Pileus Git - ~andy/linux/blob - 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 "1.2.20" 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                         /* MSI-X failed, so fall through and try MSI */
1813                         e_err("Failed to initialize MSI-X interrupts.  "
1814                               "Falling back to MSI interrupts.\n");
1815                         e1000e_reset_interrupt_capability(adapter);
1816                 }
1817                 adapter->int_mode = E1000E_INT_MODE_MSI;
1818                 /* Fall through */
1819         case E1000E_INT_MODE_MSI:
1820                 if (!pci_enable_msi(adapter->pdev)) {
1821                         adapter->flags |= FLAG_MSI_ENABLED;
1822                 } else {
1823                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1824                         e_err("Failed to initialize MSI interrupts.  Falling "
1825                               "back to legacy interrupts.\n");
1826                 }
1827                 /* Fall through */
1828         case E1000E_INT_MODE_LEGACY:
1829                 /* Don't do anything; this is the system default */
1830                 break;
1831         }
1832
1833         /* store the number of vectors being used */
1834         adapter->num_vectors = 1;
1835 }
1836
1837 /**
1838  * e1000_request_msix - Initialize MSI-X interrupts
1839  *
1840  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1841  * kernel.
1842  **/
1843 static int e1000_request_msix(struct e1000_adapter *adapter)
1844 {
1845         struct net_device *netdev = adapter->netdev;
1846         int err = 0, vector = 0;
1847
1848         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1849                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1850         else
1851                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1852         err = request_irq(adapter->msix_entries[vector].vector,
1853                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1854                           netdev);
1855         if (err)
1856                 goto out;
1857         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1858         adapter->rx_ring->itr_val = adapter->itr;
1859         vector++;
1860
1861         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1862                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1863         else
1864                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1865         err = request_irq(adapter->msix_entries[vector].vector,
1866                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1867                           netdev);
1868         if (err)
1869                 goto out;
1870         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1871         adapter->tx_ring->itr_val = adapter->itr;
1872         vector++;
1873
1874         err = request_irq(adapter->msix_entries[vector].vector,
1875                           e1000_msix_other, 0, netdev->name, netdev);
1876         if (err)
1877                 goto out;
1878
1879         e1000_configure_msix(adapter);
1880         return 0;
1881 out:
1882         return err;
1883 }
1884
1885 /**
1886  * e1000_request_irq - initialize interrupts
1887  *
1888  * Attempts to configure interrupts using the best available
1889  * capabilities of the hardware and kernel.
1890  **/
1891 static int e1000_request_irq(struct e1000_adapter *adapter)
1892 {
1893         struct net_device *netdev = adapter->netdev;
1894         int err;
1895
1896         if (adapter->msix_entries) {
1897                 err = e1000_request_msix(adapter);
1898                 if (!err)
1899                         return err;
1900                 /* fall back to MSI */
1901                 e1000e_reset_interrupt_capability(adapter);
1902                 adapter->int_mode = E1000E_INT_MODE_MSI;
1903                 e1000e_set_interrupt_capability(adapter);
1904         }
1905         if (adapter->flags & FLAG_MSI_ENABLED) {
1906                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1907                                   netdev->name, netdev);
1908                 if (!err)
1909                         return err;
1910
1911                 /* fall back to legacy interrupt */
1912                 e1000e_reset_interrupt_capability(adapter);
1913                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1914         }
1915
1916         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1917                           netdev->name, netdev);
1918         if (err)
1919                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1920
1921         return err;
1922 }
1923
1924 static void e1000_free_irq(struct e1000_adapter *adapter)
1925 {
1926         struct net_device *netdev = adapter->netdev;
1927
1928         if (adapter->msix_entries) {
1929                 int vector = 0;
1930
1931                 free_irq(adapter->msix_entries[vector].vector, netdev);
1932                 vector++;
1933
1934                 free_irq(adapter->msix_entries[vector].vector, netdev);
1935                 vector++;
1936
1937                 /* Other Causes interrupt vector */
1938                 free_irq(adapter->msix_entries[vector].vector, netdev);
1939                 return;
1940         }
1941
1942         free_irq(adapter->pdev->irq, netdev);
1943 }
1944
1945 /**
1946  * e1000_irq_disable - Mask off interrupt generation on the NIC
1947  **/
1948 static void e1000_irq_disable(struct e1000_adapter *adapter)
1949 {
1950         struct e1000_hw *hw = &adapter->hw;
1951
1952         ew32(IMC, ~0);
1953         if (adapter->msix_entries)
1954                 ew32(EIAC_82574, 0);
1955         e1e_flush();
1956
1957         if (adapter->msix_entries) {
1958                 int i;
1959                 for (i = 0; i < adapter->num_vectors; i++)
1960                         synchronize_irq(adapter->msix_entries[i].vector);
1961         } else {
1962                 synchronize_irq(adapter->pdev->irq);
1963         }
1964 }
1965
1966 /**
1967  * e1000_irq_enable - Enable default interrupt generation settings
1968  **/
1969 static void e1000_irq_enable(struct e1000_adapter *adapter)
1970 {
1971         struct e1000_hw *hw = &adapter->hw;
1972
1973         if (adapter->msix_entries) {
1974                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1975                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1976         } else {
1977                 ew32(IMS, IMS_ENABLE_MASK);
1978         }
1979         e1e_flush();
1980 }
1981
1982 /**
1983  * e1000_get_hw_control - get control of the h/w from f/w
1984  * @adapter: address of board private structure
1985  *
1986  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1987  * For ASF and Pass Through versions of f/w this means that
1988  * the driver is loaded. For AMT version (only with 82573)
1989  * of the f/w this means that the network i/f is open.
1990  **/
1991 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1992 {
1993         struct e1000_hw *hw = &adapter->hw;
1994         u32 ctrl_ext;
1995         u32 swsm;
1996
1997         /* Let firmware know the driver has taken over */
1998         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1999                 swsm = er32(SWSM);
2000                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2001         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2002                 ctrl_ext = er32(CTRL_EXT);
2003                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2004         }
2005 }
2006
2007 /**
2008  * e1000_release_hw_control - release control of the h/w to f/w
2009  * @adapter: address of board private structure
2010  *
2011  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2012  * For ASF and Pass Through versions of f/w this means that the
2013  * driver is no longer loaded. For AMT version (only with 82573) i
2014  * of the f/w this means that the network i/f is closed.
