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
 479         skb_checksum_none_assert(skb);
 480
 481         /* Ignore Checksum bit is set */
 482         if (status & E1000_RXD_STAT_IXSM)
 483                 return;
 484         /* TCP/UDP checksum error bit is set */
 485         if (errors & E1000_RXD_ERR_TCPE) {
 486                 /* let the stack verify checksum errors */
 487                 adapter->hw_csum_err++;
 488                 return;
 489         }
 490
 491         /* TCP/UDP Checksum has not been calculated */
 492         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
 493                 return;
 494
 495         /* It must be a TCP or UDP packet with a valid checksum */
 496         if (status & E1000_RXD_STAT_TCPCS) {
 497                 /* TCP checksum is good */
 498                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 499         } else {
 500                 /*
 501                  * IP fragment with UDP payload
 502                  * Hardware complements the payload checksum, so we undo it
 503                  * and then put the value in host order for further stack use.
 504                  */
 505                 __sum16 sum = (__force __sum16)htons(csum);
 506                 skb->csum = csum_unfold(~sum);
 507                 skb->ip_summed = CHECKSUM_COMPLETE;
 508         }
 509         adapter->hw_csum_good++;
 510 }
 511
 512 /**
 513  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
 514  * @adapter: address of board private structure
 515  **/
 516 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
 517                                    int cleaned_count)
 518 {
 519         struct net_device *netdev = adapter->netdev;
 520         struct pci_dev *pdev = adapter->pdev;
 521         struct e1000_ring *rx_ring = adapter->rx_ring;
 522         struct e1000_rx_desc *rx_desc;
 523         struct e1000_buffer *buffer_info;
 524         struct sk_buff *skb;
 525         unsigned int i;
 526         unsigned int bufsz = adapter->rx_buffer_len;
 527
 528         i = rx_ring->next_to_use;
 529         buffer_info = &rx_ring->buffer_info[i];
 530
 531         while (cleaned_count--) {
 532                 skb = buffer_info->skb;
 533                 if (skb) {
 534                         skb_trim(skb, 0);
 535                         goto map_skb;
 536                 }
 537
 538                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 539                 if (!skb) {
 540                         /* Better luck next round */
 541                         adapter->alloc_rx_buff_failed++;
 542                         break;
 543                 }
 544
 545                 buffer_info->skb = skb;
 546 map_skb:
 547                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 548                                                   adapter->rx_buffer_len,
 549                                                   DMA_FROM_DEVICE);
 550                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 551                         dev_err(&pdev->dev, "RX DMA map failed\n");
 552                         adapter->rx_dma_failed++;
 553                         break;
 554                 }
 555
 556                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 557                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 558
 559                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 560                         /*
 561                          * Force memory writes to complete before letting h/w
 562                          * know there are new descriptors to fetch.  (Only
 563                          * applicable for weak-ordered memory model archs,
 564                          * such as IA-64).
 565                          */
 566                         wmb();
 567                         writel(i, adapter->hw.hw_addr + rx_ring->tail);
 568                 }
 569                 i++;
 570                 if (i == rx_ring->count)
 571                         i = 0;
 572                 buffer_info = &rx_ring->buffer_info[i];
 573         }
 574
 575         rx_ring->next_to_use = i;
 576 }
 577
 578 /**
 579  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
 580  * @adapter: address of board private structure
 581  **/
 582 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 583                                       int cleaned_count)
 584 {
 585         struct net_device *netdev = adapter->netdev;
 586         struct pci_dev *pdev = adapter->pdev;
 587         union e1000_rx_desc_packet_split *rx_desc;
 588         struct e1000_ring *rx_ring = adapter->rx_ring;
 589         struct e1000_buffer *buffer_info;
 590         struct e1000_ps_page *ps_page;
 591         struct sk_buff *skb;
 592         unsigned int i, j;
 593
 594         i = rx_ring->next_to_use;
 595         buffer_info = &rx_ring->buffer_info[i];
 596
 597         while (cleaned_count--) {
 598                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 599
 600                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 601                         ps_page = &buffer_info->ps_pages[j];
 602                         if (j >= adapter->rx_ps_pages) {
 603                                 /* all unused desc entries get hw null ptr */
 604                                 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
 605                                 continue;
 606                         }
 607                         if (!ps_page->page) {
 608                                 ps_page->page = alloc_page(GFP_ATOMIC);
 609                                 if (!ps_page->page) {
 610                                         adapter->alloc_rx_buff_failed++;
 611                                         goto no_buffers;
 612                                 }
 613                                 ps_page->dma = dma_map_page(&pdev->dev,
 614                                                             ps_page->page,
 615                                                             0, PAGE_SIZE,
 616                                                             DMA_FROM_DEVICE);
 617                                 if (dma_mapping_error(&pdev->dev,
 618                                                       ps_page->dma)) {
 619                                         dev_err(&adapter->pdev->dev,
 620                                           "RX DMA page map failed\n");
 621                                         adapter->rx_dma_failed++;
 622                                         goto no_buffers;
 623                                 }
 624                         }
 625                         /*
 626                          * Refresh the desc even if buffer_addrs
 627                          * didn't change because each write-back
 628                          * erases this info.
 629                          */
 630                         rx_desc->read.buffer_addr[j+1] =
 631                              cpu_to_le64(ps_page->dma);
 632                 }
 633
 634                 skb = netdev_alloc_skb_ip_align(netdev,
 635                                                 adapter->rx_ps_bsize0);
 636
 637                 if (!skb) {
 638                         adapter->alloc_rx_buff_failed++;
 639                         break;
 640                 }
 641
 642                 buffer_info->skb = skb;
 643                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 644                                                   adapter->rx_ps_bsize0,
 645                                                   DMA_FROM_DEVICE);
 646                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 647                         dev_err(&pdev->dev, "RX DMA map failed\n");
 648                         adapter->rx_dma_failed++;
 649                         /* cleanup skb */
 650                         dev_kfree_skb_any(skb);
 651                         buffer_info->skb = NULL;
 652                         break;
 653                 }
 654
 655                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 656
 657                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 658                         /*
 659                          * Force memory writes to complete before letting h/w
 660                          * know there are new descriptors to fetch.  (Only
 661                          * applicable for weak-ordered memory model archs,
 662                          * such as IA-64).
 663                          */
 664                         wmb();
 665                         writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
 666                 }
 667
 668                 i++;
 669                 if (i == rx_ring->count)
 670                         i = 0;
 671                 buffer_info = &rx_ring->buffer_info[i];
 672         }
 673
 674 no_buffers:
 675         rx_ring->next_to_use = i;
 676 }
 677
 678 /**
 679  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
 680  * @adapter: address of board private structure
 681  * @cleaned_count: number of buffers to allocate this pass
 682  **/
 683
 684 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
 685                                          int cleaned_count)
 686 {
 687         struct net_device *netdev = adapter->netdev;
 688         struct pci_dev *pdev = adapter->pdev;
 689         struct e1000_rx_desc *rx_desc;
 690         struct e1000_ring *rx_ring = adapter->rx_ring;
 691         struct e1000_buffer *buffer_info;
 692         struct sk_buff *skb;
 693         unsigned int i;
 694         unsigned int bufsz = 256 - 16 /* for skb_reserve */;
 695
 696         i = rx_ring->next_to_use;
 697         buffer_info = &rx_ring->buffer_info[i];
 698
 699         while (cleaned_count--) {
 700                 skb = buffer_info->skb;
 701                 if (skb) {
 702                         skb_trim(skb, 0);
 703                         goto check_page;
 704                 }
 705
 706                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 707                 if (unlikely(!skb)) {
 708                         /* Better luck next round */
 709                         adapter->alloc_rx_buff_failed++;
 710                         break;
 711                 }
 712
 713                 buffer_info->skb = skb;
 714 check_page:
 715                 /* allocate a new page if necessary */
 716                 if (!buffer_info->page) {
 717                         buffer_info->page = alloc_page(GFP_ATOMIC);
 718                         if (unlikely(!buffer_info->page)) {
 719                                 adapter->alloc_rx_buff_failed++;
 720                                 break;
 721                         }
 722                 }
 723
 724                 if (!buffer_info->dma)
 725                         buffer_info->dma = dma_map_page(&pdev->dev,
 726                                                         buffer_info->page, 0,
 727                                                         PAGE_SIZE,
 728                                                         DMA_FROM_DEVICE);
 729
 730                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 731                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 732
 733                 if (unlikely(++i == rx_ring->count))
 734                         i = 0;
 735                 buffer_info = &rx_ring->buffer_info[i];
 736         }
 737
 738         if (likely(rx_ring->next_to_use != i)) {
 739                 rx_ring->next_to_use = i;
 740                 if (unlikely(i-- == 0))
 741                         i = (rx_ring->count - 1);
 742
 743                 /* Force memory writes to complete before letting h/w
 744                  * know there are new descriptors to fetch.  (Only
 745                  * applicable for weak-ordered memory model archs,
 746                  * such as IA-64). */
 747                 wmb();
 748                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 749         }
 750 }
 751
 752 /**
 753  * e1000_clean_rx_irq - Send received data up the network stack; legacy
 754  * @adapter: board private structure
 755  *
 756  * the return value indicates whether actual cleaning was done, there
 757  * is no guarantee that everything was cleaned
 758  **/
 759 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
 760                                int *work_done, int work_to_do)
 761 {
 762         struct net_device *netdev = adapter->netdev;
 763         struct pci_dev *pdev = adapter->pdev;
 764         struct e1000_hw *hw = &adapter->hw;
 765         struct e1000_ring *rx_ring = adapter->rx_ring;
 766         struct e1000_rx_desc *rx_desc, *next_rxd;
 767         struct e1000_buffer *buffer_info, *next_buffer;
 768         u32 length;
 769         unsigned int i;
 770         int cleaned_count = 0;
 771         bool cleaned = 0;
 772         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 773
 774         i = rx_ring->next_to_clean;
 775         rx_desc = E1000_RX_DESC(*rx_ring, i);
 776         buffer_info = &rx_ring->buffer_info[i];
 777
 778         while (rx_desc->status & E1000_RXD_STAT_DD) {
 779                 struct sk_buff *skb;
 780                 u8 status;
 781
 782                 if (*work_done >= work_to_do)
 783                         break;
 784                 (*work_done)++;
 785                 rmb();  /* read descriptor and rx_buffer_info after status DD */
 786
 787                 status = rx_desc->status;
 788                 skb = buffer_info->skb;
 789                 buffer_info->skb = NULL;
 790
 791                 prefetch(skb->data - NET_IP_ALIGN);
 792
 793                 i++;
 794                 if (i == rx_ring->count)
 795                         i = 0;
 796                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 797                 prefetch(next_rxd);
 798
 799                 next_buffer = &rx_ring->buffer_info[i];
 800
 801                 cleaned = 1;
 802                 cleaned_count++;
 803                 dma_unmap_single(&pdev->dev,
 804                                  buffer_info->dma,
 805                                  adapter->rx_buffer_len,
 806                                  DMA_FROM_DEVICE);
 807                 buffer_info->dma = 0;
 808
 809                 length = le16_to_cpu(rx_desc->length);
 810
 811                 /*
 812                  * !EOP means multiple descriptors were used to store a single
 813                  * packet, if that's the case we need to toss it.  In fact, we
 814                  * need to toss every packet with the EOP bit clear and the
 815                  * next frame that _does_ have the EOP bit set, as it is by
 816                  * definition only a frame fragment
 817                  */
 818                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
 819                         adapter->flags2 |= FLAG2_IS_DISCARDING;
 820
 821                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
 822                         /* All receives must fit into a single buffer */
 823                         e_dbg("Receive packet consumed multiple buffers\n");
 824                         /* recycle */
 825                         buffer_info->skb = skb;
 826                         if (status & E1000_RXD_STAT_EOP)
 827                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
 828                         goto next_desc;
 829                 }
 830
 831                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
 832                         /* recycle */
 833                         buffer_info->skb = skb;
 834                         goto next_desc;
 835                 }
 836
 837                 /* adjust length to remove Ethernet CRC */
 838                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 839                         length -= 4;
 840
 841                 total_rx_bytes += length;
 842                 total_rx_packets++;
 843
 844                 /*
 845                  * code added for copybreak, this should improve
 846                  * performance for small packets with large amounts
 847                  * of reassembly being done in the stack
 848                  */
 849                 if (length < copybreak) {
 850                         struct sk_buff *new_skb =
 851                             netdev_alloc_skb_ip_align(netdev, length);
 852                         if (new_skb) {
 853                                 skb_copy_to_linear_data_offset(new_skb,
 854                                                                -NET_IP_ALIGN,
 855                                                                (skb->data -
 856                                                                 NET_IP_ALIGN),
 857                                                                (length +
 858                                                                 NET_IP_ALIGN));
 859                                 /* save the skb in buffer_info as good */
 860                                 buffer_info->skb = skb;
 861                                 skb = new_skb;
 862                         }
 863                         /* else just continue with the old one */
 864                 }
 865                 /* end copybreak code */
 866                 skb_put(skb, length);
 867
 868                 /* Receive Checksum Offload */
 869                 e1000_rx_checksum(adapter,
 870                                   (u32)(status) |
 871                                   ((u32)(rx_desc->errors) << 24),
 872                                   le16_to_cpu(rx_desc->csum), skb);
 873
 874                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
 875
 876 next_desc:
 877                 rx_desc->status = 0;
 878
 879                 /* return some buffers to hardware, one at a time is too slow */
 880                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 881                         adapter->alloc_rx_buf(adapter, cleaned_count);
 882                         cleaned_count = 0;
 883                 }
 884
 885                 /* use prefetched values */
 886                 rx_desc = next_rxd;
 887                 buffer_info = next_buffer;
 888         }
 889         rx_ring->next_to_clean = i;
 890
 891         cleaned_count = e1000_desc_unused(rx_ring);
 892         if (cleaned_count)
 893                 adapter->alloc_rx_buf(adapter, cleaned_count);
 894
 895         adapter->total_rx_bytes += total_rx_bytes;
 896         adapter->total_rx_packets += total_rx_packets;
 897         netdev->stats.rx_bytes += total_rx_bytes;
 898         netdev->stats.rx_packets += total_rx_packets;
 899         return cleaned;
 900 }
 901
 902 static void e1000_put_txbuf(struct e1000_adapter *adapter,
 903                              struct e1000_buffer *buffer_info)
 904 {
 905         if (buffer_info->dma) {
 906                 if (buffer_info->mapped_as_page)
 907                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
 908                                        buffer_info->length, DMA_TO_DEVICE);
 909                 else
 910                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
 911                                          buffer_info->length, DMA_TO_DEVICE);
 912                 buffer_info->dma = 0;
 913         }
 914         if (buffer_info->skb) {
 915                 dev_kfree_skb_any(buffer_info->skb);
 916                 buffer_info->skb = NULL;
 917         }
 918         buffer_info->time_stamp = 0;
 919 }
 920
 921 static void e1000_print_hw_hang(struct work_struct *work)
 922 {
 923         struct e1000_adapter *adapter = container_of(work,
 924                                                      struct e1000_adapter,
 925                                                      print_hang_task);
 926         struct e1000_ring *tx_ring = adapter->tx_ring;
 927         unsigned int i = tx_ring->next_to_clean;
 928         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 929         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
 930         struct e1000_hw *hw = &adapter->hw;
 931         u16 phy_status, phy_1000t_status, phy_ext_status;
 932         u16 pci_status;
 933
 934         e1e_rphy(hw, PHY_STATUS, &phy_status);
 935         e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
 936         e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
 937
 938         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
 939
 940         /* detected Hardware unit hang */
 941         e_err("Detected Hardware Unit Hang:\n"
 942               "  TDH                  <%x>\n"
 943               "  TDT                  <%x>\n"
 944               "  next_to_use          <%x>\n"
 945               "  next_to_clean        <%x>\n"
 946               "buffer_info[next_to_clean]:\n"
 947               "  time_stamp           <%lx>\n"
 948               "  next_to_watch        <%x>\n"
 949               "  jiffies              <%lx>\n"
 950               "  next_to_watch.status <%x>\n"
 951               "MAC Status             <%x>\n"
 952               "PHY Status             <%x>\n"
 953               "PHY 1000BASE-T Status  <%x>\n"
 954               "PHY Extended Status    <%x>\n"
 955               "PCI Status             <%x>\n",
 956               readl(adapter->hw.hw_addr + tx_ring->head),
 957               readl(adapter->hw.hw_addr + tx_ring->tail),
 958               tx_ring->next_to_use,
 959               tx_ring->next_to_clean,
 960               tx_ring->buffer_info[eop].time_stamp,
 961               eop,
 962               jiffies,
 963               eop_desc->upper.fields.status,
 964               er32(STATUS),
 965               phy_status,
 966               phy_1000t_status,
 967               phy_ext_status,
 968               pci_status);
 969 }
 970
 971 /**
 972  * e1000_clean_tx_irq - Reclaim resources after transmit completes
 973  * @adapter: board private structure
 974  *
 975  * the return value indicates whether actual cleaning was done, there
 976  * is no guarantee that everything was cleaned
 977  **/
 978 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
 979 {
 980         struct net_device *netdev = adapter->netdev;
 981         struct e1000_hw *hw = &adapter->hw;
 982         struct e1000_ring *tx_ring = adapter->tx_ring;
 983         struct e1000_tx_desc *tx_desc, *eop_desc;
 984         struct e1000_buffer *buffer_info;
 985         unsigned int i, eop;
 986         unsigned int count = 0;
 987         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
 988
 989         i = tx_ring->next_to_clean;
 990         eop = tx_ring->buffer_info[i].next_to_watch;
 991         eop_desc = E1000_TX_DESC(*tx_ring, eop);
 992
 993         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
 994                (count < tx_ring->count)) {
 995                 bool cleaned = false;
 996                 rmb(); /* read buffer_info after eop_desc */
 997                 for (; !cleaned; count++) {
 998                         tx_desc = E1000_TX_DESC(*tx_ring, i);
 999                         buffer_info = &tx_ring->buffer_info[i];
1000                         cleaned = (i == eop);
1001
1002                         if (cleaned) {
1003                                 total_tx_packets += buffer_info->segs;
1004                                 total_tx_bytes += buffer_info->bytecount;
1005                         }
1006
1007                         e1000_put_txbuf(adapter, buffer_info);
1008                         tx_desc->upper.data = 0;
1009
1010                         i++;
1011                         if (i == tx_ring->count)
1012                                 i = 0;
1013                 }
1014
1015                 if (i == tx_ring->next_to_use)
1016                         break;
1017                 eop = tx_ring->buffer_info[i].next_to_watch;
1018                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1019         }
1020
1021         tx_ring->next_to_clean = i;
1022
1023 #define TX_WAKE_THRESHOLD 32
1024         if (count && netif_carrier_ok(netdev) &&
1025             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1026                 /* Make sure that anybody stopping the queue after this
1027                  * sees the new next_to_clean.
