_val = E1000_IMS_TXQ0;
1732         vector++;
1733         if (tx_ring->itr_val)
1734                 writel(1000000000 / (tx_ring->itr_val * 256),
1735                        hw->hw_addr + tx_ring->itr_register);
1736         else
1737                 writel(1, hw->hw_addr + tx_ring->itr_register);
1738         adapter->eiac_mask |= tx_ring->ims_val;
1739         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1740
1741         /* set vector for Other Causes, e.g. link changes */
1742         vector++;
1743         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1744         if (rx_ring->itr_val)
1745                 writel(1000000000 / (rx_ring->itr_val * 256),
1746                        hw->hw_addr + E1000_EITR_82574(vector));
1747         else
1748                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1749
1750         /* Cause Tx interrupts on every write back */
1751         ivar |= (1 << 31);
1752
1753         ew32(IVAR, ivar);
1754
1755         /* enable MSI-X PBA support */
1756         ctrl_ext = er32(CTRL_EXT);
1757         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1758
1759         /* Auto-Mask Other interrupts upon ICR read */
1760 #define E1000_EIAC_MASK_82574   0x01F00000
1761         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1762         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1763         ew32(CTRL_EXT, ctrl_ext);
1764         e1e_flush();
1765 }
1766
1767 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1768 {
1769         if (adapter->msix_entries) {
1770                 pci_disable_msix(adapter->pdev);
1771                 kfree(adapter->msix_entries);
1772                 adapter->msix_entries = NULL;
1773         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1774                 pci_disable_msi(adapter->pdev);
1775                 adapter->flags &= ~FLAG_MSI_ENABLED;
1776         }
1777 }
1778
1779 /**
1780  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1781  *
1782  * Attempt to configure interrupts using the best available
1783  * capabilities of the hardware and kernel.
1784  **/
1785 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1786 {
1787         int err;
1788         int numvecs, i;
1789
1790
1791         switch (adapter->int_mode) {
1792         case E1000E_INT_MODE_MSIX:
1793                 if (adapter->flags & FLAG_HAS_MSIX) {
1794                         numvecs = 3; /* RxQ0, TxQ0 and other */
1795                         adapter->msix_entries = kcalloc(numvecs,
1796                                                       sizeof(struct msix_entry),
1797                                                       GFP_KERNEL);
1798                         if (adapter->msix_entries) {
1799                                 for (i = 0; i < numvecs; i++)
1800                                         adapter->msix_entries[i].entry = i;
1801
1802                                 err = pci_enable_msix(adapter->pdev,
1803                                                       adapter->msix_entries,
1804                                                       numvecs);
1805                                 if (err == 0)
1806                                         return;
1807                         }
1808                         /* MSI-X failed, so fall through and try MSI */
1809                         e_err("Failed to initialize MSI-X interrupts.  "
1810                               "Falling back to MSI interrupts.\n");
1811                         e1000e_reset_interrupt_capability(adapter);
1812                 }
1813                 adapter->int_mode = E1000E_INT_MODE_MSI;
1814                 /* Fall through */
1815         case E1000E_INT_MODE_MSI:
1816                 if (!pci_enable_msi(adapter->pdev)) {
1817                         adapter->flags |= FLAG_MSI_ENABLED;
1818                 } else {
1819                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1820                         e_err("Failed to initialize MSI interrupts.  Falling "
1821                               "back to legacy interrupts.\n");
1822                 }
1823                 /* Fall through */
1824         case E1000E_INT_MODE_LEGACY:
1825                 /* Don't do anything; this is the system default */
1826                 break;
1827         }
1828 }
1829
1830 /**
1831  * e1000_request_msix - Initialize MSI-X interrupts
1832  *
1833  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1834  * kernel.
1835  **/
1836 static int e1000_request_msix(struct e1000_adapter *adapter)
1837 {
1838         struct net_device *netdev = adapter->netdev;
1839         int err = 0, vector = 0;
1840
1841         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1842                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1843         else
1844                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1845         err = request_irq(adapter->msix_entries[vector].vector,
1846                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1847                           netdev);
1848         if (err)
1849                 goto out;
1850         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1851         adapter->rx_ring->itr_val = adapter->itr;
1852         vector++;
1853
1854         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1855                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1856         else
1857                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1858         err = request_irq(adapter->msix_entries[vector].vector,
1859                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1860                           netdev);
1861         if (err)
1862                 goto out;
1863         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1864         adapter->tx_ring->itr_val = adapter->itr;
1865         vector++;
1866
1867         err = request_irq(adapter->msix_entries[vector].vector,
1868                           e1000_msix_other, 0, netdev->name, netdev);
1869         if (err)
1870                 goto out;
1871
1872         e1000_configure_msix(adapter);
1873         return 0;
1874 out:
1875         return err;
1876 }
1877
1878 /**
1879  * e1000_request_irq - initialize interrupts
1880  *
1881  * Attempts to configure interrupts using the best available
1882  * capabilities of the hardware and kernel.
1883  **/
1884 static int e1000_request_irq(struct e1000_adapter *adapter)
1885 {
1886         struct net_device *netdev = adapter->netdev;
1887         int err;
1888
1889         if (adapter->msix_entries) {
1890                 err = e1000_request_msix(adapter);
1891                 if (!err)
1892                         return err;
1893                 /* fall back to MSI */
1894                 e1000e_reset_interrupt_capability(adapter);
1895                 adapter->int_mode = E1000E_INT_MODE_MSI;
1896                 e1000e_set_interrupt_capability(adapter);
1897         }
1898         if (adapter->flags & FLAG_MSI_ENABLED) {
1899                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1900                                   netdev->name, netdev);
1901                 if (!err)
1902                         return err;
1903
1904                 /* fall back to legacy interrupt */
1905                 e1000e_reset_interrupt_capability(adapter);
1906                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1907         }
1908
1909         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1910                           netdev->name, netdev);
1911         if (err)
1912                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1913
1914         return err;
1915 }
1916
1917 static void e1000_free_irq(struct e1000_adapter *adapter)
1918 {
1919         struct net_device *netdev = adapter->netdev;
1920
1921         if (adapter->msix_entries) {
1922                 int vector = 0;
1923
1924                 free_irq(adapter->msix_entries[vector].vector, netdev);
1925                 vector++;
1926
1927                 free_irq(adapter->msix_entries[vector].vector, netdev);
1928                 vector++;
1929
1930                 /* Other Causes interrupt vector */
1931                 free_irq(adapter->msix_entries[vector].vector, netdev);
1932                 return;
1933         }
1934
1935         free_irq(adapter->pdev->irq, netdev);
1936 }
1937
1938 /**
1939  * e1000_irq_disable - Mask off interrupt generation on the NIC
1940  **/
1941 static void e1000_irq_disable(struct e1000_adapter *adapter)
1942 {
1943         struct e1000_hw *hw = &adapter->hw;
1944
1945         ew32(IMC, ~0);
1946         if (adapter->msix_entries)
1947                 ew32(EIAC_82574, 0);
1948         e1e_flush();
1949         synchronize_irq(adapter->pdev->irq);
1950 }
1951
1952 /**
1953  * e1000_irq_enable - Enable default interrupt generation settings
1954  **/
1955 static void e1000_irq_enable(struct e1000_adapter *adapter)
1956 {
1957         struct e1000_hw *hw = &adapter->hw;
1958
1959         if (adapter->msix_entries) {
1960                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1961                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1962         } else {
1963                 ew32(IMS, IMS_ENABLE_MASK);
1964         }
1965         e1e_flush();
1966 }
1967
1968 /**
1969  * e1000_get_hw_control - get control of the h/w from f/w
1970  * @adapter: address of board private structure
1971  *
1972  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1973  * For ASF and Pass Through versions of f/w this means that
1974  * the driver is loaded. For AMT version (only with 82573)
1975  * of the f/w this means that the network i/f is open.
1976  **/
1977 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1978 {
1979         struct e1000_hw *hw = &adapter->hw;
1980         u32 ctrl_ext;
1981         u32 swsm;
1982
1983         /* Let firmware know the driver has taken over */
1984         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1985                 swsm = er32(SWSM);
1986                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1987         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1988                 ctrl_ext = er32(CTRL_EXT);
1989                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1990         }
1991 }
1992
1993 /**
1994  * e1000_release_hw_control - release control of the h/w to f/w
1995  * @adapter: address of board private structure
1996  *
1997  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1998  * For ASF and Pass Through versions of f/w this means that the
1999  * driver is no longer loaded. For AMT version (only with 82573) i
2000  * of the f/w this means that the network i/f is closed.
2001  *
2002  **/
2003 static void e1000_release_hw_control(struct e1000_adapter *adapter)
2004 {
2005         struct e1000_hw *hw = &adapter->hw;
2006         u32 ctrl_ext;
2007         u32 swsm;
2008
2009         /* Let firmware taken over control of h/w */
2010         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2011                 swsm = er32(SWSM);
2012                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2013         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2014                 ctrl_ext = er32(CTRL_EXT);
2015                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2016         }
2017 }
2018
2019 /**
2020  * @e1000_alloc_ring - allocate memory for a ring structure
2021  **/
2022 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2023                                 struct e1000_ring *ring)
2024 {
2025         struct pci_dev *pdev = adapter->pdev;
2026
2027         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2028                                         GFP_KERNEL);
2029         if (!ring->desc)
2030                 return -ENOMEM;
2031
2032         return 0;
2033 }
2034
2035 /**
2036  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2037  * @adapter: board private structure
2038  *
2039  * Return 0 on success, negative on failure
2040  **/
2041 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
2042 {
2043         struct e1000_ring *tx_ring = adapter->tx_ring;
2044         int err = -ENOMEM, size;
2045
2046         size = sizeof(struct e1000_buffer) * tx_ring->count;
2047         tx_ring->buffer_info = vmalloc(size);
2048         if (!tx_ring->buffer_info)
2049                 goto err;
2050         memset(tx_ring->buffer_info, 0, size);
2051
2052         /* round up to nearest 4K */
2053         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2054         tx_ring->size = ALIGN(tx_ring->size, 4096);
2055
2056         err = e1000_alloc_ring_dma(adapter, tx_ring);
2057         if (err)
2058                 goto err;
2059
2060         tx_ring->next_to_use = 0;
2061         tx_ring->next_to_clean = 0;
2062
2063         return 0;
2064 err:
2065         vfree(tx_ring->buffer_info);
2066         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2067         return err;
2068 }
2069
2070 /**
2071  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2072  * @adapter: board private structure
2073  *
2074  * Returns 0 on success, negative on failure
2075  **/
2076 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
2077 {
2078         struct e1000_ring *rx_ring = adapter->rx_ring;
2079         struct e1000_buffer *buffer_info;
2080         int i, size, desc_len, err = -ENOMEM;
2081
2082         size = sizeof(struct e1000_buffer) * rx_ring->count;
2083         rx_ring->buffer_info = vmalloc(size);
2084         if (!rx_ring->buffer_info)
2085                 goto err;
2086         memset(rx_ring->buffer_info, 0, size);
2087
2088         for (i = 0; i < rx_ring->count; i++) {
2089                 buffer_info = &rx_ring->buffer_info[i];
2090                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2091                                                 sizeof(struct e1000_ps_page),
2092                                                 GFP_KERNEL);
2093                 if (!buffer_info->ps_pages)
2094                         goto err_pages;
2095         }
2096
2097         desc_len = sizeof(union e1000_rx_desc_packet_split);
2098
2099         /* Round up to nearest 4K */
2100         rx_ring->size = rx_ring->count * desc_len;
2101         rx_ring->size = ALIGN(rx_ring->size, 4096);
2102
2103         err = e1000_alloc_ring_dma(adapter, rx_ring);
2104         if (err)
2105                 goto err_pages;
2106
2107         rx_ring->next_to_clean = 0;
2108         rx_ring->next_to_use = 0;
2109         rx_ring->rx_skb_top = NULL;
2110
2111         return 0;
2112
2113 err_pages:
2114         for (i = 0; i < rx_ring->count; i++) {
2115                 buffer_info = &rx_ring->buffer_info[i];
2116                 kfree(buffer_info->ps_pages);
2117         }
2118 err:
2119         vfree(rx_ring->buffer_info);
2120         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2121         return err;
2122 }
2123
2124 /**
2125  * e1000_clean_tx_ring - Free Tx Buffers
2126  * @adapter: board private structure
2127  **/
2128 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
2129 {
2130         struct e1000_ring *tx_ring = adapter->tx_ring;
2131         struct e1000_buffer *buffer_info;
2132         unsigned long size;
2133         unsigned int i;
2134
2135         for (i = 0; i < tx_ring->count; i++) {
2136                 buffer_info = &tx_ring->buffer_info[i];
2137                 e1000_put_txbuf(adapter, buffer_info);
2138         }
2139
2140         size = sizeof(struct e1000_buffer) * tx_ring->count;
2141         memset(tx_ring->buffer_info, 0, size);
2142
2143         memset(tx_ring->desc, 0, tx_ring->size);
2144
2145         tx_ring->next_to_use = 0;
2146         tx_ring->next_to_clean = 0;
2147
2148         writel(0, adapter->hw.hw_addr + tx_ring->head);
2149         writel(0, adapter->hw.hw_addr + tx_ring->tail);
2150 }
2151
2152 /**
2153  * e1000e_free_tx_resources - Free Tx Resources per Queue
2154  * @adapter: board private structure
2155  *
2156  * Free all transmit software resources
2157  **/
2158 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
2159 {
2160         struct pci_dev *pdev = adapter->pdev;
2161         struct e1000_ring *tx_ring = adapter->tx_ring;
2162
2163         e1000_clean_tx_ring(adapter);
2164
2165         vfree(tx_ring->buffer_info);
2166         tx_ring->buffer_info = NULL;
2167
2168         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2169                           tx_ring->dma);
2170         tx_ring->desc = NULL;
2171 }
2172
2173 /**
2174  * e1000e_free_rx_resources - Free Rx Resources
2175  * @adapter: board private structure
2176  *
2177  * Free all receive software resources
2178  **/
2179
2180 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
2181 {
2182         struct pci_dev *pdev = adapter->pdev;
2183         struct e1000_ring *rx_ring = adapter->rx_ring;
2184         int i;
2185
2186         e1000_clean_rx_ring(adapter);
2187
2188         for (i = 0; i < rx_ring->count; i++) {
2189                 kfree(rx_ring->buffer_info[i].ps_pages);
2190         }
2191
2192         vfree(rx_ring->buffer_info);
2193         rx_ring->buffer_info = NULL;
2194
2195         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2196                           rx_ring->dma);
2197         rx_ring->desc = NULL;
2198 }
2199
2200 /**
2201  * e1000_update_itr - update the dynamic ITR value based on statistics
2202  * @adapter: pointer to adapter
2203  * @itr_setting: current adapter->itr
2204  * @packets: the number of packets during this measurement interval
2205  * @bytes: the number of bytes during this measurement interval
2206  *
2207  *      Stores a new ITR value based on packets and byte
2208  *      counts during the last interrupt.  The advantage of per interrupt
2209  *      computation is faster updates and more accurate ITR for the current
2210  *      traffic pattern.  Constants in this function were computed
2211  *      based on theoretical maximum wire speed and thresholds were set based
2212  *      on testing data as well as attempting to minimize response time
2213  *      while increasing bulk throughput.  This functionality is controlled
2214  *      by the InterruptThrottleRate module parameter.
2215  **/
2216 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2217                                      u16 itr_setting, int packets,
2218                                      int bytes)
2219 {
2220         unsigned int retval = itr_setting;
2221
2222         if (packets == 0)
2223                 goto update_itr_done;
2224
2225         switch (itr_setting) {
2226         case lowest_latency:
2227                 /* handle TSO and jumbo frames */
2228                 if (bytes/packets > 8000)
2229                         retval = bulk_latency;
2230                 else if ((packets < 5) && (bytes > 512)) {
2231                         retval = low_latency;
2232                 }
2233                 break;
2234         case low_latency:  /* 50 usec aka 20000 ints/s */
2235                 if (bytes > 10000) {
2236                         /* this if handles the TSO accounting */
2237                         if (bytes/packets > 8000) {
2238                                 retval = bulk_latency;
2239                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
2240                                 retval = bulk_latency;
2241                         } else if ((packets > 35)) {
2242                                 retval = lowest_latency;
2243                         }
2244                 } else if (bytes/packets > 2000) {
2245                         retval = bulk_latency;
2246                 } else if (packets <= 2 && bytes < 512) {
2247                         retval = lowest_latency;
2248                 }
2249                 break;
2250         case bulk_latency: /* 250 usec aka 4000 ints/s */
2251                 if (bytes > 25000) {
2252                         if (packets > 35) {
2253                                 retval = low_latency;
2254                         }
2255                 } else if (bytes < 6000) {
2256                         retval = low_latency;
2257                 }
2258                 break;
2259         }
2260
2261 update_itr_done:
2262         return retval;
2263 }
2264
2265 static void e1000_set_itr(struct e1000_adapter *adapter)
2266 {
2267         struct e1000_hw *hw = &adapter->hw;
2268         u16 current_itr;
2269         u32 new_itr = adapter->itr;
2270
2271         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2272         if (adapter->link_speed != SPEED_1000) {
2273                 current_itr = 0;
2274                 new_itr = 4000;
2275                 goto set_itr_now;
2276         }
2277
2278         adapter->tx_itr = e1000_update_itr(adapter,
2279                                     adapter->tx_itr,
2280                                     adapter->total_tx_packets,
2281                                     adapter->total_tx_bytes);
2282         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2283         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2284                 adapter->tx_itr = low_latency;
2285
2286         adapter->rx_itr = e1000_update_itr(adapter,
2287                                     adapter->rx_itr,
2288                                     adapter->total_rx_packets,
2289                                     adapter->total_rx_bytes);
2290         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2291         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2292                 adapter->rx_itr = low_latency;
2293
2294         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2295
2296         switch (current_itr) {
2297         /* counts and packets in update_itr are dependent on these numbers */
2298         case lowest_latency:
2299                 new_itr = 70000;
2300                 break;
2301         case low_latency:
2302                 new_itr = 20000; /* aka hwitr = ~200 */
2303                 break;
2304         case bulk_latency:
2305                 new_itr = 4000;
2306                 break;
2307         default:
2308                 break;
2309         }
2310
2311 set_itr_now:
2312         if (new_itr != adapter->itr) {
2313                 /*
2314                  * this attempts to bias the interrupt rate towards Bulk
2315                  * by adding intermediate steps when interrupt rate is
2316                  * increasing
2317                  */
2318                 new_itr = new_itr > adapter->itr ?