2015  *
2016  **/
2017 static void e1000_release_hw_control(struct e1000_adapter *adapter)
2018 {
2019         struct e1000_hw *hw = &adapter->hw;
2020         u32 ctrl_ext;
2021         u32 swsm;
2022
2023         /* Let firmware taken over control of h/w */
2024         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2025                 swsm = er32(SWSM);
2026                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2027         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2028                 ctrl_ext = er32(CTRL_EXT);
2029                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2030         }
2031 }
2032
2033 /**
2034  * @e1000_alloc_ring - allocate memory for a ring structure
2035  **/
2036 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2037                                 struct e1000_ring *ring)
2038 {
2039         struct pci_dev *pdev = adapter->pdev;
2040
2041         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2042                                         GFP_KERNEL);
2043         if (!ring->desc)
2044                 return -ENOMEM;
2045
2046         return 0;
2047 }
2048
2049 /**
2050  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2051  * @adapter: board private structure
2052  *
2053  * Return 0 on success, negative on failure
2054  **/
2055 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
2056 {
2057         struct e1000_ring *tx_ring = adapter->tx_ring;
2058         int err = -ENOMEM, size;
2059
2060         size = sizeof(struct e1000_buffer) * tx_ring->count;
2061         tx_ring->buffer_info = vzalloc(size);
2062         if (!tx_ring->buffer_info)
2063                 goto err;
2064
2065         /* round up to nearest 4K */
2066         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2067         tx_ring->size = ALIGN(tx_ring->size, 4096);
2068
2069         err = e1000_alloc_ring_dma(adapter, tx_ring);
2070         if (err)
2071                 goto err;
2072
2073         tx_ring->next_to_use = 0;
2074         tx_ring->next_to_clean = 0;
2075
2076         return 0;
2077 err:
2078         vfree(tx_ring->buffer_info);
2079         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2080         return err;
2081 }
2082
2083 /**
2084  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2085  * @adapter: board private structure
2086  *
2087  * Returns 0 on success, negative on failure
2088  **/
2089 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
2090 {
2091         struct e1000_ring *rx_ring = adapter->rx_ring;
2092         struct e1000_buffer *buffer_info;
2093         int i, size, desc_len, err = -ENOMEM;
2094
2095         size = sizeof(struct e1000_buffer) * rx_ring->count;
2096         rx_ring->buffer_info = vzalloc(size);
2097         if (!rx_ring->buffer_info)
2098                 goto err;
2099
2100         for (i = 0; i < rx_ring->count; i++) {
2101                 buffer_info = &rx_ring->buffer_info[i];
2102                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2103                                                 sizeof(struct e1000_ps_page),
2104                                                 GFP_KERNEL);
2105                 if (!buffer_info->ps_pages)
2106                         goto err_pages;
2107         }
2108
2109         desc_len = sizeof(union e1000_rx_desc_packet_split);
2110
2111         /* Round up to nearest 4K */
2112         rx_ring->size = rx_ring->count * desc_len;
2113         rx_ring->size = ALIGN(rx_ring->size, 4096);
2114
2115         err = e1000_alloc_ring_dma(adapter, rx_ring);
2116         if (err)
2117                 goto err_pages;
2118
2119         rx_ring->next_to_clean = 0;
2120         rx_ring->next_to_use = 0;
2121         rx_ring->rx_skb_top = NULL;
2122
2123         return 0;
2124
2125 err_pages:
2126         for (i = 0; i < rx_ring->count; i++) {
2127                 buffer_info = &rx_ring->buffer_info[i];
2128                 kfree(buffer_info->ps_pages);
2129         }
2130 err:
2131         vfree(rx_ring->buffer_info);
2132         e_err("Unable to allocate memory for the receive descriptor ring\n");
2133         return err;
2134 }
2135
2136 /**
2137  * e1000_clean_tx_ring - Free Tx Buffers
2138  * @adapter: board private structure
2139  **/
2140 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
2141 {
2142         struct e1000_ring *tx_ring = adapter->tx_ring;
2143         struct e1000_buffer *buffer_info;
2144         unsigned long size;
2145         unsigned int i;
2146
2147         for (i = 0; i < tx_ring->count; i++) {
2148                 buffer_info = &tx_ring->buffer_info[i];
2149                 e1000_put_txbuf(adapter, buffer_info);
2150         }
2151
2152         size = sizeof(struct e1000_buffer) * tx_ring->count;
2153         memset(tx_ring->buffer_info, 0, size);
2154
2155         memset(tx_ring->desc, 0, tx_ring->size);
2156
2157         tx_ring->next_to_use = 0;
2158         tx_ring->next_to_clean = 0;
2159
2160         writel(0, adapter->hw.hw_addr + tx_ring->head);
2161         writel(0, adapter->hw.hw_addr + tx_ring->tail);
2162 }
2163
2164 /**
2165  * e1000e_free_tx_resources - Free Tx Resources per Queue
2166  * @adapter: board private structure
2167  *
2168  * Free all transmit software resources
2169  **/
2170 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
2171 {
2172         struct pci_dev *pdev = adapter->pdev;
2173         struct e1000_ring *tx_ring = adapter->tx_ring;
2174
2175         e1000_clean_tx_ring(adapter);
2176
2177         vfree(tx_ring->buffer_info);
2178         tx_ring->buffer_info = NULL;
2179
2180         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2181                           tx_ring->dma);
2182         tx_ring->desc = NULL;
2183 }
2184
2185 /**
2186  * e1000e_free_rx_resources - Free Rx Resources
2187  * @adapter: board private structure
2188  *
2189  * Free all receive software resources
2190  **/
2191
2192 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
2193 {
2194         struct pci_dev *pdev = adapter->pdev;
2195         struct e1000_ring *rx_ring = adapter->rx_ring;
2196         int i;
2197
2198         e1000_clean_rx_ring(adapter);
2199
2200         for (i = 0; i < rx_ring->count; i++)
2201                 kfree(rx_ring->buffer_info[i].ps_pages);
2202
2203         vfree(rx_ring->buffer_info);
2204         rx_ring->buffer_info = NULL;
2205
2206         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2207                           rx_ring->dma);
2208         rx_ring->desc = NULL;
2209 }
2210
2211 /**
2212  * e1000_update_itr - update the dynamic ITR value based on statistics
2213  * @adapter: pointer to adapter
2214  * @itr_setting: current adapter->itr
2215  * @packets: the number of packets during this measurement interval
2216  * @bytes: the number of bytes during this measurement interval
2217  *
2218  *      Stores a new ITR value based on packets and byte
2219  *      counts during the last interrupt.  The advantage of per interrupt
2220  *      computation is faster updates and more accurate ITR for the current
2221  *      traffic pattern.  Constants in this function were computed
2222  *      based on theoretical maximum wire speed and thresholds were set based
2223  *      on testing data as well as attempting to minimize response time
2224  *      while increasing bulk throughput.  This functionality is controlled
2225  *      by the InterruptThrottleRate module parameter.
2226  **/
2227 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2228                                      u16 itr_setting, int packets,
2229                                      int bytes)
2230 {
2231         unsigned int retval = itr_setting;
2232
2233         if (packets == 0)
2234                 goto update_itr_done;
2235
2236         switch (itr_setting) {
2237         case lowest_latency:
2238                 /* handle TSO and jumbo frames */
2239                 if (bytes/packets > 8000)
2240                         retval = bulk_latency;
2241                 else if ((packets < 5) && (bytes > 512))
2242                         retval = low_latency;
2243                 break;
2244         case low_latency:  /* 50 usec aka 20000 ints/s */
2245                 if (bytes > 10000) {
2246                         /* this if handles the TSO accounting */
2247                         if (bytes/packets > 8000)
2248                                 retval = bulk_latency;
2249                         else if ((packets < 10) || ((bytes/packets) > 1200))
2250                                 retval = bulk_latency;
2251                         else if ((packets > 35))
2252                                 retval = lowest_latency;
2253                 } else if (bytes/packets > 2000) {
2254                         retval = bulk_latency;
2255                 } else if (packets <= 2 && bytes < 512) {
2256                         retval = lowest_latency;
2257                 }
2258                 break;
2259         case bulk_latency: /* 250 usec aka 4000 ints/s */
2260                 if (bytes > 25000) {
2261                         if (packets > 35)
2262                                 retval = low_latency;
2263                 } else if (bytes < 6000) {
2264                         retval = low_latency;
2265                 }
2266                 break;
2267         }
2268
2269 update_itr_done:
2270         return retval;
2271 }
2272
2273 static void e1000_set_itr(struct e1000_adapter *adapter)
2274 {
2275         struct e1000_hw *hw = &adapter->hw;
2276         u16 current_itr;
2277         u32 new_itr = adapter->itr;
2278
2279         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2280         if (adapter->link_speed != SPEED_1000) {
2281                 current_itr = 0;
2282                 new_itr = 4000;
2283                 goto set_itr_now;
2284         }
2285
2286         if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2287                 new_itr = 0;
2288                 goto set_itr_now;
2289         }
2290
2291         adapter->tx_itr = e1000_update_itr(adapter,
2292                                     adapter->tx_itr,
2293                                     adapter->total_tx_packets,
2294                                     adapter->total_tx_bytes);
2295         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2296         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2297                 adapter->tx_itr = low_latency;
2298
2299         adapter->rx_itr = e1000_update_itr(adapter,
2300                                     adapter->rx_itr,
2301                                     adapter->total_rx_packets,
2302                                     adapter->total_rx_bytes);
2303         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2304         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2305                 adapter->rx_itr = low_latency;
2306
2307         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2308
2309         switch (current_itr) {
2310         /* counts and packets in update_itr are dependent on these numbers */
2311         case lowest_latency:
2312                 new_itr = 70000;
2313                 break;
2314         case low_latency:
2315                 new_itr = 20000; /* aka hwitr = ~200 */
2316                 break;
2317         case bulk_latency:
2318                 new_itr = 4000;
2319                 break;
2320         default:
2321                 break;
2322         }
2323
2324 set_itr_now:
2325         if (new_itr != adapter->itr) {
2326                 /*
2327                  * this attempts to bias the interrupt rate towards Bulk
2328                  * by adding intermediate steps when interrupt rate is
2329                  * increasing
2330                  */
2331                 new_itr = new_itr > adapter->itr ?