1028                  */
1029                 smp_mb();
1030
1031                 if (netif_queue_stopped(netdev) &&
1032                     !(test_bit(__E1000_DOWN, &adapter->state))) {
1033                         netif_wake_queue(netdev);
1034                         ++adapter->restart_queue;
1035                 }
1036         }
1037
1038         if (adapter->detect_tx_hung) {
1039                 /*
1040                  * Detect a transmit hang in hardware, this serializes the
1041                  * check with the clearing of time_stamp and movement of i
1042                  */
1043                 adapter->detect_tx_hung = 0;
1044                 if (tx_ring->buffer_info[i].time_stamp &&
1045                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1046                                + (adapter->tx_timeout_factor * HZ)) &&
1047                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
1048                         schedule_work(&adapter->print_hang_task);
1049                         netif_stop_queue(netdev);
1050                 }
1051         }
1052         adapter->total_tx_bytes += total_tx_bytes;
1053         adapter->total_tx_packets += total_tx_packets;
1054         netdev->stats.tx_bytes += total_tx_bytes;
1055         netdev->stats.tx_packets += total_tx_packets;
1056         return count < tx_ring->count;
1057 }
1058
1059 /**
1060  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1061  * @adapter: board private structure
1062  *
1063  * the return value indicates whether actual cleaning was done, there
1064  * is no guarantee that everything was cleaned
1065  **/
1066 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
1067                                   int *work_done, int work_to_do)
1068 {
1069         struct e1000_hw *hw = &adapter->hw;
1070         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1071         struct net_device *netdev = adapter->netdev;
1072         struct pci_dev *pdev = adapter->pdev;
1073         struct e1000_ring *rx_ring = adapter->rx_ring;
1074         struct e1000_buffer *buffer_info, *next_buffer;
1075         struct e1000_ps_page *ps_page;
1076         struct sk_buff *skb;
1077         unsigned int i, j;
1078         u32 length, staterr;
1079         int cleaned_count = 0;
1080         bool cleaned = 0;
1081         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1082
1083         i = rx_ring->next_to_clean;
1084         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1085         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1086         buffer_info = &rx_ring->buffer_info[i];
1087
1088         while (staterr & E1000_RXD_STAT_DD) {
1089                 if (*work_done >= work_to_do)
1090                         break;
1091                 (*work_done)++;
1092                 skb = buffer_info->skb;
1093                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1094
1095                 /* in the packet split case this is header only */
1096                 prefetch(skb->data - NET_IP_ALIGN);
1097
1098                 i++;
1099                 if (i == rx_ring->count)
1100                         i = 0;
1101                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1102                 prefetch(next_rxd);
1103
1104                 next_buffer = &rx_ring->buffer_info[i];
1105
1106                 cleaned = 1;
1107                 cleaned_count++;
1108                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1109                                  adapter->rx_ps_bsize0,
1110                                  DMA_FROM_DEVICE);
1111                 buffer_info->dma = 0;
1112
1113                 /* see !EOP comment in other rx routine */
1114                 if (!(staterr & E1000_RXD_STAT_EOP))
1115                         adapter->flags2 |= FLAG2_IS_DISCARDING;
1116
1117                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1118                         e_dbg("Packet Split buffers didn't pick up the full "
1119                               "packet\n");
1120                         dev_kfree_skb_irq(skb);
1121                         if (staterr & E1000_RXD_STAT_EOP)
1122                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1123                         goto next_desc;
1124                 }
1125
1126                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
1127                         dev_kfree_skb_irq(skb);
1128                         goto next_desc;
1129                 }
1130
1131                 length = le16_to_cpu(rx_desc->wb.middle.length0);
1132
1133                 if (!length) {
1134                         e_dbg("Last part of the packet spanning multiple "
1135                               "descriptors\n");
1136                         dev_kfree_skb_irq(skb);
1137                         goto next_desc;
1138                 }
1139
1140                 /* Good Receive */
1141                 skb_put(skb, length);
1142
1143                 {
1144                 /*
1145                  * this looks ugly, but it seems compiler issues make it
1146                  * more efficient than reusing j
1147                  */
1148                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1149
1150                 /*
1151                  * page alloc/put takes too long and effects small packet
1152                  * throughput, so unsplit small packets and save the alloc/put
1153                  * only valid in softirq (napi) context to call kmap_*
1154                  */
1155                 if (l1 && (l1 <= copybreak) &&
1156                     ((length + l1) <= adapter->rx_ps_bsize0)) {
1157                         u8 *vaddr;
1158
1159                         ps_page = &buffer_info->ps_pages[0];
1160
1161                         /*
1162                          * there is no documentation about how to call
1163                          * kmap_atomic, so we can't hold the mapping
1164                          * very long
1165                          */
1166                         dma_sync_single_for_cpu(&pdev->dev, ps_page->dma,
1167                                                 PAGE_SIZE, DMA_FROM_DEVICE);
1168                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
1169                         memcpy(skb_tail_pointer(skb), vaddr, l1);
1170                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
1171                         dma_sync_single_for_device(&pdev->dev, ps_page->dma,
1172                                                    PAGE_SIZE, DMA_FROM_DEVICE);
1173
1174                         /* remove the CRC */
1175                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1176                                 l1 -= 4;
1177
1178                         skb_put(skb, l1);
1179                         goto copydone;
1180                 } /* if */
1181                 }
1182
1183                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1184                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1185                         if (!length)
1186                                 break;
1187
1188                         ps_page = &buffer_info->ps_pages[j];
1189                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1190                                        DMA_FROM_DEVICE);
1191                         ps_page->dma = 0;
1192                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1193                         ps_page->page = NULL;
1194                         skb->len += length;
1195                         skb->data_len += length;
1196                         skb->truesize += length;
1197                 }
1198
1199                 /* strip the ethernet crc, problem is we're using pages now so
1200                  * this whole operation can get a little cpu intensive
1201                  */
1202                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1203                         pskb_trim(skb, skb->len - 4);
1204
1205 copydone:
1206                 total_rx_bytes += skb->len;
1207                 total_rx_packets++;
1208
1209                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
1210                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
1211
1212                 if (rx_desc->wb.upper.header_status &
1213                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1214                         adapter->rx_hdr_split++;
1215
1216                 e1000_receive_skb(adapter, netdev, skb,
1217                                   staterr, rx_desc->wb.middle.vlan);
1218
1219 next_desc:
1220                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1221                 buffer_info->skb = NULL;
1222
1223                 /* return some buffers to hardware, one at a time is too slow */
1224                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1225                         adapter->alloc_rx_buf(adapter, cleaned_count);
1226                         cleaned_count = 0;
1227                 }
1228
1229                 /* use prefetched values */
1230                 rx_desc = next_rxd;
1231                 buffer_info = next_buffer;
1232
1233                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1234         }
1235         rx_ring->next_to_clean = i;
1236
1237         cleaned_count = e1000_desc_unused(rx_ring);
1238         if (cleaned_count)
1239                 adapter->alloc_rx_buf(adapter, cleaned_count);
1240
1241         adapter->total_rx_bytes += total_rx_bytes;
1242         adapter->total_rx_packets += total_rx_packets;
1243         netdev->stats.rx_bytes += total_rx_bytes;
1244         netdev->stats.rx_packets += total_rx_packets;
1245         return cleaned;
1246 }
1247
1248 /**
1249  * e1000_consume_page - helper function
1250  **/
1251 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1252                                u16 length)
1253 {
1254         bi->page = NULL;
1255         skb->len += length;
1256         skb->data_len += length;
1257         skb->truesize += length;
1258 }
1259
1260 /**
1261  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1262  * @adapter: board private structure
1263  *
1264  * the return value indicates whether actual cleaning was done, there
1265  * is no guarantee that everything was cleaned
1266  **/
1267
1268 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
1269                                      int *work_done, int work_to_do)
1270 {
1271         struct net_device *netdev = adapter->netdev;
1272         struct pci_dev *pdev = adapter->pdev;
1273         struct e1000_ring *rx_ring = adapter->rx_ring;
1274         struct e1000_rx_desc *rx_desc, *next_rxd;
1275         struct e1000_buffer *buffer_info, *next_buffer;
1276         u32 length;
1277         unsigned int i;
1278         int cleaned_count = 0;
1279         bool cleaned = false;
1280         unsigned int total_rx_bytes=0, total_rx_packets=0;
1281
1282         i = rx_ring->next_to_clean;
1283         rx_desc = E1000_RX_DESC(*rx_ring, i);
1284         buffer_info = &rx_ring->buffer_info[i];
1285
1286         while (rx_desc->status & E1000_RXD_STAT_DD) {
1287                 struct sk_buff *skb;
1288                 u8 status;
1289
1290                 if (*work_done >= work_to_do)
1291                         break;
1292                 (*work_done)++;
1293                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1294
1295                 status = rx_desc->status;
1296                 skb = buffer_info->skb;
1297                 buffer_info->skb = NULL;
1298
1299                 ++i;
1300                 if (i == rx_ring->count)
1301                         i = 0;
1302                 next_rxd = E1000_RX_DESC(*rx_ring, i);
1303                 prefetch(next_rxd);
1304
1305                 next_buffer = &rx_ring->buffer_info[i];
1306
1307                 cleaned = true;
1308                 cleaned_count++;
1309                 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1310                                DMA_FROM_DEVICE);
1311                 buffer_info->dma = 0;
1312
1313                 length = le16_to_cpu(rx_desc->length);
1314
1315                 /* errors is only valid for DD + EOP descriptors */
1316                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
1317                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
1318                                 /* recycle both page and skb */
1319                                 buffer_info->skb = skb;
1320                                 /* an error means any chain goes out the window
1321                                  * too */
1322                                 if (rx_ring->rx_skb_top)
1323                                         dev_kfree_skb(rx_ring->rx_skb_top);
1324                                 rx_ring->rx_skb_top = NULL;
1325                                 goto next_desc;
1326                 }
1327
1328 #define rxtop rx_ring->rx_skb_top
1329                 if (!(status & E1000_RXD_STAT_EOP)) {
1330                         /* this descriptor is only the beginning (or middle) */
1331                         if (!rxtop) {
1332                                 /* this is the beginning of a chain */
1333                                 rxtop = skb;
1334                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1335                                                    0, length);
1336                         } else {
1337                                 /* this is the middle of a chain */
1338                                 skb_fill_page_desc(rxtop,
1339                                     skb_shinfo(rxtop)->nr_frags,
1340                                     buffer_info->page, 0, length);
1341                                 /* re-use the skb, only consumed the page */
1342                                 buffer_info->skb = skb;
1343                         }
1344                         e1000_consume_page(buffer_info, rxtop, length);
1345                         goto next_desc;
1346                 } else {
1347                         if (rxtop) {
1348                                 /* end of the chain */
1349                                 skb_fill_page_desc(rxtop,
1350                                     skb_shinfo(rxtop)->nr_frags,
1351                                     buffer_info->page, 0, length);
1352                                 /* re-use the current skb, we only consumed the
1353                                  * page */
1354                                 buffer_info->skb = skb;
1355                                 skb = rxtop;
1356                                 rxtop = NULL;
1357                                 e1000_consume_page(buffer_info, skb, length);
1358                         } else {
1359                                 /* no chain, got EOP, this buf is the packet
1360                                  * copybreak to save the put_page/alloc_page */
1361                                 if (length <= copybreak &&
1362                                     skb_tailroom(skb) >= length) {
1363                                         u8 *vaddr;
1364                                         vaddr = kmap_atomic(buffer_info->page,
1365                                                            KM_SKB_DATA_SOFTIRQ);
1366                                         memcpy(skb_tail_pointer(skb), vaddr,
1367                                                length);
1368                                         kunmap_atomic(vaddr,
1369                                                       KM_SKB_DATA_SOFTIRQ);
1370                                         /* re-use the page, so don't erase
1371                                          * buffer_info->page */
1372                                         skb_put(skb, length);
1373                                 } else {
1374                                         skb_fill_page_desc(skb, 0,
1375                                                            buffer_info->page, 0,
1376                                                            length);
1377                                         e1000_consume_page(buffer_info, skb,
1378                                                            length);
1379                                 }
1380                         }
1381                 }
1382
1383                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1384                 e1000_rx_checksum(adapter,
1385                                   (u32)(status) |
1386                                   ((u32)(rx_desc->errors) << 24),
1387                                   le16_to_cpu(rx_desc->csum), skb);
1388
1389                 /* probably a little skewed due to removing CRC */
1390                 total_rx_bytes += skb->len;
1391                 total_rx_packets++;
1392
1393                 /* eth type trans needs skb->data to point to something */
1394                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1395                         e_err("pskb_may_pull failed.\n");
1396                         dev_kfree_skb(skb);
1397                         goto next_desc;
1398                 }
1399
1400                 e1000_receive_skb(adapter, netdev, skb, status,
1401                                   rx_desc->special);
1402
1403 next_desc:
1404                 rx_desc->status = 0;
1405
1406                 /* return some buffers to hardware, one at a time is too slow */
1407                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1408                         adapter->alloc_rx_buf(adapter, cleaned_count);
1409                         cleaned_count = 0;
1410                 }
1411
1412                 /* use prefetched values */
1413                 rx_desc = next_rxd;
1414                 buffer_info = next_buffer;
1415         }
1416         rx_ring->next_to_clean = i;
1417
1418         cleaned_count = e1000_desc_unused(rx_ring);
1419         if (cleaned_count)
1420                 adapter->alloc_rx_buf(adapter, cleaned_count);
1421
1422         adapter->total_rx_bytes += total_rx_bytes;
1423         adapter->total_rx_packets += total_rx_packets;
1424         netdev->stats.rx_bytes += total_rx_bytes;
1425         netdev->stats.rx_packets += total_rx_packets;
1426         return cleaned;
1427 }
1428
1429 /**
1430  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1431  * @adapter: board private structure
1432  **/
1433 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1434 {
1435         struct e1000_ring *rx_ring = adapter->rx_ring;
1436         struct e1000_buffer *buffer_info;
1437         struct e1000_ps_page *ps_page;
1438         struct pci_dev *pdev = adapter->pdev;
1439         unsigned int i, j;
1440
1441         /* Free all the Rx ring sk_buffs */
1442         for (i = 0; i < rx_ring->count; i++) {
1443                 buffer_info = &rx_ring->buffer_info[i];
1444                 if (buffer_info->dma) {
1445                         if (adapter->clean_rx == e1000_clean_rx_irq)
1446                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1447                                                  adapter->rx_buffer_len,
1448                                                  DMA_FROM_DEVICE);
1449                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1450                                 dma_unmap_page(&pdev->dev, buffer_info->dma,
1451                                                PAGE_SIZE,
1452                                                DMA_FROM_DEVICE);
1453                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1454                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1455                                                  adapter->rx_ps_bsize0,
1456                                                  DMA_FROM_DEVICE);
1457                         buffer_info->dma = 0;
1458                 }
1459
1460                 if (buffer_info->page) {
1461                         put_page(buffer_info->page);
1462                         buffer_info->page = NULL;
1463                 }
1464
1465                 if (buffer_info->skb) {
1466                         dev_kfree_skb(buffer_info->skb);
1467                         buffer_info->skb = NULL;
1468                 }
1469
1470                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1471                         ps_page = &buffer_info->ps_pages[j];
1472                         if (!ps_page->page)
1473                                 break;
1474                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1475                                        DMA_FROM_DEVICE);
1476                         ps_page->dma = 0;
1477                         put_page(ps_page->page);
1478                         ps_page->page = NULL;
1479                 }
1480         }
1481
1482         /* there also may be some cached data from a chained receive */
1483         if (rx_ring->rx_skb_top) {
1484                 dev_kfree_skb(rx_ring->rx_skb_top);
1485                 rx_ring->rx_skb_top = NULL;
1486         }
1487
1488         /* Zero out the descriptor ring */
1489         memset(rx_ring->desc, 0, rx_ring->size);
1490
1491         rx_ring->next_to_clean = 0;
1492         rx_ring->next_to_use = 0;
1493         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1494
1495         writel(0, adapter->hw.hw_addr + rx_ring->head);
1496         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1497 }
1498
1499 static void e1000e_downshift_workaround(struct work_struct *work)
1500 {
1501         struct e1000_adapter *adapter = container_of(work,
1502                                         struct e1000_adapter, downshift_task);
1503
1504         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1505 }
1506
1507 /**
1508  * e1000_intr_msi - Interrupt Handler
1509  * @irq: interrupt number
1510  * @data: pointer to a network interface device structure
1511  **/
1512 static irqreturn_t e1000_intr_msi(int irq, void *data)
1513 {
1514         struct net_device *netdev = data;
1515         struct e1000_adapter *adapter = netdev_priv(netdev);
1516         struct e1000_hw *hw = &adapter->hw;
1517         u32 icr = er32(ICR);
1518
1519         /*
1520          * read ICR disables interrupts using IAM
1521          */
1522
1523         if (icr & E1000_ICR_LSC) {
1524                 hw->mac.get_link_status = 1;
1525                 /*
1526                  * ICH8 workaround-- Call gig speed drop workaround on cable
1527                  * disconnect (LSC) before accessing any PHY registers
1528                  */
1529                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1530                     (!(er32(STATUS) & E1000_STATUS_LU)))
1531                         schedule_work(&adapter->downshift_task);
1532
1533                 /*
1534                  * 80003ES2LAN workaround-- For packet buffer work-around on
1535                  * link down event; disable receives here in the ISR and reset
1536                  * adapter in watchdog
1537                  */
1538                 if (netif_carrier_ok(netdev) &&
1539                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1540                         /* disable receives */
1541                         u32 rctl = er32(RCTL);
1542                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1543                         adapter->flags |= FLAG_RX_RESTART_NOW;
1544                 }
1545                 /* guard against interrupt when we're going down */
1546                 if (!test_bit(__E1000_DOWN, &adapter->state))
1547                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1548         }
1549
1550         if (napi_schedule_prep(&adapter->napi)) {
1551                 adapter->total_tx_bytes = 0;
1552                 adapter->total_tx_packets = 0;
1553                 adapter->total_rx_bytes = 0;
1554                 adapter->total_rx_packets = 0;
1555                 __napi_schedule(&adapter->napi);
1556         }
1557
1558         return IRQ_HANDLED;
1559 }
1560
1561 /**
1562  * e1000_intr - Interrupt Handler
1563  * @irq: interrupt number
1564  * @data: pointer to a network interface device structure
1565  **/
1566 static irqreturn_t e1000_intr(int irq, void *data)
1567 {
1568         struct net_device *netdev = data;
1569         struct e1000_adapter *adapter = netdev_priv(netdev);
1570         struct e1000_hw *hw = &adapter->hw;
1571         u32 rctl, icr = er32(ICR);
1572
1573         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1574                 return IRQ_NONE;  /* Not our interrupt */
1575
1576         /*
1577          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1578          * not set, then the adapter didn't send an interrupt
1579          */
1580         if (!(icr & E1000_ICR_INT_ASSERTED))
1581                 return IRQ_NONE;
1582
1583         /*
1584          * Interrupt Auto-Mask...upon reading ICR,
1585          * interrupts are masked.  No need for the
1586          * IMC write
1587          */
1588
1589         if (icr & E1000_ICR_LSC) {
1590                 hw->mac.get_link_status = 1;
1591                 /*
1592                  * ICH8 workaround-- Call gig speed drop workaround on cable
1593                  * disconnect (LSC) before accessing any PHY registers
1594                  */
1595                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1596                     (!(er32(STATUS) & E1000_STATUS_LU)))
1597                         schedule_work(&adapter->downshift_task);
1598
1599                 /*
1600                  * 80003ES2LAN workaround--
1601                  * For packet buffer work-around on link down event;
1602                  * disable receives here in the ISR and
1603                  * reset adapter in watchdog
1604                  */
1605                 if (netif_carrier_ok(netdev) &&
1606                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1607                         /* disable receives */
1608                         rctl = er32(RCTL);
1609                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1610                         adapter->flags |= FLAG_RX_RESTART_NOW;
1611                 }
1612                 /* guard against interrupt when we're going down */
1613                 if (!test_bit(__E1000_DOWN, &adapter->state))
1614                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1615         }
1616
1617         if (napi_schedule_prep(&adapter->napi)) {
1618                 adapter->total_tx_bytes = 0;
1619                 adapter->total_tx_packets = 0;
1620                 adapter->total_rx_bytes = 0;
1621                 adapter->total_rx_packets = 0;
1622                 __napi_schedule(&adapter->napi);
1623         }
1624
1625         return IRQ_HANDLED;
1626 }
1627
1628 static irqreturn_t e1000_msix_other(int irq, void *data)
1629 {
1630         struct net_device *netdev = data;
1631         struct e1000_adapter *adapter = netdev_priv(netdev);
1632         struct e1000_hw *hw = &adapter->hw;
1633         u32 icr = er32(ICR);
1634
1635         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1636                 if (!test_bit(__E1000_DOWN, &adapter->state))
1637                         ew32(IMS, E1000_IMS_OTHER);
1638                 return IRQ_NONE;
1639         }
1640
1641         if (icr & adapter->eiac_mask)
1642                 ew32(ICS, (icr & adapter->eiac_mask));
1643
1644         if (icr & E1000_ICR_OTHER) {
1645                 if (!(icr & E1000_ICR_LSC))
1646                         goto no_link_interrupt;
1647                 hw->mac.get_link_status = 1;
1648                 /* guard against interrupt when we're going down */
1649                 if (!test_bit(__E1000_DOWN, &adapter->state))
1650                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1651         }
1652
1653 no_link_interrupt:
1654         if (!test_bit(__E1000_DOWN, &adapter->state))
1655                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1656
1657         return IRQ_HANDLED;
1658 }
1659
1660
1661 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1662 {
1663         struct net_device *netdev = data;
1664         struct e1000_adapter *adapter = netdev_priv(netdev);
1665         struct e1000_hw *hw = &adapter->hw;
1666         struct e1000_ring *tx_ring = adapter->tx_ring;
1667
1668
1669         adapter->total_tx_bytes = 0;
1670         adapter->total_tx_packets = 0;
1671
1672         if (!e1000_clean_tx_irq(adapter))
1673                 /* Ring was not completely cleaned, so fire another interrupt */
1674                 ew32(ICS, tx_ring->ims_val);
1675
1676         return IRQ_HANDLED;
1677 }
1678
1679 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1680 {
1681         struct net_device *netdev = data;
1682         struct e1000_adapter *adapter = netdev_priv(netdev);
1683
1684         /* Write the ITR value calculated at the end of the
1685          * previous interrupt.
1686          */
1687         if (adapter->rx_ring->set_itr) {
1688                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1689                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1690                 adapter->rx_ring->set_itr = 0;
1691         }
1692
1693         if (napi_schedule_prep(&adapter->napi)) {
1694                 adapter->total_rx_bytes = 0;
1695                 adapter->total_rx_packets = 0;
1696                 __napi_schedule(&adapter->napi);
1697         }
1698         return IRQ_HANDLED;
1699 }
1700
1701 /**
1702  * e1000_configure_msix - Configure MSI-X hardware
1703  *
1704  * e1000_configure_msix sets up the hardware to properly
1705  * generate MSI-X interrupts.
1706  **/
1707 static void e1000_configure_msix(struct e1000_adapter *adapter)
1708 {
1709         struct e1000_hw *hw = &adapter->hw;
1710         struct e1000_ring *rx_ring = adapter->rx_ring;
1711         struct e1000_ring *tx_ring = adapter->tx_ring;
1712         int vector = 0;
1713         u32 ctrl_ext, ivar = 0;
1714
1715         adapter->eiac_mask = 0;
1716
1717         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1718         if (hw->mac.type == e1000_82574) {
1719                 u32 rfctl = er32(RFCTL);
1720                 rfctl |= E1000_RFCTL_ACK_DIS;
1721                 ew32(RFCTL, rfctl);
1722         }
1723
1724 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1725         /* Configure Rx vector */
1726         rx_ring->ims_val = E1000_IMS_RXQ0;
1727         adapter->eiac_mask |= rx_ring->ims_val;
1728         if (rx_ring->itr_val)
1729                 writel(1000000000 / (rx_ring->itr_val * 256),
1730                        hw->hw_addr + rx_ring->itr_register);
1731         else
1732                 writel(1, hw->hw_addr + rx_ring->itr_register);
1733         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1734
1735         /* Configure Tx vector */
1736         tx_ring->ims_val = E1000_IMS_TXQ0;
1737         vector++;
1738         if (tx_ring->itr_val)
1739                 writel(1000000000 / (tx_ring->itr_val * 256),
1740                        hw->hw_addr + tx_ring->itr_register);
1741         else
1742                 writel(1, hw->hw_addr + tx_ring->itr_register);
1743         adapter->eiac_mask |= tx_ring->ims_val;
1744         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1745
1746         /* set vector for Other Causes, e.g. link changes */
1747         vector++;
1748         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1749         if (rx_ring->itr_val)
1750                 writel(1000000000 / (rx_ring->itr_val * 256),
1751                        hw->hw_addr + E1000_EITR_82574(vector));
1752         else
1753                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1754
1755         /* Cause Tx interrupts on every write back */
1756         ivar |= (1 << 31);
1757
1758         ew32(IVAR, ivar);
1759
1760         /* enable MSI-X PBA support */
1761         ctrl_ext = er32(CTRL_EXT);
1762         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1763
1764         /* Auto-Mask Other interrupts upon ICR read */
1765 #define E1000_EIAC_MASK_82574   0x01F00000
1766         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1767         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1768         ew32(CTRL_EXT, ctrl_ext);
1769         e1e_flush();
1770 }
1771
1772 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1773 {
1774         if (adapter->msix_entries) {
1775                 pci_disable_msix(adapter->pdev);
1776                 kfree(adapter->msix_entries);
1777                 adapter->msix_entries = NULL;
1778         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1779                 pci_disable_msi(adapter->pdev);
1780                 adapter->flags &= ~FLAG_MSI_ENABLED;
1781         }
1782 }
1783
1784 /**
1785  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1786  *
1787  * Attempt to configure interrupts using the best available
1788  * capabilities of the hardware and kernel.
1789  **/
1790 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1791 {
1792         int err;
1793         int i;
1794
1795         switch (adapter->int_mode) {
1796         case E1000E_INT_MODE_MSIX:
1797                 if (adapter->flags & FLAG_HAS_MSIX) {
1798                         adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1799                         adapter->msix_entries = kcalloc(adapter->num_vectors,
1800                                                       sizeof(struct msix_entry),
1801                                                       GFP_KERNEL);
1802                         if (adapter->msix_entries) {
1803                                 for (i = 0; i < adapter->num_vectors; i++)
1804                                         adapter->msix_entries[i].entry = i;
1805
1806                                 err = pci_enable_msix(adapter->pdev,
1807                                                       adapter->msix_entries,
1808                                                       adapter->num_vectors);
1809                                 if (err == 0) {
1810                                         return;
1811                                 }
1812                         }
1813                         /* MSI-X failed, so fall through and try MSI */
1814                         e_err("Failed to initialize MSI-X interrupts.  "
1815                               "Falling back to MSI interrupts.\n");
1816                         e1000e_reset_interrupt_capability(adapter);
1817                 }
1818                 adapter->int_mode = E1000E_INT_MODE_MSI;
1819                 /* Fall through */
1820         case E1000E_INT_MODE_MSI:
1821                 if (!pci_enable_msi(adapter->pdev)) {
1822                         adapter->flags |= FLAG_MSI_ENABLED;
1823                 } else {
1824                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1825                         e_err("Failed to initialize MSI interrupts.  Falling "
1826                               "back to legacy interrupts.\n");
1827                 }
1828                 /* Fall through */
1829         case E1000E_INT_MODE_LEGACY:
1830                 /* Don't do anything; this is the system default */
1831                 break;
1832         }
1833
1834         /* store the number of vectors being used */
1835         adapter->num_vectors = 1;
1836 }
1837
1838 /**
1839  * e1000_request_msix - Initialize MSI-X interrupts
1840  *
1841  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1842  * kernel.