2319                              min(adapter->itr + (new_itr >> 2), new_itr) :
2320                              new_itr;
2321                 adapter->itr = new_itr;
2322                 adapter->rx_ring->itr_val = new_itr;
2323                 if (adapter->msix_entries)
2324                         adapter->rx_ring->set_itr = 1;
2325                 else
2326                         ew32(ITR, 1000000000 / (new_itr * 256));
2327         }
2328 }
2329
2330 /**
2331  * e1000_alloc_queues - Allocate memory for all rings
2332  * @adapter: board private structure to initialize
2333  **/
2334 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2335 {
2336         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2337         if (!adapter->tx_ring)
2338                 goto err;
2339
2340         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2341         if (!adapter->rx_ring)
2342                 goto err;
2343
2344         return 0;
2345 err:
2346         e_err("Unable to allocate memory for queues\n");
2347         kfree(adapter->rx_ring);
2348         kfree(adapter->tx_ring);
2349         return -ENOMEM;
2350 }
2351
2352 /**
2353  * e1000_clean - NAPI Rx polling callback
2354  * @napi: struct associated with this polling callback
2355  * @budget: amount of packets driver is allowed to process this poll
2356  **/
2357 static int e1000_clean(struct napi_struct *napi, int budget)
2358 {
2359         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2360         struct e1000_hw *hw = &adapter->hw;
2361         struct net_device *poll_dev = adapter->netdev;
2362         int tx_cleaned = 1, work_done = 0;
2363
2364         adapter = netdev_priv(poll_dev);
2365
2366         if (adapter->msix_entries &&
2367             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2368                 goto clean_rx;
2369
2370         tx_cleaned = e1000_clean_tx_irq(adapter);
2371
2372 clean_rx:
2373         adapter->clean_rx(adapter, &work_done, budget);
2374
2375         if (!tx_cleaned)
2376                 work_done = budget;
2377
2378         /* If budget not fully consumed, exit the polling mode */
2379         if (work_done < budget) {
2380                 if (adapter->itr_setting & 3)
2381                         e1000_set_itr(adapter);
2382                 napi_complete(napi);
2383                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2384                         if (adapter->msix_entries)
2385                                 ew32(IMS, adapter->rx_ring->ims_val);
2386                         else
2387                                 e1000_irq_enable(adapter);
2388                 }
2389         }
2390
2391         return work_done;
2392 }
2393
2394 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2395 {
2396         struct e1000_adapter *adapter = netdev_priv(netdev);
2397         struct e1000_hw *hw = &adapter->hw;
2398         u32 vfta, index;
2399
2400         /* don't update vlan cookie if already programmed */
2401         if ((adapter->hw.mng_cookie.status &
2402              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2403             (vid == adapter->mng_vlan_id))
2404                 return;
2405
2406         /* add VID to filter table */
2407         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2408                 index = (vid >> 5) & 0x7F;
2409                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2410                 vfta |= (1 << (vid & 0x1F));
2411                 hw->mac.ops.write_vfta(hw, index, vfta);
2412         }
2413 }
2414
2415 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2416 {
2417         struct e1000_adapter *adapter = netdev_priv(netdev);
2418         struct e1000_hw *hw = &adapter->hw;
2419         u32 vfta, index;
2420
2421         if (!test_bit(__E1000_DOWN, &adapter->state))
2422                 e1000_irq_disable(adapter);
2423         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2424
2425         if (!test_bit(__E1000_DOWN, &adapter->state))
2426                 e1000_irq_enable(adapter);
2427
2428         if ((adapter->hw.mng_cookie.status &
2429              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2430             (vid == adapter->mng_vlan_id)) {
2431                 /* release control to f/w */
2432                 e1000_release_hw_control(adapter);
2433                 return;
2434         }
2435
2436         /* remove VID from filter table */
2437         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2438                 index = (vid >> 5) & 0x7F;
2439                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2440                 vfta &= ~(1 << (vid & 0x1F));
2441                 hw->mac.ops.write_vfta(hw, index, vfta);
2442         }
2443 }
2444
2445 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2446 {
2447         struct net_device *netdev = adapter->netdev;
2448         u16 vid = adapter->hw.mng_cookie.vlan_id;
2449         u16 old_vid = adapter->mng_vlan_id;
2450
2451         if (!adapter->vlgrp)
2452                 return;
2453
2454         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2455                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2456                 if (adapter->hw.mng_cookie.status &
2457                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2458                         e1000_vlan_rx_add_vid(netdev, vid);
2459                         adapter->mng_vlan_id = vid;
2460                 }
2461
2462                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2463                                 (vid != old_vid) &&
2464                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2465                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2466         } else {
2467                 adapter->mng_vlan_id = vid;
2468         }
2469 }
2470
2471
2472 static void e1000_vlan_rx_register(struct net_device *netdev,
2473                                    struct vlan_group *grp)
2474 {
2475         struct e1000_adapter *adapter = netdev_priv(netdev);
2476         struct e1000_hw *hw = &adapter->hw;
2477         u32 ctrl, rctl;
2478
2479         if (!test_bit(__E1000_DOWN, &adapter->state))
2480                 e1000_irq_disable(adapter);
2481         adapter->vlgrp = grp;
2482
2483         if (grp) {
2484                 /* enable VLAN tag insert/strip */
2485                 ctrl = er32(CTRL);
2486                 ctrl |= E1000_CTRL_VME;
2487                 ew32(CTRL, ctrl);
2488
2489                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2490                         /* enable VLAN receive filtering */
2491                         rctl = er32(RCTL);
2492                         rctl &= ~E1000_RCTL_CFIEN;
2493                         ew32(RCTL, rctl);
2494                         e1000_update_mng_vlan(adapter);
2495                 }
2496         } else {
2497                 /* disable VLAN tag insert/strip */
2498                 ctrl = er32(CTRL);
2499                 ctrl &= ~E1000_CTRL_VME;
2500                 ew32(CTRL, ctrl);
2501
2502                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2503                         if (adapter->mng_vlan_id !=
2504                             (u16)E1000_MNG_VLAN_NONE) {
2505                                 e1000_vlan_rx_kill_vid(netdev,
2506                                                        adapter->mng_vlan_id);
2507                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2508                         }
2509                 }
2510         }
2511
2512         if (!test_bit(__E1000_DOWN, &adapter->state))
2513                 e1000_irq_enable(adapter);
2514 }
2515
2516 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2517 {
2518         u16 vid;
2519
2520         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2521
2522         if (!adapter->vlgrp)
2523                 return;
2524
2525         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2526                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2527                         continue;
2528                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2529         }
2530 }
2531
2532 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2533 {
2534         struct e1000_hw *hw = &adapter->hw;
2535         u32 manc, manc2h, mdef, i, j;
2536
2537         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2538                 return;
2539
2540         manc = er32(MANC);
2541
2542         /*
2543          * enable receiving management packets to the host. this will probably
2544          * generate destination unreachable messages from the host OS, but
2545          * the packets will be handled on SMBUS
2546          */
2547         manc |= E1000_MANC_EN_MNG2HOST;
2548         manc2h = er32(MANC2H);
2549
2550         switch (hw->mac.type) {
2551         default:
2552                 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2553                 break;
2554         case e1000_82574:
2555         case e1000_82583:
2556                 /*
2557                  * Check if IPMI pass-through decision filter already exists;
2558                  * if so, enable it.
2559                  */
2560                 for (i = 0, j = 0; i < 8; i++) {
2561                         mdef = er32(MDEF(i));
2562
2563                         /* Ignore filters with anything other than IPMI ports */
2564                         if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2565                                 continue;
2566
2567                         /* Enable this decision filter in MANC2H */
2568                         if (mdef)
2569                                 manc2h |= (1 << i);
2570
2571                         j |= mdef;
2572                 }
2573
2574                 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2575                         break;
2576
2577                 /* Create new decision filter in an empty filter */
2578                 for (i = 0, j = 0; i < 8; i++)
2579                         if (er32(MDEF(i)) == 0) {
2580                                 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2581                                                E1000_MDEF_PORT_664));
2582                                 manc2h |= (1 << 1);
2583                                 j++;
2584                                 break;
2585                         }
2586
2587                 if (!j)
2588                         e_warn("Unable to create IPMI pass-through filter\n");
2589                 break;
2590         }
2591
2592         ew32(MANC2H, manc2h);
2593         ew32(MANC, manc);
2594 }
2595
2596 /**
2597  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2598  * @adapter: board private structure
2599  *
2600  * Configure the Tx unit of the MAC after a reset.
2601  **/
2602 static void e1000_configure_tx(struct e1000_adapter *adapter)
2603 {
2604         struct e1000_hw *hw = &adapter->hw;
2605         struct e1000_ring *tx_ring = adapter->tx_ring;
2606         u64 tdba;
2607         u32 tdlen, tctl, tipg, tarc;
2608         u32 ipgr1, ipgr2;
2609
2610         /* Setup the HW Tx Head and Tail descriptor pointers */
2611         tdba = tx_ring->dma;
2612         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2613         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2614         ew32(TDBAH, (tdba >> 32));
2615         ew32(TDLEN, tdlen);
2616         ew32(TDH, 0);
2617         ew32(TDT, 0);
2618         tx_ring->head = E1000_TDH;
2619         tx_ring->tail = E1000_TDT;
2620
2621         /* Set the default values for the Tx Inter Packet Gap timer */
2622         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2623         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2624         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2625
2626         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2627                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2628
2629         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2630         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2631         ew32(TIPG, tipg);
2632
2633         /* Set the Tx Interrupt Delay register */
2634         ew32(TIDV, adapter->tx_int_delay);
2635         /* Tx irq moderation */
2636         ew32(TADV, adapter->tx_abs_int_delay);
2637
2638         /* Program the Transmit Control Register */
2639         tctl = er32(TCTL);
2640         tctl &= ~E1000_TCTL_CT;
2641         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2642                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2643
2644         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2645                 tarc = er32(TARC(0));
2646                 /*
2647                  * set the speed mode bit, we'll clear it if we're not at
2648                  * gigabit link later
2649                  */
2650 #define SPEED_MODE_BIT (1 << 21)
2651                 tarc |= SPEED_MODE_BIT;
2652                 ew32(TARC(0), tarc);
2653         }
2654
2655         /* errata: program both queues to unweighted RR */
2656         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2657                 tarc = er32(TARC(0));
2658                 tarc |= 1;
2659                 ew32(TARC(0), tarc);
2660                 tarc = er32(TARC(1));
2661                 tarc |= 1;
2662                 ew32(TARC(1), tarc);
2663         }
2664
2665         /* Setup Transmit Descriptor Settings for eop descriptor */
2666         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2667
2668         /* only set IDE if we are delaying interrupts using the timers */
2669         if (adapter->tx_int_delay)
2670                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2671
2672         /* enable Report Status bit */
2673         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2674
2675         ew32(TCTL, tctl);
2676
2677         e1000e_config_collision_dist(hw);
2678 }
2679
2680 /**
2681  * e1000_setup_rctl - configure the receive control registers
2682  * @adapter: Board private structure
2683  **/
2684 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2685                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2686 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2687 {
2688         struct e1000_hw *hw = &adapter->hw;
2689         u32 rctl, rfctl;
2690         u32 psrctl = 0;
2691         u32 pages = 0;
2692
2693         /* Program MC offset vector base */
2694         rctl = er32(RCTL);
2695         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2696         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2697                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2698                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2699
2700         /* Do not Store bad packets */
2701         rctl &= ~E1000_RCTL_SBP;
2702
2703         /* Enable Long Packet receive */
2704         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2705                 rctl &= ~E1000_RCTL_LPE;
2706         else
2707                 rctl |= E1000_RCTL_LPE;
2708
2709         /* Some systems expect that the CRC is included in SMBUS traffic. The
2710          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2711          * host memory when this is enabled
2712          */
2713         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2714                 rctl |= E1000_RCTL_SECRC;
2715
2716         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2717         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2718                 u16 phy_data;
2719
2720                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2721                 phy_data &= 0xfff8;
2722                 phy_data |= (1 << 2);
2723                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2724
2725                 e1e_rphy(hw, 22, &phy_data);
2726                 phy_data &= 0x0fff;
2727                 phy_data |= (1 << 14);
2728                 e1e_wphy(hw, 0x10, 0x2823);
2729                 e1e_wphy(hw, 0x11, 0x0003);
2730                 e1e_wphy(hw, 22, phy_data);
2731         }
2732
2733         /* Workaround Si errata on 82579 - configure jumbo frame flow */
2734         if (hw->mac.type == e1000_pch2lan) {
2735                 s32 ret_val;
2736
2737                 if (rctl & E1000_RCTL_LPE)
2738                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2739                 else
2740                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2741         }
2742
2743         /* Setup buffer sizes */
2744         rctl &= ~E1000_RCTL_SZ_4096;
2745         rctl |= E1000_RCTL_BSEX;
2746         switch (adapter->rx_buffer_len) {
2747         case 2048:
2748         default:
2749                 rctl |= E1000_RCTL_SZ_2048;
2750                 rctl &= ~E1000_RCTL_BSEX;
2751                 break;
2752         case 4096:
2753                 rctl |= E1000_RCTL_SZ_4096;
2754                 break;
2755         case 8192:
2756                 rctl |= E1000_RCTL_SZ_8192;
2757                 break;
2758         case 16384:
2759                 rctl |= E1000_RCTL_SZ_16384;
2760                 break;
2761         }
2762
2763         /*
2764          * 82571 and greater support packet-split where the protocol
2765          * header is placed in skb->data and the packet data is
2766          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2767          * In the case of a non-split, skb->data is linearly filled,
2768          * followed by the page buffers.  Therefore, skb->data is
2769          * sized to hold the largest protocol header.
2770          *
2771          * allocations using alloc_page take too long for regular MTU
2772          * so only enable packet split for jumbo frames
2773          *
2774          * Using pages when the page size is greater than 16k wastes
2775          * a lot of memory, since we allocate 3 pages at all times
2776          * per packet.
2777          */
2778         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2779         if (!(adapter->flags & FLAG_HAS_ERT) && (pages <= 3) &&
2780             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2781                 adapter->rx_ps_pages = pages;
2782         else
2783                 adapter->rx_ps_pages = 0;
2784
2785         if (adapter->rx_ps_pages) {
2786                 /* Configure extra packet-split registers */
2787                 rfctl = er32(RFCTL);
2788                 rfctl |= E1000_RFCTL_EXTEN;
2789                 /*
2790                  * disable packet split support for IPv6 extension headers,
2791                  * because some malformed IPv6 headers can hang the Rx
2792                  */
2793                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2794                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2795
2796                 ew32(RFCTL, rfctl);
2797
2798                 /* Enable Packet split descriptors */
2799                 rctl |= E1000_RCTL_DTYP_PS;
2800
2801                 psrctl |= adapter->rx_ps_bsize0 >>
2802                         E1000_PSRCTL_BSIZE0_SHIFT;
2803
2804                 switch (adapter->rx_ps_pages) {
2805                 case 3:
2806                         psrctl |= PAGE_SIZE <<
2807                                 E1000_PSRCTL_BSIZE3_SHIFT;
2808                 case 2:
2809                         psrctl |= PAGE_SIZE <<
2810                                 E1000_PSRCTL_BSIZE2_SHIFT;
2811                 case 1:
2812                         psrctl |= PAGE_SIZE >>
2813                                 E1000_PSRCTL_BSIZE1_SHIFT;
2814                         break;
2815                 }
2816
2817                 ew32(PSRCTL, psrctl);
2818         }
2819
2820         ew32(RCTL, rctl);
2821         /* just started the receive unit, no need to restart */
2822         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2823 }
2824
2825 /**
2826  * e1000_configure_rx - Configure Receive Unit after Reset
2827  * @adapter: board private structure
2828  *
2829  * Configure the Rx unit of the MAC after a reset.