2332                              min(adapter->itr + (new_itr >> 2), new_itr) :
2333                              new_itr;
2334                 adapter->itr = new_itr;
2335                 adapter->rx_ring->itr_val = new_itr;
2336                 if (adapter->msix_entries)
2337                         adapter->rx_ring->set_itr = 1;
2338                 else
2339                         if (new_itr)
2340                                 ew32(ITR, 1000000000 / (new_itr * 256));
2341                         else
2342                                 ew32(ITR, 0);
2343         }
2344 }
2345
2346 /**
2347  * e1000_alloc_queues - Allocate memory for all rings
2348  * @adapter: board private structure to initialize
2349  **/
2350 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2351 {
2352         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2353         if (!adapter->tx_ring)
2354                 goto err;
2355
2356         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2357         if (!adapter->rx_ring)
2358                 goto err;
2359
2360         return 0;
2361 err:
2362         e_err("Unable to allocate memory for queues\n");
2363         kfree(adapter->rx_ring);
2364         kfree(adapter->tx_ring);
2365         return -ENOMEM;
2366 }
2367
2368 /**
2369  * e1000_clean - NAPI Rx polling callback
2370  * @napi: struct associated with this polling callback
2371  * @budget: amount of packets driver is allowed to process this poll
2372  **/
2373 static int e1000_clean(struct napi_struct *napi, int budget)
2374 {
2375         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2376         struct e1000_hw *hw = &adapter->hw;
2377         struct net_device *poll_dev = adapter->netdev;
2378         int tx_cleaned = 1, work_done = 0;
2379
2380         adapter = netdev_priv(poll_dev);
2381
2382         if (adapter->msix_entries &&
2383             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2384                 goto clean_rx;
2385
2386         tx_cleaned = e1000_clean_tx_irq(adapter);
2387
2388 clean_rx:
2389         adapter->clean_rx(adapter, &work_done, budget);
2390
2391         if (!tx_cleaned)
2392                 work_done = budget;
2393
2394         /* If budget not fully consumed, exit the polling mode */
2395         if (work_done < budget) {
2396                 if (adapter->itr_setting & 3)
2397                         e1000_set_itr(adapter);
2398                 napi_complete(napi);
2399                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2400                         if (adapter->msix_entries)
2401                                 ew32(IMS, adapter->rx_ring->ims_val);
2402                         else
2403                                 e1000_irq_enable(adapter);
2404                 }
2405         }
2406
2407         return work_done;
2408 }
2409
2410 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2411 {
2412         struct e1000_adapter *adapter = netdev_priv(netdev);
2413         struct e1000_hw *hw = &adapter->hw;
2414         u32 vfta, index;
2415
2416         /* don't update vlan cookie if already programmed */
2417         if ((adapter->hw.mng_cookie.status &
2418              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2419             (vid == adapter->mng_vlan_id))
2420                 return;
2421
2422         /* add VID to filter table */
2423         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2424                 index = (vid >> 5) & 0x7F;
2425                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2426                 vfta |= (1 << (vid & 0x1F));
2427                 hw->mac.ops.write_vfta(hw, index, vfta);
2428         }
2429 }
2430
2431 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2432 {
2433         struct e1000_adapter *adapter = netdev_priv(netdev);
2434         struct e1000_hw *hw = &adapter->hw;
2435         u32 vfta, index;
2436
2437         if (!test_bit(__E1000_DOWN, &adapter->state))
2438                 e1000_irq_disable(adapter);
2439         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2440
2441         if (!test_bit(__E1000_DOWN, &adapter->state))
2442                 e1000_irq_enable(adapter);
2443
2444         if ((adapter->hw.mng_cookie.status &
2445              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2446             (vid == adapter->mng_vlan_id)) {
2447                 /* release control to f/w */
2448                 e1000_release_hw_control(adapter);
2449                 return;
2450         }
2451
2452         /* remove VID from filter table */
2453         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2454                 index = (vid >> 5) & 0x7F;
2455                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2456                 vfta &= ~(1 << (vid & 0x1F));
2457                 hw->mac.ops.write_vfta(hw, index, vfta);
2458         }
2459 }
2460
2461 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2462 {
2463         struct net_device *netdev = adapter->netdev;
2464         u16 vid = adapter->hw.mng_cookie.vlan_id;
2465         u16 old_vid = adapter->mng_vlan_id;
2466
2467         if (!adapter->vlgrp)
2468                 return;
2469
2470         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2471                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2472                 if (adapter->hw.mng_cookie.status &
2473                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2474                         e1000_vlan_rx_add_vid(netdev, vid);
2475                         adapter->mng_vlan_id = vid;
2476                 }
2477
2478                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2479                                 (vid != old_vid) &&
2480                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2481                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2482         } else {
2483                 adapter->mng_vlan_id = vid;
2484         }
2485 }
2486
2487
2488 static void e1000_vlan_rx_register(struct net_device *netdev,
2489                                    struct vlan_group *grp)
2490 {
2491         struct e1000_adapter *adapter = netdev_priv(netdev);
2492         struct e1000_hw *hw = &adapter->hw;
2493         u32 ctrl, rctl;
2494
2495         if (!test_bit(__E1000_DOWN, &adapter->state))
2496                 e1000_irq_disable(adapter);
2497         adapter->vlgrp = grp;
2498
2499         if (grp) {
2500                 /* enable VLAN tag insert/strip */
2501                 ctrl = er32(CTRL);
2502                 ctrl |= E1000_CTRL_VME;
2503                 ew32(CTRL, ctrl);
2504
2505                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2506                         /* enable VLAN receive filtering */
2507                         rctl = er32(RCTL);
2508                         rctl &= ~E1000_RCTL_CFIEN;
2509                         ew32(RCTL, rctl);
2510                         e1000_update_mng_vlan(adapter);
2511                 }
2512         } else {
2513                 /* disable VLAN tag insert/strip */
2514                 ctrl = er32(CTRL);
2515                 ctrl &= ~E1000_CTRL_VME;
2516                 ew32(CTRL, ctrl);
2517
2518                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2519                         if (adapter->mng_vlan_id !=
2520                             (u16)E1000_MNG_VLAN_NONE) {
2521                                 e1000_vlan_rx_kill_vid(netdev,
2522                                                        adapter->mng_vlan_id);
2523                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2524                         }
2525                 }
2526         }
2527
2528         if (!test_bit(__E1000_DOWN, &adapter->state))
2529                 e1000_irq_enable(adapter);
2530 }
2531
2532 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2533 {
2534         u16 vid;
2535
2536         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2537
2538         if (!adapter->vlgrp)
2539                 return;
2540
2541         for (vid = 0; vid < VLAN_N_VID; vid++) {
2542                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2543                         continue;
2544                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2545         }
2546 }
2547
2548 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2549 {
2550         struct e1000_hw *hw = &adapter->hw;
2551         u32 manc, manc2h, mdef, i, j;
2552
2553         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2554                 return;
2555
2556         manc = er32(MANC);
2557
2558         /*
2559          * enable receiving management packets to the host. this will probably
2560          * generate destination unreachable messages from the host OS, but
2561          * the packets will be handled on SMBUS
2562          */
2563         manc |= E1000_MANC_EN_MNG2HOST;
2564         manc2h = er32(MANC2H);
2565
2566         switch (hw->mac.type) {
2567         default:
2568                 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2569                 break;
2570         case e1000_82574:
2571         case e1000_82583:
2572                 /*
2573                  * Check if IPMI pass-through decision filter already exists;
2574                  * if so, enable it.
2575                  */
2576                 for (i = 0, j = 0; i < 8; i++) {
2577                         mdef = er32(MDEF(i));
2578
2579                         /* Ignore filters with anything other than IPMI ports */
2580                         if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2581                                 continue;
2582
2583                         /* Enable this decision filter in MANC2H */
2584                         if (mdef)
2585                                 manc2h |= (1 << i);
2586
2587                         j |= mdef;
2588                 }
2589
2590                 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2591                         break;
2592
2593                 /* Create new decision filter in an empty filter */
2594                 for (i = 0, j = 0; i < 8; i++)
2595                         if (er32(MDEF(i)) == 0) {
2596                                 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2597                                                E1000_MDEF_PORT_664));
2598                                 manc2h |= (1 << 1);
2599                                 j++;
2600                                 break;
2601                         }
2602
2603                 if (!j)
2604                         e_warn("Unable to create IPMI pass-through filter\n");
2605                 break;
2606         }
2607
2608         ew32(MANC2H, manc2h);
2609         ew32(MANC, manc);
2610 }
2611
2612 /**
2613  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2614  * @adapter: board private structure
2615  *
2616  * Configure the Tx unit of the MAC after a reset.