1843  **/
1844 static int e1000_request_msix(struct e1000_adapter *adapter)
1845 {
1846         struct net_device *netdev = adapter->netdev;
1847         int err = 0, vector = 0;
1848
1849         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1850                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1851         else
1852                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1853         err = request_irq(adapter->msix_entries[vector].vector,
1854                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1855                           netdev);
1856         if (err)
1857                 goto out;
1858         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1859         adapter->rx_ring->itr_val = adapter->itr;
1860         vector++;
1861
1862         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1863                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1864         else
1865                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1866         err = request_irq(adapter->msix_entries[vector].vector,
1867                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1868                           netdev);
1869         if (err)
1870                 goto out;
1871         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1872         adapter->tx_ring->itr_val = adapter->itr;
1873         vector++;
1874
1875         err = request_irq(adapter->msix_entries[vector].vector,
1876                           e1000_msix_other, 0, netdev->name, netdev);
1877         if (err)
1878                 goto out;
1879
1880         e1000_configure_msix(adapter);
1881         return 0;
1882 out:
1883         return err;
1884 }
1885
1886 /**
1887  * e1000_request_irq - initialize interrupts
1888  *
1889  * Attempts to configure interrupts using the best available
1890  * capabilities of the hardware and kernel.
1891  **/
1892 static int e1000_request_irq(struct e1000_adapter *adapter)
1893 {
1894         struct net_device *netdev = adapter->netdev;
1895         int err;
1896
1897         if (adapter->msix_entries) {
1898                 err = e1000_request_msix(adapter);
1899                 if (!err)
1900                         return err;
1901                 /* fall back to MSI */
1902                 e1000e_reset_interrupt_capability(adapter);
1903                 adapter->int_mode = E1000E_INT_MODE_MSI;
1904                 e1000e_set_interrupt_capability(adapter);
1905         }
1906         if (adapter->flags & FLAG_MSI_ENABLED) {
1907                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1908                                   netdev->name, netdev);
1909                 if (!err)
1910                         return err;
1911
1912                 /* fall back to legacy interrupt */
1913                 e1000e_reset_interrupt_capability(adapter);
1914                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1915         }
1916
1917         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1918                           netdev->name, netdev);
1919         if (err)
1920                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1921
1922         return err;
1923 }
1924
1925 static void e1000_free_irq(struct e1000_adapter *adapter)
1926 {
1927         struct net_device *netdev = adapter->netdev;
1928
1929         if (adapter->msix_entries) {
1930                 int vector = 0;
1931
1932                 free_irq(adapter->msix_entries[vector].vector, netdev);
1933                 vector++;
1934
1935                 free_irq(adapter->msix_entries[vector].vector, netdev);
1936                 vector++;
1937
1938                 /* Other Causes interrupt vector */
1939                 free_irq(adapter->msix_entries[vector].vector, netdev);
1940                 return;
1941         }
1942
1943         free_irq(adapter->pdev->irq, netdev);
1944 }
1945
1946 /**
1947  * e1000_irq_disable - Mask off interrupt generation on the NIC
1948  **/
1949 static void e1000_irq_disable(struct e1000_adapter *adapter)
1950 {
1951         struct e1000_hw *hw = &adapter->hw;
1952
1953         ew32(IMC, ~0);
1954         if (adapter->msix_entries)
1955                 ew32(EIAC_82574, 0);
1956         e1e_flush();
1957
1958         if (adapter->msix_entries) {
1959                 int i;
1960                 for (i = 0; i < adapter->num_vectors; i++)
1961                         synchronize_irq(adapter->msix_entries[i].vector);
1962         } else {
1963                 synchronize_irq(adapter->pdev->irq);
1964         }
1965 }
1966
1967 /**
1968  * e1000_irq_enable - Enable default interrupt generation settings
1969  **/
1970 static void e1000_irq_enable(struct e1000_adapter *adapter)
1971 {
1972         struct e1000_hw *hw = &adapter->hw;
1973
1974         if (adapter->msix_entries) {
1975                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1976                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1977         } else {
1978                 ew32(IMS, IMS_ENABLE_MASK);
1979         }
1980         e1e_flush();
1981 }
1982
1983 /**
1984  * e1000_get_hw_control - get control of the h/w from f/w
1985  * @adapter: address of board private structure
1986  *
1987  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1988  * For ASF and Pass Through versions of f/w this means that
1989  * the driver is loaded. For AMT version (only with 82573)
1990  * of the f/w this means that the network i/f is open.
1991  **/
1992 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1993 {
1994         struct e1000_hw *hw = &adapter->hw;
1995         u32 ctrl_ext;
1996         u32 swsm;
1997
1998         /* Let firmware know the driver has taken over */
1999         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2000                 swsm = er32(SWSM);
2001                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2002         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2003                 ctrl_ext = er32(CTRL_EXT);
2004                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2005         }
2006 }
2007
2008 /**
2009  * e1000_release_hw_control - release control of the h/w to f/w
2010  * @adapter: address of board private structure
2011  *
2012  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2013  * For ASF and Pass Through versions of f/w this means that the
2014  * driver is no longer loaded. For AMT version (only with 82573) i
2015  * of the f/w this means that the network i/f is closed.
2016  *
2017  **/
2018 static void e1000_release_hw_control(struct e1000_adapter *adapter)
2019 {
2020         struct e1000_hw *hw = &adapter->hw;
2021         u32 ctrl_ext;
2022         u32 swsm;
2023
2024         /* Let firmware taken over control of h/w */
2025         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2026                 swsm = er32(SWSM);
2027                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2028         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2029                 ctrl_ext = er32(CTRL_EXT);
2030                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2031         }
2032 }
2033
2034 /**
2035  * @e1000_alloc_ring - allocate memory for a ring structure
2036  **/
2037 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2038                                 struct e1000_ring *ring)
2039 {
2040         struct pci_dev *pdev = adapter->pdev;
2041
2042         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2043                                         GFP_KERNEL);
2044         if (!ring->desc)
2045                 return -ENOMEM;
2046
2047         return 0;
2048 }
2049
2050 /**
2051  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2052  * @adapter: board private structure
2053  *
2054  * Return 0 on success, negative on failure
2055  **/
2056 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
2057 {
2058         struct e1000_ring *tx_ring = adapter->tx_ring;
2059         int err = -ENOMEM, size;
2060
2061         size = sizeof(struct e1000_buffer) * tx_ring->count;
2062         tx_ring->buffer_info = vmalloc(size);
2063         if (!tx_ring->buffer_info)
2064                 goto err;
2065         memset(tx_ring->buffer_info, 0, size);
2066
2067         /* round up to nearest 4K */
2068         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2069         tx_ring->size = ALIGN(tx_ring->size, 4096);
2070
2071         err = e1000_alloc_ring_dma(adapter, tx_ring);
2072         if (err)
2073                 goto err;
2074
2075         tx_ring->next_to_use = 0;
2076         tx_ring->next_to_clean = 0;
2077
2078         return 0;
2079 err:
2080         vfree(tx_ring->buffer_info);
2081         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2082         return err;
2083 }
2084
2085 /**
2086  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2087  * @adapter: board private structure
2088  *
2089  * Returns 0 on success, negative on failure
2090  **/
2091 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
2092 {
2093         struct e1000_ring *rx_ring = adapter->rx_ring;
2094         struct e1000_buffer *buffer_info;
2095         int i, size, desc_len, err = -ENOMEM;
2096
2097         size = sizeof(struct e1000_buffer) * rx_ring->count;
2098         rx_ring->buffer_info = vmalloc(size);
2099         if (!rx_ring->buffer_info)
2100                 goto err;
2101         memset(rx_ring->buffer_info, 0, size);
2102
2103         for (i = 0; i < rx_ring->count; i++) {
2104                 buffer_info = &rx_ring->buffer_info[i];
2105                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2106                                                 sizeof(struct e1000_ps_page),
2107                                                 GFP_KERNEL);
2108                 if (!buffer_info->ps_pages)
2109                         goto err_pages;
2110         }
2111
2112         desc_len = sizeof(union e1000_rx_desc_packet_split);
2113
2114         /* Round up to nearest 4K */
2115         rx_ring->size = rx_ring->count * desc_len;
2116         rx_ring->size = ALIGN(rx_ring->size, 4096);
2117
2118         err = e1000_alloc_ring_dma(adapter, rx_ring);
2119         if (err)
2120                 goto err_pages;
2121
2122         rx_ring->next_to_clean = 0;
2123         rx_ring->next_to_use = 0;
2124         rx_ring->rx_skb_top = NULL;
2125
2126         return 0;
2127
2128 err_pages:
2129         for (i = 0; i < rx_ring->count; i++) {
2130                 buffer_info = &rx_ring->buffer_info[i];
2131                 kfree(buffer_info->ps_pages);
2132         }
2133 err:
2134         vfree(rx_ring->buffer_info);
2135         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2136         return err;
2137 }
2138
2139 /**
2140  * e1000_clean_tx_ring - Free Tx Buffers
2141  * @adapter: board private structure
2142  **/
2143 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
2144 {
2145         struct e1000_ring *tx_ring = adapter->tx_ring;
2146         struct e1000_buffer *buffer_info;
2147         unsigned long size;
2148         unsigned int i;
2149
2150         for (i = 0; i < tx_ring->count; i++) {
2151                 buffer_info = &tx_ring->buffer_info[i];
2152                 e1000_put_txbuf(adapter, buffer_info);
2153         }
2154
2155         size = sizeof(struct e1000_buffer) * tx_ring->count;
2156         memset(tx_ring->buffer_info, 0, size);
2157
2158         memset(tx_ring->desc, 0, tx_ring->size);
2159
2160         tx_ring->next_to_use = 0;
2161         tx_ring->next_to_clean = 0;
2162
2163         writel(0, adapter->hw.hw_addr + tx_ring->head);
2164         writel(0, adapter->hw.hw_addr + tx_ring->tail);
2165 }
2166
2167 /**
2168  * e1000e_free_tx_resources - Free Tx Resources per Queue
2169  * @adapter: board private structure
2170  *
2171  * Free all transmit software resources
2172  **/
2173 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
2174 {
2175         struct pci_dev *pdev = adapter->pdev;
2176         struct e1000_ring *tx_ring = adapter->tx_ring;
2177
2178         e1000_clean_tx_ring(adapter);
2179
2180         vfree(tx_ring->buffer_info);
2181         tx_ring->buffer_info = NULL;
2182
2183         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2184                           tx_ring->dma);
2185         tx_ring->desc = NULL;
2186 }
2187
2188 /**
2189  * e1000e_free_rx_resources - Free Rx Resources
2190  * @adapter: board private structure
2191  *
2192  * Free all receive software resources
2193  **/
2194
2195 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
2196 {
2197         struct pci_dev *pdev = adapter->pdev;
2198         struct e1000_ring *rx_ring = adapter->rx_ring;
2199         int i;
2200
2201         e1000_clean_rx_ring(adapter);
2202
2203         for (i = 0; i < rx_ring->count; i++) {
2204                 kfree(rx_ring->buffer_info[i].ps_pages);
2205         }
2206
2207         vfree(rx_ring->buffer_info);
2208         rx_ring->buffer_info = NULL;
2209
2210         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2211                           rx_ring->dma);
2212         rx_ring->desc = NULL;
2213 }
2214
2215 /**
2216  * e1000_update_itr - update the dynamic ITR value based on statistics
2217  * @adapter: pointer to adapter
2218  * @itr_setting: current adapter->itr
2219  * @packets: the number of packets during this measurement interval
2220  * @bytes: the number of bytes during this measurement interval
2221  *
2222  *      Stores a new ITR value based on packets and byte
2223  *      counts during the last interrupt.  The advantage of per interrupt
2224  *      computation is faster updates and more accurate ITR for the current
2225  *      traffic pattern.  Constants in this function were computed
2226  *      based on theoretical maximum wire speed and thresholds were set based
2227  *      on testing data as well as attempting to minimize response time
2228  *      while increasing bulk throughput.  This functionality is controlled
2229  *      by the InterruptThrottleRate module parameter.
2230  **/
2231 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2232                                      u16 itr_setting, int packets,
2233                                      int bytes)
2234 {
2235         unsigned int retval = itr_setting;
2236
2237         if (packets == 0)
2238                 goto update_itr_done;
2239
2240         switch (itr_setting) {
2241         case lowest_latency:
2242                 /* handle TSO and jumbo frames */
2243                 if (bytes/packets > 8000)
2244                         retval = bulk_latency;
2245                 else if ((packets < 5) && (bytes > 512)) {
2246                         retval = low_latency;
2247                 }
2248                 break;
2249         case low_latency:  /* 50 usec aka 20000 ints/s */
2250                 if (bytes > 10000) {
2251                         /* this if handles the TSO accounting */
2252                         if (bytes/packets > 8000) {
2253                                 retval = bulk_latency;
2254                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
2255                                 retval = bulk_latency;
2256                         } else if ((packets > 35)) {
2257                                 retval = lowest_latency;
2258                         }
2259                 } else if (bytes/packets > 2000) {
2260                         retval = bulk_latency;
2261                 } else if (packets <= 2 && bytes < 512) {
2262                         retval = lowest_latency;
2263                 }
2264                 break;
2265         case bulk_latency: /* 250 usec aka 4000 ints/s */
2266                 if (bytes > 25000) {
2267                         if (packets > 35) {
2268                                 retval = low_latency;
2269                         }
2270                 } else if (bytes < 6000) {
2271                         retval = low_latency;
2272                 }
2273                 break;
2274         }
2275
2276 update_itr_done:
2277         return retval;
2278 }
2279
2280 static void e1000_set_itr(struct e1000_adapter *adapter)
2281 {
2282         struct e1000_hw *hw = &adapter->hw;
2283         u16 current_itr;
2284         u32 new_itr = adapter->itr;
2285
2286         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2287         if (adapter->link_speed != SPEED_1000) {
2288                 current_itr = 0;
2289                 new_itr = 4000;
2290                 goto set_itr_now;
2291         }
2292
2293         adapter->tx_itr = e1000_update_itr(adapter,
2294                                     adapter->tx_itr,
2295                                     adapter->total_tx_packets,
2296                                     adapter->total_tx_bytes);
2297         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2298         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2299                 adapter->tx_itr = low_latency;
2300
2301         adapter->rx_itr = e1000_update_itr(adapter,
2302                                     adapter->rx_itr,
2303                                     adapter->total_rx_packets,
2304                                     adapter->total_rx_bytes);
2305         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2306         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2307                 adapter->rx_itr = low_latency;
2308
2309         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2310
2311         switch (current_itr) {
2312         /* counts and packets in update_itr are dependent on these numbers */
2313         case lowest_latency:
2314                 new_itr = 70000;
2315                 break;
2316         case low_latency:
2317                 new_itr = 20000; /* aka hwitr = ~200 */
2318                 break;
2319         case bulk_latency:
2320                 new_itr = 4000;
2321                 break;
2322         default:
2323                 break;
2324         }
2325
2326 set_itr_now:
2327         if (new_itr != adapter->itr) {
2328                 /*
2329                  * this attempts to bias the interrupt rate towards Bulk
2330                  * by adding intermediate steps when interrupt rate is
2331                  * increasing
2332                  */
2333                 new_itr = new_itr > adapter->itr ?
2334                              min(adapter->itr + (new_itr >> 2), new_itr) :
2335                              new_itr;
2336                 adapter->itr = new_itr;
2337                 adapter->rx_ring->itr_val = new_itr;
2338                 if (adapter->msix_entries)
2339                         adapter->rx_ring->set_itr = 1;
2340                 else
2341                         ew32(ITR, 1000000000 / (new_itr * 256));
2342         }
2343 }
2344
2345 /**
2346  * e1000_alloc_queues - Allocate memory for all rings
2347  * @adapter: board private structure to initialize
2348  **/
2349 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2350 {
2351         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2352         if (!adapter->tx_ring)
2353                 goto err;
2354
2355         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2356         if (!adapter->rx_ring)
2357                 goto err;
2358
2359         return 0;
2360 err:
2361         e_err("Unable to allocate memory for queues\n");
2362         kfree(adapter->rx_ring);
2363         kfree(adapter->tx_ring);
2364         return -ENOMEM;
2365 }
2366
2367 /**
2368  * e1000_clean - NAPI Rx polling callback
2369  * @napi: struct associated with this polling callback
2370  * @budget: amount of packets driver is allowed to process this poll
2371  **/
2372 static int e1000_clean(struct napi_struct *napi, int budget)
2373 {
2374         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2375         struct e1000_hw *hw = &adapter->hw;
2376         struct net_device *poll_dev = adapter->netdev;
2377         int tx_cleaned = 1, work_done = 0;
2378
2379         adapter = netdev_priv(poll_dev);
2380
2381         if (adapter->msix_entries &&
2382             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2383                 goto clean_rx;
2384
2385         tx_cleaned = e1000_clean_tx_irq(adapter);
2386
2387 clean_rx:
2388         adapter->clean_rx(adapter, &work_done, budget);
2389
2390         if (!tx_cleaned)
2391                 work_done = budget;
2392
2393         /* If budget not fully consumed, exit the polling mode */
2394         if (work_done < budget) {
2395                 if (adapter->itr_setting & 3)
2396                         e1000_set_itr(adapter);
2397                 napi_complete(napi);
2398                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2399                         if (adapter->msix_entries)
2400                                 ew32(IMS, adapter->rx_ring->ims_val);
2401                         else
2402                                 e1000_irq_enable(adapter);
2403                 }
2404         }
2405
2406         return work_done;
2407 }
2408
2409 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2410 {
2411         struct e1000_adapter *adapter = netdev_priv(netdev);
2412         struct e1000_hw *hw = &adapter->hw;
2413         u32 vfta, index;
2414
2415         /* don't update vlan cookie if already programmed */
2416         if ((adapter->hw.mng_cookie.status &
2417              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2418             (vid == adapter->mng_vlan_id))
2419                 return;
2420
2421         /* add VID to filter table */
2422         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2423                 index = (vid >> 5) & 0x7F;
2424                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2425                 vfta |= (1 << (vid & 0x1F));
2426                 hw->mac.ops.write_vfta(hw, index, vfta);
2427         }
2428 }
2429
2430 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2431 {
2432         struct e1000_adapter *adapter = netdev_priv(netdev);
2433         struct e1000_hw *hw = &adapter->hw;
2434         u32 vfta, index;
2435
2436         if (!test_bit(__E1000_DOWN, &adapter->state))
2437                 e1000_irq_disable(adapter);
2438         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2439
2440         if (!test_bit(__E1000_DOWN, &adapter->state))
2441                 e1000_irq_enable(adapter);
2442
2443         if ((adapter->hw.mng_cookie.status &
2444              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2445             (vid == adapter->mng_vlan_id)) {
2446                 /* release control to f/w */
2447                 e1000_release_hw_control(adapter);
2448                 return;
2449         }
2450
2451         /* remove VID from filter table */
2452         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2453                 index = (vid >> 5) & 0x7F;
2454                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2455                 vfta &= ~(1 << (vid & 0x1F));
2456                 hw->mac.ops.write_vfta(hw, index, vfta);
2457         }
2458 }
2459
2460 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2461 {
2462         struct net_device *netdev = adapter->netdev;
2463         u16 vid = adapter->hw.mng_cookie.vlan_id;
2464         u16 old_vid = adapter->mng_vlan_id;
2465
2466         if (!adapter->vlgrp)
2467                 return;
2468
2469         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2470                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2471                 if (adapter->hw.mng_cookie.status &
2472                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2473                         e1000_vlan_rx_add_vid(netdev, vid);
2474                         adapter->mng_vlan_id = vid;
2475                 }
2476
2477                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2478                                 (vid != old_vid) &&
2479                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2480                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2481         } else {
2482                 adapter->mng_vlan_id = vid;
2483         }
2484 }
2485
2486
2487 static void e1000_vlan_rx_register(struct net_device *netdev,
2488                                    struct vlan_group *grp)
2489 {
2490         struct e1000_adapter *adapter = netdev_priv(netdev);
2491         struct e1000_hw *hw = &adapter->hw;
2492         u32 ctrl, rctl;
2493
2494         if (!test_bit(__E1000_DOWN, &adapter->state))
2495                 e1000_irq_disable(adapter);
2496         adapter->vlgrp = grp;
2497
2498         if (grp) {
2499                 /* enable VLAN tag insert/strip */
2500                 ctrl = er32(CTRL);
2501                 ctrl |= E1000_CTRL_VME;
2502                 ew32(CTRL, ctrl);
2503
2504                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2505                         /* enable VLAN receive filtering */
2506                         rctl = er32(RCTL);
2507                         rctl &= ~E1000_RCTL_CFIEN;
2508                         ew32(RCTL, rctl);
2509                         e1000_update_mng_vlan(adapter);
2510                 }
2511         } else {
2512                 /* disable VLAN tag insert/strip */
2513                 ctrl = er32(CTRL);
2514                 ctrl &= ~E1000_CTRL_VME;
2515                 ew32(CTRL, ctrl);
2516
2517                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2518                         if (adapter->mng_vlan_id !=
2519                             (u16)E1000_MNG_VLAN_NONE) {
2520                                 e1000_vlan_rx_kill_vid(netdev,
2521                                                        adapter->mng_vlan_id);
2522                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2523                         }
2524                 }
2525         }
2526
2527         if (!test_bit(__E1000_DOWN, &adapter->state))
2528                 e1000_irq_enable(adapter);
2529 }
2530
2531 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2532 {
2533         u16 vid;
2534
2535         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2536
2537         if (!adapter->vlgrp)
2538                 return;
2539
2540         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2541                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2542                         continue;
2543                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2544         }
2545 }
2546
2547 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2548 {
2549         struct e1000_hw *hw = &adapter->hw;
2550         u32 manc, manc2h, mdef, i, j;
2551
2552         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2553                 return;
2554
2555         manc = er32(MANC);
2556
2557         /*
2558          * enable receiving management packets to the host. this will probably
2559          * generate destination unreachable messages from the host OS, but
2560          * the packets will be handled on SMBUS
2561          */
2562         manc |= E1000_MANC_EN_MNG2HOST;
2563         manc2h = er32(MANC2H);
2564
2565         switch (hw->mac.type) {
2566         default:
2567                 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2568                 break;
2569         case e1000_82574:
2570         case e1000_82583:
2571                 /*
2572                  * Check if IPMI pass-through decision filter already exists;
2573                  * if so, enable it.
2574                  */
2575                 for (i = 0, j = 0; i < 8; i++) {
2576                         mdef = er32(MDEF(i));
2577
2578                         /* Ignore filters with anything other than IPMI ports */
2579                         if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2580                                 continue;
2581
2582                         /* Enable this decision filter in MANC2H */
2583                         if (mdef)
2584                                 manc2h |= (1 << i);
2585
2586                         j |= mdef;
2587                 }
2588
2589                 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2590                         break;
2591
2592                 /* Create new decision filter in an empty filter */
2593                 for (i = 0, j = 0; i < 8; i++)
2594                         if (er32(MDEF(i)) == 0) {
2595                                 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2596                                                E1000_MDEF_PORT_664));
2597                                 manc2h |= (1 << 1);
2598                                 j++;
2599                                 break;
2600                         }
2601
2602                 if (!j)
2603                         e_warn("Unable to create IPMI pass-through filter\n");
2604                 break;
2605         }
2606
2607         ew32(MANC2H, manc2h);
2608         ew32(MANC, manc);
2609 }
2610
2611 /**
2612  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2613  * @adapter: board private structure
2614  *
2615  * Configure the Tx unit of the MAC after a reset.