2830  **/
2831 static void e1000_configure_rx(struct e1000_adapter *adapter)
2832 {
2833         struct e1000_hw *hw = &adapter->hw;
2834         struct e1000_ring *rx_ring = adapter->rx_ring;
2835         u64 rdba;
2836         u32 rdlen, rctl, rxcsum, ctrl_ext;
2837
2838         if (adapter->rx_ps_pages) {
2839                 /* this is a 32 byte descriptor */
2840                 rdlen = rx_ring->count *
2841                         sizeof(union e1000_rx_desc_packet_split);
2842                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2843                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2844         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2845                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2846                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2847                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2848         } else {
2849                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2850                 adapter->clean_rx = e1000_clean_rx_irq;
2851                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2852         }
2853
2854         /* disable receives while setting up the descriptors */
2855         rctl = er32(RCTL);
2856         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2857         e1e_flush();
2858         msleep(10);
2859
2860         /* set the Receive Delay Timer Register */
2861         ew32(RDTR, adapter->rx_int_delay);
2862
2863         /* irq moderation */
2864         ew32(RADV, adapter->rx_abs_int_delay);
2865         if (adapter->itr_setting != 0)
2866                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2867
2868         ctrl_ext = er32(CTRL_EXT);
2869         /* Auto-Mask interrupts upon ICR access */
2870         ctrl_ext |= E1000_CTRL_EXT_IAME;
2871         ew32(IAM, 0xffffffff);
2872         ew32(CTRL_EXT, ctrl_ext);
2873         e1e_flush();
2874
2875         /*
2876          * Setup the HW Rx Head and Tail Descriptor Pointers and
2877          * the Base and Length of the Rx Descriptor Ring
2878          */
2879         rdba = rx_ring->dma;
2880         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2881         ew32(RDBAH, (rdba >> 32));
2882         ew32(RDLEN, rdlen);
2883         ew32(RDH, 0);
2884         ew32(RDT, 0);
2885         rx_ring->head = E1000_RDH;
2886         rx_ring->tail = E1000_RDT;
2887
2888         /* Enable Receive Checksum Offload for TCP and UDP */
2889         rxcsum = er32(RXCSUM);
2890         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2891                 rxcsum |= E1000_RXCSUM_TUOFL;
2892
2893                 /*
2894                  * IPv4 payload checksum for UDP fragments must be
2895                  * used in conjunction with packet-split.
2896                  */
2897                 if (adapter->rx_ps_pages)
2898                         rxcsum |= E1000_RXCSUM_IPPCSE;
2899         } else {
2900                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2901                 /* no need to clear IPPCSE as it defaults to 0 */
2902         }
2903         ew32(RXCSUM, rxcsum);
2904
2905         /*
2906          * Enable early receives on supported devices, only takes effect when
2907          * packet size is equal or larger than the specified value (in 8 byte
2908          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2909          */
2910         if (adapter->flags & FLAG_HAS_ERT) {
2911                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2912                         u32 rxdctl = er32(RXDCTL(0));
2913                         ew32(RXDCTL(0), rxdctl | 0x3);
2914                         ew32(ERT, E1000_ERT_2048 | (1 << 13));
2915                         /*
2916                          * With jumbo frames and early-receive enabled,
2917                          * excessive C-state transition latencies result in
2918                          * dropped transactions.
2919                          */
2920                         pm_qos_update_request(
2921                                 adapter->netdev->pm_qos_req, 55);
2922                 } else {
2923                         pm_qos_update_request(
2924                                 adapter->netdev->pm_qos_req,
2925                                 PM_QOS_DEFAULT_VALUE);
2926                 }
2927         }
2928
2929         /* Enable Receives */
2930         ew32(RCTL, rctl);
2931 }
2932
2933 /**
2934  *  e1000_update_mc_addr_list - Update Multicast addresses
2935  *  @hw: pointer to the HW structure
2936  *  @mc_addr_list: array of multicast addresses to program
2937  *  @mc_addr_count: number of multicast addresses to program
2938  *
2939  *  Updates the Multicast Table Array.
2940  *  The caller must have a packed mc_addr_list of multicast addresses.
2941  **/
2942 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2943                                       u32 mc_addr_count)
2944 {
2945         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
2946 }
2947
2948 /**
2949  * e1000_set_multi - Multicast and Promiscuous mode set
2950  * @netdev: network interface device structure
2951  *
2952  * The set_multi entry point is called whenever the multicast address
2953  * list or the network interface flags are updated.  This routine is
2954  * responsible for configuring the hardware for proper multicast,
2955  * promiscuous mode, and all-multi behavior.
2956  **/
2957 static void e1000_set_multi(struct net_device *netdev)
2958 {
2959         struct e1000_adapter *adapter = netdev_priv(netdev);
2960         struct e1000_hw *hw = &adapter->hw;
2961         struct netdev_hw_addr *ha;
2962         u8  *mta_list;
2963         u32 rctl;
2964         int i;
2965
2966         /* Check for Promiscuous and All Multicast modes */
2967
2968         rctl = er32(RCTL);
2969
2970         if (netdev->flags & IFF_PROMISC) {
2971                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2972                 rctl &= ~E1000_RCTL_VFE;
2973         } else {
2974                 if (netdev->flags & IFF_ALLMULTI) {
2975                         rctl |= E1000_RCTL_MPE;
2976                         rctl &= ~E1000_RCTL_UPE;
2977                 } else {
2978                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2979                 }
2980                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2981                         rctl |= E1000_RCTL_VFE;
2982         }
2983
2984         ew32(RCTL, rctl);
2985
2986         if (!netdev_mc_empty(netdev)) {
2987                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
2988                 if (!mta_list)
2989                         return;
2990
2991                 /* prepare a packed array of only addresses. */
2992                 i = 0;
2993                 netdev_for_each_mc_addr(ha, netdev)
2994                         memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
2995
2996                 e1000_update_mc_addr_list(hw, mta_list, i);
2997                 kfree(mta_list);
2998         } else {
2999                 /*
3000                  * if we're called from probe, we might not have
3001                  * anything to do here, so clear out the list
3002                  */
3003                 e1000_update_mc_addr_list(hw, NULL, 0);
3004         }
3005 }
3006
3007 /**
3008  * e1000_configure - configure the hardware for Rx and Tx
3009  * @adapter: private board structure
3010  **/
3011 static void e1000_configure(struct e1000_adapter *adapter)
3012 {
3013         e1000_set_multi(adapter->netdev);
3014
3015         e1000_restore_vlan(adapter);
3016         e1000_init_manageability_pt(adapter);
3017
3018         e1000_configure_tx(adapter);
3019         e1000_setup_rctl(adapter);
3020         e1000_configure_rx(adapter);
3021         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
3022 }
3023
3024 /**
3025  * e1000e_power_up_phy - restore link in case the phy was powered down
3026  * @adapter: address of board private structure
3027  *
3028  * The phy may be powered down to save power and turn off link when the
3029  * driver is unloaded and wake on lan is not enabled (among others)
3030  * *** this routine MUST be followed by a call to e1000e_reset ***
3031  **/
3032 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3033 {
3034         if (adapter->hw.phy.ops.power_up)
3035                 adapter->hw.phy.ops.power_up(&adapter->hw);
3036
3037         adapter->hw.mac.ops.setup_link(&adapter->hw);
3038 }
3039
3040 /**
3041  * e1000_power_down_phy - Power down the PHY
3042  *
3043  * Power down the PHY so no link is implied when interface is down.
3044  * The PHY cannot be powered down if management or WoL is active.
3045  */
3046 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3047 {
3048         /* WoL is enabled */
3049         if (adapter->wol)
3050                 return;
3051
3052         if (adapter->hw.phy.ops.power_down)
3053                 adapter->hw.phy.ops.power_down(&adapter->hw);
3054 }
3055
3056 /**
3057  * e1000e_reset - bring the hardware into a known good state
3058  *
3059  * This function boots the hardware and enables some settings that
3060  * require a configuration cycle of the hardware - those cannot be
3061  * set/changed during runtime. After reset the device needs to be
3062  * properly configured for Rx, Tx etc.
3063  */
3064 void e1000e_reset(struct e1000_adapter *adapter)
3065 {
3066         struct e1000_mac_info *mac = &adapter->hw.mac;
3067         struct e1000_fc_info *fc = &adapter->hw.fc;
3068         struct e1000_hw *hw = &adapter->hw;
3069         u32 tx_space, min_tx_space, min_rx_space;
3070         u32 pba = adapter->pba;
3071         u16 hwm;
3072
3073         /* reset Packet Buffer Allocation to default */
3074         ew32(PBA, pba);
3075
3076         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3077                 /*
3078                  * To maintain wire speed transmits, the Tx FIFO should be
3079                  * large enough to accommodate two full transmit packets,
3080                  * rounded up to the next 1KB and expressed in KB.  Likewise,
3081                  * the Rx FIFO should be large enough to accommodate at least
3082                  * one full receive packet and is similarly rounded up and
3083                  * expressed in KB.
3084                  */
3085                 pba = er32(PBA);
3086                 /* upper 16 bits has Tx packet buffer allocation size in KB */
3087                 tx_space = pba >> 16;
3088                 /* lower 16 bits has Rx packet buffer allocation size in KB */
3089                 pba &= 0xffff;
3090                 /*
3091                  * the Tx fifo also stores 16 bytes of information about the tx
3092                  * but don't include ethernet FCS because hardware appends it
3093                  */
3094                 min_tx_space = (adapter->max_frame_size +
3095                                 sizeof(struct e1000_tx_desc) -
3096                                 ETH_FCS_LEN) * 2;
3097                 min_tx_space = ALIGN(min_tx_space, 1024);
3098                 min_tx_space >>= 10;
3099                 /* software strips receive CRC, so leave room for it */
3100                 min_rx_space = adapter->max_frame_size;
3101                 min_rx_space = ALIGN(min_rx_space, 1024);
3102                 min_rx_space >>= 10;
3103
3104                 /*
3105                  * If current Tx allocation is less than the min Tx FIFO size,
3106                  * and the min Tx FIFO size is less than the current Rx FIFO
3107                  * allocation, take space away from current Rx allocation
3108                  */
3109                 if ((tx_space < min_tx_space) &&
3110                     ((min_tx_space - tx_space) < pba)) {
3111                         pba -= min_tx_space - tx_space;
3112
3113                         /*
3114                          * if short on Rx space, Rx wins and must trump tx
3115                          * adjustment or use Early Receive if available
3116                          */
3117                         if ((pba < min_rx_space) &&
3118                             (!(adapter->flags & FLAG_HAS_ERT)))
3119                                 /* ERT enabled in e1000_configure_rx */
3120                                 pba = min_rx_space;
3121                 }
3122
3123                 ew32(PBA, pba);
3124         }
3125
3126
3127         /*
3128          * flow control settings
3129          *
3130          * The high water mark must be low enough to fit one full frame
3131          * (or the size used for early receive) above it in the Rx FIFO.
3132          * Set it to the lower of:
3133          * - 90% of the Rx FIFO size, and
3134          * - the full Rx FIFO size minus the early receive size (for parts
3135          *   with ERT support assuming ERT set to E1000_ERT_2048), or
3136          * - the full Rx FIFO size minus one full frame
3137          */
3138         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3139                 fc->pause_time = 0xFFFF;
3140         else
3141                 fc->pause_time = E1000_FC_PAUSE_TIME;
3142         fc->send_xon = 1;
3143         fc->current_mode = fc->requested_mode;
3144
3145         switch (hw->mac.type) {
3146         default:
3147                 if ((adapter->flags & FLAG_HAS_ERT) &&
3148                     (adapter->netdev->mtu > ETH_DATA_LEN))
3149                         hwm = min(((pba << 10) * 9 / 10),
3150                                   ((pba << 10) - (E1000_ERT_2048 << 3)));
3151                 else
3152                         hwm = min(((pba << 10) * 9 / 10),
3153                                   ((pba << 10) - adapter->max_frame_size));
3154
3155                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3156                 fc->low_water = fc->high_water - 8;
3157                 break;
3158         case e1000_pchlan:
3159                 /*
3160                  * Workaround PCH LOM adapter hangs with certain network
3161                  * loads.  If hangs persist, try disabling Tx flow control.
3162                  */
3163                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3164                         fc->high_water = 0x3500;
3165                         fc->low_water  = 0x1500;
3166                 } else {
3167                         fc->high_water = 0x5000;
3168                         fc->low_water  = 0x3000;
3169                 }
3170                 fc->refresh_time = 0x1000;
3171                 break;
3172         case e1000_pch2lan:
3173                 fc->high_water = 0x05C20;
3174                 fc->low_water = 0x05048;
3175                 fc->pause_time = 0x0650;
3176                 fc->refresh_time = 0x0400;
3177                 break;
3178         }
3179
3180         /* Allow time for pending master requests to run */
3181         mac->ops.reset_hw(hw);
3182
3183         /*
3184          * For parts with AMT enabled, let the firmware know
3185          * that the network interface is in control
3186          */
3187         if (adapter->flags & FLAG_HAS_AMT)
3188                 e1000_get_hw_control(adapter);
3189
3190         ew32(WUC, 0);
3191
3192         if (mac->ops.init_hw(hw))
3193                 e_err("Hardware Error\n");
3194
3195         e1000_update_mng_vlan(adapter);
3196
3197         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3198         ew32(VET, ETH_P_8021Q);
3199
3200         e1000e_reset_adaptive(hw);
3201         e1000_get_phy_info(hw);
3202
3203         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3204             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3205                 u16 phy_data = 0;
3206                 /*
3207                  * speed up time to link by disabling smart power down, ignore
3208                  * the return value of this function because there is nothing
3209                  * different we would do if it failed
3210                  */
3211                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3212                 phy_data &= ~IGP02E1000_PM_SPD;
3213                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3214         }
3215 }
3216
3217 int e1000e_up(struct e1000_adapter *adapter)
3218 {
3219         struct e1000_hw *hw = &adapter->hw;
3220
3221         /* DMA latency requirement to workaround early-receive/jumbo issue */
3222         if (adapter->flags & FLAG_HAS_ERT)
3223                 adapter->netdev->pm_qos_req =
3224                         pm_qos_add_request(PM_QOS_CPU_DMA_LATENCY,
3225                                        PM_QOS_DEFAULT_VALUE);
3226
3227         /* hardware has been reset, we need to reload some things */
3228         e1000_configure(adapter);
3229
3230         clear_bit(__E1000_DOWN, &adapter->state);
3231
3232         napi_enable(&adapter->napi);
3233         if (adapter->msix_entries)
3234                 e1000_configure_msix(adapter);
3235         e1000_irq_enable(adapter);
3236
3237         netif_wake_queue(adapter->netdev);
3238
3239         /* fire a link change interrupt to start the watchdog */
3240         if (adapter->msix_entries)
3241                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3242         else
3243                 ew32(ICS, E1000_ICS_LSC);
3244
3245         return 0;
3246 }
3247
3248 void e1000e_down(struct e1000_adapter *adapter)
3249 {
3250         struct net_device *netdev = adapter->netdev;
3251         struct e1000_hw *hw = &adapter->hw;
3252         u32 tctl, rctl;
3253
3254         /*
3255          * signal that we're down so the interrupt handler does not
3256          * reschedule our watchdog timer
3257          */
3258         set_bit(__E1000_DOWN, &adapter->state);
3259
3260         /* disable receives in the hardware */
3261         rctl = er32(RCTL);
3262         ew32(RCTL, rctl & ~E1000_RCTL_EN);
3263         /* flush and sleep below */
3264
3265         netif_stop_queue(netdev);
3266
3267         /* disable transmits in the hardware */
3268         tctl = er32(TCTL);
3269         tctl &= ~E1000_TCTL_EN;
3270         ew32(TCTL, tctl);
3271         /* flush both disables and wait for them to finish */
3272         e1e_flush();
3273         msleep(10);
3274
3275         napi_disable(&adapter->napi);
3276         e1000_irq_disable(adapter);
3277
3278         del_timer_sync(&adapter->watchdog_timer);
3279         del_timer_sync(&adapter->phy_info_timer);
3280
3281         netif_carrier_off(netdev);
3282         adapter->link_speed = 0;
3283         adapter->link_duplex = 0;
3284
3285         if (!pci_channel_offline(adapter->pdev))
3286                 e1000e_reset(adapter);
3287         e1000_clean_tx_ring(adapter);
3288         e1000_clean_rx_ring(adapter);
3289
3290         if (adapter->flags & FLAG_HAS_ERT) {
3291                 pm_qos_remove_request(
3292                               adapter->netdev->pm_qos_req);
3293                 adapter->netdev->pm_qos_req = NULL;
3294         }
3295
3296         /*
3297          * TODO: for power management, we could drop the link and
3298          * pci_disable_device here.
3299          */
3300 }
3301
3302 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3303 {
3304         might_sleep();
3305         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3306                 msleep(1);
3307         e1000e_down(adapter);
3308         e1000e_up(adapter);
3309         clear_bit(__E1000_RESETTING, &adapter->state);
3310 }
3311
3312 /**
3313  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3314  * @adapter: board private structure to initialize
3315  *
3316  * e1000_sw_init initializes the Adapter private data structure.
3317  * Fields are initialized based on PCI device information and
3318  * OS network device settings (MTU size).