2617  **/
2618 static void e1000_configure_tx(struct e1000_adapter *adapter)
2619 {
2620         struct e1000_hw *hw = &adapter->hw;
2621         struct e1000_ring *tx_ring = adapter->tx_ring;
2622         u64 tdba;
2623         u32 tdlen, tctl, tipg, tarc;
2624         u32 ipgr1, ipgr2;
2625
2626         /* Setup the HW Tx Head and Tail descriptor pointers */
2627         tdba = tx_ring->dma;
2628         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2629         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2630         ew32(TDBAH, (tdba >> 32));
2631         ew32(TDLEN, tdlen);
2632         ew32(TDH, 0);
2633         ew32(TDT, 0);
2634         tx_ring->head = E1000_TDH;
2635         tx_ring->tail = E1000_TDT;
2636
2637         /* Set the default values for the Tx Inter Packet Gap timer */
2638         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2639         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2640         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2641
2642         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2643                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2644
2645         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2646         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2647         ew32(TIPG, tipg);
2648
2649         /* Set the Tx Interrupt Delay register */
2650         ew32(TIDV, adapter->tx_int_delay);
2651         /* Tx irq moderation */
2652         ew32(TADV, adapter->tx_abs_int_delay);
2653
2654         if (adapter->flags2 & FLAG2_DMA_BURST) {
2655                 u32 txdctl = er32(TXDCTL(0));
2656                 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2657                             E1000_TXDCTL_WTHRESH);
2658                 /*
2659                  * set up some performance related parameters to encourage the
2660                  * hardware to use the bus more efficiently in bursts, depends
2661                  * on the tx_int_delay to be enabled,
2662                  * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
2663                  * hthresh = 1 ==> prefetch when one or more available
2664                  * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2665                  * BEWARE: this seems to work but should be considered first if
2666                  * there are tx hangs or other tx related bugs
2667                  */
2668                 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2669                 ew32(TXDCTL(0), txdctl);
2670                 /* erratum work around: set txdctl the same for both queues */
2671                 ew32(TXDCTL(1), txdctl);
2672         }
2673
2674         /* Program the Transmit Control Register */
2675         tctl = er32(TCTL);
2676         tctl &= ~E1000_TCTL_CT;
2677         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2678                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2679
2680         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2681                 tarc = er32(TARC(0));
2682                 /*
2683                  * set the speed mode bit, we'll clear it if we're not at
2684                  * gigabit link later
2685                  */
2686 #define SPEED_MODE_BIT (1 << 21)
2687                 tarc |= SPEED_MODE_BIT;
2688                 ew32(TARC(0), tarc);
2689         }
2690
2691         /* errata: program both queues to unweighted RR */
2692         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2693                 tarc = er32(TARC(0));
2694                 tarc |= 1;
2695                 ew32(TARC(0), tarc);
2696                 tarc = er32(TARC(1));
2697                 tarc |= 1;
2698                 ew32(TARC(1), tarc);
2699         }
2700
2701         /* Setup Transmit Descriptor Settings for eop descriptor */
2702         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2703
2704         /* only set IDE if we are delaying interrupts using the timers */
2705         if (adapter->tx_int_delay)
2706                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2707
2708         /* enable Report Status bit */
2709         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2710
2711         ew32(TCTL, tctl);
2712
2713         e1000e_config_collision_dist(hw);
2714 }
2715
2716 /**
2717  * e1000_setup_rctl - configure the receive control registers
2718  * @adapter: Board private structure
2719  **/
2720 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2721                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2722 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2723 {
2724         struct e1000_hw *hw = &adapter->hw;
2725         u32 rctl, rfctl;
2726         u32 psrctl = 0;
2727         u32 pages = 0;
2728
2729         /* Workaround Si errata on 82579 - configure jumbo frame flow */
2730         if (hw->mac.type == e1000_pch2lan) {
2731                 s32 ret_val;
2732
2733                 if (adapter->netdev->mtu > ETH_DATA_LEN)
2734                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2735                 else
2736                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2737
2738                 if (ret_val)
2739                         e_dbg("failed to enable jumbo frame workaround mode\n");
2740         }
2741
2742         /* Program MC offset vector base */
2743         rctl = er32(RCTL);
2744         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2745         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2746                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2747                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2748
2749         /* Do not Store bad packets */
2750         rctl &= ~E1000_RCTL_SBP;
2751
2752         /* Enable Long Packet receive */
2753         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2754                 rctl &= ~E1000_RCTL_LPE;
2755         else
2756                 rctl |= E1000_RCTL_LPE;
2757
2758         /* Some systems expect that the CRC is included in SMBUS traffic. The
2759          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2760          * host memory when this is enabled
2761          */
2762         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2763                 rctl |= E1000_RCTL_SECRC;
2764
2765         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2766         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2767                 u16 phy_data;
2768
2769                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2770                 phy_data &= 0xfff8;
2771                 phy_data |= (1 << 2);
2772                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2773
2774                 e1e_rphy(hw, 22, &phy_data);
2775                 phy_data &= 0x0fff;
2776                 phy_data |= (1 << 14);
2777                 e1e_wphy(hw, 0x10, 0x2823);
2778                 e1e_wphy(hw, 0x11, 0x0003);
2779                 e1e_wphy(hw, 22, phy_data);
2780         }
2781
2782         /* Setup buffer sizes */
2783         rctl &= ~E1000_RCTL_SZ_4096;
2784         rctl |= E1000_RCTL_BSEX;
2785         switch (adapter->rx_buffer_len) {
2786         case 2048:
2787         default:
2788                 rctl |= E1000_RCTL_SZ_2048;
2789                 rctl &= ~E1000_RCTL_BSEX;
2790                 break;
2791         case 4096:
2792                 rctl |= E1000_RCTL_SZ_4096;
2793                 break;
2794         case 8192:
2795                 rctl |= E1000_RCTL_SZ_8192;
2796                 break;
2797         case 16384:
2798                 rctl |= E1000_RCTL_SZ_16384;
2799                 break;
2800         }
2801
2802         /*
2803          * 82571 and greater support packet-split where the protocol
2804          * header is placed in skb->data and the packet data is
2805          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2806          * In the case of a non-split, skb->data is linearly filled,
2807          * followed by the page buffers.  Therefore, skb->data is
2808          * sized to hold the largest protocol header.
2809          *
2810          * allocations using alloc_page take too long for regular MTU
2811          * so only enable packet split for jumbo frames
2812          *
2813          * Using pages when the page size is greater than 16k wastes
2814          * a lot of memory, since we allocate 3 pages at all times
2815          * per packet.
2816          */
2817         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2818         if (!(adapter->flags & FLAG_HAS_ERT) && (pages <= 3) &&
2819             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2820                 adapter->rx_ps_pages = pages;
2821         else
2822                 adapter->rx_ps_pages = 0;
2823
2824         if (adapter->rx_ps_pages) {
2825                 /* Configure extra packet-split registers */
2826                 rfctl = er32(RFCTL);
2827                 rfctl |= E1000_RFCTL_EXTEN;
2828                 /*
2829                  * disable packet split support for IPv6 extension headers,
2830                  * because some malformed IPv6 headers can hang the Rx
2831                  */
2832                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2833                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2834
2835                 ew32(RFCTL, rfctl);
2836
2837                 /* Enable Packet split descriptors */
2838                 rctl |= E1000_RCTL_DTYP_PS;
2839
2840                 psrctl |= adapter->rx_ps_bsize0 >>
2841                         E1000_PSRCTL_BSIZE0_SHIFT;
2842
2843                 switch (adapter->rx_ps_pages) {
2844                 case 3:
2845                         psrctl |= PAGE_SIZE <<
2846                                 E1000_PSRCTL_BSIZE3_SHIFT;
2847                 case 2:
2848                         psrctl |= PAGE_SIZE <<
2849                                 E1000_PSRCTL_BSIZE2_SHIFT;
2850                 case 1:
2851                         psrctl |= PAGE_SIZE >>
2852                                 E1000_PSRCTL_BSIZE1_SHIFT;
2853                         break;
2854                 }
2855
2856                 ew32(PSRCTL, psrctl);
2857         }
2858
2859         ew32(RCTL, rctl);
2860         /* just started the receive unit, no need to restart */
2861         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2862 }
2863
2864 /**
2865  * e1000_configure_rx - Configure Receive Unit after Reset
2866  * @adapter: board private structure
2867  *
2868  * Configure the Rx unit of the MAC after a reset.