2616  **/
2617 static void e1000_configure_tx(struct e1000_adapter *adapter)
2618 {
2619         struct e1000_hw *hw = &adapter->hw;
2620         struct e1000_ring *tx_ring = adapter->tx_ring;
2621         u64 tdba;
2622         u32 tdlen, tctl, tipg, tarc;
2623         u32 ipgr1, ipgr2;
2624
2625         /* Setup the HW Tx Head and Tail descriptor pointers */
2626         tdba = tx_ring->dma;
2627         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2628         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2629         ew32(TDBAH, (tdba >> 32));
2630         ew32(TDLEN, tdlen);
2631         ew32(TDH, 0);
2632         ew32(TDT, 0);
2633         tx_ring->head = E1000_TDH;
2634         tx_ring->tail = E1000_TDT;
2635
2636         /* Set the default values for the Tx Inter Packet Gap timer */
2637         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2638         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2639         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2640
2641         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2642                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2643
2644         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2645         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2646         ew32(TIPG, tipg);
2647
2648         /* Set the Tx Interrupt Delay register */
2649         ew32(TIDV, adapter->tx_int_delay);
2650         /* Tx irq moderation */
2651         ew32(TADV, adapter->tx_abs_int_delay);
2652
2653         if (adapter->flags2 & FLAG2_DMA_BURST) {
2654                 u32 txdctl = er32(TXDCTL(0));
2655                 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2656                             E1000_TXDCTL_WTHRESH);
2657                 /*
2658                  * set up some performance related parameters to encourage the
2659                  * hardware to use the bus more efficiently in bursts, depends
2660                  * on the tx_int_delay to be enabled,
2661                  * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
2662                  * hthresh = 1 ==> prefetch when one or more available
2663                  * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2664                  * BEWARE: this seems to work but should be considered first if
2665                  * there are tx hangs or other tx related bugs
2666                  */
2667                 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2668                 ew32(TXDCTL(0), txdctl);
2669                 /* erratum work around: set txdctl the same for both queues */
2670                 ew32(TXDCTL(1), txdctl);
2671         }
2672
2673         /* Program the Transmit Control Register */
2674         tctl = er32(TCTL);
2675         tctl &= ~E1000_TCTL_CT;
2676         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2677                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2678
2679         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2680                 tarc = er32(TARC(0));
2681                 /*
2682                  * set the speed mode bit, we'll clear it if we're not at
2683                  * gigabit link later
2684                  */
2685 #define SPEED_MODE_BIT (1 << 21)
2686                 tarc |= SPEED_MODE_BIT;
2687                 ew32(TARC(0), tarc);
2688         }
2689
2690         /* errata: program both queues to unweighted RR */
2691         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2692                 tarc = er32(TARC(0));
2693                 tarc |= 1;
2694                 ew32(TARC(0), tarc);
2695                 tarc = er32(TARC(1));
2696                 tarc |= 1;
2697                 ew32(TARC(1), tarc);
2698         }
2699
2700         /* Setup Transmit Descriptor Settings for eop descriptor */
2701         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2702
2703         /* only set IDE if we are delaying interrupts using the timers */
2704         if (adapter->tx_int_delay)
2705                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2706
2707         /* enable Report Status bit */
2708         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2709
2710         ew32(TCTL, tctl);
2711
2712         e1000e_config_collision_dist(hw);
2713 }
2714
2715 /**
2716  * e1000_setup_rctl - configure the receive control registers
2717  * @adapter: Board private structure
2718  **/
2719 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2720                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2721 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2722 {
2723         struct e1000_hw *hw = &adapter->hw;
2724         u32 rctl, rfctl;
2725         u32 psrctl = 0;
2726         u32 pages = 0;
2727
2728         /* Workaround Si errata on 82579 - configure jumbo frame flow */
2729         if (hw->mac.type == e1000_pch2lan) {
2730                 s32 ret_val;
2731
2732                 if (adapter->netdev->mtu > ETH_DATA_LEN)
2733                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2734                 else
2735                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2736         }
2737
2738         /* Program MC offset vector base */
2739         rctl = er32(RCTL);
2740         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2741         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2742                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2743                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2744
2745         /* Do not Store bad packets */
2746         rctl &= ~E1000_RCTL_SBP;
2747
2748         /* Enable Long Packet receive */
2749         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2750                 rctl &= ~E1000_RCTL_LPE;
2751         else
2752                 rctl |= E1000_RCTL_LPE;
2753
2754         /* Some systems expect that the CRC is included in SMBUS traffic. The
2755          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2756          * host memory when this is enabled
2757          */
2758         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2759                 rctl |= E1000_RCTL_SECRC;
2760
2761         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2762         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2763                 u16 phy_data;
2764
2765                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2766                 phy_data &= 0xfff8;
2767                 phy_data |= (1 << 2);
2768                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2769
2770                 e1e_rphy(hw, 22, &phy_data);
2771                 phy_data &= 0x0fff;
2772                 phy_data |= (1 << 14);
2773                 e1e_wphy(hw, 0x10, 0x2823);
2774                 e1e_wphy(hw, 0x11, 0x0003);
2775                 e1e_wphy(hw, 22, phy_data);
2776         }
2777
2778         /* Setup buffer sizes */
2779         rctl &= ~E1000_RCTL_SZ_4096;
2780         rctl |= E1000_RCTL_BSEX;
2781         switch (adapter->rx_buffer_len) {
2782         case 2048:
2783         default:
2784                 rctl |= E1000_RCTL_SZ_2048;
2785                 rctl &= ~E1000_RCTL_BSEX;
2786                 break;
2787         case 4096:
2788                 rctl |= E1000_RCTL_SZ_4096;
2789                 break;
2790         case 8192:
2791                 rctl |= E1000_RCTL_SZ_8192;
2792                 break;
2793         case 16384:
2794                 rctl |= E1000_RCTL_SZ_16384;
2795                 break;
2796         }
2797
2798         /*
2799          * 82571 and greater support packet-split where the protocol
2800          * header is placed in skb->data and the packet data is
2801          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2802          * In the case of a non-split, skb->data is linearly filled,
2803          * followed by the page buffers.  Therefore, skb->data is
2804          * sized to hold the largest protocol header.
2805          *
2806          * allocations using alloc_page take too long for regular MTU
2807          * so only enable packet split for jumbo frames
2808          *
2809          * Using pages when the page size is greater than 16k wastes
2810          * a lot of memory, since we allocate 3 pages at all times
2811          * per packet.
2812          */
2813         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2814         if (!(adapter->flags & FLAG_HAS_ERT) && (pages <= 3) &&
2815             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2816                 adapter->rx_ps_pages = pages;
2817         else
2818                 adapter->rx_ps_pages = 0;
2819
2820         if (adapter->rx_ps_pages) {
2821                 /* Configure extra packet-split registers */
2822                 rfctl = er32(RFCTL);
2823                 rfctl |= E1000_RFCTL_EXTEN;
2824                 /*
2825                  * disable packet split support for IPv6 extension headers,
2826                  * because some malformed IPv6 headers can hang the Rx
2827                  */
2828                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2829                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2830
2831                 ew32(RFCTL, rfctl);
2832
2833                 /* Enable Packet split descriptors */
2834                 rctl |= E1000_RCTL_DTYP_PS;
2835
2836                 psrctl |= adapter->rx_ps_bsize0 >>
2837                         E1000_PSRCTL_BSIZE0_SHIFT;
2838
2839                 switch (adapter->rx_ps_pages) {
2840                 case 3:
2841                         psrctl |= PAGE_SIZE <<
2842                                 E1000_PSRCTL_BSIZE3_SHIFT;
2843                 case 2:
2844                         psrctl |= PAGE_SIZE <<
2845                                 E1000_PSRCTL_BSIZE2_SHIFT;
2846                 case 1:
2847                         psrctl |= PAGE_SIZE >>
2848                                 E1000_PSRCTL_BSIZE1_SHIFT;
2849                         break;
2850                 }
2851
2852                 ew32(PSRCTL, psrctl);
2853         }
2854
2855         ew32(RCTL, rctl);
2856         /* just started the receive unit, no need to restart */
2857         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2858 }
2859
2860 /**
2861  * e1000_configure_rx - Configure Receive Unit after Reset
2862  * @adapter: board private structure
2863  *
2864  * Configure the Rx unit of the MAC after a reset.
2865  **/
2866 static void e1000_configure_rx(struct e1000_adapter *adapter)
2867 {
2868         struct e1000_hw *hw = &adapter->hw;
2869         struct e1000_ring *rx_ring = adapter->rx_ring;
2870         u64 rdba;
2871         u32 rdlen, rctl, rxcsum, ctrl_ext;
2872
2873         if (adapter->rx_ps_pages) {
2874                 /* this is a 32 byte descriptor */
2875                 rdlen = rx_ring->count *
2876                         sizeof(union e1000_rx_desc_packet_split);
2877                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2878                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2879         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2880                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2881                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2882                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2883         } else {
2884                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2885                 adapter->clean_rx = e1000_clean_rx_irq;
2886                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2887         }
2888
2889         /* disable receives while setting up the descriptors */
2890         rctl = er32(RCTL);
2891         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2892         e1e_flush();
2893         msleep(10);
2894
2895         if (adapter->flags2 & FLAG2_DMA_BURST) {
2896                 /*
2897                  * set the writeback threshold (only takes effect if the RDTR
2898                  * is set). set GRAN=1 and write back up to 0x4 worth, and
2899                  * enable prefetching of 0x20 rx descriptors
2900                  * granularity = 01
2901                  * wthresh = 04,
2902                  * hthresh = 04,
2903                  * pthresh = 0x20
2904                  */
2905                 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
2906                 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
2907
2908                 /*
2909                  * override the delay timers for enabling bursting, only if
2910                  * the value was not set by the user via module options
2911                  */
2912                 if (adapter->rx_int_delay == DEFAULT_RDTR)
2913                         adapter->rx_int_delay = BURST_RDTR;
2914                 if (adapter->rx_abs_int_delay == DEFAULT_RADV)
2915                         adapter->rx_abs_int_delay = BURST_RADV;
2916         }
2917
2918         /* set the Receive Delay Timer Register */
2919         ew32(RDTR, adapter->rx_int_delay);
2920
2921         /* irq moderation */
2922         ew32(RADV, adapter->rx_abs_int_delay);
2923         if (adapter->itr_setting != 0)
2924                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2925
2926         ctrl_ext = er32(CTRL_EXT);
2927         /* Auto-Mask interrupts upon ICR access */
2928         ctrl_ext |= E1000_CTRL_EXT_IAME;
2929         ew32(IAM, 0xffffffff);
2930         ew32(CTRL_EXT, ctrl_ext);
2931         e1e_flush();
2932
2933         /*
2934          * Setup the HW Rx Head and Tail Descriptor Pointers and
2935          * the Base and Length of the Rx Descriptor Ring
2936          */
2937         rdba = rx_ring->dma;
2938         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2939         ew32(RDBAH, (rdba >> 32));
2940         ew32(RDLEN, rdlen);
2941         ew32(RDH, 0);
2942         ew32(RDT, 0);
2943         rx_ring->head = E1000_RDH;
2944         rx_ring->tail = E1000_RDT;
2945
2946         /* Enable Receive Checksum Offload for TCP and UDP */
2947         rxcsum = er32(RXCSUM);
2948         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2949                 rxcsum |= E1000_RXCSUM_TUOFL;
2950
2951                 /*
2952                  * IPv4 payload checksum for UDP fragments must be
2953                  * used in conjunction with packet-split.
2954                  */
2955                 if (adapter->rx_ps_pages)
2956                         rxcsum |= E1000_RXCSUM_IPPCSE;
2957         } else {
2958                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2959                 /* no need to clear IPPCSE as it defaults to 0 */
2960         }
2961         ew32(RXCSUM, rxcsum);
2962
2963         /*
2964          * Enable early receives on supported devices, only takes effect when
2965          * packet size is equal or larger than the specified value (in 8 byte
2966          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2967          */
2968         if (adapter->flags & FLAG_HAS_ERT) {
2969                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2970                         u32 rxdctl = er32(RXDCTL(0));
2971                         ew32(RXDCTL(0), rxdctl | 0x3);
2972                         ew32(ERT, E1000_ERT_2048 | (1 << 13));
2973                         /*
2974                          * With jumbo frames and early-receive enabled,
2975                          * excessive C-state transition latencies result in
2976                          * dropped transactions.
2977                          */
2978                         pm_qos_update_request(
2979                                 &adapter->netdev->pm_qos_req, 55);
2980                 } else {
2981                         pm_qos_update_request(
2982                                 &adapter->netdev->pm_qos_req,
2983                                 PM_QOS_DEFAULT_VALUE);
2984                 }
2985         }
2986
2987         /* Enable Receives */
2988         ew32(RCTL, rctl);
2989 }
2990
2991 /**
2992  *  e1000_update_mc_addr_list - Update Multicast addresses
2993  *  @hw: pointer to the HW structure
2994  *  @mc_addr_list: array of multicast addresses to program
2995  *  @mc_addr_count: number of multicast addresses to program
2996  *
2997  *  Updates the Multicast Table Array.
2998  *  The caller must have a packed mc_addr_list of multicast addresses.
2999  **/
3000 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
3001                                       u32 mc_addr_count)
3002 {
3003         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
3004 }
3005
3006 /**
3007  * e1000_set_multi - Multicast and Promiscuous mode set
3008  * @netdev: network interface device structure
3009  *
3010  * The set_multi entry point is called whenever the multicast address
3011  * list or the network interface flags are updated.  This routine is
3012  * responsible for configuring the hardware for proper multicast,
3013  * promiscuous mode, and all-multi behavior.
3014  **/
3015 static void e1000_set_multi(struct net_device *netdev)
3016 {
3017         struct e1000_adapter *adapter = netdev_priv(netdev);
3018         struct e1000_hw *hw = &adapter->hw;
3019         struct netdev_hw_addr *ha;
3020         u8  *mta_list;
3021         u32 rctl;
3022         int i;
3023
3024         /* Check for Promiscuous and All Multicast modes */
3025
3026         rctl = er32(RCTL);
3027
3028         if (netdev->flags & IFF_PROMISC) {
3029                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3030                 rctl &= ~E1000_RCTL_VFE;
3031         } else {
3032                 if (netdev->flags & IFF_ALLMULTI) {
3033                         rctl |= E1000_RCTL_MPE;
3034                         rctl &= ~E1000_RCTL_UPE;
3035                 } else {
3036                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3037                 }
3038                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
3039                         rctl |= E1000_RCTL_VFE;
3040         }
3041
3042         ew32(RCTL, rctl);
3043
3044         if (!netdev_mc_empty(netdev)) {
3045                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3046                 if (!mta_list)
3047                         return;
3048
3049                 /* prepare a packed array of only addresses. */
3050                 i = 0;
3051                 netdev_for_each_mc_addr(ha, netdev)
3052                         memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3053
3054                 e1000_update_mc_addr_list(hw, mta_list, i);
3055                 kfree(mta_list);
3056         } else {
3057                 /*
3058                  * if we're called from probe, we might not have
3059                  * anything to do here, so clear out the list
3060                  */
3061                 e1000_update_mc_addr_list(hw, NULL, 0);
3062         }
3063 }
3064
3065 /**
3066  * e1000_configure - configure the hardware for Rx and Tx
3067  * @adapter: private board structure
3068  **/
3069 static void e1000_configure(struct e1000_adapter *adapter)
3070 {
3071         e1000_set_multi(adapter->netdev);
3072
3073         e1000_restore_vlan(adapter);
3074         e1000_init_manageability_pt(adapter);
3075
3076         e1000_configure_tx(adapter);
3077         e1000_setup_rctl(adapter);
3078         e1000_configure_rx(adapter);
3079         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
3080 }
3081
3082 /**
3083  * e1000e_power_up_phy - restore link in case the phy was powered down
3084  * @adapter: address of board private structure
3085  *
3086  * The phy may be powered down to save power and turn off link when the
3087  * driver is unloaded and wake on lan is not enabled (among others)
3088  * *** this routine MUST be followed by a call to e1000e_reset ***
3089  **/
3090 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3091 {
3092         if (adapter->hw.phy.ops.power_up)
3093                 adapter->hw.phy.ops.power_up(&adapter->hw);
3094
3095         adapter->hw.mac.ops.setup_link(&adapter->hw);
3096 }
3097
3098 /**
3099  * e1000_power_down_phy - Power down the PHY
3100  *
3101  * Power down the PHY so no link is implied when interface is down.
3102  * The PHY cannot be powered down if management or WoL is active.
3103  */
3104 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3105 {
3106         /* WoL is enabled */
3107         if (adapter->wol)
3108                 return;
3109
3110         if (adapter->hw.phy.ops.power_down)
3111                 adapter->hw.phy.ops.power_down(&adapter->hw);
3112 }
3113
3114 /**
3115  * e1000e_reset - bring the hardware into a known good state
3116  *
3117  * This function boots the hardware and enables some settings that
3118  * require a configuration cycle of the hardware - those cannot be
3119  * set/changed during runtime. After reset the device needs to be
3120  * properly configured for Rx, Tx etc.
3121  */
3122 void e1000e_reset(struct e1000_adapter *adapter)
3123 {
3124         struct e1000_mac_info *mac = &adapter->hw.mac;
3125         struct e1000_fc_info *fc = &adapter->hw.fc;
3126         struct e1000_hw *hw = &adapter->hw;
3127         u32 tx_space, min_tx_space, min_rx_space;
3128         u32 pba = adapter->pba;
3129         u16 hwm;
3130
3131         /* reset Packet Buffer Allocation to default */
3132         ew32(PBA, pba);
3133
3134         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3135                 /*
3136                  * To maintain wire speed transmits, the Tx FIFO should be
3137                  * large enough to accommodate two full transmit packets,
3138                  * rounded up to the next 1KB and expressed in KB.  Likewise,
3139                  * the Rx FIFO should be large enough to accommodate at least
3140                  * one full receive packet and is similarly rounded up and
3141                  * expressed in KB.
3142                  */
3143                 pba = er32(PBA);
3144                 /* upper 16 bits has Tx packet buffer allocation size in KB */
3145                 tx_space = pba >> 16;
3146                 /* lower 16 bits has Rx packet buffer allocation size in KB */
3147                 pba &= 0xffff;
3148                 /*
3149                  * the Tx fifo also stores 16 bytes of information about the tx
3150                  * but don't include ethernet FCS because hardware appends it
3151                  */
3152                 min_tx_space = (adapter->max_frame_size +
3153                                 sizeof(struct e1000_tx_desc) -
3154                                 ETH_FCS_LEN) * 2;
3155                 min_tx_space = ALIGN(min_tx_space, 1024);
3156                 min_tx_space >>= 10;
3157                 /* software strips receive CRC, so leave room for it */
3158                 min_rx_space = adapter->max_frame_size;
3159                 min_rx_space = ALIGN(min_rx_space, 1024);
3160                 min_rx_space >>= 10;
3161
3162                 /*
3163                  * If current Tx allocation is less than the min Tx FIFO size,
3164                  * and the min Tx FIFO size is less than the current Rx FIFO
3165                  * allocation, take space away from current Rx allocation
3166                  */
3167                 if ((tx_space < min_tx_space) &&
3168                     ((min_tx_space - tx_space) < pba)) {
3169                         pba -= min_tx_space - tx_space;
3170
3171                         /*
3172                          * if short on Rx space, Rx wins and must trump tx
3173                          * adjustment or use Early Receive if available
3174                          */
3175                         if ((pba < min_rx_space) &&
3176                             (!(adapter->flags & FLAG_HAS_ERT)))
3177                                 /* ERT enabled in e1000_configure_rx */
3178                                 pba = min_rx_space;
3179                 }
3180
3181                 ew32(PBA, pba);
3182         }
3183
3184
3185         /*
3186          * flow control settings
3187          *
3188          * The high water mark must be low enough to fit one full frame
3189          * (or the size used for early receive) above it in the Rx FIFO.
3190          * Set it to the lower of:
3191          * - 90% of the Rx FIFO size, and
3192          * - the full Rx FIFO size minus the early receive size (for parts
3193          *   with ERT support assuming ERT set to E1000_ERT_2048), or
3194          * - the full Rx FIFO size minus one full frame
3195          */
3196         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3197                 fc->pause_time = 0xFFFF;
3198         else
3199                 fc->pause_time = E1000_FC_PAUSE_TIME;
3200         fc->send_xon = 1;
3201         fc->current_mode = fc->requested_mode;
3202
3203         switch (hw->mac.type) {
3204         default:
3205                 if ((adapter->flags & FLAG_HAS_ERT) &&
3206                     (adapter->netdev->mtu > ETH_DATA_LEN))
3207                         hwm = min(((pba << 10) * 9 / 10),
3208                                   ((pba << 10) - (E1000_ERT_2048 << 3)));
3209                 else
3210                         hwm = min(((pba << 10) * 9 / 10),
3211                                   ((pba << 10) - adapter->max_frame_size));
3212
3213                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3214                 fc->low_water = fc->high_water - 8;
3215                 break;
3216         case e1000_pchlan:
3217                 /*
3218                  * Workaround PCH LOM adapter hangs with certain network
3219                  * loads.  If hangs persist, try disabling Tx flow control.