3319  **/
3320 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3321 {
3322         struct net_device *netdev = adapter->netdev;
3323
3324         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3325         adapter->rx_ps_bsize0 = 128;
3326         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3327         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3328
3329         e1000e_set_interrupt_capability(adapter);
3330
3331         if (e1000_alloc_queues(adapter))
3332                 return -ENOMEM;
3333
3334         /* Explicitly disable IRQ since the NIC can be in any state. */
3335         e1000_irq_disable(adapter);
3336
3337         set_bit(__E1000_DOWN, &adapter->state);
3338         return 0;
3339 }
3340
3341 /**
3342  * e1000_intr_msi_test - Interrupt Handler
3343  * @irq: interrupt number
3344  * @data: pointer to a network interface device structure
3345  **/
3346 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3347 {
3348         struct net_device *netdev = data;
3349         struct e1000_adapter *adapter = netdev_priv(netdev);
3350         struct e1000_hw *hw = &adapter->hw;
3351         u32 icr = er32(ICR);
3352
3353         e_dbg("icr is %08X\n", icr);
3354         if (icr & E1000_ICR_RXSEQ) {
3355                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3356                 wmb();
3357         }
3358
3359         return IRQ_HANDLED;
3360 }
3361
3362 /**
3363  * e1000_test_msi_interrupt - Returns 0 for successful test
3364  * @adapter: board private struct
3365  *
3366  * code flow taken from tg3.c
3367  **/
3368 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3369 {
3370         struct net_device *netdev = adapter->netdev;
3371         struct e1000_hw *hw = &adapter->hw;
3372         int err;
3373
3374         /* poll_enable hasn't been called yet, so don't need disable */
3375         /* clear any pending events */
3376         er32(ICR);
3377
3378         /* free the real vector and request a test handler */
3379         e1000_free_irq(adapter);
3380         e1000e_reset_interrupt_capability(adapter);
3381
3382         /* Assume that the test fails, if it succeeds then the test
3383          * MSI irq handler will unset this flag */
3384         adapter->flags |= FLAG_MSI_TEST_FAILED;
3385
3386         err = pci_enable_msi(adapter->pdev);
3387         if (err)
3388                 goto msi_test_failed;
3389
3390         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3391                           netdev->name, netdev);
3392         if (err) {
3393                 pci_disable_msi(adapter->pdev);
3394                 goto msi_test_failed;
3395         }
3396
3397         wmb();
3398
3399         e1000_irq_enable(adapter);
3400
3401         /* fire an unusual interrupt on the test handler */
3402         ew32(ICS, E1000_ICS_RXSEQ);
3403         e1e_flush();
3404         msleep(50);
3405
3406         e1000_irq_disable(adapter);
3407
3408         rmb();
3409
3410         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3411                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3412                 err = -EIO;
3413                 e_info("MSI interrupt test failed!\n");
3414         }
3415
3416         free_irq(adapter->pdev->irq, netdev);
3417         pci_disable_msi(adapter->pdev);
3418
3419         if (err == -EIO)
3420                 goto msi_test_failed;
3421
3422         /* okay so the test worked, restore settings */
3423         e_dbg("MSI interrupt test succeeded!\n");
3424 msi_test_failed:
3425         e1000e_set_interrupt_capability(adapter);
3426         e1000_request_irq(adapter);
3427         return err;
3428 }
3429
3430 /**
3431  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3432  * @adapter: board private struct
3433  *
3434  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3435  **/
3436 static int e1000_test_msi(struct e1000_adapter *adapter)
3437 {
3438         int err;
3439         u16 pci_cmd;
3440
3441         if (!(adapter->flags & FLAG_MSI_ENABLED))
3442                 return 0;
3443
3444         /* disable SERR in case the MSI write causes a master abort */
3445         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3446         if (pci_cmd & PCI_COMMAND_SERR)
3447                 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3448                                       pci_cmd & ~PCI_COMMAND_SERR);
3449
3450         err = e1000_test_msi_interrupt(adapter);
3451
3452         /* re-enable SERR */
3453         if (pci_cmd & PCI_COMMAND_SERR) {
3454                 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3455                 pci_cmd |= PCI_COMMAND_SERR;
3456                 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3457         }
3458
3459         /* success ! */
3460         if (!err)
3461                 return 0;
3462
3463         /* EIO means MSI test failed */
3464         if (err != -EIO)
3465                 return err;
3466
3467         /* back to INTx mode */
3468         e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3469
3470         e1000_free_irq(adapter);
3471
3472         err = e1000_request_irq(adapter);
3473
3474         return err;
3475 }
3476
3477 /**
3478  * e1000_open - Called when a network interface is made active
3479  * @netdev: network interface device structure
3480  *
3481  * Returns 0 on success, negative value on failure
3482  *
3483  * The open entry point is called when a network interface is made
3484  * active by the system (IFF_UP).  At this point all resources needed
3485  * for transmit and receive operations are allocated, the interrupt
3486  * handler is registered with the OS, the watchdog timer is started,
3487  * and the stack is notified that the interface is ready.
3488  **/
3489 static int e1000_open(struct net_device *netdev)
3490 {
3491         struct e1000_adapter *adapter = netdev_priv(netdev);
3492         struct e1000_hw *hw = &adapter->hw;
3493         struct pci_dev *pdev = adapter->pdev;
3494         int err;
3495
3496         /* disallow open during test */
3497         if (test_bit(__E1000_TESTING, &adapter->state))
3498                 return -EBUSY;
3499
3500         pm_runtime_get_sync(&pdev->dev);
3501
3502         netif_carrier_off(netdev);
3503
3504         /* allocate transmit descriptors */
3505         err = e1000e_setup_tx_resources(adapter);
3506         if (err)
3507                 goto err_setup_tx;
3508
3509         /* allocate receive descriptors */
3510         err = e1000e_setup_rx_resources(adapter);
3511         if (err)
3512                 goto err_setup_rx;
3513
3514         /*
3515          * If AMT is enabled, let the firmware know that the network
3516          * interface is now open and reset the part to a known state.
3517          */
3518         if (adapter->flags & FLAG_HAS_AMT) {
3519                 e1000_get_hw_control(adapter);
3520                 e1000e_reset(adapter);
3521         }
3522
3523         e1000e_power_up_phy(adapter);
3524
3525         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3526         if ((adapter->hw.mng_cookie.status &
3527              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3528                 e1000_update_mng_vlan(adapter);
3529
3530         /*
3531          * before we allocate an interrupt, we must be ready to handle it.
3532          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3533          * as soon as we call pci_request_irq, so we have to setup our
3534          * clean_rx handler before we do so.
3535          */
3536         e1000_configure(adapter);
3537
3538         err = e1000_request_irq(adapter);
3539         if (err)
3540                 goto err_req_irq;
3541
3542         /*
3543          * Work around PCIe errata with MSI interrupts causing some chipsets to
3544          * ignore e1000e MSI messages, which means we need to test our MSI
3545          * interrupt now
3546          */
3547         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3548                 err = e1000_test_msi(adapter);
3549                 if (err) {
3550                         e_err("Interrupt allocation failed\n");
3551                         goto err_req_irq;
3552                 }
3553         }
3554
3555         /* From here on the code is the same as e1000e_up() */
3556         clear_bit(__E1000_DOWN, &adapter->state);
3557
3558         napi_enable(&adapter->napi);
3559
3560         e1000_irq_enable(adapter);
3561
3562         netif_start_queue(netdev);
3563
3564         adapter->idle_check = true;
3565         pm_runtime_put(&pdev->dev);
3566
3567         /* fire a link status change interrupt to start the watchdog */
3568         if (adapter->msix_entries)
3569                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3570         else
3571                 ew32(ICS, E1000_ICS_LSC);
3572
3573         return 0;
3574
3575 err_req_irq:
3576         e1000_release_hw_control(adapter);
3577         e1000_power_down_phy(adapter);
3578         e1000e_free_rx_resources(adapter);
3579 err_setup_rx:
3580         e1000e_free_tx_resources(adapter);
3581 err_setup_tx:
3582         e1000e_reset(adapter);
3583         pm_runtime_put_sync(&pdev->dev);
3584
3585         return err;
3586 }
3587
3588 /**
3589  * e1000_close - Disables a network interface
3590  * @netdev: network interface device structure
3591  *
3592  * Returns 0, this is not allowed to fail
3593  *
3594  * The close entry point is called when an interface is de-activated
3595  * by the OS.  The hardware is still under the drivers control, but
3596  * needs to be disabled.  A global MAC reset is issued to stop the
3597  * hardware, and all transmit and receive resources are freed.
3598  **/
3599 static int e1000_close(struct net_device *netdev)
3600 {
3601         struct e1000_adapter *adapter = netdev_priv(netdev);
3602         struct pci_dev *pdev = adapter->pdev;
3603
3604         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3605
3606         pm_runtime_get_sync(&pdev->dev);
3607
3608         if (!test_bit(__E1000_DOWN, &adapter->state)) {
3609                 e1000e_down(adapter);
3610                 e1000_free_irq(adapter);
3611         }
3612         e1000_power_down_phy(adapter);
3613
3614         e1000e_free_tx_resources(adapter);
3615         e1000e_free_rx_resources(adapter);
3616
3617         /*
3618          * kill manageability vlan ID if supported, but not if a vlan with
3619          * the same ID is registered on the host OS (let 8021q kill it)
3620          */
3621         if ((adapter->hw.mng_cookie.status &
3622                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3623              !(adapter->vlgrp &&
3624                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3625                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3626
3627         /*
3628          * If AMT is enabled, let the firmware know that the network
3629          * interface is now closed
3630          */
3631         if (adapter->flags & FLAG_HAS_AMT)
3632                 e1000_release_hw_control(adapter);
3633
3634         pm_runtime_put_sync(&pdev->dev);
3635
3636         return 0;
3637 }
3638 /**
3639  * e1000_set_mac - Change the Ethernet Address of the NIC
3640  * @netdev: network interface device structure
3641  * @p: pointer to an address structure
3642  *
3643  * Returns 0 on success, negative on failure
3644  **/
3645 static int e1000_set_mac(struct net_device *netdev, void *p)
3646 {
3647         struct e1000_adapter *adapter = netdev_priv(netdev);
3648         struct sockaddr *addr = p;
3649
3650         if (!is_valid_ether_addr(addr->sa_data))
3651                 return -EADDRNOTAVAIL;
3652
3653         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3654         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3655
3656         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3657
3658         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3659                 /* activate the work around */
3660                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3661
3662                 /*
3663                  * Hold a copy of the LAA in RAR[14] This is done so that
3664                  * between the time RAR[0] gets clobbered  and the time it
3665                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3666                  * of the RARs and no incoming packets directed to this port
3667                  * are dropped. Eventually the LAA will be in RAR[0] and
3668                  * RAR[14]
3669                  */
3670                 e1000e_rar_set(&adapter->hw,
3671                               adapter->hw.mac.addr,
3672                               adapter->hw.mac.rar_entry_count - 1);
3673         }
3674
3675         return 0;
3676 }
3677
3678 /**
3679  * e1000e_update_phy_task - work thread to update phy
3680  * @work: pointer to our work struct
3681  *
3682  * this worker thread exists because we must acquire a
3683  * semaphore to read the phy, which we could msleep while
3684  * waiting for it, and we can't msleep in a timer.
3685  **/
3686 static void e1000e_update_phy_task(struct work_struct *work)
3687 {
3688         struct e1000_adapter *adapter = container_of(work,
3689                                         struct e1000_adapter, update_phy_task);
3690         e1000_get_phy_info(&adapter->hw);
3691 }
3692
3693 /*
3694  * Need to wait a few seconds after link up to get diagnostic information from
3695  * the phy
3696  */
3697 static void e1000_update_phy_info(unsigned long data)
3698 {
3699         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3700         schedule_work(&adapter->update_phy_task);
3701 }
3702
3703 /**
3704  * e1000e_update_phy_stats - Update the PHY statistics counters
3705  * @adapter: board private structure
3706  **/
3707 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
3708 {
3709         struct e1000_hw *hw = &adapter->hw;
3710         s32 ret_val;
3711         u16 phy_data;
3712
3713         ret_val = hw->phy.ops.acquire(hw);
3714         if (ret_val)
3715                 return;
3716
3717         hw->phy.addr = 1;
3718
3719 #define HV_PHY_STATS_PAGE       778
3720         /*
3721          * A page set is expensive so check if already on desired page.
3722          * If not, set to the page with the PHY status registers.
3723          */
3724         ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
3725                                            &phy_data);
3726         if (ret_val)
3727                 goto release;
3728         if (phy_data != (HV_PHY_STATS_PAGE << IGP_PAGE_SHIFT)) {
3729                 ret_val = e1000e_write_phy_reg_mdic(hw,
3730                                                     IGP01E1000_PHY_PAGE_SELECT,
3731                                                     (HV_PHY_STATS_PAGE <<
3732                                                      IGP_PAGE_SHIFT));
3733                 if (ret_val)
3734                         goto release;
3735         }
3736
3737         /* Read/clear the upper 16-bit registers and read/accumulate lower */
3738
3739         /* Single Collision Count */
3740         e1000e_read_phy_reg_mdic(hw, HV_SCC_UPPER & MAX_PHY_REG_ADDRESS,
3741                                  &phy_data);
3742         ret_val = e1000e_read_phy_reg_mdic(hw,
3743                                            HV_SCC_LOWER & MAX_PHY_REG_ADDRESS,
3744                                            &phy_data);
3745         if (!ret_val)
3746                 adapter->stats.scc += phy_data;
3747
3748         /* Excessive Collision Count */
3749         e1000e_read_phy_reg_mdic(hw, HV_ECOL_UPPER & MAX_PHY_REG_ADDRESS,
3750                                  &phy_data);
3751         ret_val = e1000e_read_phy_reg_mdic(hw,
3752                                            HV_ECOL_LOWER & MAX_PHY_REG_ADDRESS,
3753                                            &phy_data);
3754         if (!ret_val)
3755                 adapter->stats.ecol += phy_data;
3756
3757         /* Multiple Collision Count */
3758         e1000e_read_phy_reg_mdic(hw, HV_MCC_UPPER & MAX_PHY_REG_ADDRESS,
3759                                  &phy_data);
3760         ret_val = e1000e_read_phy_reg_mdic(hw,
3761                                            HV_MCC_LOWER & MAX_PHY_REG_ADDRESS,
3762                                            &phy_data);
3763         if (!ret_val)
3764                 adapter->stats.mcc += phy_data;
3765
3766         /* Late Collision Count */
3767         e1000e_read_phy_reg_mdic(hw, HV_LATECOL_UPPER & MAX_PHY_REG_ADDRESS,
3768                                  &phy_data);
3769         ret_val = e1000e_read_phy_reg_mdic(hw,
3770                                            HV_LATECOL_LOWER &
3771                                            MAX_PHY_REG_ADDRESS,
3772                                            &phy_data);
3773         if (!ret_val)
3774                 adapter->stats.latecol += phy_data;
3775
3776         /* Collision Count - also used for adaptive IFS */
3777         e1000e_read_phy_reg_mdic(hw, HV_COLC_UPPER & MAX_PHY_REG_ADDRESS,
3778                                  &phy_data);
3779         ret_val = e1000e_read_phy_reg_mdic(hw,
3780                                            HV_COLC_LOWER & MAX_PHY_REG_ADDRESS,
3781                                            &phy_data);
3782         if (!ret_val)
3783                 hw->mac.collision_delta = phy_data;
3784
3785         /* Defer Count */
3786         e1000e_read_phy_reg_mdic(hw, HV_DC_UPPER & MAX_PHY_REG_ADDRESS,
3787                                  &phy_data);
3788         ret_val = e1000e_read_phy_reg_mdic(hw,
3789                                            HV_DC_LOWER & MAX_PHY_REG_ADDRESS,
3790                                            &phy_data);
3791         if (!ret_val)
3792                 adapter->stats.dc += phy_data;
3793
3794         /* Transmit with no CRS */
3795         e1000e_read_phy_reg_mdic(hw, HV_TNCRS_UPPER & MAX_PHY_REG_ADDRESS,
3796                                  &phy_data);
3797         ret_val = e1000e_read_phy_reg_mdic(hw,
3798                                            HV_TNCRS_LOWER & MAX_PHY_REG_ADDRESS,
3799                                            &phy_data);
3800         if (!ret_val)
3801                 adapter->stats.tncrs += phy_data;
3802
3803 release:
3804         hw->phy.ops.release(hw);
3805 }
3806
3807 /**
3808  * e1000e_update_stats - Update the board statistics counters
3809  * @adapter: board private structure
3810  **/
3811 void e1000e_update_stats(struct e1000_adapter *adapter)
3812 {
3813         struct net_device *netdev = adapter->netdev;
3814         struct e1000_hw *hw = &adapter->hw;
3815         struct pci_dev *pdev = adapter->pdev;
3816
3817         /*
3818          * Prevent stats update while adapter is being reset, or if the pci
3819          * connection is down.