2869  **/
2870 static void e1000_configure_rx(struct e1000_adapter *adapter)
2871 {
2872         struct e1000_hw *hw = &adapter->hw;
2873         struct e1000_ring *rx_ring = adapter->rx_ring;
2874         u64 rdba;
2875         u32 rdlen, rctl, rxcsum, ctrl_ext;
2876
2877         if (adapter->rx_ps_pages) {
2878                 /* this is a 32 byte descriptor */
2879                 rdlen = rx_ring->count *
2880                         sizeof(union e1000_rx_desc_packet_split);
2881                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2882                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2883         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2884                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2885                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2886                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2887         } else {
2888                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2889                 adapter->clean_rx = e1000_clean_rx_irq;
2890                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2891         }
2892
2893         /* disable receives while setting up the descriptors */
2894         rctl = er32(RCTL);
2895         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2896         e1e_flush();
2897         msleep(10);
2898
2899         if (adapter->flags2 & FLAG2_DMA_BURST) {
2900                 /*
2901                  * set the writeback threshold (only takes effect if the RDTR
2902                  * is set). set GRAN=1 and write back up to 0x4 worth, and
2903                  * enable prefetching of 0x20 rx descriptors
2904                  * granularity = 01
2905                  * wthresh = 04,
2906                  * hthresh = 04,
2907                  * pthresh = 0x20
2908                  */
2909                 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
2910                 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
2911
2912                 /*
2913                  * override the delay timers for enabling bursting, only if
2914                  * the value was not set by the user via module options
2915                  */
2916                 if (adapter->rx_int_delay == DEFAULT_RDTR)
2917                         adapter->rx_int_delay = BURST_RDTR;
2918                 if (adapter->rx_abs_int_delay == DEFAULT_RADV)
2919                         adapter->rx_abs_int_delay = BURST_RADV;
2920         }
2921
2922         /* set the Receive Delay Timer Register */
2923         ew32(RDTR, adapter->rx_int_delay);
2924
2925         /* irq moderation */
2926         ew32(RADV, adapter->rx_abs_int_delay);
2927         if ((adapter->itr_setting != 0) && (adapter->itr != 0))
2928                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2929
2930         ctrl_ext = er32(CTRL_EXT);
2931         /* Auto-Mask interrupts upon ICR access */
2932         ctrl_ext |= E1000_CTRL_EXT_IAME;
2933         ew32(IAM, 0xffffffff);
2934         ew32(CTRL_EXT, ctrl_ext);
2935         e1e_flush();
2936
2937         /*
2938          * Setup the HW Rx Head and Tail Descriptor Pointers and
2939          * the Base and Length of the Rx Descriptor Ring
2940          */
2941         rdba = rx_ring->dma;
2942         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2943         ew32(RDBAH, (rdba >> 32));
2944         ew32(RDLEN, rdlen);
2945         ew32(RDH, 0);
2946         ew32(RDT, 0);
2947         rx_ring->head = E1000_RDH;
2948         rx_ring->tail = E1000_RDT;
2949
2950         /* Enable Receive Checksum Offload for TCP and UDP */
2951         rxcsum = er32(RXCSUM);
2952         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2953                 rxcsum |= E1000_RXCSUM_TUOFL;
2954
2955                 /*
2956                  * IPv4 payload checksum for UDP fragments must be
2957                  * used in conjunction with packet-split.
2958                  */
2959                 if (adapter->rx_ps_pages)
2960                         rxcsum |= E1000_RXCSUM_IPPCSE;
2961         } else {
2962                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2963                 /* no need to clear IPPCSE as it defaults to 0 */
2964         }
2965         ew32(RXCSUM, rxcsum);
2966
2967         /*
2968          * Enable early receives on supported devices, only takes effect when
2969          * packet size is equal or larger than the specified value (in 8 byte
2970          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2971          */
2972         if ((adapter->flags & FLAG_HAS_ERT) ||
2973             (adapter->hw.mac.type == e1000_pch2lan)) {
2974                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2975                         u32 rxdctl = er32(RXDCTL(0));
2976                         ew32(RXDCTL(0), rxdctl | 0x3);
2977                         if (adapter->flags & FLAG_HAS_ERT)
2978                                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2979                         /*
2980                          * With jumbo frames and early-receive enabled,
2981                          * excessive C-state transition latencies result in
2982                          * dropped transactions.
2983                          */
2984                         pm_qos_update_request(
2985                                 &adapter->netdev->pm_qos_req, 55);
2986                 } else {
2987                         pm_qos_update_request(
2988                                 &adapter->netdev->pm_qos_req,
2989                                 PM_QOS_DEFAULT_VALUE);
2990                 }
2991         }
2992
2993         /* Enable Receives */
2994         ew32(RCTL, rctl);
2995 }
2996
2997 /**
2998  *  e1000_update_mc_addr_list - Update Multicast addresses
2999  *  @hw: pointer to the HW structure
3000  *  @mc_addr_list: array of multicast addresses to program
3001  *  @mc_addr_count: number of multicast addresses to program
3002  *
3003  *  Updates the Multicast Table Array.
3004  *  The caller must have a packed mc_addr_list of multicast addresses.
3005  **/
3006 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
3007                                       u32 mc_addr_count)
3008 {
3009         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
3010 }
3011
3012 /**
3013  * e1000_set_multi - Multicast and Promiscuous mode set
3014  * @netdev: network interface device structure
3015  *
3016  * The set_multi entry point is called whenever the multicast address
3017  * list or the network interface flags are updated.  This routine is
3018  * responsible for configuring the hardware for proper multicast,
3019  * promiscuous mode, and all-multi behavior.
3020  **/
3021 static void e1000_set_multi(struct net_device *netdev)
3022 {
3023         struct e1000_adapter *adapter = netdev_priv(netdev);
3024         struct e1000_hw *hw = &adapter->hw;
3025         struct netdev_hw_addr *ha;
3026         u8  *mta_list;
3027         u32 rctl;
3028         int i;
3029
3030         /* Check for Promiscuous and All Multicast modes */
3031
3032         rctl = er32(RCTL);
3033
3034         if (netdev->flags & IFF_PROMISC) {
3035                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3036                 rctl &= ~E1000_RCTL_VFE;
3037         } else {
3038                 if (netdev->flags & IFF_ALLMULTI) {
3039                         rctl |= E1000_RCTL_MPE;
3040                         rctl &= ~E1000_RCTL_UPE;
3041                 } else {
3042                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3043                 }
3044                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
3045                         rctl |= E1000_RCTL_VFE;
3046         }
3047
3048         ew32(RCTL, rctl);
3049
3050         if (!netdev_mc_empty(netdev)) {
3051                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3052                 if (!mta_list)
3053                         return;
3054
3055                 /* prepare a packed array of only addresses. */
3056                 i = 0;
3057                 netdev_for_each_mc_addr(ha, netdev)
3058                         memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3059
3060                 e1000_update_mc_addr_list(hw, mta_list, i);
3061                 kfree(mta_list);
3062         } else {
3063                 /*
3064                  * if we're called from probe, we might not have
3065                  * anything to do here, so clear out the list
3066                  */
3067                 e1000_update_mc_addr_list(hw, NULL, 0);
3068         }
3069 }
3070
3071 /**
3072  * e1000_configure - configure the hardware for Rx and Tx
3073  * @adapter: private board structure
3074  **/
3075 static void e1000_configure(struct e1000_adapter *adapter)
3076 {
3077         e1000_set_multi(adapter->netdev);
3078
3079         e1000_restore_vlan(adapter);
3080         e1000_init_manageability_pt(adapter);
3081
3082         e1000_configure_tx(adapter);
3083         e1000_setup_rctl(adapter);
3084         e1000_configure_rx(adapter);
3085         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
3086 }
3087
3088 /**
3089  * e1000e_power_up_phy - restore link in case the phy was powered down
3090  * @adapter: address of board private structure
3091  *
3092  * The phy may be powered down to save power and turn off link when the
3093  * driver is unloaded and wake on lan is not enabled (among others)
3094  * *** this routine MUST be followed by a call to e1000e_reset ***
3095  **/
3096 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3097 {
3098         if (adapter->hw.phy.ops.power_up)
3099                 adapter->hw.phy.ops.power_up(&adapter->hw);
3100
3101         adapter->hw.mac.ops.setup_link(&adapter->hw);
3102 }
3103
3104 /**
3105  * e1000_power_down_phy - Power down the PHY
3106  *
3107  * Power down the PHY so no link is implied when interface is down.