3220                  */
3221                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3222                         fc->high_water = 0x3500;
3223                         fc->low_water  = 0x1500;
3224                 } else {
3225                         fc->high_water = 0x5000;
3226                         fc->low_water  = 0x3000;
3227                 }
3228                 fc->refresh_time = 0x1000;
3229                 break;
3230         case e1000_pch2lan:
3231                 fc->high_water = 0x05C20;
3232                 fc->low_water = 0x05048;
3233                 fc->pause_time = 0x0650;
3234                 fc->refresh_time = 0x0400;
3235                 break;
3236         }
3237
3238         /* Allow time for pending master requests to run */
3239         mac->ops.reset_hw(hw);
3240
3241         /*
3242          * For parts with AMT enabled, let the firmware know
3243          * that the network interface is in control
3244          */
3245         if (adapter->flags & FLAG_HAS_AMT)
3246                 e1000_get_hw_control(adapter);
3247
3248         ew32(WUC, 0);
3249
3250         if (mac->ops.init_hw(hw))
3251                 e_err("Hardware Error\n");
3252
3253         e1000_update_mng_vlan(adapter);
3254
3255         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3256         ew32(VET, ETH_P_8021Q);
3257
3258         e1000e_reset_adaptive(hw);
3259         e1000_get_phy_info(hw);
3260
3261         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3262             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3263                 u16 phy_data = 0;
3264                 /*
3265                  * speed up time to link by disabling smart power down, ignore
3266                  * the return value of this function because there is nothing
3267                  * different we would do if it failed
3268                  */
3269                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3270                 phy_data &= ~IGP02E1000_PM_SPD;
3271                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3272         }
3273 }
3274
3275 int e1000e_up(struct e1000_adapter *adapter)
3276 {
3277         struct e1000_hw *hw = &adapter->hw;
3278
3279         /* hardware has been reset, we need to reload some things */
3280         e1000_configure(adapter);
3281
3282         clear_bit(__E1000_DOWN, &adapter->state);
3283
3284         napi_enable(&adapter->napi);
3285         if (adapter->msix_entries)
3286                 e1000_configure_msix(adapter);
3287         e1000_irq_enable(adapter);
3288
3289         netif_wake_queue(adapter->netdev);
3290
3291         /* fire a link change interrupt to start the watchdog */
3292         if (adapter->msix_entries)
3293                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3294         else
3295                 ew32(ICS, E1000_ICS_LSC);
3296
3297         return 0;
3298 }
3299
3300 void e1000e_down(struct e1000_adapter *adapter)
3301 {
3302         struct net_device *netdev = adapter->netdev;
3303         struct e1000_hw *hw = &adapter->hw;
3304         u32 tctl, rctl;
3305
3306         /*
3307          * signal that we're down so the interrupt handler does not
3308          * reschedule our watchdog timer
3309          */
3310         set_bit(__E1000_DOWN, &adapter->state);
3311
3312         /* disable receives in the hardware */
3313         rctl = er32(RCTL);
3314         ew32(RCTL, rctl & ~E1000_RCTL_EN);
3315         /* flush and sleep below */
3316
3317         netif_stop_queue(netdev);
3318
3319         /* disable transmits in the hardware */
3320         tctl = er32(TCTL);
3321         tctl &= ~E1000_TCTL_EN;
3322         ew32(TCTL, tctl);
3323         /* flush both disables and wait for them to finish */
3324         e1e_flush();
3325         msleep(10);
3326
3327         napi_disable(&adapter->napi);
3328         e1000_irq_disable(adapter);
3329
3330         del_timer_sync(&adapter->watchdog_timer);
3331         del_timer_sync(&adapter->phy_info_timer);
3332
3333         netif_carrier_off(netdev);
3334         adapter->link_speed = 0;
3335         adapter->link_duplex = 0;
3336
3337         if (!pci_channel_offline(adapter->pdev))
3338                 e1000e_reset(adapter);
3339         e1000_clean_tx_ring(adapter);
3340         e1000_clean_rx_ring(adapter);
3341
3342         /*
3343          * TODO: for power management, we could drop the link and
3344          * pci_disable_device here.
3345          */
3346 }
3347
3348 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3349 {
3350         might_sleep();
3351         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3352                 msleep(1);
3353         e1000e_down(adapter);
3354         e1000e_up(adapter);
3355         clear_bit(__E1000_RESETTING, &adapter->state);
3356 }
3357
3358 /**
3359  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3360  * @adapter: board private structure to initialize
3361  *
3362  * e1000_sw_init initializes the Adapter private data structure.
3363  * Fields are initialized based on PCI device information and
3364  * OS network device settings (MTU size).
3365  **/
3366 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3367 {
3368         struct net_device *netdev = adapter->netdev;
3369
3370         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3371         adapter->rx_ps_bsize0 = 128;
3372         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3373         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3374
3375         e1000e_set_interrupt_capability(adapter);
3376
3377         if (e1000_alloc_queues(adapter))
3378                 return -ENOMEM;
3379
3380         /* Explicitly disable IRQ since the NIC can be in any state. */
3381         e1000_irq_disable(adapter);
3382
3383         set_bit(__E1000_DOWN, &adapter->state);
3384         return 0;
3385 }
3386
3387 /**
3388  * e1000_intr_msi_test - Interrupt Handler
3389  * @irq: interrupt number
3390  * @data: pointer to a network interface device structure
3391  **/
3392 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3393 {
3394         struct net_device *netdev = data;
3395         struct e1000_adapter *adapter = netdev_priv(netdev);
3396         struct e1000_hw *hw = &adapter->hw;
3397         u32 icr = er32(ICR);
3398
3399         e_dbg("icr is %08X\n", icr);
3400         if (icr & E1000_ICR_RXSEQ) {
3401                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3402                 wmb();
3403         }
3404
3405         return IRQ_HANDLED;
3406 }
3407
3408 /**
3409  * e1000_test_msi_interrupt - Returns 0 for successful test
3410  * @adapter: board private struct
3411  *
3412  * code flow taken from tg3.c
3413  **/
3414 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3415 {
3416         struct net_device *netdev = adapter->netdev;
3417         struct e1000_hw *hw = &adapter->hw;
3418         int err;
3419
3420         /* poll_enable hasn't been called yet, so don't need disable */
3421         /* clear any pending events */
3422         er32(ICR);
3423
3424         /* free the real vector and request a test handler */
3425         e1000_free_irq(adapter);
3426         e1000e_reset_interrupt_capability(adapter);
3427
3428         /* Assume that the test fails, if it succeeds then the test
3429          * MSI irq handler will unset this flag */
3430         adapter->flags |= FLAG_MSI_TEST_FAILED;
3431
3432         err = pci_enable_msi(adapter->pdev);
3433         if (err)
3434                 goto msi_test_failed;
3435
3436         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3437                           netdev->name, netdev);
3438         if (err) {
3439                 pci_disable_msi(adapter->pdev);
3440                 goto msi_test_failed;
3441         }
3442
3443         wmb();
3444
3445         e1000_irq_enable(adapter);
3446
3447         /* fire an unusual interrupt on the test handler */
3448         ew32(ICS, E1000_ICS_RXSEQ);
3449         e1e_flush();
3450         msleep(50);
3451
3452         e1000_irq_disable(adapter);
3453
3454         rmb();
3455
3456         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3457                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3458                 e_info("MSI interrupt test failed, using legacy interrupt.\n");
3459         } else
3460                 e_dbg("MSI interrupt test succeeded!\n");
3461
3462         free_irq(adapter->pdev->irq, netdev);
3463         pci_disable_msi(adapter->pdev);
3464
3465 msi_test_failed:
3466         e1000e_set_interrupt_capability(adapter);
3467         return e1000_request_irq(adapter);
3468 }
3469
3470 /**
3471  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3472  * @adapter: board private struct
3473  *
3474  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3475  **/
3476 static int e1000_test_msi(struct e1000_adapter *adapter)
3477 {
3478         int err;
3479         u16 pci_cmd;
3480
3481         if (!(adapter->flags & FLAG_MSI_ENABLED))
3482                 return 0;
3483
3484         /* disable SERR in case the MSI write causes a master abort */
3485         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3486         if (pci_cmd & PCI_COMMAND_SERR)
3487                 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3488                                       pci_cmd & ~PCI_COMMAND_SERR);
3489
3490         err = e1000_test_msi_interrupt(adapter);
3491
3492         /* re-enable SERR */
3493         if (pci_cmd & PCI_COMMAND_SERR) {
3494                 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3495                 pci_cmd |= PCI_COMMAND_SERR;
3496                 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3497         }
3498
3499         return err;
3500 }
3501
3502 /**
3503  * e1000_open - Called when a network interface is made active
3504  * @netdev: network interface device structure
3505  *
3506  * Returns 0 on success, negative value on failure
3507  *
3508  * The open entry point is called when a network interface is made
3509  * active by the system (IFF_UP).  At this point all resources needed
3510  * for transmit and receive operations are allocated, the interrupt
3511  * handler is registered with the OS, the watchdog timer is started,
3512  * and the stack is notified that the interface is ready.
3513  **/
3514 static int e1000_open(struct net_device *netdev)
3515 {
3516         struct e1000_adapter *adapter = netdev_priv(netdev);
3517         struct e1000_hw *hw = &adapter->hw;
3518         struct pci_dev *pdev = adapter->pdev;
3519         int err;
3520
3521         /* disallow open during test */
3522         if (test_bit(__E1000_TESTING, &adapter->state))
3523                 return -EBUSY;
3524
3525         pm_runtime_get_sync(&pdev->dev);
3526
3527         netif_carrier_off(netdev);
3528
3529         /* allocate transmit descriptors */
3530         err = e1000e_setup_tx_resources(adapter);
3531         if (err)
3532                 goto err_setup_tx;
3533
3534         /* allocate receive descriptors */
3535         err = e1000e_setup_rx_resources(adapter);
3536         if (err)
3537                 goto err_setup_rx;
3538
3539         /*
3540          * If AMT is enabled, let the firmware know that the network
3541          * interface is now open and reset the part to a known state.
3542          */
3543         if (adapter->flags & FLAG_HAS_AMT) {
3544                 e1000_get_hw_control(adapter);
3545                 e1000e_reset(adapter);
3546         }
3547
3548         e1000e_power_up_phy(adapter);
3549
3550         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3551         if ((adapter->hw.mng_cookie.status &
3552              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3553                 e1000_update_mng_vlan(adapter);
3554
3555         /* DMA latency requirement to workaround early-receive/jumbo issue */
3556         if (adapter->flags & FLAG_HAS_ERT)
3557                 pm_qos_add_request(&adapter->netdev->pm_qos_req,
3558                                    PM_QOS_CPU_DMA_LATENCY,
3559                                    PM_QOS_DEFAULT_VALUE);
3560
3561         /*
3562          * before we allocate an interrupt, we must be ready to handle it.
3563          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3564          * as soon as we call pci_request_irq, so we have to setup our
3565          * clean_rx handler before we do so.
3566          */
3567         e1000_configure(adapter);
3568
3569         err = e1000_request_irq(adapter);
3570         if (err)
3571                 goto err_req_irq;
3572
3573         /*
3574          * Work around PCIe errata with MSI interrupts causing some chipsets to
3575          * ignore e1000e MSI messages, which means we need to test our MSI
3576          * interrupt now
3577          */
3578         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3579                 err = e1000_test_msi(adapter);
3580                 if (err) {
3581                         e_err("Interrupt allocation failed\n");
3582                         goto err_req_irq;
3583                 }
3584         }
3585
3586         /* From here on the code is the same as e1000e_up() */
3587         clear_bit(__E1000_DOWN, &adapter->state);
3588
3589         napi_enable(&adapter->napi);
3590
3591         e1000_irq_enable(adapter);
3592
3593         netif_start_queue(netdev);
3594
3595         adapter->idle_check = true;
3596         pm_runtime_put(&pdev->dev);
3597
3598         /* fire a link status change interrupt to start the watchdog */
3599         if (adapter->msix_entries)
3600                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3601         else
3602                 ew32(ICS, E1000_ICS_LSC);
3603
3604         return 0;
3605
3606 err_req_irq:
3607         e1000_release_hw_control(adapter);
3608         e1000_power_down_phy(adapter);
3609         e1000e_free_rx_resources(adapter);
3610 err_setup_rx:
3611         e1000e_free_tx_resources(adapter);
3612 err_setup_tx:
3613         e1000e_reset(adapter);
3614         pm_runtime_put_sync(&pdev->dev);
3615
3616         return err;
3617 }
3618
3619 /**
3620  * e1000_close - Disables a network interface
3621  * @netdev: network interface device structure
3622  *
3623  * Returns 0, this is not allowed to fail
3624  *
3625  * The close entry point is called when an interface is de-activated
3626  * by the OS.  The hardware is still under the drivers control, but
3627  * needs to be disabled.  A global MAC reset is issued to stop the
3628  * hardware, and all transmit and receive resources are freed.
3629  **/
3630 static int e1000_close(struct net_device *netdev)
3631 {
3632         struct e1000_adapter *adapter = netdev_priv(netdev);
3633         struct pci_dev *pdev = adapter->pdev;
3634
3635         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3636
3637         pm_runtime_get_sync(&pdev->dev);
3638
3639         if (!test_bit(__E1000_DOWN, &adapter->state)) {
3640                 e1000e_down(adapter);
3641                 e1000_free_irq(adapter);
3642         }
3643         e1000_power_down_phy(adapter);
3644
3645         e1000e_free_tx_resources(adapter);
3646         e1000e_free_rx_resources(adapter);
3647
3648         /*
3649          * kill manageability vlan ID if supported, but not if a vlan with
3650          * the same ID is registered on the host OS (let 8021q kill it)
3651          */
3652         if ((adapter->hw.mng_cookie.status &
3653                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3654              !(adapter->vlgrp &&
3655                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3656                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3657
3658         /*
3659          * If AMT is enabled, let the firmware know that the network
3660          * interface is now closed
3661          */
3662         if (adapter->flags & FLAG_HAS_AMT)
3663                 e1000_release_hw_control(adapter);
3664
3665         if (adapter->flags & FLAG_HAS_ERT)
3666                 pm_qos_remove_request(&adapter->netdev->pm_qos_req);
3667
3668         pm_runtime_put_sync(&pdev->dev);
3669
3670         return 0;
3671 }
3672 /**
3673  * e1000_set_mac - Change the Ethernet Address of the NIC
3674  * @netdev: network interface device structure
3675  * @p: pointer to an address structure
3676  *
3677  * Returns 0 on success, negative on failure
3678  **/
3679 static int e1000_set_mac(struct net_device *netdev, void *p)
3680 {
3681         struct e1000_adapter *adapter = netdev_priv(netdev);
3682         struct sockaddr *addr = p;
3683
3684         if (!is_valid_ether_addr(addr->sa_data))
3685                 return -EADDRNOTAVAIL;
3686
3687         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3688         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3689
3690         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3691
3692         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3693                 /* activate the work around */
3694                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3695
3696                 /*
3697                  * Hold a copy of the LAA in RAR[14] This is done so that
3698                  * between the time RAR[0] gets clobbered  and the time it
3699                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3700                  * of the RARs and no incoming packets directed to this port
3701                  * are dropped. Eventually the LAA will be in RAR[0] and
3702                  * RAR[14]
3703                  */
3704                 e1000e_rar_set(&adapter->hw,
3705                               adapter->hw.mac.addr,
3706                               adapter->hw.mac.rar_entry_count - 1);
3707         }
3708
3709         return 0;
3710 }
3711
3712 /**
3713  * e1000e_update_phy_task - work thread to update phy
3714  * @work: pointer to our work struct
3715  *
3716  * this worker thread exists because we must acquire a
3717  * semaphore to read the phy, which we could msleep while
3718  * waiting for it, and we can't msleep in a timer.
3719  **/
3720 static void e1000e_update_phy_task(struct work_struct *work)
3721 {
3722         struct e1000_adapter *adapter = container_of(work,
3723                                         struct e1000_adapter, update_phy_task);
3724         e1000_get_phy_info(&adapter->hw);
3725 }
3726
3727 /*
3728  * Need to wait a few seconds after link up to get diagnostic information from
3729  * the phy
3730  */
3731 static void e1000_update_phy_info(unsigned long data)
3732 {
3733         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3734         schedule_work(&adapter->update_phy_task);
3735 }
3736
3737 /**
3738  * e1000e_update_phy_stats - Update the PHY statistics counters
3739  * @adapter: board private structure
3740  **/
3741 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
3742 {
3743         struct e1000_hw *hw = &adapter->hw;
3744         s32 ret_val;
3745         u16 phy_data;
3746
3747         ret_val = hw->phy.ops.acquire(hw);
3748         if (ret_val)
3749                 return;
3750
3751         hw->phy.addr = 1;
3752
3753 #define HV_PHY_STATS_PAGE       778
3754         /*
3755          * A page set is expensive so check if already on desired page.
3756          * If not, set to the page with the PHY status registers.
3757          */
3758         ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
3759                                            &phy_data);
3760         if (ret_val)
3761                 goto release;
3762         if (phy_data != (HV_PHY_STATS_PAGE << IGP_PAGE_SHIFT)) {
3763                 ret_val = e1000e_write_phy_reg_mdic(hw,
3764                                                     IGP01E1000_PHY_PAGE_SELECT,
3765                                                     (HV_PHY_STATS_PAGE <<
3766                                                      IGP_PAGE_SHIFT));
3767                 if (ret_val)
3768                         goto release;
3769         }
3770
3771         /* Read/clear the upper 16-bit registers and read/accumulate lower */
3772
3773         /* Single Collision Count */
3774         e1000e_read_phy_reg_mdic(hw, HV_SCC_UPPER & MAX_PHY_REG_ADDRESS,
3775                                  &phy_data);
3776         ret_val = e1000e_read_phy_reg_mdic(hw,
3777                                            HV_SCC_LOWER & MAX_PHY_REG_ADDRESS,
3778                                            &phy_data);
3779         if (!ret_val)
3780                 adapter->stats.scc += phy_data;
3781
3782         /* Excessive Collision Count */
3783         e1000e_read_phy_reg_mdic(hw, HV_ECOL_UPPER & MAX_PHY_REG_ADDRESS,
3784                                  &phy_data);
3785         ret_val = e1000e_read_phy_reg_mdic(hw,
3786                                            HV_ECOL_LOWER & MAX_PHY_REG_ADDRESS,
3787                                            &phy_data);
3788         if (!ret_val)
3789                 adapter->stats.ecol += phy_data;
3790
3791         /* Multiple Collision Count */
3792         e1000e_read_phy_reg_mdic(hw, HV_MCC_UPPER & MAX_PHY_REG_ADDRESS,
3793                                  &phy_data);
3794         ret_val = e1000e_read_phy_reg_mdic(hw,
3795                                            HV_MCC_LOWER & MAX_PHY_REG_ADDRESS,
3796                                            &phy_data);
3797         if (!ret_val)
3798                 adapter->stats.mcc += phy_data;
3799
3800         /* Late Collision Count */
3801         e1000e_read_phy_reg_mdic(hw, HV_LATECOL_UPPER & MAX_PHY_REG_ADDRESS,
3802                                  &phy_data);
3803         ret_val = e1000e_read_phy_reg_mdic(hw,
3804                                            HV_LATECOL_LOWER &
3805                                            MAX_PHY_REG_ADDRESS,
3806                                            &phy_data);
3807         if (!ret_val)
3808                 adapter->stats.latecol += phy_data;
3809
3810         /* Collision Count - also used for adaptive IFS */
3811         e1000e_read_phy_reg_mdic(hw, HV_COLC_UPPER & MAX_PHY_REG_ADDRESS,
3812                                  &phy_data);
3813         ret_val = e1000e_read_phy_reg_mdic(hw,
3814                                            HV_COLC_LOWER & MAX_PHY_REG_ADDRESS,
3815                                            &phy_data);
3816         if (!ret_val)
3817                 hw->mac.collision_delta = phy_data;
3818
3819         /* Defer Count */
3820         e1000e_read_phy_reg_mdic(hw, HV_DC_UPPER & MAX_PHY_REG_ADDRESS,
3821                                  &phy_data);
3822         ret_val = e1000e_read_phy_reg_mdic(hw,
3823                                            HV_DC_LOWER & MAX_PHY_REG_ADDRESS,
3824                                            &phy_data);
3825         if (!ret_val)
3826                 adapter->stats.dc += phy_data;
3827
3828         /* Transmit with no CRS */
3829         e1000e_read_phy_reg_mdic(hw, HV_TNCRS_UPPER & MAX_PHY_REG_ADDRESS,
3830                                  &phy_data);
3831         ret_val = e1000e_read_phy_reg_mdic(hw,
3832                                            HV_TNCRS_LOWER & MAX_PHY_REG_ADDRESS,
3833                                            &phy_data);
3834         if (!ret_val)
3835                 adapter->stats.tncrs += phy_data;
3836
3837 release:
3838         hw->phy.ops.release(hw);
3839 }
3840
3841 /**
3842  * e1000e_update_stats - Update the board statistics counters
3843  * @adapter: board private structure
3844  **/
3845 void e1000e_update_stats(struct e1000_adapter *adapter)
3846 {
3847         struct net_device *netdev = adapter->netdev;
3848         struct e1000_hw *hw = &adapter->hw;
3849         struct pci_dev *pdev = adapter->pdev;
3850
3851         /*
3852          * Prevent stats update while adapter is being reset, or if the pci
3853          * connection is down.
3854          */
3855         if (adapter->link_speed == 0)
3856                 return;
3857         if (pci_channel_offline(pdev))
3858                 return;
3859
3860         adapter->stats.crcerrs += er32(CRCERRS);
3861         adapter->stats.gprc += er32(GPRC);
3862         adapter->stats.gorc += er32(GORCL);
3863         er32(GORCH); /* Clear gorc */
3864         adapter->stats.bprc += er32(BPRC);
3865         adapter->stats.mprc += er32(MPRC);
3866         adapter->stats.roc += er32(ROC);
3867
3868         adapter->stats.mpc += er32(MPC);
3869
3870         /* Half-duplex statistics */
3871         if (adapter->link_duplex == HALF_DUPLEX) {
3872                 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
3873                         e1000e_update_phy_stats(adapter);
3874                 } else {
3875                         adapter->stats.scc += er32(SCC);
3876                         adapter->stats.ecol += er32(ECOL);
3877                         adapter->stats.mcc += er32(MCC);
3878                         adapter->stats.latecol += er32(LATECOL);
3879                         adapter->stats.dc += er32(DC);
3880
3881                         hw->mac.collision_delta = er32(COLC);
3882
3883                         if ((hw->mac.type != e1000_82574) &&
3884                             (hw->mac.type != e1000_82583))
3885                                 adapter->stats.tncrs += er32(TNCRS);
3886                 }
3887                 adapter->stats.colc += hw->mac.collision_delta;
3888         }
3889
3890         adapter->stats.xonrxc += er32(XONRXC);
3891         adapter->stats.xontxc += er32(XONTXC);
3892         adapter->stats.xoffrxc += er32(XOFFRXC);
3893         adapter->stats.xofftxc += er32(XOFFTXC);
3894         adapter->stats.gptc += er32(GPTC);
3895         adapter->stats.gotc += er32(GOTCL);
3896         er32(GOTCH); /* Clear gotc */
3897         adapter->stats.rnbc += er32(RNBC);
3898         adapter->stats.ruc += er32(RUC);
3899
3900         adapter->stats.mptc += er32(MPTC);
3901         adapter->stats.bptc += er32(BPTC);
3902
3903         /* used for adaptive IFS */
3904
3905         hw->mac.tx_packet_delta = er32(TPT);
3906         adapter->stats.tpt += hw->mac.tx_packet_delta;
3907
3908         adapter->stats.algnerrc += er32(ALGNERRC);
3909         adapter->stats.rxerrc += er32(RXERRC);
3910         adapter->stats.cexterr += er32(CEXTERR);
3911         adapter->stats.tsctc += er32(TSCTC);
3912         adapter->stats.tsctfc += er32(TSCTFC);
3913
3914         /* Fill out the OS statistics structure */
3915         netdev->stats.multicast = adapter->stats.mprc;
3916         netdev->stats.collisions = adapter->stats.colc;
3917
3918         /* Rx Errors */
3919
3920         /*
3921          * RLEC on some newer hardware can be incorrect so build
3922          * our own version based on RUC and ROC
3923          */
3924         netdev->stats.rx_errors = adapter->stats.rxerrc +
3925                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3926                 adapter->stats.ruc + adapter->stats.roc +
3927                 adapter->stats.cexterr;
3928         netdev->stats.rx_length_errors = adapter->stats.ruc +
3929                                               adapter->stats.roc;
3930         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3931         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3932         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3933
3934         /* Tx Errors */
3935         netdev->stats.tx_errors = adapter->stats.ecol +
3936                                        adapter->stats.latecol;
3937         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3938         netdev->stats.tx_window_errors = adapter->stats.latecol;
3939         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3940
3941         /* Tx Dropped needs to be maintained elsewhere */
3942
3943         /* Management Stats */
3944         adapter->stats.mgptc += er32(MGTPTC);
3945         adapter->stats.mgprc += er32(MGTPRC);
3946         adapter->stats.mgpdc += er32(MGTPDC);
3947 }
3948
3949 /**
3950  * e1000_phy_read_status - Update the PHY register status snapshot
3951  * @adapter: board private structure
3952  **/
3953 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3954 {
3955         struct e1000_hw *hw = &adapter->hw;
3956         struct e1000_phy_regs *phy = &adapter->phy_regs;
3957         int ret_val;
3958
3959         if ((er32(STATUS) & E1000_STATUS_LU) &&
3960             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3961                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3962                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3963                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3964                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3965                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3966                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3967                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3968                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3969                 if (ret_val)
3970                         e_warn("Error reading PHY register\n");
3971         } else {
3972                 /*
3973                  * Do not read PHY registers if link is not up
3974                  * Set values to typical power-on defaults
3975                  */
3976                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3977                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3978                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3979                              BMSR_ERCAP);
3980                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3981                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3982                 phy->lpa = 0;
3983                 phy->expansion = EXPANSION_ENABLENPAGE;
3984                 phy->ctrl1000 = ADVERTISE_1000FULL;
3985                 phy->stat1000 = 0;
3986                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3987         }
3988 }
3989
3990 static void e1000_print_link_info(struct e1000_adapter *adapter)
3991 {
3992         struct e1000_hw *hw = &adapter->hw;
3993         u32 ctrl = er32(CTRL);
3994
3995         /* Link status message must follow this format for user tools */
3996         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
3997                "Flow Control: %s\n",
3998                adapter->netdev->name,
3999                adapter->link_speed,
4000                (adapter->link_duplex == FULL_DUPLEX) ?