3820          */
3821         if (adapter->link_speed == 0)
3822                 return;
3823         if (pci_channel_offline(pdev))
3824                 return;
3825
3826         adapter->stats.crcerrs += er32(CRCERRS);
3827         adapter->stats.gprc += er32(GPRC);
3828         adapter->stats.gorc += er32(GORCL);
3829         er32(GORCH); /* Clear gorc */
3830         adapter->stats.bprc += er32(BPRC);
3831         adapter->stats.mprc += er32(MPRC);
3832         adapter->stats.roc += er32(ROC);
3833
3834         adapter->stats.mpc += er32(MPC);
3835
3836         /* Half-duplex statistics */
3837         if (adapter->link_duplex == HALF_DUPLEX) {
3838                 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
3839                         e1000e_update_phy_stats(adapter);
3840                 } else {
3841                         adapter->stats.scc += er32(SCC);
3842                         adapter->stats.ecol += er32(ECOL);
3843                         adapter->stats.mcc += er32(MCC);
3844                         adapter->stats.latecol += er32(LATECOL);
3845                         adapter->stats.dc += er32(DC);
3846
3847                         hw->mac.collision_delta = er32(COLC);
3848
3849                         if ((hw->mac.type != e1000_82574) &&
3850                             (hw->mac.type != e1000_82583))
3851                                 adapter->stats.tncrs += er32(TNCRS);
3852                 }
3853                 adapter->stats.colc += hw->mac.collision_delta;
3854         }
3855
3856         adapter->stats.xonrxc += er32(XONRXC);
3857         adapter->stats.xontxc += er32(XONTXC);
3858         adapter->stats.xoffrxc += er32(XOFFRXC);
3859         adapter->stats.xofftxc += er32(XOFFTXC);
3860         adapter->stats.gptc += er32(GPTC);
3861         adapter->stats.gotc += er32(GOTCL);
3862         er32(GOTCH); /* Clear gotc */
3863         adapter->stats.rnbc += er32(RNBC);
3864         adapter->stats.ruc += er32(RUC);
3865
3866         adapter->stats.mptc += er32(MPTC);
3867         adapter->stats.bptc += er32(BPTC);
3868
3869         /* used for adaptive IFS */
3870
3871         hw->mac.tx_packet_delta = er32(TPT);
3872         adapter->stats.tpt += hw->mac.tx_packet_delta;
3873
3874         adapter->stats.algnerrc += er32(ALGNERRC);
3875         adapter->stats.rxerrc += er32(RXERRC);
3876         adapter->stats.cexterr += er32(CEXTERR);
3877         adapter->stats.tsctc += er32(TSCTC);
3878         adapter->stats.tsctfc += er32(TSCTFC);
3879
3880         /* Fill out the OS statistics structure */
3881         netdev->stats.multicast = adapter->stats.mprc;
3882         netdev->stats.collisions = adapter->stats.colc;
3883
3884         /* Rx Errors */
3885
3886         /*
3887          * RLEC on some newer hardware can be incorrect so build
3888          * our own version based on RUC and ROC
3889          */
3890         netdev->stats.rx_errors = adapter->stats.rxerrc +
3891                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3892                 adapter->stats.ruc + adapter->stats.roc +
3893                 adapter->stats.cexterr;
3894         netdev->stats.rx_length_errors = adapter->stats.ruc +
3895                                               adapter->stats.roc;
3896         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3897         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3898         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3899
3900         /* Tx Errors */
3901         netdev->stats.tx_errors = adapter->stats.ecol +
3902                                        adapter->stats.latecol;
3903         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3904         netdev->stats.tx_window_errors = adapter->stats.latecol;
3905         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3906
3907         /* Tx Dropped needs to be maintained elsewhere */
3908
3909         /* Management Stats */
3910         adapter->stats.mgptc += er32(MGTPTC);
3911         adapter->stats.mgprc += er32(MGTPRC);
3912         adapter->stats.mgpdc += er32(MGTPDC);
3913 }
3914
3915 /**
3916  * e1000_phy_read_status - Update the PHY register status snapshot
3917  * @adapter: board private structure
3918  **/
3919 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3920 {
3921         struct e1000_hw *hw = &adapter->hw;
3922         struct e1000_phy_regs *phy = &adapter->phy_regs;
3923         int ret_val;
3924
3925         if ((er32(STATUS) & E1000_STATUS_LU) &&
3926             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3927                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3928                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3929                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3930                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3931                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3932                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3933                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3934                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3935                 if (ret_val)
3936                         e_warn("Error reading PHY register\n");
3937         } else {
3938                 /*
3939                  * Do not read PHY registers if link is not up
3940                  * Set values to typical power-on defaults
3941                  */
3942                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3943                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3944                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3945                              BMSR_ERCAP);
3946                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3947                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3948                 phy->lpa = 0;
3949                 phy->expansion = EXPANSION_ENABLENPAGE;
3950                 phy->ctrl1000 = ADVERTISE_1000FULL;
3951                 phy->stat1000 = 0;
3952                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3953         }
3954 }
3955
3956 static void e1000_print_link_info(struct e1000_adapter *adapter)
3957 {
3958         struct e1000_hw *hw = &adapter->hw;
3959         u32 ctrl = er32(CTRL);
3960
3961         /* Link status message must follow this format for user tools */
3962         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
3963                "Flow Control: %s\n",
3964                adapter->netdev->name,
3965                adapter->link_speed,
3966                (adapter->link_duplex == FULL_DUPLEX) ?
3967                                 "Full Duplex" : "Half Duplex",
3968                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3969                                 "RX/TX" :
3970                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3971                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3972 }
3973
3974 static bool e1000e_has_link(struct e1000_adapter *adapter)
3975 {
3976         struct e1000_hw *hw = &adapter->hw;
3977         bool link_active = 0;
3978         s32 ret_val = 0;
3979
3980         /*
3981          * get_link_status is set on LSC (link status) interrupt or
3982          * Rx sequence error interrupt.  get_link_status will stay
3983          * false until the check_for_link establishes link
3984          * for copper adapters ONLY
3985          */
3986         switch (hw->phy.media_type) {
3987         case e1000_media_type_copper:
3988                 if (hw->mac.get_link_status) {
3989                         ret_val = hw->mac.ops.check_for_link(hw);
3990                         link_active = !hw->mac.get_link_status;
3991                 } else {
3992                         link_active = 1;
3993                 }
3994                 break;
3995         case e1000_media_type_fiber:
3996                 ret_val = hw->mac.ops.check_for_link(hw);
3997                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3998                 break;
3999         case e1000_media_type_internal_serdes:
4000                 ret_val = hw->mac.ops.check_for_link(hw);
4001                 link_active = adapter->hw.mac.serdes_has_link;
4002                 break;
4003         default:
4004         case e1000_media_type_unknown:
4005                 break;
4006         }
4007
4008         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4009             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4010                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4011                 e_info("Gigabit has been disabled, downgrading speed\n");
4012         }
4013
4014         return link_active;
4015 }
4016
4017 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4018 {
4019         /* make sure the receive unit is started */
4020         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4021             (adapter->flags & FLAG_RX_RESTART_NOW)) {
4022                 struct e1000_hw *hw = &adapter->hw;
4023                 u32 rctl = er32(RCTL);
4024                 ew32(RCTL, rctl | E1000_RCTL_EN);
4025                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
4026         }
4027 }
4028
4029 /**
4030  * e1000_watchdog - Timer Call-back
4031  * @data: pointer to adapter cast into an unsigned long
4032  **/
4033 static void e1000_watchdog(unsigned long data)
4034 {
4035         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4036
4037         /* Do the rest outside of interrupt context */
4038         schedule_work(&adapter->watchdog_task);
4039
4040         /* TODO: make this use queue_delayed_work() */
4041 }
4042
4043 static void e1000_watchdog_task(struct work_struct *work)
4044 {
4045         struct e1000_adapter *adapter = container_of(work,
4046                                         struct e1000_adapter, watchdog_task);
4047         struct net_device *netdev = adapter->netdev;
4048         struct e1000_mac_info *mac = &adapter->hw.mac;
4049         struct e1000_phy_info *phy = &adapter->hw.phy;
4050         struct e1000_ring *tx_ring = adapter->tx_ring;
4051         struct e1000_hw *hw = &adapter->hw;
4052         u32 link, tctl;
4053         int tx_pending = 0;
4054
4055         link = e1000e_has_link(adapter);
4056         if ((netif_carrier_ok(netdev)) && link) {
4057                 /* Cancel scheduled suspend requests. */
4058                 pm_runtime_resume(netdev->dev.parent);
4059
4060                 e1000e_enable_receives(adapter);
4061                 goto link_up;
4062         }
4063
4064         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4065             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4066                 e1000_update_mng_vlan(adapter);
4067
4068         if (link) {
4069                 if (!netif_carrier_ok(netdev)) {
4070                         bool txb2b = 1;
4071
4072                         /* Cancel scheduled suspend requests. */
4073                         pm_runtime_resume(netdev->dev.parent);
4074
4075                         /* update snapshot of PHY registers on LSC */
4076                         e1000_phy_read_status(adapter);
4077                         mac->ops.get_link_up_info(&adapter->hw,
4078                                                    &adapter->link_speed,
4079                                                    &adapter->link_duplex);
4080                         e1000_print_link_info(adapter);
4081                         /*
4082                          * On supported PHYs, check for duplex mismatch only
4083                          * if link has autonegotiated at 10/100 half
4084                          */
4085                         if ((hw->phy.type == e1000_phy_igp_3 ||
4086                              hw->phy.type == e1000_phy_bm) &&
4087                             (hw->mac.autoneg == true) &&
4088                             (adapter->link_speed == SPEED_10 ||
4089                              adapter->link_speed == SPEED_100) &&
4090                             (adapter->link_duplex == HALF_DUPLEX)) {
4091                                 u16 autoneg_exp;
4092
4093                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4094
4095                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4096                                         e_info("Autonegotiated half duplex but"
4097                                                " link partner cannot autoneg. "
4098                                                " Try forcing full duplex if "
4099                                                "link gets many collisions.\n");
4100                         }
4101
4102                         /* adjust timeout factor according to speed/duplex */
4103                         adapter->tx_timeout_factor = 1;
4104                         switch (adapter->link_speed) {
4105                         case SPEED_10:
4106                                 txb2b = 0;
4107                                 adapter->tx_timeout_factor = 16;
4108                                 break;
4109                         case SPEED_100:
4110                                 txb2b = 0;
4111                                 adapter->tx_timeout_factor = 10;
4112                                 break;
4113                         }
4114
4115                         /*
4116                          * workaround: re-program speed mode bit after
4117                          * link-up event
4118                          */
4119                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4120                             !txb2b) {
4121                                 u32 tarc0;
4122                                 tarc0 = er32(TARC(0));
4123                                 tarc0 &= ~SPEED_MODE_BIT;
4124                                 ew32(TARC(0), tarc0);
4125                         }
4126
4127                         /*
4128                          * disable TSO for pcie and 10/100 speeds, to avoid
4129                          * some hardware issues
4130                          */
4131                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
4132                                 switch (adapter->link_speed) {
4133                                 case SPEED_10:
4134                                 case SPEED_100:
4135                                         e_info("10/100 speed: disabling TSO\n");
4136                                         netdev->features &= ~NETIF_F_TSO;
4137                                         netdev->features &= ~NETIF_F_TSO6;
4138                                         break;
4139                                 case SPEED_1000:
4140                                         netdev->features |= NETIF_F_TSO;
4141                                         netdev->features |= NETIF_F_TSO6;
4142                                         break;
4143                                 default:
4144                                         /* oops */
4145                                         break;
4146                                 }
4147                         }
4148
4149                         /*
4150                          * enable transmits in the hardware, need to do this
4151                          * after setting TARC(0)
4152                          */
4153                         tctl = er32(TCTL);
4154                         tctl |= E1000_TCTL_EN;
4155                         ew32(TCTL, tctl);
4156
4157                         /*
4158                          * Perform any post-link-up configuration before
4159                          * reporting link up.
4160                          */
4161                         if (phy->ops.cfg_on_link_up)
4162                                 phy->ops.cfg_on_link_up(hw);
4163
4164                         netif_carrier_on(netdev);
4165
4166                         if (!test_bit(__E1000_DOWN, &adapter->state))
4167                                 mod_timer(&adapter->phy_info_timer,
4168                                           round_jiffies(jiffies + 2 * HZ));
4169                 }
4170         } else {
4171                 if (netif_carrier_ok(netdev)) {
4172                         adapter->link_speed = 0;
4173                         adapter->link_duplex = 0;
4174                         /* Link status message must follow this format */
4175                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4176                                adapter->netdev->name);
4177                         netif_carrier_off(netdev);
4178                         if (!test_bit(__E1000_DOWN, &adapter->state))
4179                                 mod_timer(&adapter->phy_info_timer,
4180                                           round_jiffies(jiffies + 2 * HZ));
4181
4182                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4183                                 schedule_work(&adapter->reset_task);
4184                         else
4185                                 pm_schedule_suspend(netdev->dev.parent,
4186                                                         LINK_TIMEOUT);
4187                 }
4188         }
4189
4190 link_up:
4191         e1000e_update_stats(adapter);
4192
4193         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4194         adapter->tpt_old = adapter->stats.tpt;
4195         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4196         adapter->colc_old = adapter->stats.colc;
4197
4198         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4199         adapter->gorc_old = adapter->stats.gorc;
4200         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4201         adapter->gotc_old = adapter->stats.gotc;
4202
4203         e1000e_update_adaptive(&adapter->hw);
4204
4205         if (!netif_carrier_ok(netdev)) {
4206                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
4207                                tx_ring->count);
4208                 if (tx_pending) {
4209                         /*
4210                          * We've lost link, so the controller stops DMA,
4211                          * but we've got queued Tx work that's never going
4212                          * to get done, so reset controller to flush Tx.
4213                          * (Do the reset outside of interrupt context).
4214                          */
4215                         adapter->tx_timeout_count++;
4216                         schedule_work(&adapter->reset_task);
4217                         /* return immediately since reset is imminent */
4218                         return;
4219                 }
4220         }
4221
4222         /* Simple mode for Interrupt Throttle Rate (ITR) */
4223         if (adapter->itr_setting == 4) {
4224                 /*
4225                  * Symmetric Tx/Rx gets a reduced ITR=2000;
4226                  * Total asymmetrical Tx or Rx gets ITR=8000;
4227                  * everyone else is between 2000-8000.
4228                  */
4229                 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4230                 u32 dif = (adapter->gotc > adapter->gorc ?
4231                             adapter->gotc - adapter->gorc :
4232                             adapter->gorc - adapter->gotc) / 10000;
4233                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4234
4235                 ew32(ITR, 1000000000 / (itr * 256));
4236         }
4237
4238         /* Cause software interrupt to ensure Rx ring is cleaned */
4239         if (adapter->msix_entries)
4240                 ew32(ICS, adapter->rx_ring->ims_val);
4241         else
4242                 ew32(ICS, E1000_ICS_RXDMT0);
4243
4244         /* Force detection of hung controller every watchdog period */
4245         adapter->detect_tx_hung = 1;
4246
4247         /*
4248          * With 82571 controllers, LAA may be overwritten due to controller
4249          * reset from the other port. Set the appropriate LAA in RAR[0]
4250          */
4251         if (e1000e_get_laa_state_82571(hw))
4252                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4253
4254         /* Reset the timer */
4255         if (!test_bit(__E1000_DOWN, &adapter->state))
4256                 mod_timer(&adapter->watchdog_timer,
4257                           round_jiffies(jiffies + 2 * HZ));
4258 }
4259
4260 #define E1000_TX_FLAGS_CSUM             0x00000001
4261 #define E1000_TX_FLAGS_VLAN             0x00000002
4262 #define E1000_TX_FLAGS_TSO              0x00000004
4263 #define E1000_TX_FLAGS_IPV4             0x00000008
4264 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
4265 #define E1000_TX_FLAGS_VLAN_SHIFT       16
4266
4267 static int e1000_tso(struct e1000_adapter *adapter,
4268                      struct sk_buff *skb)
4269 {
4270         struct e1000_ring *tx_ring = adapter->tx_ring;
4271         struct e1000_context_desc *context_desc;
4272         struct e1000_buffer *buffer_info;
4273         unsigned int i;
4274         u32 cmd_length = 0;
4275         u16 ipcse = 0, tucse, mss;
4276         u8 ipcss, ipcso, tucss, tucso, hdr_len;
4277         int err;
4278
4279         if (!skb_is_gso(skb))
4280                 return 0;
4281
4282         if (skb_header_cloned(skb)) {
4283                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4284                 if (err)
4285                         return err;
4286         }
4287
4288         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4289         mss = skb_shinfo(skb)->gso_size;
4290         if (skb->protocol == htons(ETH_P_IP)) {
4291                 struct iphdr *iph = ip_hdr(skb);
4292                 iph->tot_len = 0;
4293                 iph->check = 0;
4294                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4295                                                          0, IPPROTO_TCP, 0);
4296                 cmd_length = E1000_TXD_CMD_IP;
4297                 ipcse = skb_transport_offset(skb) - 1;
4298         } else if (skb_is_gso_v6(skb)) {
4299                 ipv6_hdr(skb)->payload_len = 0;
4300                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4301                                                        &ipv6_hdr(skb)->daddr,
4302                                                        0, IPPROTO_TCP, 0);
4303                 ipcse = 0;
4304         }
4305         ipcss = skb_network_offset(skb);
4306         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4307         tucss = skb_transport_offset(skb);
4308         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4309         tucse = 0;
4310
4311         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4312                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4313
4314         i = tx_ring->next_to_use;
4315         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4316         buffer_info = &tx_ring->buffer_info[i];
4317
4318         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4319         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4320         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4321         context_desc->upper_setup.tcp_fields.tucss = tucss;
4322         context_desc->upper_setup.tcp_fields.tucso = tucso;
4323         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4324         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4325         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4326         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4327
4328         buffer_info->time_stamp = jiffies;
4329         buffer_info->next_to_watch = i;
4330
4331         i++;
4332         if (i == tx_ring->count)
4333                 i = 0;
4334         tx_ring->next_to_use = i;
4335
4336         return 1;
4337 }
4338
4339 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
4340 {
4341         struct e1000_ring *tx_ring = adapter->tx_ring;
4342         struct e1000_context_desc *context_desc;
4343         struct e1000_buffer *buffer_info;
4344         unsigned int i;
4345         u8 css;
4346         u32 cmd_len = E1000_TXD_CMD_DEXT;
4347         __be16 protocol;
4348
4349         if (skb->ip_summed != CHECKSUM_PARTIAL)
4350                 return 0;
4351
4352         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4353                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4354         else
4355                 protocol = skb->protocol;
4356
4357         switch (protocol) {
4358         case cpu_to_be16(ETH_P_IP):
4359                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4360                         cmd_len |= E1000_TXD_CMD_TCP;
4361                 break;
4362         case cpu_to_be16(ETH_P_IPV6):
4363                 /* XXX not handling all IPV6 headers */
4364                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4365                         cmd_len |= E1000_TXD_CMD_TCP;
4366                 break;
4367         default:
4368                 if (unlikely(net_ratelimit()))
4369                         e_warn("checksum_partial proto=%x!\n",
4370                                be16_to_cpu(protocol));
4371                 break;
4372         }
4373
4374         css = skb_transport_offset(skb);
4375
4376         i = tx_ring->next_to_use;
4377         buffer_info = &tx_ring->buffer_info[i];
4378         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4379
4380         context_desc->lower_setup.ip_config = 0;
4381         context_desc->upper_setup.tcp_fields.tucss = css;
4382         context_desc->upper_setup.tcp_fields.tucso =
4383                                 css + skb->csum_offset;
4384         context_desc->upper_setup.tcp_fields.tucse = 0;
4385         context_desc->tcp_seg_setup.data = 0;
4386         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4387
4388         buffer_info->time_stamp = jiffies;
4389         buffer_info->next_to_watch = i;
4390
4391         i++;
4392         if (i == tx_ring->count)
4393                 i = 0;
4394         tx_ring->next_to_use = i;
4395
4396         return 1;
4397 }
4398
4399 #define E1000_MAX_PER_TXD       8192
4400 #define E1000_MAX_TXD_PWR       12
4401
4402 static int e1000_tx_map(struct e1000_adapter *adapter,
4403                         struct sk_buff *skb, unsigned int first,
4404                         unsigned int max_per_txd, unsigned int nr_frags,
4405                         unsigned int mss)
4406 {
4407         struct e1000_ring *tx_ring = adapter->tx_ring;
4408         struct pci_dev *pdev = adapter->pdev;
4409         struct e1000_buffer *buffer_info;
4410         unsigned int len = skb_headlen(skb);
4411         unsigned int offset = 0, size, count = 0, i;
4412         unsigned int f, bytecount, segs;
4413
4414         i = tx_ring->next_to_use;
4415
4416         while (len) {
4417                 buffer_info = &tx_ring->buffer_info[i];
4418                 size = min(len, max_per_txd);
4419
4420                 buffer_info->length = size;
4421                 buffer_info->time_stamp = jiffies;
4422                 buffer_info->next_to_watch = i;
4423                 buffer_info->dma = dma_map_single(&pdev->dev,
4424                                                   skb->data + offset,
4425                                                   size, DMA_TO_DEVICE);
4426                 buffer_info->mapped_as_page = false;
4427                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4428                         goto dma_error;
4429
4430                 len -= size;
4431                 offset += size;
4432                 count++;
4433
4434                 if (len) {
4435                         i++;
4436                         if (i == tx_ring->count)
4437                                 i = 0;
4438                 }
4439         }
4440
4441         for (f = 0; f < nr_frags; f++) {
4442                 struct skb_frag_struct *frag;
4443
4444                 frag = &skb_shinfo(skb)->frags[f];
4445                 len = frag->size;
4446                 offset = frag->page_offset;
4447
4448                 while (len) {
4449                         i++;
4450                         if (i == tx_ring->count)
4451                                 i = 0;
4452
4453                         buffer_info = &tx_ring->buffer_info[i];
4454                         size = min(len, max_per_txd);
4455
4456                         buffer_info->length = size;
4457                         buffer_info->time_stamp = jiffies;
4458                         buffer_info->next_to_watch = i;
4459                         buffer_info->dma = dma_map_page(&pdev->dev, frag->page,
4460                                                         offset, size,
4461                                                         DMA_TO_DEVICE);
4462                         buffer_info->mapped_as_page = true;
4463                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4464                                 goto dma_error;
4465
4466                         len -= size;
4467                         offset += size;
4468                         count++;
4469                 }
4470         }
4471
4472         segs = skb_shinfo(skb)->gso_segs ?: 1;
4473         /* multiply data chunks by size of headers */
4474         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4475
4476         tx_ring->buffer_info[i].skb = skb;
4477         tx_ring->buffer_info[i].segs = segs;
4478         tx_ring->buffer_info[i].bytecount = bytecount;
4479         tx_ring->buffer_info[first].next_to_watch = i;
4480
4481         return count;
4482
4483 dma_error:
4484         dev_err(&pdev->dev, "TX DMA map failed\n");
4485         buffer_info->dma = 0;
4486         if (count)
4487                 count--;
4488
4489         while (count--) {
4490                 if (i==0)
4491                         i += tx_ring->count;
4492                 i--;
4493                 buffer_info = &tx_ring->buffer_info[i];
4494                 e1000_put_txbuf(adapter, buffer_info);;
4495         }
4496
4497         return 0;
4498 }
4499
4500 static void e1000_tx_queue(struct e1000_adapter *adapter,
4501                            int tx_flags, int count)
4502 {
4503         struct e1000_ring *tx_ring = adapter->tx_ring;
4504         struct e1000_tx_desc *tx_desc = NULL;
4505         struct e1000_buffer *buffer_info;
4506         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4507         unsigned int i;
4508
4509         if (tx_flags & E1000_TX_FLAGS_TSO) {
4510                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4511                              E1000_TXD_CMD_TSE;
4512                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4513
4514                 if (tx_flags & E1000_TX_FLAGS_IPV4)
4515                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4516         }
4517
4518         if (tx_flags & E1000_TX_FLAGS_CSUM) {
4519                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4520                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4521         }
4522
4523         if (tx_flags & E1000_TX_FLAGS_VLAN) {
4524                 txd_lower |= E1000_TXD_CMD_VLE;
4525                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4526         }
4527
4528         i = tx_ring->next_to_use;
4529
4530         while (count--) {
4531                 buffer_info = &tx_ring->buffer_info[i];
4532                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4533                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4534                 tx_desc->lower.data =
4535                         cpu_to_le32(txd_lower | buffer_info->length);
4536                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4537
4538                 i++;
4539                 if (i == tx_ring->count)
4540                         i = 0;
4541         }
4542
4543         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4544
4545         /*
4546          * Force memory writes to complete before letting h/w
4547          * know there are new descriptors to fetch.  (Only
4548          * applicable for weak-ordered memory model archs,
4549          * such as IA-64).