3108  * The PHY cannot be powered down if management or WoL is active.
3109  */
3110 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3111 {
3112         /* WoL is enabled */
3113         if (adapter->wol)
3114                 return;
3115
3116         if (adapter->hw.phy.ops.power_down)
3117                 adapter->hw.phy.ops.power_down(&adapter->hw);
3118 }
3119
3120 /**
3121  * e1000e_reset - bring the hardware into a known good state
3122  *
3123  * This function boots the hardware and enables some settings that
3124  * require a configuration cycle of the hardware - those cannot be
3125  * set/changed during runtime. After reset the device needs to be
3126  * properly configured for Rx, Tx etc.
3127  */
3128 void e1000e_reset(struct e1000_adapter *adapter)
3129 {
3130         struct e1000_mac_info *mac = &adapter->hw.mac;
3131         struct e1000_fc_info *fc = &adapter->hw.fc;
3132         struct e1000_hw *hw = &adapter->hw;
3133         u32 tx_space, min_tx_space, min_rx_space;
3134         u32 pba = adapter->pba;
3135         u16 hwm;
3136
3137         /* reset Packet Buffer Allocation to default */
3138         ew32(PBA, pba);
3139
3140         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3141                 /*
3142                  * To maintain wire speed transmits, the Tx FIFO should be
3143                  * large enough to accommodate two full transmit packets,
3144                  * rounded up to the next 1KB and expressed in KB.  Likewise,
3145                  * the Rx FIFO should be large enough to accommodate at least
3146                  * one full receive packet and is similarly rounded up and
3147                  * expressed in KB.
3148                  */
3149                 pba = er32(PBA);
3150                 /* upper 16 bits has Tx packet buffer allocation size in KB */
3151                 tx_space = pba >> 16;
3152                 /* lower 16 bits has Rx packet buffer allocation size in KB */
3153                 pba &= 0xffff;
3154                 /*
3155                  * the Tx fifo also stores 16 bytes of information about the tx
3156                  * but don't include ethernet FCS because hardware appends it
3157                  */
3158                 min_tx_space = (adapter->max_frame_size +
3159                                 sizeof(struct e1000_tx_desc) -
3160                                 ETH_FCS_LEN) * 2;
3161                 min_tx_space = ALIGN(min_tx_space, 1024);
3162                 min_tx_space >>= 10;
3163                 /* software strips receive CRC, so leave room for it */
3164                 min_rx_space = adapter->max_frame_size;
3165                 min_rx_space = ALIGN(min_rx_space, 1024);
3166                 min_rx_space >>= 10;
3167
3168                 /*
3169                  * If current Tx allocation is less than the min Tx FIFO size,
3170                  * and the min Tx FIFO size is less than the current Rx FIFO
3171                  * allocation, take space away from current Rx allocation
3172                  */
3173                 if ((tx_space < min_tx_space) &&
3174                     ((min_tx_space - tx_space) < pba)) {
3175                         pba -= min_tx_space - tx_space;
3176
3177                         /*
3178                          * if short on Rx space, Rx wins and must trump tx
3179                          * adjustment or use Early Receive if available
3180                          */
3181                         if ((pba < min_rx_space) &&
3182                             (!(adapter->flags & FLAG_HAS_ERT)))
3183                                 /* ERT enabled in e1000_configure_rx */
3184                                 pba = min_rx_space;
3185                 }
3186
3187                 ew32(PBA, pba);
3188         }
3189
3190
3191         /*
3192          * flow control settings
3193          *
3194          * The high water mark must be low enough to fit one full frame
3195          * (or the size used for early receive) above it in the Rx FIFO.
3196          * Set it to the lower of:
3197          * - 90% of the Rx FIFO size, and
3198          * - the full Rx FIFO size minus the early receive size (for parts
3199          *   with ERT support assuming ERT set to E1000_ERT_2048), or
3200          * - the full Rx FIFO size minus one full frame
3201          */
3202         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3203                 fc->pause_time = 0xFFFF;
3204         else
3205                 fc->pause_time = E1000_FC_PAUSE_TIME;
3206         fc->send_xon = 1;
3207         fc->current_mode = fc->requested_mode;
3208
3209         switch (hw->mac.type) {
3210         default:
3211                 if ((adapter->flags & FLAG_HAS_ERT) &&
3212                     (adapter->netdev->mtu > ETH_DATA_LEN))
3213                         hwm = min(((pba << 10) * 9 / 10),
3214                                   ((pba << 10) - (E1000_ERT_2048 << 3)));
3215                 else
3216                         hwm = min(((pba << 10) * 9 / 10),
3217                                   ((pba << 10) - adapter->max_frame_size));
3218
3219                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3220                 fc->low_water = fc->high_water - 8;
3221                 break;
3222         case e1000_pchlan:
3223                 /*
3224                  * Workaround PCH LOM adapter hangs with certain network
3225                  * loads.  If hangs persist, try disabling Tx flow control.
3226                  */
3227                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3228                         fc->high_water = 0x3500;
3229                         fc->low_water  = 0x1500;
3230                 } else {
3231                         fc->high_water = 0x5000;
3232                         fc->low_water  = 0x3000;
3233                 }
3234                 fc->refresh_time = 0x1000;
3235                 break;
3236         case e1000_pch2lan:
3237                 fc->high_water = 0x05C20;
3238                 fc->low_water = 0x05048;
3239                 fc->pause_time = 0x0650;
3240                 fc->refresh_time = 0x0400;
3241                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3242                         pba = 14;
3243                         ew32(PBA, pba);
3244                 }
3245                 break;
3246         }
3247
3248         /*
3249          * Disable Adaptive Interrupt Moderation if 2 full packets cannot
3250          * fit in receive buffer and early-receive not supported.
3251          */
3252         if (adapter->itr_setting & 0x3) {
3253                 if (((adapter->max_frame_size * 2) > (pba << 10)) &&
3254                     !(adapter->flags & FLAG_HAS_ERT)) {
3255                         if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
3256                                 dev_info(&adapter->pdev->dev,
3257                                         "Interrupt Throttle Rate turned off\n");
3258                                 adapter->flags2 |= FLAG2_DISABLE_AIM;
3259                                 ew32(ITR, 0);
3260                         }
3261                 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
3262                         dev_info(&adapter->pdev->dev,
3263                                  "Interrupt Throttle Rate turned on\n");
3264                         adapter->flags2 &= ~FLAG2_DISABLE_AIM;
3265                         adapter->itr = 20000;
3266                         ew32(ITR, 1000000000 / (adapter->itr * 256));
3267                 }
3268         }
3269
3270         /* Allow time for pending master requests to run */
3271         mac->ops.reset_hw(hw);
3272
3273         /*
3274          * For parts with AMT enabled, let the firmware know
3275          * that the network interface is in control
3276          */
3277         if (adapter->flags & FLAG_HAS_AMT)
3278                 e1000_get_hw_control(adapter);
3279
3280         ew32(WUC, 0);
3281
3282         if (mac->ops.init_hw(hw))
3283                 e_err("Hardware Error\n");
3284
3285         e1000_update_mng_vlan(adapter);
3286
3287         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3288         ew32(VET, ETH_P_8021Q);
3289
3290         e1000e_reset_adaptive(hw);
3291         e1000_get_phy_info(hw);
3292
3293         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3294             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3295                 u16 phy_data = 0;
3296                 /*
3297                  * speed up time to link by disabling smart power down, ignore
3298                  * the return value of this function because there is nothing
3299                  * different we would do if it failed
3300                  */
3301                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3302                 phy_data &= ~IGP02E1000_PM_SPD;
3303                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3304         }
3305 }
3306
3307 int e1000e_up(struct e1000_adapter *adapter)
3308 {
3309         struct e1000_hw *hw = &adapter->hw;
3310
3311         /* hardware has been reset, we need to reload some things */
3312         e1000_configure(adapter);
3313
3314         clear_bit(__E1000_DOWN, &adapter->state);
3315
3316         napi_enable(&adapter->napi);
3317         if (adapter->msix_entries)
3318                 e1000_configure_msix(adapter);
3319         e1000_irq_enable(adapter);
3320
3321         netif_wake_queue(adapter->netdev);
3322
3323         /* fire a link change interrupt to start the watchdog */
3324         if (adapter->msix_entries)
3325                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3326         else
3327                 ew32(ICS, E1000_ICS_LSC);
3328
3329         return 0;
3330 }
3331
3332 void e1000e_down(struct e1000_adapter *adapter)
3333 {
3334         struct net_device *netdev = adapter->netdev;
3335         struct e1000_hw *hw = &adapter->hw;
3336         u32 tctl, rctl;
3337
3338         /*
3339          * signal that we're down so the interrupt handler does not
3340          * reschedule our watchdog timer
3341          */
3342         set_bit(__E1000_DOWN, &adapter->state);
3343
3344         /* disable receives in the hardware */
3345         rctl = er32(RCTL);
3346         ew32(RCTL, rctl & ~E1000_RCTL_EN);
3347         /* flush and sleep below */
3348
3349         netif_stop_queue(netdev);
3350
3351         /* disable transmits in the hardware */
3352         tctl = er32(TCTL);
3353         tctl &= ~E1000_TCTL_EN;
3354         ew32(TCTL, tctl);
3355         /* flush both disables and wait for them to finish */
3356         e1e_flush();
3357         msleep(10);
3358
3359         napi_disable(&adapter->napi);
3360         e1000_irq_disable(adapter);
3361
3362         del_timer_sync(&adapter->watchdog_timer);
3363         del_timer_sync(&adapter->phy_info_timer);
3364
3365         netif_carrier_off(netdev);
3366         adapter->link_speed = 0;
3367         adapter->link_duplex = 0;
3368
3369         if (!pci_channel_offline(adapter->pdev))
3370                 e1000e_reset(adapter);
3371         e1000_clean_tx_ring(adapter);
3372         e1000_clean_rx_ring(adapter);
3373
3374         /*
3375          * TODO: for power management, we could drop the link and
3376          * pci_disable_device here.