4001                                 "Full Duplex" : "Half Duplex",
4002                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
4003                                 "RX/TX" :
4004                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
4005                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
4006 }
4007
4008 static bool e1000e_has_link(struct e1000_adapter *adapter)
4009 {
4010         struct e1000_hw *hw = &adapter->hw;
4011         bool link_active = 0;
4012         s32 ret_val = 0;
4013
4014         /*
4015          * get_link_status is set on LSC (link status) interrupt or
4016          * Rx sequence error interrupt.  get_link_status will stay
4017          * false until the check_for_link establishes link
4018          * for copper adapters ONLY
4019          */
4020         switch (hw->phy.media_type) {
4021         case e1000_media_type_copper:
4022                 if (hw->mac.get_link_status) {
4023                         ret_val = hw->mac.ops.check_for_link(hw);
4024                         link_active = !hw->mac.get_link_status;
4025                 } else {
4026                         link_active = 1;
4027                 }
4028                 break;
4029         case e1000_media_type_fiber:
4030                 ret_val = hw->mac.ops.check_for_link(hw);
4031                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4032                 break;
4033         case e1000_media_type_internal_serdes:
4034                 ret_val = hw->mac.ops.check_for_link(hw);
4035                 link_active = adapter->hw.mac.serdes_has_link;
4036                 break;
4037         default:
4038         case e1000_media_type_unknown:
4039                 break;
4040         }
4041
4042         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4043             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4044                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4045                 e_info("Gigabit has been disabled, downgrading speed\n");
4046         }
4047
4048         return link_active;
4049 }
4050
4051 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4052 {
4053         /* make sure the receive unit is started */
4054         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4055             (adapter->flags & FLAG_RX_RESTART_NOW)) {
4056                 struct e1000_hw *hw = &adapter->hw;
4057                 u32 rctl = er32(RCTL);
4058                 ew32(RCTL, rctl | E1000_RCTL_EN);
4059                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
4060         }
4061 }
4062
4063 /**
4064  * e1000_watchdog - Timer Call-back
4065  * @data: pointer to adapter cast into an unsigned long
4066  **/
4067 static void e1000_watchdog(unsigned long data)
4068 {
4069         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4070
4071         /* Do the rest outside of interrupt context */
4072         schedule_work(&adapter->watchdog_task);
4073
4074         /* TODO: make this use queue_delayed_work() */
4075 }
4076
4077 static void e1000_watchdog_task(struct work_struct *work)
4078 {
4079         struct e1000_adapter *adapter = container_of(work,
4080                                         struct e1000_adapter, watchdog_task);
4081         struct net_device *netdev = adapter->netdev;
4082         struct e1000_mac_info *mac = &adapter->hw.mac;
4083         struct e1000_phy_info *phy = &adapter->hw.phy;
4084         struct e1000_ring *tx_ring = adapter->tx_ring;
4085         struct e1000_hw *hw = &adapter->hw;
4086         u32 link, tctl;
4087         int tx_pending = 0;
4088
4089         link = e1000e_has_link(adapter);
4090         if ((netif_carrier_ok(netdev)) && link) {
4091                 /* Cancel scheduled suspend requests. */
4092                 pm_runtime_resume(netdev->dev.parent);
4093
4094                 e1000e_enable_receives(adapter);
4095                 goto link_up;
4096         }
4097
4098         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4099             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4100                 e1000_update_mng_vlan(adapter);
4101
4102         if (link) {
4103                 if (!netif_carrier_ok(netdev)) {
4104                         bool txb2b = 1;
4105
4106                         /* Cancel scheduled suspend requests. */
4107                         pm_runtime_resume(netdev->dev.parent);
4108
4109                         /* update snapshot of PHY registers on LSC */
4110                         e1000_phy_read_status(adapter);
4111                         mac->ops.get_link_up_info(&adapter->hw,
4112                                                    &adapter->link_speed,
4113                                                    &adapter->link_duplex);
4114                         e1000_print_link_info(adapter);
4115                         /*
4116                          * On supported PHYs, check for duplex mismatch only
4117                          * if link has autonegotiated at 10/100 half
4118                          */
4119                         if ((hw->phy.type == e1000_phy_igp_3 ||
4120                              hw->phy.type == e1000_phy_bm) &&
4121                             (hw->mac.autoneg == true) &&
4122                             (adapter->link_speed == SPEED_10 ||
4123                              adapter->link_speed == SPEED_100) &&
4124                             (adapter->link_duplex == HALF_DUPLEX)) {
4125                                 u16 autoneg_exp;
4126
4127                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4128
4129                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4130                                         e_info("Autonegotiated half duplex but"
4131                                                " link partner cannot autoneg. "
4132                                                " Try forcing full duplex if "
4133                                                "link gets many collisions.\n");
4134                         }
4135
4136                         /* adjust timeout factor according to speed/duplex */
4137                         adapter->tx_timeout_factor = 1;
4138                         switch (adapter->link_speed) {
4139                         case SPEED_10:
4140                                 txb2b = 0;
4141                                 adapter->tx_timeout_factor = 16;
4142                                 break;
4143                         case SPEED_100:
4144                                 txb2b = 0;
4145                                 adapter->tx_timeout_factor = 10;
4146                                 break;
4147                         }
4148
4149                         /*
4150                          * workaround: re-program speed mode bit after
4151                          * link-up event
4152                          */
4153                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4154                             !txb2b) {
4155                                 u32 tarc0;
4156                                 tarc0 = er32(TARC(0));
4157                                 tarc0 &= ~SPEED_MODE_BIT;
4158                                 ew32(TARC(0), tarc0);
4159                         }
4160
4161                         /*
4162                          * disable TSO for pcie and 10/100 speeds, to avoid
4163                          * some hardware issues
4164                          */
4165                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
4166                                 switch (adapter->link_speed) {
4167                                 case SPEED_10:
4168                                 case SPEED_100:
4169                                         e_info("10/100 speed: disabling TSO\n");
4170                                         netdev->features &= ~NETIF_F_TSO;
4171                                         netdev->features &= ~NETIF_F_TSO6;
4172                                         break;
4173                                 case SPEED_1000:
4174                                         netdev->features |= NETIF_F_TSO;
4175                                         netdev->features |= NETIF_F_TSO6;
4176                                         break;
4177                                 default:
4178                                         /* oops */
4179                                         break;
4180                                 }
4181                         }
4182
4183                         /*
4184                          * enable transmits in the hardware, need to do this
4185                          * after setting TARC(0)
4186                          */
4187                         tctl = er32(TCTL);
4188                         tctl |= E1000_TCTL_EN;
4189                         ew32(TCTL, tctl);
4190
4191                         /*
4192                          * Perform any post-link-up configuration before
4193                          * reporting link up.
4194                          */
4195                         if (phy->ops.cfg_on_link_up)
4196                                 phy->ops.cfg_on_link_up(hw);
4197
4198                         netif_carrier_on(netdev);
4199
4200                         if (!test_bit(__E1000_DOWN, &adapter->state))
4201                                 mod_timer(&adapter->phy_info_timer,
4202                                           round_jiffies(jiffies + 2 * HZ));
4203                 }
4204         } else {
4205                 if (netif_carrier_ok(netdev)) {
4206                         adapter->link_speed = 0;
4207                         adapter->link_duplex = 0;
4208                         /* Link status message must follow this format */
4209                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4210                                adapter->netdev->name);
4211                         netif_carrier_off(netdev);
4212                         if (!test_bit(__E1000_DOWN, &adapter->state))
4213                                 mod_timer(&adapter->phy_info_timer,
4214                                           round_jiffies(jiffies + 2 * HZ));
4215
4216                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4217                                 schedule_work(&adapter->reset_task);
4218                         else
4219                                 pm_schedule_suspend(netdev->dev.parent,
4220                                                         LINK_TIMEOUT);
4221                 }
4222         }
4223
4224 link_up:
4225         e1000e_update_stats(adapter);
4226
4227         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4228         adapter->tpt_old = adapter->stats.tpt;
4229         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4230         adapter->colc_old = adapter->stats.colc;
4231
4232         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4233         adapter->gorc_old = adapter->stats.gorc;
4234         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4235         adapter->gotc_old = adapter->stats.gotc;
4236
4237         e1000e_update_adaptive(&adapter->hw);
4238
4239         if (!netif_carrier_ok(netdev)) {
4240                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
4241                                tx_ring->count);
4242                 if (tx_pending) {
4243                         /*
4244                          * We've lost link, so the controller stops DMA,
4245                          * but we've got queued Tx work that's never going
4246                          * to get done, so reset controller to flush Tx.
4247                          * (Do the reset outside of interrupt context).
4248                          */
4249                         adapter->tx_timeout_count++;
4250                         schedule_work(&adapter->reset_task);
4251                         /* return immediately since reset is imminent */
4252                         return;
4253                 }
4254         }
4255
4256         /* Simple mode for Interrupt Throttle Rate (ITR) */
4257         if (adapter->itr_setting == 4) {
4258                 /*
4259                  * Symmetric Tx/Rx gets a reduced ITR=2000;
4260                  * Total asymmetrical Tx or Rx gets ITR=8000;
4261                  * everyone else is between 2000-8000.
4262                  */
4263                 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4264                 u32 dif = (adapter->gotc > adapter->gorc ?
4265                             adapter->gotc - adapter->gorc :
4266                             adapter->gorc - adapter->gotc) / 10000;
4267                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4268
4269                 ew32(ITR, 1000000000 / (itr * 256));
4270         }
4271
4272         /* Cause software interrupt to ensure Rx ring is cleaned */
4273         if (adapter->msix_entries)
4274                 ew32(ICS, adapter->rx_ring->ims_val);
4275         else
4276                 ew32(ICS, E1000_ICS_RXDMT0);
4277
4278         /* Force detection of hung controller every watchdog period */
4279         adapter->detect_tx_hung = 1;
4280
4281         /* flush partial descriptors to memory before detecting tx hang */
4282         if (adapter->flags2 & FLAG2_DMA_BURST) {
4283                 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
4284                 ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
4285                 /*
4286                  * no need to flush the writes because the timeout code does
4287                  * an er32 first thing
4288                  */
4289         }
4290
4291         /*
4292          * With 82571 controllers, LAA may be overwritten due to controller
4293          * reset from the other port. Set the appropriate LAA in RAR[0]
4294          */
4295         if (e1000e_get_laa_state_82571(hw))
4296                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4297
4298         /* Reset the timer */
4299         if (!test_bit(__E1000_DOWN, &adapter->state))
4300                 mod_timer(&adapter->watchdog_timer,
4301                           round_jiffies(jiffies + 2 * HZ));
4302 }
4303
4304 #define E1000_TX_FLAGS_CSUM             0x00000001
4305 #define E1000_TX_FLAGS_VLAN             0x00000002
4306 #define E1000_TX_FLAGS_TSO              0x00000004
4307 #define E1000_TX_FLAGS_IPV4             0x00000008
4308 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
4309 #define E1000_TX_FLAGS_VLAN_SHIFT       16
4310
4311 static int e1000_tso(struct e1000_adapter *adapter,
4312                      struct sk_buff *skb)
4313 {
4314         struct e1000_ring *tx_ring = adapter->tx_ring;
4315         struct e1000_context_desc *context_desc;
4316         struct e1000_buffer *buffer_info;
4317         unsigned int i;
4318         u32 cmd_length = 0;
4319         u16 ipcse = 0, tucse, mss;
4320         u8 ipcss, ipcso, tucss, tucso, hdr_len;
4321         int err;
4322
4323         if (!skb_is_gso(skb))
4324                 return 0;
4325
4326         if (skb_header_cloned(skb)) {
4327                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4328                 if (err)
4329                         return err;
4330         }
4331
4332         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4333         mss = skb_shinfo(skb)->gso_size;
4334         if (skb->protocol == htons(ETH_P_IP)) {
4335                 struct iphdr *iph = ip_hdr(skb);
4336                 iph->tot_len = 0;
4337                 iph->check = 0;
4338                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4339                                                          0, IPPROTO_TCP, 0);
4340                 cmd_length = E1000_TXD_CMD_IP;
4341                 ipcse = skb_transport_offset(skb) - 1;
4342         } else if (skb_is_gso_v6(skb)) {
4343                 ipv6_hdr(skb)->payload_len = 0;
4344                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4345                                                        &ipv6_hdr(skb)->daddr,
4346                                                        0, IPPROTO_TCP, 0);
4347                 ipcse = 0;
4348         }
4349         ipcss = skb_network_offset(skb);
4350         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4351         tucss = skb_transport_offset(skb);
4352         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4353         tucse = 0;
4354
4355         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4356                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4357
4358         i = tx_ring->next_to_use;
4359         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4360         buffer_info = &tx_ring->buffer_info[i];
4361
4362         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4363         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4364         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4365         context_desc->upper_setup.tcp_fields.tucss = tucss;
4366         context_desc->upper_setup.tcp_fields.tucso = tucso;
4367         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4368         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4369         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4370         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4371
4372         buffer_info->time_stamp = jiffies;
4373         buffer_info->next_to_watch = i;
4374
4375         i++;
4376         if (i == tx_ring->count)
4377                 i = 0;
4378         tx_ring->next_to_use = i;
4379
4380         return 1;
4381 }
4382
4383 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
4384 {
4385         struct e1000_ring *tx_ring = adapter->tx_ring;
4386         struct e1000_context_desc *context_desc;
4387         struct e1000_buffer *buffer_info;
4388         unsigned int i;
4389         u8 css;
4390         u32 cmd_len = E1000_TXD_CMD_DEXT;
4391         __be16 protocol;
4392
4393         if (skb->ip_summed != CHECKSUM_PARTIAL)
4394                 return 0;
4395
4396         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4397                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4398         else
4399                 protocol = skb->protocol;
4400
4401         switch (protocol) {
4402         case cpu_to_be16(ETH_P_IP):
4403                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4404                         cmd_len |= E1000_TXD_CMD_TCP;
4405                 break;
4406         case cpu_to_be16(ETH_P_IPV6):
4407                 /* XXX not handling all IPV6 headers */
4408                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4409                         cmd_len |= E1000_TXD_CMD_TCP;
4410                 break;
4411         default:
4412                 if (unlikely(net_ratelimit()))
4413                         e_warn("checksum_partial proto=%x!\n",
4414                                be16_to_cpu(protocol));
4415                 break;
4416         }
4417
4418         css = skb_transport_offset(skb);
4419
4420         i = tx_ring->next_to_use;
4421         buffer_info = &tx_ring->buffer_info[i];
4422         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4423
4424         context_desc->lower_setup.ip_config = 0;
4425         context_desc->upper_setup.tcp_fields.tucss = css;
4426         context_desc->upper_setup.tcp_fields.tucso =
4427                                 css + skb->csum_offset;
4428         context_desc->upper_setup.tcp_fields.tucse = 0;
4429         context_desc->tcp_seg_setup.data = 0;
4430         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4431
4432         buffer_info->time_stamp = jiffies;
4433         buffer_info->next_to_watch = i;
4434
4435         i++;
4436         if (i == tx_ring->count)
4437                 i = 0;
4438         tx_ring->next_to_use = i;
4439
4440         return 1;
4441 }
4442
4443 #define E1000_MAX_PER_TXD       8192
4444 #define E1000_MAX_TXD_PWR       12
4445
4446 static int e1000_tx_map(struct e1000_adapter *adapter,
4447                         struct sk_buff *skb, unsigned int first,
4448                         unsigned int max_per_txd, unsigned int nr_frags,
4449                         unsigned int mss)
4450 {
4451         struct e1000_ring *tx_ring = adapter->tx_ring;
4452         struct pci_dev *pdev = adapter->pdev;
4453         struct e1000_buffer *buffer_info;
4454         unsigned int len = skb_headlen(skb);
4455         unsigned int offset = 0, size, count = 0, i;
4456         unsigned int f, bytecount, segs;
4457
4458         i = tx_ring->next_to_use;
4459
4460         while (len) {
4461                 buffer_info = &tx_ring->buffer_info[i];
4462                 size = min(len, max_per_txd);
4463
4464                 buffer_info->length = size;
4465                 buffer_info->time_stamp = jiffies;
4466                 buffer_info->next_to_watch = i;
4467                 buffer_info->dma = dma_map_single(&pdev->dev,
4468                                                   skb->data + offset,
4469                                                   size, DMA_TO_DEVICE);
4470                 buffer_info->mapped_as_page = false;
4471                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4472                         goto dma_error;
4473
4474                 len -= size;
4475                 offset += size;
4476                 count++;
4477
4478                 if (len) {
4479                         i++;
4480                         if (i == tx_ring->count)
4481                                 i = 0;
4482                 }
4483         }
4484
4485         for (f = 0; f < nr_frags; f++) {
4486                 struct skb_frag_struct *frag;
4487
4488                 frag = &skb_shinfo(skb)->frags[f];
4489                 len = frag->size;
4490                 offset = frag->page_offset;
4491
4492                 while (len) {
4493                         i++;
4494                         if (i == tx_ring->count)
4495                                 i = 0;
4496
4497                         buffer_info = &tx_ring->buffer_info[i];
4498                         size = min(len, max_per_txd);
4499
4500                         buffer_info->length = size;
4501                         buffer_info->time_stamp = jiffies;
4502                         buffer_info->next_to_watch = i;
4503                         buffer_info->dma = dma_map_page(&pdev->dev, frag->page,
4504                                                         offset, size,
4505                                                         DMA_TO_DEVICE);
4506                         buffer_info->mapped_as_page = true;
4507                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4508                                 goto dma_error;
4509
4510                         len -= size;
4511                         offset += size;
4512                         count++;
4513                 }
4514         }
4515
4516         segs = skb_shinfo(skb)->gso_segs ?: 1;
4517         /* multiply data chunks by size of headers */
4518         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4519
4520         tx_ring->buffer_info[i].skb = skb;
4521         tx_ring->buffer_info[i].segs = segs;
4522         tx_ring->buffer_info[i].bytecount = bytecount;
4523         tx_ring->buffer_info[first].next_to_watch = i;
4524
4525         return count;
4526
4527 dma_error:
4528         dev_err(&pdev->dev, "TX DMA map failed\n");
4529         buffer_info->dma = 0;
4530         if (count)
4531                 count--;
4532
4533         while (count--) {
4534                 if (i==0)
4535                         i += tx_ring->count;
4536                 i--;
4537                 buffer_info = &tx_ring->buffer_info[i];
4538                 e1000_put_txbuf(adapter, buffer_info);;
4539         }
4540
4541         return 0;
4542 }
4543
4544 static void e1000_tx_queue(struct e1000_adapter *adapter,
4545                            int tx_flags, int count)
4546 {
4547         struct e1000_ring *tx_ring = adapter->tx_ring;
4548         struct e1000_tx_desc *tx_desc = NULL;
4549         struct e1000_buffer *buffer_info;
4550         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4551         unsigned int i;
4552
4553         if (tx_flags & E1000_TX_FLAGS_TSO) {
4554                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4555                              E1000_TXD_CMD_TSE;
4556                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4557
4558                 if (tx_flags & E1000_TX_FLAGS_IPV4)
4559                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4560         }
4561
4562         if (tx_flags & E1000_TX_FLAGS_CSUM) {
4563                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4564                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4565         }
4566
4567         if (tx_flags & E1000_TX_FLAGS_VLAN) {
4568                 txd_lower |= E1000_TXD_CMD_VLE;
4569                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4570         }
4571
4572         i = tx_ring->next_to_use;
4573
4574         while (count--) {
4575                 buffer_info = &tx_ring->buffer_info[i];
4576                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4577                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4578                 tx_desc->lower.data =
4579                         cpu_to_le32(txd_lower | buffer_info->length);
4580                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4581
4582                 i++;
4583                 if (i == tx_ring->count)
4584                         i = 0;
4585         }
4586
4587         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4588
4589         /*
4590          * Force memory writes to complete before letting h/w
4591          * know there are new descriptors to fetch.  (Only
4592          * applicable for weak-ordered memory model archs,
4593          * such as IA-64).
4594          */
4595         wmb();
4596
4597         tx_ring->next_to_use = i;
4598         writel(i, adapter->hw.hw_addr + tx_ring->tail);
4599         /*
4600          * we need this if more than one processor can write to our tail
4601          * at a time, it synchronizes IO on IA64/Altix systems
4602          */
4603         mmiowb();
4604 }
4605
4606 #define MINIMUM_DHCP_PACKET_SIZE 282
4607 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4608                                     struct sk_buff *skb)
4609 {
4610         struct e1000_hw *hw =  &adapter->hw;
4611         u16 length, offset;
4612
4613         if (vlan_tx_tag_present(skb)) {
4614                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4615                     (adapter->hw.mng_cookie.status &
4616                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4617                         return 0;
4618         }
4619
4620         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4621                 return 0;
4622
4623         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4624                 return 0;
4625
4626         {
4627                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4628                 struct udphdr *udp;
4629
4630                 if (ip->protocol != IPPROTO_UDP)
4631                         return 0;
4632
4633                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4634                 if (ntohs(udp->dest) != 67)
4635                         return 0;
4636
4637                 offset = (u8 *)udp + 8 - skb->data;
4638                 length = skb->len - offset;
4639                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4640         }
4641
4642         return 0;
4643 }
4644
4645 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4646 {
4647         struct e1000_adapter *adapter = netdev_priv(netdev);
4648
4649         netif_stop_queue(netdev);
4650         /*
4651          * Herbert's original patch had:
4652          *  smp_mb__after_netif_stop_queue();
4653          * but since that doesn't exist yet, just open code it.