4550          */
4551         wmb();
4552
4553         tx_ring->next_to_use = i;
4554         writel(i, adapter->hw.hw_addr + tx_ring->tail);
4555         /*
4556          * we need this if more than one processor can write to our tail
4557          * at a time, it synchronizes IO on IA64/Altix systems
4558          */
4559         mmiowb();
4560 }
4561
4562 #define MINIMUM_DHCP_PACKET_SIZE 282
4563 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4564                                     struct sk_buff *skb)
4565 {
4566         struct e1000_hw *hw =  &adapter->hw;
4567         u16 length, offset;
4568
4569         if (vlan_tx_tag_present(skb)) {
4570                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4571                     (adapter->hw.mng_cookie.status &
4572                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4573                         return 0;
4574         }
4575
4576         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4577                 return 0;
4578
4579         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4580                 return 0;
4581
4582         {
4583                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4584                 struct udphdr *udp;
4585
4586                 if (ip->protocol != IPPROTO_UDP)
4587                         return 0;
4588
4589                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4590                 if (ntohs(udp->dest) != 67)
4591                         return 0;
4592
4593                 offset = (u8 *)udp + 8 - skb->data;
4594                 length = skb->len - offset;
4595                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4596         }
4597
4598         return 0;
4599 }
4600
4601 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4602 {
4603         struct e1000_adapter *adapter = netdev_priv(netdev);
4604
4605         netif_stop_queue(netdev);
4606         /*
4607          * Herbert's original patch had:
4608          *  smp_mb__after_netif_stop_queue();
4609          * but since that doesn't exist yet, just open code it.
4610          */
4611         smp_mb();
4612
4613         /*
4614          * We need to check again in a case another CPU has just
4615          * made room available.
4616          */
4617         if (e1000_desc_unused(adapter->tx_ring) < size)
4618                 return -EBUSY;
4619
4620         /* A reprieve! */
4621         netif_start_queue(netdev);
4622         ++adapter->restart_queue;
4623         return 0;
4624 }
4625
4626 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4627 {
4628         struct e1000_adapter *adapter = netdev_priv(netdev);
4629
4630         if (e1000_desc_unused(adapter->tx_ring) >= size)
4631                 return 0;
4632         return __e1000_maybe_stop_tx(netdev, size);
4633 }
4634
4635 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4636 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4637                                     struct net_device *netdev)
4638 {
4639         struct e1000_adapter *adapter = netdev_priv(netdev);
4640         struct e1000_ring *tx_ring = adapter->tx_ring;
4641         unsigned int first;
4642         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4643         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4644         unsigned int tx_flags = 0;
4645         unsigned int len = skb_headlen(skb);
4646         unsigned int nr_frags;
4647         unsigned int mss;
4648         int count = 0;
4649         int tso;
4650         unsigned int f;
4651
4652         if (test_bit(__E1000_DOWN, &adapter->state)) {
4653                 dev_kfree_skb_any(skb);
4654                 return NETDEV_TX_OK;
4655         }
4656
4657         if (skb->len <= 0) {
4658                 dev_kfree_skb_any(skb);
4659                 return NETDEV_TX_OK;
4660         }
4661
4662         mss = skb_shinfo(skb)->gso_size;
4663         /*
4664          * The controller does a simple calculation to
4665          * make sure there is enough room in the FIFO before
4666          * initiating the DMA for each buffer.  The calc is:
4667          * 4 = ceil(buffer len/mss).  To make sure we don't
4668          * overrun the FIFO, adjust the max buffer len if mss
4669          * drops.
4670          */
4671         if (mss) {
4672                 u8 hdr_len;
4673                 max_per_txd = min(mss << 2, max_per_txd);
4674                 max_txd_pwr = fls(max_per_txd) - 1;
4675
4676                 /*
4677                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4678                  * points to just header, pull a few bytes of payload from
4679                  * frags into skb->data
4680                  */
4681                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4682                 /*
4683                  * we do this workaround for ES2LAN, but it is un-necessary,
4684                  * avoiding it could save a lot of cycles
4685                  */
4686                 if (skb->data_len && (hdr_len == len)) {
4687                         unsigned int pull_size;
4688
4689                         pull_size = min((unsigned int)4, skb->data_len);
4690                         if (!__pskb_pull_tail(skb, pull_size)) {
4691                                 e_err("__pskb_pull_tail failed.\n");
4692                                 dev_kfree_skb_any(skb);
4693                                 return NETDEV_TX_OK;
4694                         }
4695                         len = skb_headlen(skb);
4696                 }
4697         }
4698
4699         /* reserve a descriptor for the offload context */
4700         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4701                 count++;
4702         count++;
4703
4704         count += TXD_USE_COUNT(len, max_txd_pwr);
4705
4706         nr_frags = skb_shinfo(skb)->nr_frags;
4707         for (f = 0; f < nr_frags; f++)
4708                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4709                                        max_txd_pwr);
4710
4711         if (adapter->hw.mac.tx_pkt_filtering)
4712                 e1000_transfer_dhcp_info(adapter, skb);
4713
4714         /*
4715          * need: count + 2 desc gap to keep tail from touching
4716          * head, otherwise try next time
4717          */
4718         if (e1000_maybe_stop_tx(netdev, count + 2))
4719                 return NETDEV_TX_BUSY;
4720
4721         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4722                 tx_flags |= E1000_TX_FLAGS_VLAN;
4723                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4724         }
4725
4726         first = tx_ring->next_to_use;
4727
4728         tso = e1000_tso(adapter, skb);
4729         if (tso < 0) {
4730                 dev_kfree_skb_any(skb);
4731                 return NETDEV_TX_OK;
4732         }
4733
4734         if (tso)
4735                 tx_flags |= E1000_TX_FLAGS_TSO;
4736         else if (e1000_tx_csum(adapter, skb))
4737                 tx_flags |= E1000_TX_FLAGS_CSUM;
4738
4739         /*
4740          * Old method was to assume IPv4 packet by default if TSO was enabled.
4741          * 82571 hardware supports TSO capabilities for IPv6 as well...
4742          * no longer assume, we must.
4743          */
4744         if (skb->protocol == htons(ETH_P_IP))
4745                 tx_flags |= E1000_TX_FLAGS_IPV4;
4746
4747         /* if count is 0 then mapping error has occured */
4748         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4749         if (count) {
4750                 e1000_tx_queue(adapter, tx_flags, count);
4751                 /* Make sure there is space in the ring for the next send. */
4752                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4753
4754         } else {
4755                 dev_kfree_skb_any(skb);
4756                 tx_ring->buffer_info[first].time_stamp = 0;
4757                 tx_ring->next_to_use = first;
4758         }
4759
4760         return NETDEV_TX_OK;
4761 }
4762
4763 /**
4764  * e1000_tx_timeout - Respond to a Tx Hang
4765  * @netdev: network interface device structure
4766  **/
4767 static void e1000_tx_timeout(struct net_device *netdev)
4768 {
4769         struct e1000_adapter *adapter = netdev_priv(netdev);
4770
4771         /* Do the reset outside of interrupt context */
4772         adapter->tx_timeout_count++;
4773         schedule_work(&adapter->reset_task);
4774 }
4775
4776 static void e1000_reset_task(struct work_struct *work)
4777 {
4778         struct e1000_adapter *adapter;
4779         adapter = container_of(work, struct e1000_adapter, reset_task);
4780
4781         e1000e_dump(adapter);
4782         e_err("Reset adapter\n");
4783         e1000e_reinit_locked(adapter);
4784 }
4785
4786 /**
4787  * e1000_get_stats - Get System Network Statistics
4788  * @netdev: network interface device structure
4789  *
4790  * Returns the address of the device statistics structure.
4791  * The statistics are actually updated from the timer callback.
4792  **/
4793 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4794 {
4795         /* only return the current stats */
4796         return &netdev->stats;
4797 }
4798
4799 /**
4800  * e1000_change_mtu - Change the Maximum Transfer Unit
4801  * @netdev: network interface device structure
4802  * @new_mtu: new value for maximum frame size
4803  *
4804  * Returns 0 on success, negative on failure
4805  **/
4806 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4807 {
4808         struct e1000_adapter *adapter = netdev_priv(netdev);
4809         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4810
4811         /* Jumbo frame support */
4812         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
4813             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4814                 e_err("Jumbo Frames not supported.\n");
4815                 return -EINVAL;
4816         }
4817
4818         /* Supported frame sizes */
4819         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4820             (max_frame > adapter->max_hw_frame_size)) {
4821                 e_err("Unsupported MTU setting\n");
4822                 return -EINVAL;
4823         }
4824
4825         /* 82573 Errata 17 */
4826         if (((adapter->hw.mac.type == e1000_82573) ||
4827              (adapter->hw.mac.type == e1000_82574)) &&
4828             (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
4829                 adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
4830                 e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
4831         }
4832
4833         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4834                 msleep(1);
4835         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
4836         adapter->max_frame_size = max_frame;
4837         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4838         netdev->mtu = new_mtu;
4839         if (netif_running(netdev))
4840                 e1000e_down(adapter);
4841
4842         /*
4843          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4844          * means we reserve 2 more, this pushes us to allocate from the next
4845          * larger slab size.