3377          */
3378 }
3379
3380 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3381 {
3382         might_sleep();
3383         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3384                 msleep(1);
3385         e1000e_down(adapter);
3386         e1000e_up(adapter);
3387         clear_bit(__E1000_RESETTING, &adapter->state);
3388 }
3389
3390 /**
3391  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3392  * @adapter: board private structure to initialize
3393  *
3394  * e1000_sw_init initializes the Adapter private data structure.
3395  * Fields are initialized based on PCI device information and
3396  * OS network device settings (MTU size).
3397  **/
3398 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3399 {
3400         struct net_device *netdev = adapter->netdev;
3401
3402         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3403         adapter->rx_ps_bsize0 = 128;
3404         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3405         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3406
3407         e1000e_set_interrupt_capability(adapter);
3408
3409         if (e1000_alloc_queues(adapter))
3410                 return -ENOMEM;
3411
3412         /* Explicitly disable IRQ since the NIC can be in any state. */
3413         e1000_irq_disable(adapter);
3414
3415         set_bit(__E1000_DOWN, &adapter->state);
3416         return 0;
3417 }
3418
3419 /**
3420  * e1000_intr_msi_test - Interrupt Handler
3421  * @irq: interrupt number
3422  * @data: pointer to a network interface device structure
3423  **/
3424 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3425 {
3426         struct net_device *netdev = data;
3427         struct e1000_adapter *adapter = netdev_priv(netdev);
3428         struct e1000_hw *hw = &adapter->hw;
3429         u32 icr = er32(ICR);
3430
3431         e_dbg("icr is %08X\n", icr);
3432         if (icr & E1000_ICR_RXSEQ) {
3433                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3434                 wmb();
3435         }
3436
3437         return IRQ_HANDLED;
3438 }
3439
3440 /**
3441  * e1000_test_msi_interrupt - Returns 0 for successful test
3442  * @adapter: board private struct
3443  *
3444  * code flow taken from tg3.c
3445  **/
3446 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3447 {
3448         struct net_device *netdev = adapter->netdev;
3449         struct e1000_hw *hw = &adapter->hw;
3450         int err;
3451
3452         /* poll_enable hasn't been called yet, so don't need disable */
3453         /* clear any pending events */
3454         er32(ICR);
3455
3456         /* free the real vector and request a test handler */
3457         e1000_free_irq(adapter);
3458         e1000e_reset_interrupt_capability(adapter);
3459
3460         /* Assume that the test fails, if it succeeds then the test
3461          * MSI irq handler will unset this flag */
3462         adapter->flags |= FLAG_MSI_TEST_FAILED;
3463
3464         err = pci_enable_msi(adapter->pdev);
3465         if (err)
3466                 goto msi_test_failed;
3467
3468         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3469                           netdev->name, netdev);
3470         if (err) {
3471                 pci_disable_msi(adapter->pdev);
3472                 goto msi_test_failed;
3473         }
3474
3475         wmb();
3476
3477         e1000_irq_enable(adapter);
3478
3479         /* fire an unusual interrupt on the test handler */
3480         ew32(ICS, E1000_ICS_RXSEQ);
3481         e1e_flush();
3482         msleep(50);
3483
3484         e1000_irq_disable(adapter);
3485
3486         rmb();
3487
3488         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3489                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3490                 e_info("MSI interrupt test failed, using legacy interrupt.\n");
3491         } else
3492                 e_dbg("MSI interrupt test succeeded!\n");
3493
3494         free_irq(adapter->pdev->irq, netdev);
3495         pci_disable_msi(adapter->pdev);
3496
3497 msi_test_failed:
3498         e1000e_set_interrupt_capability(adapter);
3499         return e1000_request_irq(adapter);
3500 }
3501
3502 /**
3503  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3504  * @adapter: board private struct
3505  *
3506  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3507  **/
3508 static int e1000_test_msi(struct e1000_adapter *adapter)
3509 {
3510         int err;
3511         u16 pci_cmd;
3512
3513         if (!(adapter->flags & FLAG_MSI_ENABLED))
3514                 return 0;
3515
3516         /* disable SERR in case the MSI write causes a master abort */
3517         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3518         if (pci_cmd & PCI_COMMAND_SERR)
3519                 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3520                                       pci_cmd & ~PCI_COMMAND_SERR);
3521
3522         err = e1000_test_msi_interrupt(adapter);
3523
3524         /* re-enable SERR */
3525         if (pci_cmd & PCI_COMMAND_SERR) {
3526                 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3527                 pci_cmd |= PCI_COMMAND_SERR;
3528                 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3529         }
3530
3531         return err;
3532 }
3533
3534 /**
3535  * e1000_open - Called when a network interface is made active
3536  * @netdev: network interface device structure
3537  *
3538  * Returns 0 on success, negative value on failure
3539  *
3540  * The open entry point is called when a network interface is made
3541  * active by the system (IFF_UP).  At this point all resources needed
3542  * for transmit and receive operations are allocated, the interrupt
3543  * handler is registered with the OS, the watchdog timer is started,
3544  * and the stack is notified that the interface is ready.
3545  **/
3546 static int e1000_open(struct net_device *netdev)
3547 {
3548         struct e1000_adapter *adapter = netdev_priv(netdev);
3549         struct e1000_hw *hw = &adapter->hw;
3550         struct pci_dev *pdev = adapter->pdev;
3551         int err;
3552
3553         /* disallow open during test */
3554         if (test_bit(__E1000_TESTING, &adapter->state))
3555                 return -EBUSY;
3556
3557         pm_runtime_get_sync(&pdev->dev);
3558
3559         netif_carrier_off(netdev);
3560
3561         /* allocate transmit descriptors */
3562         err = e1000e_setup_tx_resources(adapter);
3563         if (err)
3564                 goto err_setup_tx;
3565
3566         /* allocate receive descriptors */
3567         err = e1000e_setup_rx_resources(adapter);
3568         if (err)
3569                 goto err_setup_rx;
3570
3571         /*
3572          * If AMT is enabled, let the firmware know that the network
3573          * interface is now open and reset the part to a known state.
3574          */
3575         if (adapter->flags & FLAG_HAS_AMT) {
3576                 e1000_get_hw_control(adapter);
3577                 e1000e_reset(adapter);
3578         }
3579
3580         e1000e_power_up_phy(adapter);
3581
3582         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3583         if ((adapter->hw.mng_cookie.status &
3584              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3585                 e1000_update_mng_vlan(adapter);
3586
3587         /* DMA latency requirement to workaround early-receive/jumbo issue */
3588         if ((adapter->flags & FLAG_HAS_ERT) ||
3589             (adapter->hw.mac.type == e1000_pch2lan))
3590                 pm_qos_add_request(&adapter->netdev->pm_qos_req,
3591                                    PM_QOS_CPU_DMA_LATENCY,
3592                                    PM_QOS_DEFAULT_VALUE);
3593
3594         /*
3595          * before we allocate an interrupt, we must be ready to handle it.