4654          */
4655         smp_mb();
4656
4657         /*
4658          * We need to check again in a case another CPU has just
4659          * made room available.
4660          */
4661         if (e1000_desc_unused(adapter->tx_ring) < size)
4662                 return -EBUSY;
4663
4664         /* A reprieve! */
4665         netif_start_queue(netdev);
4666         ++adapter->restart_queue;
4667         return 0;
4668 }
4669
4670 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4671 {
4672         struct e1000_adapter *adapter = netdev_priv(netdev);
4673
4674         if (e1000_desc_unused(adapter->tx_ring) >= size)
4675                 return 0;
4676         return __e1000_maybe_stop_tx(netdev, size);
4677 }
4678
4679 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4680 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4681                                     struct net_device *netdev)
4682 {
4683         struct e1000_adapter *adapter = netdev_priv(netdev);
4684         struct e1000_ring *tx_ring = adapter->tx_ring;
4685         unsigned int first;
4686         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4687         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4688         unsigned int tx_flags = 0;
4689         unsigned int len = skb_headlen(skb);
4690         unsigned int nr_frags;
4691         unsigned int mss;
4692         int count = 0;
4693         int tso;
4694         unsigned int f;
4695
4696         if (test_bit(__E1000_DOWN, &adapter->state)) {
4697                 dev_kfree_skb_any(skb);
4698                 return NETDEV_TX_OK;
4699         }
4700
4701         if (skb->len <= 0) {
4702                 dev_kfree_skb_any(skb);
4703                 return NETDEV_TX_OK;
4704         }
4705
4706         mss = skb_shinfo(skb)->gso_size;
4707         /*
4708          * The controller does a simple calculation to
4709          * make sure there is enough room in the FIFO before
4710          * initiating the DMA for each buffer.  The calc is:
4711          * 4 = ceil(buffer len/mss).  To make sure we don't
4712          * overrun the FIFO, adjust the max buffer len if mss
4713          * drops.
4714          */
4715         if (mss) {
4716                 u8 hdr_len;
4717                 max_per_txd = min(mss << 2, max_per_txd);
4718                 max_txd_pwr = fls(max_per_txd) - 1;
4719
4720                 /*
4721                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4722                  * points to just header, pull a few bytes of payload from
4723                  * frags into skb->data
4724                  */
4725                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4726                 /*
4727                  * we do this workaround for ES2LAN, but it is un-necessary,
4728                  * avoiding it could save a lot of cycles
4729                  */
4730                 if (skb->data_len && (hdr_len == len)) {
4731                         unsigned int pull_size;
4732
4733                         pull_size = min((unsigned int)4, skb->data_len);
4734                         if (!__pskb_pull_tail(skb, pull_size)) {
4735                                 e_err("__pskb_pull_tail failed.\n");
4736                                 dev_kfree_skb_any(skb);
4737                                 return NETDEV_TX_OK;
4738                         }
4739                         len = skb_headlen(skb);
4740                 }
4741         }
4742
4743         /* reserve a descriptor for the offload context */
4744         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4745                 count++;
4746         count++;
4747
4748         count += TXD_USE_COUNT(len, max_txd_pwr);
4749
4750         nr_frags = skb_shinfo(skb)->nr_frags;
4751         for (f = 0; f < nr_frags; f++)
4752                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4753                                        max_txd_pwr);
4754
4755         if (adapter->hw.mac.tx_pkt_filtering)
4756                 e1000_transfer_dhcp_info(adapter, skb);
4757
4758         /*
4759          * need: count + 2 desc gap to keep tail from touching
4760          * head, otherwise try next time
4761          */
4762         if (e1000_maybe_stop_tx(netdev, count + 2))
4763                 return NETDEV_TX_BUSY;
4764
4765         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4766                 tx_flags |= E1000_TX_FLAGS_VLAN;
4767                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4768         }
4769
4770         first = tx_ring->next_to_use;
4771
4772         tso = e1000_tso(adapter, skb);
4773         if (tso < 0) {
4774                 dev_kfree_skb_any(skb);
4775                 return NETDEV_TX_OK;
4776         }
4777
4778         if (tso)
4779                 tx_flags |= E1000_TX_FLAGS_TSO;
4780         else if (e1000_tx_csum(adapter, skb))
4781                 tx_flags |= E1000_TX_FLAGS_CSUM;
4782
4783         /*
4784          * Old method was to assume IPv4 packet by default if TSO was enabled.
4785          * 82571 hardware supports TSO capabilities for IPv6 as well...
4786          * no longer assume, we must.
4787          */
4788         if (skb->protocol == htons(ETH_P_IP))
4789                 tx_flags |= E1000_TX_FLAGS_IPV4;
4790
4791         /* if count is 0 then mapping error has occured */
4792         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4793         if (count) {
4794                 e1000_tx_queue(adapter, tx_flags, count);
4795                 /* Make sure there is space in the ring for the next send. */
4796                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4797
4798         } else {
4799                 dev_kfree_skb_any(skb);
4800                 tx_ring->buffer_info[first].time_stamp = 0;
4801                 tx_ring->next_to_use = first;
4802         }
4803
4804         return NETDEV_TX_OK;
4805 }
4806
4807 /**
4808  * e1000_tx_timeout - Respond to a Tx Hang
4809  * @netdev: network interface device structure
4810  **/
4811 static void e1000_tx_timeout(struct net_device *netdev)
4812 {
4813         struct e1000_adapter *adapter = netdev_priv(netdev);
4814
4815         /* Do the reset outside of interrupt context */
4816         adapter->tx_timeout_count++;
4817         schedule_work(&adapter->reset_task);
4818 }
4819
4820 static void e1000_reset_task(struct work_struct *work)
4821 {
4822         struct e1000_adapter *adapter;
4823         adapter = container_of(work, struct e1000_adapter, reset_task);
4824
4825         e1000e_dump(adapter);
4826         e_err("Reset adapter\n");
4827         e1000e_reinit_locked(adapter);
4828 }
4829
4830 /**
4831  * e1000_get_stats - Get System Network Statistics
4832  * @netdev: network interface device structure
4833  *
4834  * Returns the address of the device statistics structure.
4835  * The statistics are actually updated from the timer callback.
4836  **/
4837 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4838 {
4839         /* only return the current stats */
4840         return &netdev->stats;
4841 }
4842
4843 /**
4844  * e1000_change_mtu - Change the Maximum Transfer Unit
4845  * @netdev: network interface device structure
4846  * @new_mtu: new value for maximum frame size
4847  *
4848  * Returns 0 on success, negative on failure
4849  **/
4850 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4851 {
4852         struct e1000_adapter *adapter = netdev_priv(netdev);
4853         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4854
4855         /* Jumbo frame support */
4856         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
4857             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4858                 e_err("Jumbo Frames not supported.\n");
4859                 return -EINVAL;
4860         }
4861
4862         /* Supported frame sizes */
4863         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4864             (max_frame > adapter->max_hw_frame_size)) {
4865                 e_err("Unsupported MTU setting\n");
4866                 return -EINVAL;
4867         }
4868
4869         /* Jumbo frame workaround on 82579 requires CRC be stripped */
4870         if ((adapter->hw.mac.type == e1000_pch2lan) &&
4871             !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
4872             (new_mtu > ETH_DATA_LEN)) {
4873                 e_err("Jumbo Frames not supported on 82579 when CRC "
4874                       "stripping is disabled.\n");
4875                 return -EINVAL;
4876         }
4877
4878         /* 82573 Errata 17 */
4879         if (((adapter->hw.mac.type == e1000_82573) ||
4880              (adapter->hw.mac.type == e1000_82574)) &&
4881             (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
4882                 adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
4883                 e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
4884         }
4885
4886         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4887                 msleep(1);
4888         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
4889         adapter->max_frame_size = max_frame;
4890         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4891         netdev->mtu = new_mtu;
4892         if (netif_running(netdev))
4893                 e1000e_down(adapter);
4894
4895         /*
4896          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4897          * means we reserve 2 more, this pushes us to allocate from the next
4898          * larger slab size.
4899          * i.e. RXBUFFER_2048 --> size-4096 slab
4900          * However with the new *_jumbo_rx* routines, jumbo receives will use
4901          * fragmented skbs
4902          */
4903
4904         if (max_frame <= 2048)
4905                 adapter->rx_buffer_len = 2048;
4906         else
4907                 adapter->rx_buffer_len = 4096;
4908
4909         /* adjust allocation if LPE protects us, and we aren't using SBP */
4910         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4911              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4912                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4913                                          + ETH_FCS_LEN;
4914
4915         if (netif_running(netdev))
4916                 e1000e_up(adapter);
4917         else
4918                 e1000e_reset(adapter);
4919
4920         clear_bit(__E1000_RESETTING, &adapter->state);
4921
4922         return 0;
4923 }
4924
4925 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4926                            int cmd)
4927 {
4928         struct e1000_adapter *adapter = netdev_priv(netdev);
4929         struct mii_ioctl_data *data = if_mii(ifr);
4930
4931         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4932                 return -EOPNOTSUPP;
4933
4934         switch (cmd) {
4935         case SIOCGMIIPHY:
4936                 data->phy_id = adapter->hw.phy.addr;
4937                 break;
4938         case SIOCGMIIREG:
4939                 e1000_phy_read_status(adapter);
4940
4941                 switch (data->reg_num & 0x1F) {
4942                 case MII_BMCR:
4943                         data->val_out = adapter->phy_regs.bmcr;
4944                         break;
4945                 case MII_BMSR:
4946                         data->val_out = adapter->phy_regs.bmsr;
4947                         break;
4948                 case MII_PHYSID1:
4949                         data->val_out = (adapter->hw.phy.id >> 16);
4950                         break;
4951                 case MII_PHYSID2:
4952                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
4953                         break;
4954                 case MII_ADVERTISE:
4955                         data->val_out = adapter->phy_regs.advertise;
4956                         break;
4957                 case MII_LPA:
4958                         data->val_out = adapter->phy_regs.lpa;
4959                         break;
4960                 case MII_EXPANSION:
4961                         data->val_out = adapter->phy_regs.expansion;
4962                         break;
4963                 case MII_CTRL1000:
4964                         data->val_out = adapter->phy_regs.ctrl1000;
4965                         break;
4966                 case MII_STAT1000:
4967                         data->val_out = adapter->phy_regs.stat1000;
4968                         break;
4969                 case MII_ESTATUS:
4970                         data->val_out = adapter->phy_regs.estatus;
4971                         break;
4972                 default:
4973                         return -EIO;
4974                 }
4975                 break;
4976         case SIOCSMIIREG:
4977         default:
4978                 return -EOPNOTSUPP;
4979         }
4980         return 0;
4981 }
4982
4983 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4984 {
4985         switch (cmd) {
4986         case SIOCGMIIPHY:
4987         case SIOCGMIIREG:
4988         case SIOCSMIIREG:
4989                 return e1000_mii_ioctl(netdev, ifr, cmd);
4990         default:
4991                 return -EOPNOTSUPP;
4992         }
4993 }
4994
4995 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
4996 {
4997         struct e1000_hw *hw = &adapter->hw;
4998         u32 i, mac_reg;
4999         u16 phy_reg;
5000         int retval = 0;
5001
5002         /* copy MAC RARs to PHY RARs */
5003         e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5004
5005         /* copy MAC MTA to PHY MTA */
5006         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5007                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5008                 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
5009                 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
5010         }
5011
5012         /* configure PHY Rx Control register */
5013         e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
5014         mac_reg = er32(RCTL);
5015         if (mac_reg & E1000_RCTL_UPE)
5016                 phy_reg |= BM_RCTL_UPE;
5017         if (mac_reg & E1000_RCTL_MPE)
5018                 phy_reg |= BM_RCTL_MPE;
5019         phy_reg &= ~(BM_RCTL_MO_MASK);
5020         if (mac_reg & E1000_RCTL_MO_3)
5021                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5022                                 << BM_RCTL_MO_SHIFT);
5023         if (mac_reg & E1000_RCTL_BAM)
5024                 phy_reg |= BM_RCTL_BAM;
5025         if (mac_reg & E1000_RCTL_PMCF)
5026                 phy_reg |= BM_RCTL_PMCF;
5027         mac_reg = er32(CTRL);
5028         if (mac_reg & E1000_CTRL_RFCE)
5029                 phy_reg |= BM_RCTL_RFCE;
5030         e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
5031
5032         /* enable PHY wakeup in MAC register */
5033         ew32(WUFC, wufc);
5034         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
5035
5036         /* configure and enable PHY wakeup in PHY registers */
5037         e1e_wphy(&adapter->hw, BM_WUFC, wufc);
5038         e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
5039
5040         /* activate PHY wakeup */
5041         retval = hw->phy.ops.acquire(hw);
5042         if (retval) {
5043                 e_err("Could not acquire PHY\n");
5044                 return retval;
5045         }
5046         e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
5047                                  (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
5048         retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
5049         if (retval) {
5050                 e_err("Could not read PHY page 769\n");
5051                 goto out;
5052         }
5053         phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5054         retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
5055         if (retval)
5056                 e_err("Could not set PHY Host Wakeup bit\n");
5057 out:
5058         hw->phy.ops.release(hw);
5059
5060         return retval;
5061 }
5062
5063 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5064                             bool runtime)
5065 {
5066         struct net_device *netdev = pci_get_drvdata(pdev);
5067         struct e1000_adapter *adapter = netdev_priv(netdev);
5068         struct e1000_hw *hw = &adapter->hw;
5069         u32 ctrl, ctrl_ext, rctl, status;
5070         /* Runtime suspend should only enable wakeup for link changes */
5071         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5072         int retval = 0;
5073
5074         netif_device_detach(netdev);
5075
5076         if (netif_running(netdev)) {
5077                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5078                 e1000e_down(adapter);
5079                 e1000_free_irq(adapter);
5080         }
5081         e1000e_reset_interrupt_capability(adapter);
5082
5083         retval = pci_save_state(pdev);
5084         if (retval)
5085                 return retval;
5086
5087         status = er32(STATUS);
5088         if (status & E1000_STATUS_LU)
5089                 wufc &= ~E1000_WUFC_LNKC;
5090
5091         if (wufc) {
5092                 e1000_setup_rctl(adapter);
5093                 e1000_set_multi(netdev);
5094
5095                 /* turn on all-multi mode if wake on multicast is enabled */
5096                 if (wufc & E1000_WUFC_MC) {
5097                         rctl = er32(RCTL);
5098                         rctl |= E1000_RCTL_MPE;
5099                         ew32(RCTL, rctl);
5100                 }
5101
5102                 ctrl = er32(CTRL);
5103                 /* advertise wake from D3Cold */
5104                 #define E1000_CTRL_ADVD3WUC 0x00100000
5105                 /* phy power management enable */
5106                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5107                 ctrl |= E1000_CTRL_ADVD3WUC;
5108                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5109                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5110                 ew32(CTRL, ctrl);
5111
5112                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5113                     adapter->hw.phy.media_type ==
5114                     e1000_media_type_internal_serdes) {
5115                         /* keep the laser running in D3 */
5116                         ctrl_ext = er32(CTRL_EXT);
5117                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5118                         ew32(CTRL_EXT, ctrl_ext);
5119                 }
5120
5121                 if (adapter->flags & FLAG_IS_ICH)
5122                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
5123
5124                 /* Allow time for pending master requests to run */
5125                 e1000e_disable_pcie_master(&adapter->hw);
5126
5127                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5128                         /* enable wakeup by the PHY */
5129                         retval = e1000_init_phy_wakeup(adapter, wufc);
5130                         if (retval)
5131                                 return retval;
5132                 } else {
5133                         /* enable wakeup by the MAC */
5134                         ew32(WUFC, wufc);
5135                         ew32(WUC, E1000_WUC_PME_EN);
5136                 }
5137         } else {
5138                 ew32(WUC, 0);
5139                 ew32(WUFC, 0);
5140         }
5141
5142         *enable_wake = !!wufc;
5143
5144         /* make sure adapter isn't asleep if manageability is enabled */
5145         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5146             (hw->mac.ops.check_mng_mode(hw)))
5147                 *enable_wake = true;
5148
5149         if (adapter->hw.phy.type == e1000_phy_igp_3)
5150                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5151
5152         /*
5153          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5154          * would have already happened in close and is redundant.
5155          */
5156         e1000_release_hw_control(adapter);
5157
5158         pci_disable_device(pdev);
5159
5160         return 0;
5161 }
5162
5163 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5164 {
5165         if (sleep && wake) {
5166                 pci_prepare_to_sleep(pdev);
5167                 return;
5168         }
5169
5170         pci_wake_from_d3(pdev, wake);
5171         pci_set_power_state(pdev, PCI_D3hot);
5172 }
5173
5174 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5175                                     bool wake)
5176 {
5177         struct net_device *netdev = pci_get_drvdata(pdev);
5178         struct e1000_adapter *adapter = netdev_priv(netdev);
5179
5180         /*
5181          * The pci-e switch on some quad port adapters will report a
5182          * correctable error when the MAC transitions from D0 to D3.  To
5183          * prevent this we need to mask off the correctable errors on the
5184          * downstream port of the pci-e switch.
5185          */
5186         if (adapter->flags & FLAG_IS_QUAD_PORT) {
5187                 struct pci_dev *us_dev = pdev->bus->self;
5188                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
5189                 u16 devctl;
5190
5191                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5192                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5193                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
5194
5195                 e1000_power_off(pdev, sleep, wake);
5196
5197                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5198         } else {
5199                 e1000_power_off(pdev, sleep, wake);
5200         }
5201 }
5202
5203 #ifdef CONFIG_PCIEASPM
5204 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5205 {
5206         pci_disable_link_state(pdev, state);
5207 }
5208 #else
5209 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5210 {
5211         int pos;
5212         u16 reg16;
5213
5214         /*
5215          * Both device and parent should have the same ASPM setting.
5216          * Disable ASPM in downstream component first and then upstream.
5217          */
5218         pos = pci_pcie_cap(pdev);
5219         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5220         reg16 &= ~state;
5221         pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5222
5223         if (!pdev->bus->self)
5224                 return;
5225
5226         pos = pci_pcie_cap(pdev->bus->self);
5227         pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5228         reg16 &= ~state;
5229         pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5230 }
5231 #endif
5232 void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5233 {
5234         dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5235                  (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5236                  (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5237
5238         __e1000e_disable_aspm(pdev, state);
5239 }
5240
5241 #ifdef CONFIG_PM_OPS
5242 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5243 {
5244         return !!adapter->tx_ring->buffer_info;
5245 }
5246
5247 static int __e1000_resume(struct pci_dev *pdev)
5248 {
5249         struct net_device *netdev = pci_get_drvdata(pdev);
5250         struct e1000_adapter *adapter = netdev_priv(netdev);
5251         struct e1000_hw *hw = &adapter->hw;
5252         u32 err;
5253
5254         pci_set_power_state(pdev, PCI_D0);
5255         pci_restore_state(pdev);
5256         pci_save_state(pdev);
5257         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5258                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5259
5260         e1000e_set_interrupt_capability(adapter);
5261         if (netif_running(netdev)) {
5262                 err = e1000_request_irq(adapter);
5263                 if (err)
5264                         return err;
5265         }
5266
5267         e1000e_power_up_phy(adapter);
5268
5269         /* report the system wakeup cause from S3/S4 */
5270         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5271                 u16 phy_data;
5272
5273                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5274                 if (phy_data) {
5275                         e_info("PHY Wakeup cause - %s\n",
5276                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5277                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5278                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5279                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5280                                 phy_data & E1000_WUS_LNKC ? "Link Status "
5281                                 " Change" : "other");
5282                 }
5283                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5284         } else {
5285                 u32 wus = er32(WUS);
5286                 if (wus) {
5287                         e_info("MAC Wakeup cause - %s\n",
5288                                 wus & E1000_WUS_EX ? "Unicast Packet" :
5289                                 wus & E1000_WUS_MC ? "Multicast Packet" :
5290                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
5291                                 wus & E1000_WUS_MAG ? "Magic Packet" :
5292                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
5293                                 "other");
5294                 }
5295                 ew32(WUS, ~0);
5296         }
5297
5298         e1000e_reset(adapter);
5299
5300         e1000_init_manageability_pt(adapter);
5301
5302         if (netif_running(netdev))
5303                 e1000e_up(adapter);
5304
5305         netif_device_attach(netdev);
5306
5307         /*
5308          * If the controller has AMT, do not set DRV_LOAD until the interface
5309          * is up.  For all other cases, let the f/w know that the h/w is now
5310          * under the control of the driver.
5311          */
5312         if (!(adapter->flags & FLAG_HAS_AMT))
5313                 e1000_get_hw_control(adapter);
5314
5315         return 0;
5316 }
5317
5318 #ifdef CONFIG_PM_SLEEP
5319 static int e1000_suspend(struct device *dev)
5320 {
5321         struct pci_dev *pdev = to_pci_dev(dev);
5322         int retval;
5323         bool wake;
5324
5325         retval = __e1000_shutdown(pdev, &wake, false);
5326         if (!retval)
5327                 e1000_complete_shutdown(pdev, true, wake);
5328
5329         return retval;
5330 }
5331
5332 static int e1000_resume(struct device *dev)
5333 {
5334         struct pci_dev *pdev = to_pci_dev(dev);
5335         struct net_device *netdev = pci_get_drvdata(pdev);
5336         struct e1000_adapter *adapter = netdev_priv(netdev);
5337
5338         if (e1000e_pm_ready(adapter))
5339                 adapter->idle_check = true;
5340
5341         return __e1000_resume(pdev);
5342 }
5343 #endif /* CONFIG_PM_SLEEP */
5344
5345 #ifdef CONFIG_PM_RUNTIME
5346 static int e1000_runtime_suspend(struct device *dev)
5347 {
5348         struct pci_dev *pdev = to_pci_dev(dev);
5349         struct net_device *netdev = pci_get_drvdata(pdev);
5350         struct e1000_adapter *adapter = netdev_priv(netdev);
5351
5352         if (e1000e_pm_ready(adapter)) {
5353                 bool wake;
5354
5355                 __e1000_shutdown(pdev, &wake, true);
5356         }
5357
5358         return 0;
5359 }
5360
5361 static int e1000_idle(struct device *dev)
5362 {
5363         struct pci_dev *pdev = to_pci_dev(dev);
5364         struct net_device *netdev = pci_get_drvdata(pdev);
5365         struct e1000_adapter *adapter = netdev_priv(netdev);
5366
5367         if (!e1000e_pm_ready(adapter))
5368                 return 0;
5369
5370         if (adapter->idle_check) {
5371                 adapter->idle_check = false;
5372                 if (!e1000e_has_link(adapter))
5373                         pm_schedule_suspend(dev, MSEC_PER_SEC);
5374         }
5375
5376         return -EBUSY;
5377 }
5378
5379 static int e1000_runtime_resume(struct device *dev)
5380 {
5381         struct pci_dev *pdev = to_pci_dev(dev);
5382         struct net_device *netdev = pci_get_drvdata(pdev);
5383         struct e1000_adapter *adapter = netdev_priv(netdev);
5384
5385         if (!e1000e_pm_ready(adapter))
5386                 return 0;
5387
5388         adapter->idle_check = !dev->power.runtime_auto;
5389         return __e1000_resume(pdev);
5390 }
5391 #endif /* CONFIG_PM_RUNTIME */
5392 #endif /* CONFIG_PM_OPS */
5393
5394 static void e1000_shutdown(struct pci_dev *pdev)
5395 {
5396         bool wake = false;
5397
5398         __e1000_shutdown(pdev, &wake, false);
5399
5400         if (system_state == SYSTEM_POWER_OFF)
5401                 e1000_complete_shutdown(pdev, false, wake);
5402 }
5403
5404 #ifdef CONFIG_NET_POLL_CONTROLLER
5405 /*
5406  * Polling 'interrupt' - used by things like netconsole to send skbs
5407  * without having to re-enable interrupts. It's not called while
5408  * the interrupt routine is executing.