4846          * i.e. RXBUFFER_2048 --> size-4096 slab
4847          * However with the new *_jumbo_rx* routines, jumbo receives will use
4848          * fragmented skbs
4849          */
4850
4851         if (max_frame <= 2048)
4852                 adapter->rx_buffer_len = 2048;
4853         else
4854                 adapter->rx_buffer_len = 4096;
4855
4856         /* adjust allocation if LPE protects us, and we aren't using SBP */
4857         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4858              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4859                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4860                                          + ETH_FCS_LEN;
4861
4862         if (netif_running(netdev))
4863                 e1000e_up(adapter);
4864         else
4865                 e1000e_reset(adapter);
4866
4867         clear_bit(__E1000_RESETTING, &adapter->state);
4868
4869         return 0;
4870 }
4871
4872 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4873                            int cmd)
4874 {
4875         struct e1000_adapter *adapter = netdev_priv(netdev);
4876         struct mii_ioctl_data *data = if_mii(ifr);
4877
4878         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4879                 return -EOPNOTSUPP;
4880
4881         switch (cmd) {
4882         case SIOCGMIIPHY:
4883                 data->phy_id = adapter->hw.phy.addr;
4884                 break;
4885         case SIOCGMIIREG:
4886                 e1000_phy_read_status(adapter);
4887
4888                 switch (data->reg_num & 0x1F) {
4889                 case MII_BMCR:
4890                         data->val_out = adapter->phy_regs.bmcr;
4891                         break;
4892                 case MII_BMSR:
4893                         data->val_out = adapter->phy_regs.bmsr;
4894                         break;
4895                 case MII_PHYSID1:
4896                         data->val_out = (adapter->hw.phy.id >> 16);
4897                         break;
4898                 case MII_PHYSID2:
4899                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
4900                         break;
4901                 case MII_ADVERTISE:
4902                         data->val_out = adapter->phy_regs.advertise;
4903                         break;
4904                 case MII_LPA:
4905                         data->val_out = adapter->phy_regs.lpa;
4906                         break;
4907                 case MII_EXPANSION:
4908                         data->val_out = adapter->phy_regs.expansion;
4909                         break;
4910                 case MII_CTRL1000:
4911                         data->val_out = adapter->phy_regs.ctrl1000;
4912                         break;
4913                 case MII_STAT1000:
4914                         data->val_out = adapter->phy_regs.stat1000;
4915                         break;
4916                 case MII_ESTATUS:
4917                         data->val_out = adapter->phy_regs.estatus;
4918                         break;
4919                 default:
4920                         return -EIO;
4921                 }
4922                 break;
4923         case SIOCSMIIREG:
4924         default:
4925                 return -EOPNOTSUPP;
4926         }
4927         return 0;
4928 }
4929
4930 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4931 {
4932         switch (cmd) {
4933         case SIOCGMIIPHY:
4934         case SIOCGMIIREG:
4935         case SIOCSMIIREG:
4936                 return e1000_mii_ioctl(netdev, ifr, cmd);
4937         default:
4938                 return -EOPNOTSUPP;
4939         }
4940 }
4941
4942 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
4943 {
4944         struct e1000_hw *hw = &adapter->hw;
4945         u32 i, mac_reg;
4946         u16 phy_reg;
4947         int retval = 0;
4948
4949         /* copy MAC RARs to PHY RARs */
4950         e1000_copy_rx_addrs_to_phy_ich8lan(hw);
4951
4952         /* copy MAC MTA to PHY MTA */
4953         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
4954                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
4955                 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
4956                 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
4957         }
4958
4959         /* configure PHY Rx Control register */
4960         e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
4961         mac_reg = er32(RCTL);
4962         if (mac_reg & E1000_RCTL_UPE)
4963                 phy_reg |= BM_RCTL_UPE;
4964         if (mac_reg & E1000_RCTL_MPE)
4965                 phy_reg |= BM_RCTL_MPE;
4966         phy_reg &= ~(BM_RCTL_MO_MASK);
4967         if (mac_reg & E1000_RCTL_MO_3)
4968                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
4969                                 << BM_RCTL_MO_SHIFT);
4970         if (mac_reg & E1000_RCTL_BAM)
4971                 phy_reg |= BM_RCTL_BAM;
4972         if (mac_reg & E1000_RCTL_PMCF)
4973                 phy_reg |= BM_RCTL_PMCF;
4974         mac_reg = er32(CTRL);
4975         if (mac_reg & E1000_CTRL_RFCE)
4976                 phy_reg |= BM_RCTL_RFCE;
4977         e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
4978
4979         /* enable PHY wakeup in MAC register */
4980         ew32(WUFC, wufc);
4981         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
4982
4983         /* configure and enable PHY wakeup in PHY registers */
4984         e1e_wphy(&adapter->hw, BM_WUFC, wufc);
4985         e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
4986
4987         /* activate PHY wakeup */
4988         retval = hw->phy.ops.acquire(hw);
4989         if (retval) {
4990                 e_err("Could not acquire PHY\n");
4991                 return retval;
4992         }
4993         e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4994                                  (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
4995         retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
4996         if (retval) {
4997                 e_err("Could not read PHY page 769\n");
4998                 goto out;
4999         }
5000         phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5001         retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
5002         if (retval)
5003                 e_err("Could not set PHY Host Wakeup bit\n");
5004 out:
5005         hw->phy.ops.release(hw);
5006
5007         return retval;
5008 }
5009
5010 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5011                             bool runtime)
5012 {
5013         struct net_device *netdev = pci_get_drvdata(pdev);
5014         struct e1000_adapter *adapter = netdev_priv(netdev);
5015         struct e1000_hw *hw = &adapter->hw;
5016         u32 ctrl, ctrl_ext, rctl, status;
5017         /* Runtime suspend should only enable wakeup for link changes */
5018         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5019         int retval = 0;
5020
5021         netif_device_detach(netdev);
5022
5023         if (netif_running(netdev)) {
5024                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5025                 e1000e_down(adapter);
5026                 e1000_free_irq(adapter);
5027         }
5028         e1000e_reset_interrupt_capability(adapter);
5029
5030         retval = pci_save_state(pdev);
5031         if (retval)
5032                 return retval;
5033
5034         status = er32(STATUS);
5035         if (status & E1000_STATUS_LU)
5036                 wufc &= ~E1000_WUFC_LNKC;
5037
5038         if (wufc) {
5039                 e1000_setup_rctl(adapter);
5040                 e1000_set_multi(netdev);
5041
5042                 /* turn on all-multi mode if wake on multicast is enabled */
5043                 if (wufc & E1000_WUFC_MC) {
5044                         rctl = er32(RCTL);
5045                         rctl |= E1000_RCTL_MPE;
5046                         ew32(RCTL, rctl);
5047                 }
5048
5049                 ctrl = er32(CTRL);
5050                 /* advertise wake from D3Cold */
5051                 #define E1000_CTRL_ADVD3WUC 0x00100000
5052                 /* phy power management enable */
5053                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5054                 ctrl |= E1000_CTRL_ADVD3WUC;
5055                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5056                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5057                 ew32(CTRL, ctrl);
5058
5059                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5060                     adapter->hw.phy.media_type ==
5061                     e1000_media_type_internal_serdes) {
5062                         /* keep the laser running in D3 */
5063                         ctrl_ext = er32(CTRL_EXT);
5064                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5065                         ew32(CTRL_EXT, ctrl_ext);
5066                 }
5067
5068                 if (adapter->flags & FLAG_IS_ICH)
5069                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
5070
5071                 /* Allow time for pending master requests to run */
5072                 e1000e_disable_pcie_master(&adapter->hw);
5073
5074                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5075                         /* enable wakeup by the PHY */
5076                         retval = e1000_init_phy_wakeup(adapter, wufc);
5077                         if (retval)
5078                                 return retval;
5079                 } else {
5080                         /* enable wakeup by the MAC */
5081                         ew32(WUFC, wufc);
5082                         ew32(WUC, E1000_WUC_PME_EN);
5083                 }
5084         } else {
5085                 ew32(WUC, 0);
5086                 ew32(WUFC, 0);
5087         }
5088
5089         *enable_wake = !!wufc;
5090
5091         /* make sure adapter isn't asleep if manageability is enabled */
5092         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5093             (hw->mac.ops.check_mng_mode(hw)))
5094                 *enable_wake = true;
5095
5096         if (adapter->hw.phy.type == e1000_phy_igp_3)
5097                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5098
5099         /*
5100          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5101          * would have already happened in close and is redundant.
5102          */
5103         e1000_release_hw_control(adapter);
5104
5105         pci_disable_device(pdev);
5106
5107         return 0;
5108 }
5109
5110 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5111 {
5112         if (sleep && wake) {
5113                 pci_prepare_to_sleep(pdev);
5114                 return;
5115         }
5116
5117         pci_wake_from_d3(pdev, wake);
5118         pci_set_power_state(pdev, PCI_D3hot);
5119 }
5120
5121 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5122                                     bool wake)
5123 {
5124         struct net_device *netdev = pci_get_drvdata(pdev);
5125         struct e1000_adapter *adapter = netdev_priv(netdev);
5126
5127         /*
5128          * The pci-e switch on some quad port adapters will report a
5129          * correctable error when the MAC transitions from D0 to D3.  To
5130          * prevent this we need to mask off the correctable errors on the
5131          * downstream port of the pci-e switch.
5132          */
5133         if (adapter->flags & FLAG_IS_QUAD_PORT) {
5134                 struct pci_dev *us_dev = pdev->bus->self;
5135                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
5136                 u16 devctl;
5137
5138                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5139                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5140                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
5141
5142                 e1000_power_off(pdev, sleep, wake);
5143
5144                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5145         } else {
5146                 e1000_power_off(pdev, sleep, wake);
5147         }
5148 }
5149
5150 #ifdef CONFIG_PCIEASPM
5151 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5152 {
5153         pci_disable_link_state(pdev, state);
5154 }
5155 #else
5156 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5157 {
5158         int pos;
5159         u16 reg16;
5160
5161         /*
5162          * Both device and parent should have the same ASPM setting.
5163          * Disable ASPM in downstream component first and then upstream.
5164          */
5165         pos = pci_pcie_cap(pdev);
5166         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5167         reg16 &= ~state;
5168         pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5169
5170         if (!pdev->bus->self)
5171                 return;
5172
5173         pos = pci_pcie_cap(pdev->bus->self);
5174         pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5175         reg16 &= ~state;
5176         pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5177 }
5178 #endif
5179 void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5180 {
5181         dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5182                  (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5183                  (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5184
5185         __e1000e_disable_aspm(pdev, state);
5186 }
5187
5188 #ifdef CONFIG_PM_OPS
5189 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5190 {
5191         return !!adapter->tx_ring->buffer_info;
5192 }
5193
5194 static int __e1000_resume(struct pci_dev *pdev)
5195 {
5196         struct net_device *netdev = pci_get_drvdata(pdev);
5197         struct e1000_adapter *adapter = netdev_priv(netdev);
5198         struct e1000_hw *hw = &adapter->hw;
5199         u32 err;
5200
5201         pci_set_power_state(pdev, PCI_D0);
5202         pci_restore_state(pdev);
5203         pci_save_state(pdev);
5204         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5205                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5206
5207         e1000e_set_interrupt_capability(adapter);
5208         if (netif_running(netdev)) {
5209                 err = e1000_request_irq(adapter);
5210                 if (err)
5211                         return err;
5212         }
5213
5214         e1000e_power_up_phy(adapter);
5215
5216         /* report the system wakeup cause from S3/S4 */
5217         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5218                 u16 phy_data;
5219
5220                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5221                 if (phy_data) {
5222                         e_info("PHY Wakeup cause - %s\n",
5223                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5224                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5225                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5226                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5227                                 phy_data & E1000_WUS_LNKC ? "Link Status "
5228                                 " Change" : "other");
5229                 }
5230                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5231         } else {
5232                 u32 wus = er32(WUS);
5233                 if (wus) {
5234                         e_info("MAC Wakeup cause - %s\n",
5235                                 wus & E1000_WUS_EX ? "Unicast Packet" :
5236                                 wus & E1000_WUS_MC ? "Multicast Packet" :
5237                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
5238                                 wus & E1000_WUS_MAG ? "Magic Packet" :
5239                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
5240                                 "other");
5241                 }
5242                 ew32(WUS, ~0);
5243         }
5244
5245         e1000e_reset(adapter);
5246
5247         e1000_init_manageability_pt(adapter);
5248
5249         if (netif_running(netdev))
5250                 e1000e_up(adapter);
5251
5252         netif_device_attach(netdev);
5253
5254         /*
5255          * If the controller has AMT, do not set DRV_LOAD until the interface
5256          * is up.  For all other cases, let the f/w know that the h/w is now
5257          * under the control of the driver.
5258          */
5259         if (!(adapter->flags & FLAG_HAS_AMT))
5260                 e1000_get_hw_control(adapter);
5261
5262         return 0;
5263 }
5264
5265 #ifdef CONFIG_PM_SLEEP
5266 static int e1000_suspend(struct device *dev)
5267 {
5268         struct pci_dev *pdev = to_pci_dev(dev);
5269         int retval;
5270         bool wake;
5271
5272         retval = __e1000_shutdown(pdev, &wake, false);
5273         if (!retval)
5274                 e1000_complete_shutdown(pdev, true, wake);
5275
5276         return retval;
5277 }
5278
5279 static int e1000_resume(struct device *dev)
5280 {
5281         struct pci_dev *pdev = to_pci_dev(dev);
5282         struct net_device *netdev = pci_get_drvdata(pdev);
5283         struct e1000_adapter *adapter = netdev_priv(netdev);
5284
5285         if (e1000e_pm_ready(adapter))
5286                 adapter->idle_check = true;
5287
5288         return __e1000_resume(pdev);
5289 }
5290 #endif /* CONFIG_PM_SLEEP */
5291
5292 #ifdef CONFIG_PM_RUNTIME
5293 static int e1000_runtime_suspend(struct device *dev)
5294 {
5295         struct pci_dev *pdev = to_pci_dev(dev);
5296         struct net_device *netdev = pci_get_drvdata(pdev);
5297         struct e1000_adapter *adapter = netdev_priv(netdev);
5298
5299         if (e1000e_pm_ready(adapter)) {
5300                 bool wake;
5301
5302                 __e1000_shutdown(pdev, &wake, true);
5303         }
5304
5305         return 0;
5306 }
5307
5308 static int e1000_idle(struct device *dev)
5309 {
5310         struct pci_dev *pdev = to_pci_dev(dev);
5311         struct net_device *netdev = pci_get_drvdata(pdev);
5312         struct e1000_adapter *adapter = netdev_priv(netdev);
5313
5314         if (!e1000e_pm_ready(adapter))
5315                 return 0;
5316
5317         if (adapter->idle_check) {
5318                 adapter->idle_check = false;
5319                 if (!e1000e_has_link(adapter))
5320                         pm_schedule_suspend(dev, MSEC_PER_SEC);
5321         }
5322
5323         return -EBUSY;
5324 }
5325
5326 static int e1000_runtime_resume(struct device *dev)
5327 {
5328         struct pci_dev *pdev = to_pci_dev(dev);
5329         struct net_device *netdev = pci_get_drvdata(pdev);
5330         struct e1000_adapter *adapter = netdev_priv(netdev);
5331
5332         if (!e1000e_pm_ready(adapter))
5333                 return 0;
5334
5335         adapter->idle_check = !dev->power.runtime_auto;
5336         return __e1000_resume(pdev);
5337 }
5338 #endif /* CONFIG_PM_RUNTIME */
5339 #endif /* CONFIG_PM_OPS */
5340
5341 static void e1000_shutdown(struct pci_dev *pdev)
5342 {
5343         bool wake = false;
5344
5345         __e1000_shutdown(pdev, &wake, false);
5346
5347         if (system_state == SYSTEM_POWER_OFF)
5348                 e1000_complete_shutdown(pdev, false, wake);
5349 }
5350
5351 #ifdef CONFIG_NET_POLL_CONTROLLER
5352 /*
5353  * Polling 'interrupt' - used by things like netconsole to send skbs
5354  * without having to re-enable interrupts. It's not called while
5355  * the interrupt routine is executing.
5356  */
5357 static void e1000_netpoll(struct net_device *netdev)
5358 {
5359         struct e1000_adapter *adapter = netdev_priv(netdev);
5360
5361         disable_irq(adapter->pdev->irq);
5362         e1000_intr(adapter->pdev->irq, netdev);
5363
5364         enable_irq(adapter->pdev->irq);
5365 }
5366 #endif
5367
5368 /**
5369  * e1000_io_error_detected - called when PCI error is detected
5370  * @pdev: Pointer to PCI device
5371  * @state: The current pci connection state
5372  *
5373  * This function is called after a PCI bus error affecting
5374  * this device has been detected.
5375  */
5376 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5377                                                 pci_channel_state_t state)
5378 {
5379         struct net_device *netdev = pci_get_drvdata(pdev);
5380         struct e1000_adapter *adapter = netdev_priv(netdev);
5381
5382         netif_device_detach(netdev);
5383
5384         if (state == pci_channel_io_perm_failure)
5385                 return PCI_ERS_RESULT_DISCONNECT;
5386
5387         if (netif_running(netdev))
5388                 e1000e_down(adapter);
5389         pci_disable_device(pdev);
5390
5391         /* Request a slot slot reset. */
5392         return PCI_ERS_RESULT_NEED_RESET;
5393 }
5394
5395 /**
5396  * e1000_io_slot_reset - called after the pci bus has been reset.
5397  * @pdev: Pointer to PCI device
5398  *
5399  * Restart the card from scratch, as if from a cold-boot. Implementation
5400  * resembles the first-half of the e1000_resume routine.
5401  */
5402 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5403 {
5404         struct net_device *netdev = pci_get_drvdata(pdev);
5405         struct e1000_adapter *adapter = netdev_priv(netdev);
5406         struct e1000_hw *hw = &adapter->hw;
5407         int err;
5408         pci_ers_result_t result;
5409
5410         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5411                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5412         err = pci_enable_device_mem(pdev);
5413         if (err) {
5414                 dev_err(&pdev->dev,
5415                         "Cannot re-enable PCI device after reset.\n");
5416                 result = PCI_ERS_RESULT_DISCONNECT;
5417         } else {
5418                 pci_set_master(pdev);
5419                 pdev->state_saved = true;
5420                 pci_restore_state(pdev);
5421
5422                 pci_enable_wake(pdev, PCI_D3hot, 0);
5423                 pci_enable_wake(pdev, PCI_D3cold, 0);
5424
5425                 e1000e_reset(adapter);
5426                 ew32(WUS, ~0);
5427                 result = PCI_ERS_RESULT_RECOVERED;
5428         }
5429
5430         pci_cleanup_aer_uncorrect_error_status(pdev);
5431
5432         return result;
5433 }
5434
5435 /**
5436  * e1000_io_resume - called when traffic can start flowing again.
5437  * @pdev: Pointer to PCI device
5438  *
5439  * This callback is called when the error recovery driver tells us that
5440  * its OK to resume normal operation. Implementation resembles the
5441  * second-half of the e1000_resume routine.
5442  */
5443 static void e1000_io_resume(struct pci_dev *pdev)
5444 {
5445         struct net_device *netdev = pci_get_drvdata(pdev);
5446         struct e1000_adapter *adapter = netdev_priv(netdev);
5447
5448         e1000_init_manageability_pt(adapter);
5449
5450         if (netif_running(netdev)) {
5451                 if (e1000e_up(adapter)) {
5452                         dev_err(&pdev->dev,
5453                                 "can't bring device back up after reset\n");
5454                         return;
5455                 }
5456         }
5457
5458         netif_device_attach(netdev);
5459
5460         /*
5461          * If the controller has AMT, do not set DRV_LOAD until the interface
5462          * is up.  For all other cases, let the f/w know that the h/w is now
5463          * under the control of the driver.
5464          */
5465         if (!(adapter->flags & FLAG_HAS_AMT))
5466                 e1000_get_hw_control(adapter);
5467
5468 }
5469
5470 static void e1000_print_device_info(struct e1000_adapter *adapter)
5471 {
5472         struct e1000_hw *hw = &adapter->hw;
5473         struct net_device *netdev = adapter->netdev;
5474         u32 pba_num;
5475
5476         /* print bus type/speed/width info */
5477         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
5478                /* bus width */
5479                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5480                 "Width x1"),
5481                /* MAC address */
5482                netdev->dev_addr);
5483         e_info("Intel(R) PRO/%s Network Connection\n",
5484                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5485         e1000e_read_pba_num(hw, &pba_num);
5486         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
5487                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
5488 }
5489
5490 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5491 {
5492         struct e1000_hw *hw = &adapter->hw;
5493         int ret_val;
5494         u16 buf = 0;
5495
5496         if (hw->mac.type != e1000_82573)
5497                 return;
5498
5499         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
5500         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
5501                 /* Deep Smart Power Down (DSPD) */
5502                 dev_warn(&adapter->pdev->dev,
5503                          "Warning: detected DSPD enabled in EEPROM\n");
5504         }
5505 }
5506
5507 static const struct net_device_ops e1000e_netdev_ops = {
5508         .ndo_open               = e1000_open,
5509         .ndo_stop               = e1000_close,
5510         .ndo_start_xmit         = e1000_xmit_frame,
5511         .ndo_get_stats          = e1000_get_stats,
5512         .ndo_set_multicast_list = e1000_set_multi,
5513         .ndo_set_mac_address    = e1000_set_mac,
5514         .ndo_change_mtu         = e1000_change_mtu,
5515         .ndo_do_ioctl           = e1000_ioctl,
5516         .ndo_tx_timeout         = e1000_tx_timeout,
5517         .ndo_validate_addr      = eth_validate_addr,
5518
5519         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
5520         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
5521         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
5522 #ifdef CONFIG_NET_POLL_CONTROLLER
5523         .ndo_poll_controller    = e1000_netpoll,
5524 #endif
5525 };
5526
5527 /**
5528  * e1000_probe - Device Initialization Routine
5529  * @pdev: PCI device information struct
5530  * @ent: entry in e1000_pci_tbl
5531  *
5532  * Returns 0 on success, negative on failure
5533  *
5534  * e1000_probe initializes an adapter identified by a pci_dev structure.
5535  * The OS initialization, configuring of the adapter private structure,
5536  * and a hardware reset occur.
5537  **/
5538 static int __devinit e1000_probe(struct pci_dev *pdev,
5539                                  const struct pci_device_id *ent)
5540 {
5541         struct net_device *netdev;
5542         struct e1000_adapter *adapter;
5543         struct e1000_hw *hw;
5544         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
5545         resource_size_t mmio_start, mmio_len;
5546         resource_size_t flash_start, flash_len;
5547
5548         static int cards_found;
5549         int i, err, pci_using_dac;
5550         u16 eeprom_data = 0;
5551         u16 eeprom_apme_mask = E1000_EEPROM_APME;
5552
5553         if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
5554                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5555
5556         err = pci_enable_device_mem(pdev);
5557         if (err)
5558                 return err;
5559
5560         pci_using_dac = 0;
5561         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
5562         if (!err) {
5563                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
5564                 if (!err)
5565                         pci_using_dac = 1;
5566         } else {
5567                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
5568                 if (err) {
5569                         err = dma_set_coherent_mask(&pdev->dev,
5570                                                     DMA_BIT_MASK(32));
5571                         if (err) {
5572                                 dev_err(&pdev->dev, "No usable DMA "
5573                                         "configuration, aborting\n");
5574                                 goto err_dma;
5575                         }
5576                 }
5577         }
5578
5579         err = pci_request_selected_regions_exclusive(pdev,
5580                                           pci_select_bars(pdev, IORESOURCE_MEM),
5581                                           e1000e_driver_name);
5582         if (err)
5583                 goto err_pci_reg;
5584
5585         /* AER (Advanced Error Reporting) hooks */
5586         pci_enable_pcie_error_reporting(pdev);
5587
5588         pci_set_master(pdev);
5589         /* PCI config space info */
5590         err = pci_save_state(pdev);
5591         if (err)
5592                 goto err_alloc_etherdev;
5593
5594         err = -ENOMEM;
5595         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
5596         if (!netdev)
5597                 goto err_alloc_etherdev;
5598
5599         SET_NETDEV_DEV(netdev, &pdev->dev);
5600
5601         netdev->irq = pdev->irq;
5602
5603         pci_set_drvdata(pdev, netdev);
5604         adapter = netdev_priv(netdev);
5605         hw = &adapter->hw;
5606         adapter->netdev = netdev;
5607         adapter->pdev = pdev;
5608         adapter->ei = ei;
5609         adapter->pba = ei->pba;
5610         adapter->flags = ei->flags;
5611         adapter->flags2 = ei->flags2;
5612         adapter->hw.adapter = adapter;
5613         adapter->hw.mac.type = ei->mac;
5614         adapter->max_hw_frame_size = ei->max_hw_frame_size;
5615         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5616
5617         mmio_start = pci_resource_start(pdev, 0);
5618         mmio_len = pci_resource_len(pdev, 0);
5619
5620         err = -EIO;
5621         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5622         if (!adapter->hw.hw_addr)
5623                 goto err_ioremap;
5624
5625         if ((adapter->flags & FLAG_HAS_FLASH) &&
5626             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5627                 flash_start = pci_resource_start(pdev, 1);
5628                 flash_len = pci_resource_len(pdev, 1);
5629                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5630                 if (!adapter->hw.flash_address)
5631                         goto err_flashmap;
5632         }
5633
5634         /* construct the net_device struct */
5635         netdev->netdev_ops              = &e1000e_netdev_ops;
5636         e1000e_set_ethtool_ops(netdev);
5637         netdev->watchdog_timeo          = 5 * HZ;
5638         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5639         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5640
5641         netdev->mem_start = mmio_start;
5642         netdev->mem_end = mmio_start + mmio_len;
5643
5644         adapter->bd_number = cards_found++;
5645
5646         e1000e_check_options(adapter);
5647
5648         /* setup adapter struct */
5649         err = e1000_sw_init(adapter);
5650         if (err)
5651                 goto err_sw_init;
5652
5653         err = -EIO;
5654
5655         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5656         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5657         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5658
5659         err = ei->get_variants(adapter);
5660         if (err)
5661                 goto err_hw_init;
5662
5663         if ((adapter->flags & FLAG_IS_ICH) &&
5664             (adapter->flags & FLAG_READ_ONLY_NVM))
5665                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5666
5667         hw->mac.ops.get_bus_info(&adapter->hw);
5668
5669         adapter->hw.phy.autoneg_wait_to_complete = 0;
5670
5671         /* Copper options */
5672         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5673                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5674                 adapter->hw.phy.disable_polarity_correction = 0;
5675                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5676         }
5677
5678         if (e1000_check_reset_block(&adapter->hw))
5679                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5680
5681         netdev->features = NETIF_F_SG |
5682                            NETIF_F_HW_CSUM |
5683                            NETIF_F_HW_VLAN_TX |
5684                            NETIF_F_HW_VLAN_RX;
5685
5686         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5687                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5688
5689         netdev->features |= NETIF_F_TSO;
5690         netdev->features |= NETIF_F_TSO6;
5691
5692         netdev->vlan_features |= NETIF_F_TSO;
5693         netdev->vlan_features |= NETIF_F_TSO6;
5694         netdev->vlan_features |= NETIF_F_HW_CSUM;
5695         netdev->vlan_features |= NETIF_F_SG;
5696
5697         if (pci_using_dac)
5698                 netdev->features |= NETIF_F_HIGHDMA;
5699
5700         if (e1000e_enable_mng_pass_thru(&adapter->hw))
5701                 adapter->flags |= FLAG_MNG_PT_ENABLED;
5702
5703         /*
5704          * before reading the NVM, reset the controller to
5705          * put the device in a known good starting state
5706          */
5707         adapter->hw.mac.ops.reset_hw(&adapter->hw);
5708
5709         /*
5710          * systems with ASPM and others may see the checksum fail on the first
5711          * attempt. Let's give it a few tries
5712          */
5713         for (i = 0;; i++) {
5714                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5715                         break;
5716                 if (i == 2) {
5717                         e_err("The NVM Checksum Is Not Valid\n");
5718                         err = -EIO;
5719                         goto err_eeprom;
5720                 }
5721         }
5722
5723         e1000_eeprom_checks(adapter);
5724
5725         /* copy the MAC address */
5726         if (e1000e_read_mac_addr(&adapter->hw))
5727                 e_err("NVM Read Error while reading MAC address\n");
5728
5729         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5730         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5731
5732         if (!is_valid_ether_addr(netdev->perm_addr)) {
5733                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5734                 err = -EIO;
5735                 goto err_eeprom;
5736         }
5737
5738         init_timer(&adapter->watchdog_timer);
5739         adapter->watchdog_timer.function = &e1000_watchdog;
5740         adapter->watchdog_timer.data = (unsigned long) adapter;
5741
5742         init_timer(&adapter->phy_info_timer);
5743         adapter->phy_info_timer.function = &e1000_update_phy_info;
5744         adapter->phy_info_timer.data = (unsigned long) adapter;
5745
5746         INIT_WORK(&adapter->reset_task, e1000_reset_task);
5747         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
5748         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
5749         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
5750         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
5751
5752         /* Initialize link parameters. User can change them with ethtool */
5753         adapter->hw.mac.autoneg = 1;
5754         adapter->fc_autoneg = 1;
5755         adapter->hw.fc.requested_mode = e1000_fc_default;
5756         adapter->hw.fc.current_mode = e1000_fc_default;
5757         adapter->hw.phy.autoneg_advertised = 0x2f;
5758
5759         /* ring size defaults */
5760         adapter->rx_ring->count = 256;
5761         adapter->tx_ring->count = 256;
5762
5763         /*
5764          * Initial Wake on LAN setting - If APM wake is enabled in
5765          * the EEPROM, enable the ACPI Magic Packet filter
5766          */
5767         if (adapter->flags & FLAG_APME_IN_WUC) {
5768                 /* APME bit in EEPROM is mapped to WUC.APME */
5769                 eeprom_data = er32(WUC);
5770                 eeprom_apme_mask = E1000_WUC_APME;
5771                 if (eeprom_data & E1000_WUC_PHY_WAKE)
5772                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
5773         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5774                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5775                     (adapter->hw.bus.func == 1))
5776                         e1000_read_nvm(&adapter->hw,
5777                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5778                 else
5779                         e1000_read_nvm(&adapter->hw,
5780                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5781         }
5782
5783         /* fetch WoL from EEPROM */
5784         if (eeprom_data & eeprom_apme_mask)
5785                 adapter->eeprom_wol |= E1000_WUFC_MAG;
5786
5787         /*
5788          * now that we have the eeprom settings, apply the special cases
5789          * where the eeprom may be wrong or the board simply won't support
5790          * wake on lan on a particular port
5791          */
5792         if (!(adapter->flags & FLAG_HAS_WOL))
5793                 adapter->eeprom_wol = 0;
5794
5795         /* initialize the wol settings based on the eeprom settings */
5796         adapter->wol = adapter->eeprom_wol;
5797         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5798
5799         /* save off EEPROM version number */
5800         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
5801
5802         /* reset the hardware with the new settings */
5803         e1000e_reset(adapter);
5804
5805         /*
5806          * If the controller has AMT, do not set DRV_LOAD until the interface
5807          * is up.  For all other cases, let the f/w know that the h/w is now
5808          * under the control of the driver.
5809          */
5810         if (!(adapter->flags & FLAG_HAS_AMT))
5811                 e1000_get_hw_control(adapter);
5812
5813         strcpy(netdev->name, "eth%d");
5814         err = register_netdev(netdev);
5815         if (err)
5816                 goto err_register;
5817
5818         /* carrier off reporting is important to ethtool even BEFORE open */
5819         netif_carrier_off(netdev);
5820
5821         e1000_print_device_info(adapter);
5822
5823         if (pci_dev_run_wake(pdev)) {
5824                 pm_runtime_set_active(&pdev->dev);
5825                 pm_runtime_enable(&pdev->dev);
5826         }
5827         pm_schedule_suspend(&pdev->dev, MSEC_PER_SEC);
5828
5829         return 0;
5830
5831 err_register:
5832         if (!(adapter->flags & FLAG_HAS_AMT))
5833                 e1000_release_hw_control(adapter);
5834 err_eeprom:
5835         if (!e1000_check_reset_block(&adapter->hw))
5836                 e1000_phy_hw_reset(&adapter->hw);
5837 err_hw_init:
5838
5839         kfree(adapter->tx_ring);
5840         kfree(adapter->rx_ring);
5841 err_sw_init:
5842         if (adapter->hw.flash_address)
5843                 iounmap(adapter->hw.flash_address);
5844         e1000e_reset_interrupt_capability(adapter);
5845 err_flashmap:
5846         iounmap(adapter->hw.hw_addr);
5847 err_ioremap:
5848         free_netdev(netdev);
5849 err_alloc_etherdev:
5850         pci_release_selected_regions(pdev,
5851                                      pci_select_bars(pdev, IORESOURCE_MEM));
5852 err_pci_reg:
5853 err_dma:
5854         pci_disable_device(pdev);
5855         return err;
5856 }
5857
5858 /**
5859  * e1000_remove - Device Removal Routine
5860  * @pdev: PCI device information struct
5861  *
5862  * e1000_remove is called by the PCI subsystem to alert the driver
5863  * that it should release a PCI device.  The could be caused by a
5864  * Hot-Plug event, or because the driver is going to be removed from
5865  * memory.
5866  **/
5867 static void __devexit e1000_remove(struct pci_dev *pdev)
5868 {
5869         struct net_device *netdev = pci_get_drvdata(pdev);
5870         struct e1000_adapter *adapter = netdev_priv(netdev);
5871         bool down = test_bit(__E1000_DOWN, &adapter->state);
5872
5873         pm_runtime_get_sync(&pdev->dev);
5874
5875         /*
5876          * flush_scheduled work may reschedule our watchdog task, so
5877          * explicitly disable watchdog tasks from being rescheduled
5878          */
5879         if (!down)
5880                 set_bit(__E1000_DOWN, &adapter->state);
5881         del_timer_sync(&adapter->watchdog_timer);
5882         del_timer_sync(&adapter->phy_info_timer);
5883
5884         cancel_work_sync(&adapter->reset_task);
5885         cancel_work_sync(&adapter->watchdog_task);
5886         cancel_work_sync(&adapter->downshift_task);
5887         cancel_work_sync(&adapter->update_phy_task);
5888         cancel_work_sync(&adapter->print_hang_task);
5889         flush_scheduled_work();
5890
5891         if (!(netdev->flags & IFF_UP))
5892                 e1000_power_down_phy(adapter);
5893
5894         /* Don't lie to e1000_close() down the road. */
5895         if (!down)
5896                 clear_bit(__E1000_DOWN, &adapter->state);
5897         unregister_netdev(netdev);
5898
5899         if (pci_dev_run_wake(pdev)) {
5900                 pm_runtime_disable(&pdev->dev);
5901                 pm_runtime_set_suspended(&pdev->dev);
5902         }
5903         pm_runtime_put_noidle(&pdev->dev);
5904
5905         /*
5906          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5907          * would have already happened in close and is redundant.
5908          */
5909         e1000_release_hw_control(adapter);
5910
5911         e1000e_reset_interrupt_capability(adapter);
5912         kfree(adapter->tx_ring);
5913         kfree(adapter->rx_ring);
5914
5915         iounmap(adapter->hw.hw_addr);
5916         if (adapter->hw.flash_address)
5917                 iounmap(adapter->hw.flash_address);
5918         pci_release_selected_regions(pdev,
5919                                      pci_select_bars(pdev, IORESOURCE_MEM));
5920
5921         free_netdev(netdev);
5922
5923         /* AER disable */
5924         pci_disable_pcie_error_reporting(pdev);
5925
5926         pci_disable_device(pdev);
5927 }
5928
5929 /* PCI Error Recovery (ERS) */
5930 static struct pci_error_handlers e1000_err_handler = {
5931         .error_detected = e1000_io_error_detected,
5932         .slot_reset = e1000_io_slot_reset,
5933         .resume = e1000_io_resume,
5934 };
5935
5936 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
5937         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
5938         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
5939         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
5940         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
5941         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
5942         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
5943         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
5944         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
5945         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
5946
5947         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
5948         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
5949         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
5950         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
5951
5952         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
5953         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
5954         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
5955
5956         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
5957         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
5958         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
5959
5960         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
5961           board_80003es2lan },
5962         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
5963           board_80003es2lan },
5964         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
5965           board_80003es2lan },
5966         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
5967           board_80003es2lan },
5968
5969         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
5970         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
5971         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
5972         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
5973         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
5974         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
5975         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
5976         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
5977
5978         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
5979         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
5980         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
5981         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
5982         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
5983         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
5984         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
5985         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
5986         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
5987
5988         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
5989         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
5990         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
5991
5992         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
5993         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
5994         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
5995
5996         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
5997         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
5998         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
5999         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
6000
6001         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
6002         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
6003
6004         { }     /* terminate list */
6005 };
6006 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
6007
6008 #ifdef CONFIG_PM_OPS
6009 static const struct dev_pm_ops e1000_pm_ops = {
6010         SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
6011         SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
6012                                 e1000_runtime_resume, e1000_idle)
6013 };
6014 #endif
6015
6016 /* PCI Device API Driver */
6017 static struct pci_driver e1000_driver = {
6018         .name     = e1000e_driver_name,
6019         .id_table = e1000_pci_tbl,
6020         .probe    = e1000_probe,
6021         .remove   = __devexit_p(e1000_remove),
6022 #ifdef CONFIG_PM_OPS
6023         .driver.pm = &e1000_pm_ops,
6024 #endif
6025         .shutdown = e1000_shutdown,
6026         .err_handler = &e1000_err_handler
6027 };
6028
6029 /**
6030  * e1000_init_module - Driver Registration Routine
6031  *
6032  * e1000_init_module is the first routine called when the driver is
6033  * loaded. All it does is register with the PCI subsystem.
6034  **/
6035 static int __init e1000_init_module(void)
6036 {
6037         int ret;
6038         pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
6039                 e1000e_driver_version);
6040         pr_info("Copyright (c) 1999 - 2010 Intel Corporation.\n");
6041         ret = pci_register_driver(&e1000_driver);
6042
6043         return ret;
6044 }
6045 module_init(e1000_init_module);
6046
6047 /**
6048  * e1000_exit_module - Driver Exit Cleanup Routine
6049  *
6050  * e1000_exit_module is called just before the driver is removed
6051  * from memory.
6052  **/
6053 static void __exit e1000_exit_module(void)
6054 {
6055         pci_unregister_driver(&e1000_driver);
6056 }
6057 module_exit(e1000_exit_module);
6058
6059
6060 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
6061 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
6062 MODULE_LICENSE("GPL");
6063 MODULE_VERSION(DRV_VERSION);
6064
6065 /* e1000_main.c */