5409  */
5410 static void e1000_netpoll(struct net_device *netdev)
5411 {
5412         struct e1000_adapter *adapter = netdev_priv(netdev);
5413
5414         disable_irq(adapter->pdev->irq);
5415         e1000_intr(adapter->pdev->irq, netdev);
5416
5417         enable_irq(adapter->pdev->irq);
5418 }
5419 #endif
5420
5421 /**
5422  * e1000_io_error_detected - called when PCI error is detected
5423  * @pdev: Pointer to PCI device
5424  * @state: The current pci connection state
5425  *
5426  * This function is called after a PCI bus error affecting
5427  * this device has been detected.
5428  */
5429 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5430                                                 pci_channel_state_t state)
5431 {
5432         struct net_device *netdev = pci_get_drvdata(pdev);
5433         struct e1000_adapter *adapter = netdev_priv(netdev);
5434
5435         netif_device_detach(netdev);
5436
5437         if (state == pci_channel_io_perm_failure)
5438                 return PCI_ERS_RESULT_DISCONNECT;
5439
5440         if (netif_running(netdev))
5441                 e1000e_down(adapter);
5442         pci_disable_device(pdev);
5443
5444         /* Request a slot slot reset. */
5445         return PCI_ERS_RESULT_NEED_RESET;
5446 }
5447
5448 /**
5449  * e1000_io_slot_reset - called after the pci bus has been reset.
5450  * @pdev: Pointer to PCI device
5451  *
5452  * Restart the card from scratch, as if from a cold-boot. Implementation
5453  * resembles the first-half of the e1000_resume routine.
5454  */
5455 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5456 {
5457         struct net_device *netdev = pci_get_drvdata(pdev);
5458         struct e1000_adapter *adapter = netdev_priv(netdev);
5459         struct e1000_hw *hw = &adapter->hw;
5460         int err;
5461         pci_ers_result_t result;
5462
5463         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5464                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5465         err = pci_enable_device_mem(pdev);
5466         if (err) {
5467                 dev_err(&pdev->dev,
5468                         "Cannot re-enable PCI device after reset.\n");
5469                 result = PCI_ERS_RESULT_DISCONNECT;
5470         } else {
5471                 pci_set_master(pdev);
5472                 pdev->state_saved = true;
5473                 pci_restore_state(pdev);
5474
5475                 pci_enable_wake(pdev, PCI_D3hot, 0);
5476                 pci_enable_wake(pdev, PCI_D3cold, 0);
5477
5478                 e1000e_reset(adapter);
5479                 ew32(WUS, ~0);
5480                 result = PCI_ERS_RESULT_RECOVERED;
5481         }
5482
5483         pci_cleanup_aer_uncorrect_error_status(pdev);
5484
5485         return result;
5486 }
5487
5488 /**
5489  * e1000_io_resume - called when traffic can start flowing again.
5490  * @pdev: Pointer to PCI device
5491  *
5492  * This callback is called when the error recovery driver tells us that
5493  * its OK to resume normal operation. Implementation resembles the
5494  * second-half of the e1000_resume routine.
5495  */
5496 static void e1000_io_resume(struct pci_dev *pdev)
5497 {
5498         struct net_device *netdev = pci_get_drvdata(pdev);
5499         struct e1000_adapter *adapter = netdev_priv(netdev);
5500
5501         e1000_init_manageability_pt(adapter);
5502
5503         if (netif_running(netdev)) {
5504                 if (e1000e_up(adapter)) {
5505                         dev_err(&pdev->dev,
5506                                 "can't bring device back up after reset\n");
5507                         return;
5508                 }
5509         }
5510
5511         netif_device_attach(netdev);
5512
5513         /*
5514          * If the controller has AMT, do not set DRV_LOAD until the interface
5515          * is up.  For all other cases, let the f/w know that the h/w is now
5516          * under the control of the driver.
5517          */
5518         if (!(adapter->flags & FLAG_HAS_AMT))
5519                 e1000_get_hw_control(adapter);
5520
5521 }
5522
5523 static void e1000_print_device_info(struct e1000_adapter *adapter)
5524 {
5525         struct e1000_hw *hw = &adapter->hw;
5526         struct net_device *netdev = adapter->netdev;
5527         u32 pba_num;
5528
5529         /* print bus type/speed/width info */
5530         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
5531                /* bus width */
5532                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5533                 "Width x1"),
5534                /* MAC address */
5535                netdev->dev_addr);
5536         e_info("Intel(R) PRO/%s Network Connection\n",
5537                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5538         e1000e_read_pba_num(hw, &pba_num);
5539         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
5540                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
5541 }
5542
5543 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5544 {
5545         struct e1000_hw *hw = &adapter->hw;
5546         int ret_val;
5547         u16 buf = 0;
5548
5549         if (hw->mac.type != e1000_82573)
5550                 return;
5551
5552         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
5553         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
5554                 /* Deep Smart Power Down (DSPD) */
5555                 dev_warn(&adapter->pdev->dev,
5556                          "Warning: detected DSPD enabled in EEPROM\n");
5557         }
5558 }
5559
5560 static const struct net_device_ops e1000e_netdev_ops = {
5561         .ndo_open               = e1000_open,
5562         .ndo_stop               = e1000_close,
5563         .ndo_start_xmit         = e1000_xmit_frame,
5564         .ndo_get_stats          = e1000_get_stats,
5565         .ndo_set_multicast_list = e1000_set_multi,
5566         .ndo_set_mac_address    = e1000_set_mac,
5567         .ndo_change_mtu         = e1000_change_mtu,
5568         .ndo_do_ioctl           = e1000_ioctl,
5569         .ndo_tx_timeout         = e1000_tx_timeout,
5570         .ndo_validate_addr      = eth_validate_addr,
5571
5572         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
5573         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
5574         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
5575 #ifdef CONFIG_NET_POLL_CONTROLLER
5576         .ndo_poll_controller    = e1000_netpoll,
5577 #endif
5578 };
5579
5580 /**
5581  * e1000_probe - Device Initialization Routine
5582  * @pdev: PCI device information struct
5583  * @ent: entry in e1000_pci_tbl
5584  *
5585  * Returns 0 on success, negative on failure
5586  *
5587  * e1000_probe initializes an adapter identified by a pci_dev structure.
5588  * The OS initialization, configuring of the adapter private structure,
5589  * and a hardware reset occur.
5590  **/
5591 static int __devinit e1000_probe(struct pci_dev *pdev,
5592                                  const struct pci_device_id *ent)
5593 {
5594         struct net_device *netdev;
5595         struct e1000_adapter *adapter;
5596         struct e1000_hw *hw;
5597         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
5598         resource_size_t mmio_start, mmio_len;
5599         resource_size_t flash_start, flash_len;
5600
5601         static int cards_found;
5602         int i, err, pci_using_dac;
5603         u16 eeprom_data = 0;
5604         u16 eeprom_apme_mask = E1000_EEPROM_APME;
5605
5606         if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
5607                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5608
5609         err = pci_enable_device_mem(pdev);
5610         if (err)
5611                 return err;
5612
5613         pci_using_dac = 0;
5614         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
5615         if (!err) {
5616                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
5617                 if (!err)
5618                         pci_using_dac = 1;
5619         } else {
5620                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
5621                 if (err) {
5622                         err = dma_set_coherent_mask(&pdev->dev,
5623                                                     DMA_BIT_MASK(32));
5624                         if (err) {
5625                                 dev_err(&pdev->dev, "No usable DMA "
5626                                         "configuration, aborting\n");
5627                                 goto err_dma;
5628                         }
5629                 }
5630         }
5631
5632         err = pci_request_selected_regions_exclusive(pdev,
5633                                           pci_select_bars(pdev, IORESOURCE_MEM),
5634                                           e1000e_driver_name);
5635         if (err)
5636                 goto err_pci_reg;
5637
5638         /* AER (Advanced Error Reporting) hooks */
5639         pci_enable_pcie_error_reporting(pdev);
5640
5641         pci_set_master(pdev);
5642         /* PCI config space info */
5643         err = pci_save_state(pdev);
5644         if (err)
5645                 goto err_alloc_etherdev;
5646
5647         err = -ENOMEM;
5648         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
5649         if (!netdev)
5650                 goto err_alloc_etherdev;
5651
5652         SET_NETDEV_DEV(netdev, &pdev->dev);
5653
5654         netdev->irq = pdev->irq;
5655
5656         pci_set_drvdata(pdev, netdev);
5657         adapter = netdev_priv(netdev);
5658         hw = &adapter->hw;
5659         adapter->netdev = netdev;
5660         adapter->pdev = pdev;
5661         adapter->ei = ei;
5662         adapter->pba = ei->pba;
5663         adapter->flags = ei->flags;
5664         adapter->flags2 = ei->flags2;
5665         adapter->hw.adapter = adapter;
5666         adapter->hw.mac.type = ei->mac;
5667         adapter->max_hw_frame_size = ei->max_hw_frame_size;
5668         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5669
5670         mmio_start = pci_resource_start(pdev, 0);
5671         mmio_len = pci_resource_len(pdev, 0);
5672
5673         err = -EIO;
5674         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5675         if (!adapter->hw.hw_addr)
5676                 goto err_ioremap;
5677
5678         if ((adapter->flags & FLAG_HAS_FLASH) &&
5679             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5680                 flash_start = pci_resource_start(pdev, 1);
5681                 flash_len = pci_resource_len(pdev, 1);
5682                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5683                 if (!adapter->hw.flash_address)
5684                         goto err_flashmap;
5685         }
5686
5687         /* construct the net_device struct */
5688         netdev->netdev_ops              = &e1000e_netdev_ops;
5689         e1000e_set_ethtool_ops(netdev);
5690         netdev->watchdog_timeo          = 5 * HZ;
5691         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5692         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5693
5694         netdev->mem_start = mmio_start;
5695         netdev->mem_end = mmio_start + mmio_len;
5696
5697         adapter->bd_number = cards_found++;
5698
5699         e1000e_check_options(adapter);
5700
5701         /* setup adapter struct */
5702         err = e1000_sw_init(adapter);
5703         if (err)
5704                 goto err_sw_init;
5705
5706         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5707         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5708         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5709
5710         err = ei->get_variants(adapter);
5711         if (err)
5712                 goto err_hw_init;
5713
5714         if ((adapter->flags & FLAG_IS_ICH) &&
5715             (adapter->flags & FLAG_READ_ONLY_NVM))
5716                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5717
5718         hw->mac.ops.get_bus_info(&adapter->hw);
5719
5720         adapter->hw.phy.autoneg_wait_to_complete = 0;
5721
5722         /* Copper options */
5723         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5724                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5725                 adapter->hw.phy.disable_polarity_correction = 0;
5726                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5727         }
5728
5729         if (e1000_check_reset_block(&adapter->hw))
5730                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5731
5732         netdev->features = NETIF_F_SG |
5733                            NETIF_F_HW_CSUM |
5734                            NETIF_F_HW_VLAN_TX |
5735                            NETIF_F_HW_VLAN_RX;
5736
5737         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5738                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5739
5740         netdev->features |= NETIF_F_TSO;
5741         netdev->features |= NETIF_F_TSO6;
5742
5743         netdev->vlan_features |= NETIF_F_TSO;
5744         netdev->vlan_features |= NETIF_F_TSO6;
5745         netdev->vlan_features |= NETIF_F_HW_CSUM;
5746         netdev->vlan_features |= NETIF_F_SG;
5747
5748         if (pci_using_dac) {
5749                 netdev->features |= NETIF_F_HIGHDMA;
5750                 netdev->vlan_features |= NETIF_F_HIGHDMA;
5751         }
5752
5753         if (e1000e_enable_mng_pass_thru(&adapter->hw))
5754                 adapter->flags |= FLAG_MNG_PT_ENABLED;
5755
5756         /*
5757          * before reading the NVM, reset the controller to
5758          * put the device in a known good starting state
5759          */
5760         adapter->hw.mac.ops.reset_hw(&adapter->hw);
5761
5762         /*
5763          * systems with ASPM and others may see the checksum fail on the first
5764          * attempt. Let's give it a few tries
5765          */
5766         for (i = 0;; i++) {
5767                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5768                         break;
5769                 if (i == 2) {
5770                         e_err("The NVM Checksum Is Not Valid\n");
5771                         err = -EIO;
5772                         goto err_eeprom;
5773                 }
5774         }
5775
5776         e1000_eeprom_checks(adapter);
5777
5778         /* copy the MAC address */
5779         if (e1000e_read_mac_addr(&adapter->hw))
5780                 e_err("NVM Read Error while reading MAC address\n");
5781
5782         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5783         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5784
5785         if (!is_valid_ether_addr(netdev->perm_addr)) {
5786                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5787                 err = -EIO;
5788                 goto err_eeprom;
5789         }
5790
5791         init_timer(&adapter->watchdog_timer);
5792         adapter->watchdog_timer.function = e1000_watchdog;
5793         adapter->watchdog_timer.data = (unsigned long) adapter;
5794
5795         init_timer(&adapter->phy_info_timer);
5796         adapter->phy_info_timer.function = e1000_update_phy_info;
5797         adapter->phy_info_timer.data = (unsigned long) adapter;
5798
5799         INIT_WORK(&adapter->reset_task, e1000_reset_task);
5800         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
5801         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
5802         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
5803         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
5804
5805         /* Initialize link parameters. User can change them with ethtool */
5806         adapter->hw.mac.autoneg = 1;
5807         adapter->fc_autoneg = 1;
5808         adapter->hw.fc.requested_mode = e1000_fc_default;
5809         adapter->hw.fc.current_mode = e1000_fc_default;
5810         adapter->hw.phy.autoneg_advertised = 0x2f;
5811
5812         /* ring size defaults */
5813         adapter->rx_ring->count = 256;
5814         adapter->tx_ring->count = 256;
5815
5816         /*
5817          * Initial Wake on LAN setting - If APM wake is enabled in
5818          * the EEPROM, enable the ACPI Magic Packet filter
5819          */
5820         if (adapter->flags & FLAG_APME_IN_WUC) {
5821                 /* APME bit in EEPROM is mapped to WUC.APME */
5822                 eeprom_data = er32(WUC);
5823                 eeprom_apme_mask = E1000_WUC_APME;
5824                 if (eeprom_data & E1000_WUC_PHY_WAKE)
5825                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
5826         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5827                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5828                     (adapter->hw.bus.func == 1))
5829                         e1000_read_nvm(&adapter->hw,
5830                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5831                 else
5832                         e1000_read_nvm(&adapter->hw,
5833                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5834         }
5835
5836         /* fetch WoL from EEPROM */
5837         if (eeprom_data & eeprom_apme_mask)
5838                 adapter->eeprom_wol |= E1000_WUFC_MAG;
5839
5840         /*
5841          * now that we have the eeprom settings, apply the special cases
5842          * where the eeprom may be wrong or the board simply won't support
5843          * wake on lan on a particular port
5844          */
5845         if (!(adapter->flags & FLAG_HAS_WOL))
5846                 adapter->eeprom_wol = 0;
5847
5848         /* initialize the wol settings based on the eeprom settings */
5849         adapter->wol = adapter->eeprom_wol;
5850         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5851
5852         /* save off EEPROM version number */
5853         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
5854
5855         /* reset the hardware with the new settings */
5856         e1000e_reset(adapter);
5857
5858         /*
5859          * If the controller has AMT, do not set DRV_LOAD until the interface
5860          * is up.  For all other cases, let the f/w know that the h/w is now
5861          * under the control of the driver.
5862          */
5863         if (!(adapter->flags & FLAG_HAS_AMT))
5864                 e1000_get_hw_control(adapter);
5865
5866         strcpy(netdev->name, "eth%d");
5867         err = register_netdev(netdev);
5868         if (err)
5869                 goto err_register;
5870
5871         /* carrier off reporting is important to ethtool even BEFORE open */
5872         netif_carrier_off(netdev);
5873
5874         e1000_print_device_info(adapter);
5875
5876         if (pci_dev_run_wake(pdev))
5877                 pm_runtime_put_noidle(&pdev->dev);
5878
5879         return 0;
5880
5881 err_register:
5882         if (!(adapter->flags & FLAG_HAS_AMT))
5883                 e1000_release_hw_control(adapter);
5884 err_eeprom:
5885         if (!e1000_check_reset_block(&adapter->hw))
5886                 e1000_phy_hw_reset(&adapter->hw);
5887 err_hw_init:
5888
5889         kfree(adapter->tx_ring);
5890         kfree(adapter->rx_ring);
5891 err_sw_init:
5892         if (adapter->hw.flash_address)
5893                 iounmap(adapter->hw.flash_address);
5894         e1000e_reset_interrupt_capability(adapter);
5895 err_flashmap:
5896         iounmap(adapter->hw.hw_addr);
5897 err_ioremap:
5898         free_netdev(netdev);
5899 err_alloc_etherdev:
5900         pci_release_selected_regions(pdev,
5901                                      pci_select_bars(pdev, IORESOURCE_MEM));
5902 err_pci_reg:
5903 err_dma:
5904         pci_disable_device(pdev);
5905         return err;
5906 }
5907
5908 /**
5909  * e1000_remove - Device Removal Routine
5910  * @pdev: PCI device information struct
5911  *
5912  * e1000_remove is called by the PCI subsystem to alert the driver
5913  * that it should release a PCI device.  The could be caused by a
5914  * Hot-Plug event, or because the driver is going to be removed from
5915  * memory.
5916  **/
5917 static void __devexit e1000_remove(struct pci_dev *pdev)
5918 {
5919         struct net_device *netdev = pci_get_drvdata(pdev);
5920         struct e1000_adapter *adapter = netdev_priv(netdev);
5921         bool down = test_bit(__E1000_DOWN, &adapter->state);
5922
5923         /*
5924          * flush_scheduled work may reschedule our watchdog task, so
5925          * explicitly disable watchdog tasks from being rescheduled
5926          */
5927         if (!down)
5928                 set_bit(__E1000_DOWN, &adapter->state);
5929         del_timer_sync(&adapter->watchdog_timer);
5930         del_timer_sync(&adapter->phy_info_timer);
5931
5932         cancel_work_sync(&adapter->reset_task);
5933         cancel_work_sync(&adapter->watchdog_task);
5934         cancel_work_sync(&adapter->downshift_task);
5935         cancel_work_sync(&adapter->update_phy_task);
5936         cancel_work_sync(&adapter->print_hang_task);
5937         flush_scheduled_work();
5938
5939         if (!(netdev->flags & IFF_UP))
5940                 e1000_power_down_phy(adapter);
5941
5942         /* Don't lie to e1000_close() down the road. */
5943         if (!down)
5944                 clear_bit(__E1000_DOWN, &adapter->state);
5945         unregister_netdev(netdev);
5946
5947         if (pci_dev_run_wake(pdev))
5948                 pm_runtime_get_noresume(&pdev->dev);
5949
5950         /*
5951          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5952          * would have already happened in close and is redundant.
5953          */
5954         e1000_release_hw_control(adapter);
5955
5956         e1000e_reset_interrupt_capability(adapter);
5957         kfree(adapter->tx_ring);
5958         kfree(adapter->rx_ring);
5959
5960         iounmap(adapter->hw.hw_addr);
5961         if (adapter->hw.flash_address)
5962                 iounmap(adapter->hw.flash_address);
5963         pci_release_selected_regions(pdev,
5964                                      pci_select_bars(pdev, IORESOURCE_MEM));
5965
5966         free_netdev(netdev);
5967
5968         /* AER disable */
5969         pci_disable_pcie_error_reporting(pdev);
5970
5971         pci_disable_device(pdev);
5972 }
5973
5974 /* PCI Error Recovery (ERS) */
5975 static struct pci_error_handlers e1000_err_handler = {
5976         .error_detected = e1000_io_error_detected,
5977         .slot_reset = e1000_io_slot_reset,
5978         .resume = e1000_io_resume,
5979 };
5980
5981 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
5982         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
5983         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
5984         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
5985         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
5986         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
5987         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
5988         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
5989         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
5990         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
5991
5992         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
5993         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
5994         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
5995         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
5996
5997         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
5998         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
5999         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
6000
6001         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
6002         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
6003         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
6004
6005         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
6006           board_80003es2lan },
6007         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
6008           board_80003es2lan },
6009         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
6010           board_80003es2lan },
6011         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
6012           board_80003es2lan },
6013
6014         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
6015         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
6016         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
6017         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
6018         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
6019         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
6020         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
6021         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
6022
6023         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
6024         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
6025         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
6026         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
6027         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
6028         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
6029         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
6030         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
6031         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
6032
6033         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
6034         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
6035         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
6036
6037         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
6038         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
6039         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
6040
6041         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
6042         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
6043         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
6044         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
6045
6046         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
6047         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
6048
6049         { }     /* terminate list */
6050 };
6051 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
6052
6053 #ifdef CONFIG_PM_OPS
6054 static const struct dev_pm_ops e1000_pm_ops = {
6055         SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
6056         SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
6057                                 e1000_runtime_resume, e1000_idle)
6058 };
6059 #endif
6060
6061 /* PCI Device API Driver */
6062 static struct pci_driver e1000_driver = {
6063         .name     = e1000e_driver_name,
6064         .id_table = e1000_pci_tbl,
6065         .probe    = e1000_probe,
6066         .remove   = __devexit_p(e1000_remove),
6067 #ifdef CONFIG_PM_OPS
6068         .driver.pm = &e1000_pm_ops,
6069 #endif
6070         .shutdown = e1000_shutdown,
6071         .err_handler = &e1000_err_handler
6072 };
6073
6074 /**
6075  * e1000_init_module - Driver Registration Routine
6076  *
6077  * e1000_init_module is the first routine called when the driver is
6078  * loaded. All it does is register with the PCI subsystem.
6079  **/
6080 static int __init e1000_init_module(void)
6081 {
6082         int ret;
6083         pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
6084                 e1000e_driver_version);
6085         pr_info("Copyright (c) 1999 - 2010 Intel Corporation.\n");
6086         ret = pci_register_driver(&e1000_driver);
6087
6088         return ret;
6089 }
6090 module_init(e1000_init_module);
6091
6092 /**
6093  * e1000_exit_module - Driver Exit Cleanup Routine
6094  *
6095  * e1000_exit_module is called just before the driver is removed
6096  * from memory.
6097  **/
6098 static void __exit e1000_exit_module(void)
6099 {
6100         pci_unregister_driver(&e1000_driver);
6101 }
6102 module_exit(e1000_exit_module);
6103
6104
6105 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
6106 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
6107 MODULE_LICENSE("GPL");
6108 MODULE_VERSION(DRV_VERSION);
6109
6110 /* e1000_main.c */