3596          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3597          * as soon as we call pci_request_irq, so we have to setup our
3598          * clean_rx handler before we do so.
3599          */
3600         e1000_configure(adapter);
3601
3602         err = e1000_request_irq(adapter);
3603         if (err)
3604                 goto err_req_irq;
3605
3606         /*
3607          * Work around PCIe errata with MSI interrupts causing some chipsets to
3608          * ignore e1000e MSI messages, which means we need to test our MSI
3609          * interrupt now
3610          */
3611         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3612                 err = e1000_test_msi(adapter);
3613                 if (err) {
3614                         e_err("Interrupt allocation failed\n");
3615                         goto err_req_irq;
3616                 }
3617         }
3618
3619         /* From here on the code is the same as e1000e_up() */
3620         clear_bit(__E1000_DOWN, &adapter->state);
3621
3622         napi_enable(&adapter->napi);
3623
3624         e1000_irq_enable(adapter);
3625
3626         netif_start_queue(netdev);
3627
3628         adapter->idle_check = true;
3629         pm_runtime_put(&pdev->dev);
3630
3631         /* fire a link status change interrupt to start the watchdog */
3632         if (adapter->msix_entries)
3633                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3634         else
3635                 ew32(ICS, E1000_ICS_LSC);
3636
3637         return 0;
3638
3639 err_req_irq:
3640         e1000_release_hw_control(adapter);
3641         e1000_power_down_phy(adapter);
3642         e1000e_free_rx_resources(adapter);
3643 err_setup_rx:
3644         e1000e_free_tx_resources(adapter);
3645 err_setup_tx:
3646         e1000e_reset(adapter);
3647         pm_runtime_put_sync(&pdev->dev);
3648
3649         return err;
3650 }
3651
3652 /**
3653  * e1000_close - Disables a network interface
3654  * @netdev: network interface device structure
3655  *
3656  * Returns 0, this is not allowed to fail
3657  *
3658  * The close entry point is called when an interface is de-activated
3659  * by the OS.  The hardware is still under the drivers control, but
3660  * needs to be disabled.  A global MAC reset is issued to stop the
3661  * hardware, and all transmit and receive resources are freed.
3662  **/
3663 static int e1000_close(struct net_device *netdev)
3664 {
3665         struct e1000_adapter *adapter = netdev_priv(netdev);
3666         struct pci_dev *pdev = adapter->pdev;
3667
3668         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3669
3670         pm_runtime_get_sync(&pdev->dev);
3671
3672         if (!test_bit(__E1000_DOWN, &adapter->state)) {
3673                 e1000e_down(adapter);
3674                 e1000_free_irq(adapter);
3675         }
3676         e1000_power_down_phy(adapter);
3677
3678         e1000e_free_tx_resources(adapter);
3679         e1000e_free_rx_resources(adapter);
3680
3681         /*
3682          * kill manageability vlan ID if supported, but not if a vlan with
3683          * the same ID is registered on the host OS (let 8021q kill it)
3684          */
3685         if ((adapter->hw.mng_cookie.status &
3686                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3687              !(adapter->vlgrp &&
3688                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3689                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3690
3691         /*
3692          * If AMT is enabled, let the firmware know that the network
3693          * interface is now closed
3694          */
3695         if (adapter->flags & FLAG_HAS_AMT)
3696                 e1000_release_hw_control(adapter);
3697
3698         if ((adapter->flags & FLAG_HAS_ERT) ||
3699             (adapter->hw.mac.type == e1000_pch2lan))
3700                 pm_qos_remove_request(&adapter->netdev->pm_qos_req);
3701
3702         pm_runtime_put_sync(&pdev->dev);
3703
3704         return 0;
3705 }
3706 /**
3707  * e1000_set_mac - Change the Ethernet Address of the NIC
3708  * @netdev: network interface device structure
3709  * @p: pointer to an address structure
3710  *
3711  * Returns 0 on success, negative on failure
3712  **/
3713 static int e1000_set_mac(struct net_device *netdev, void *p)
3714 {
3715         struct e1000_adapter *adapter = netdev_priv(netdev);
3716         struct sockaddr *addr = p;
3717
3718         if (!is_valid_ether_addr(addr->sa_data))
3719                 return -EADDRNOTAVAIL;
3720
3721         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3722         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3723
3724         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3725
3726         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3727                 /* activate the work around */
3728                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3729
3730                 /*
3731                  * Hold a copy of the LAA in RAR[14] This is done so that
3732                  * between the time RAR[0] gets clobbered  and the time it
3733                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3734                  * of the RARs and no incoming packets directed to this port
3735                  * are dropped. Eventually the LAA will be in RAR[0] and
3736                  * RAR[14]
3737                  */
3738                 e1000e_rar_set(&adapter->hw,
3739                               adapter->hw.mac.addr,
3740                               adapter->hw.mac.rar_entry_count - 1);
3741         }
3742
3743         return 0;
3744 }
3745
3746 /**
3747  * e1000e_update_phy_task - work thread to update phy
3748  * @work: pointer to our work struct
3749  *
3750  * this worker thread exists because we must acquire a
3751  * semaphore to read the phy, which we could msleep while
3752  * waiting for it, and we can't msleep in a timer.
3753  **/
3754 static void e1000e_update_phy_task(struct work_struct *work)
3755 {
3756         struct e1000_adapter *adapter = container_of(work,
3757                                         struct e1000_adapter, update_phy_task);
3758         e1000_get_phy_info(&adapter->hw);
3759 }
3760
3761 /*
3762  * Need to wait a few seconds after link up to get diagnostic information from
3763  * the phy
3764  */
3765 static void e1000_update_phy_info(unsigned long data)
3766 {
3767         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3768         schedule_work(&adapter->update_phy_task);
3769 }
3770
3771 /**
3772  * e1000e_update_phy_stats - Update the PHY statistics counters
3773  * @adapter: board private structure
3774  **/
3775 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
3776 {
3777         struct e1000_hw *hw = &adapter->hw;
3778         s32 ret_val;
3779         u16 phy_data;
3780
3781         ret_val = hw->phy.ops.acquire(hw);
3782         if (ret_val)
3783                 return;
3784
3785         hw->phy.addr = 1;
3786
3787 #define HV_PHY_STATS_PAGE       778
3788         /*
3789          * A page set is expensive so check if already on desired page.
3790          * If not, set to the page with the PHY status registers.
3791          */
3792         ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
3793                                            &phy_data);
3794         if (ret_val)
3795                 goto release;
3796         if (phy_data != (HV_PHY_STATS_PAGE << IGP_PAGE_SHIFT)) {
3797                 ret_val = e1000e_write_phy_reg_mdic(hw,
3798                                                     IGP01E1000_PHY_PAGE_SELECT,
3799                                                     (HV_PHY_STATS_PAGE <<
3800                                                      IGP_PAGE_SHIFT));
3801                 if (ret_val)
3802                         goto release;
3803         }
3804
3805         /* Read/clear the upper 16-bit registers and read/accumulate lower */
3806
3807         /* Single Collision Count */
3808         e1000e_read_phy_reg_mdic(hw, HV_SCC_UPPER & MAX_PHY_REG_ADDRESS,
3809                                  &phy_data);
3810         ret_val = e1000e_read_phy_reg_mdic(hw,
3811                                            HV_SCC_LOWER & MAX_PHY_REG_ADDRESS,
3812                                            &phy_data);
3813         if (!ret_val)
3814                 adapter->stats.scc += phy_data;
3815
3816         /* Excessive Collision Count */
3817         e1000e_read_phy_reg_mdic(hw, HV_ECOL_UPPER & MAX_PHY_REG_ADDRESS,
3818                                  &phy_data);
3819         ret_val = e1000e_read_phy_reg_mdic(hw,
3820                                            HV_ECOL_LOWER & MAX_PHY_REG_ADDRESS,
3821                                            &phy_data);
3822         if (!ret_val)
3823                 adapter->stats.ecol += phy_data;
3824
3825         /* Multiple Collision Count */
3826         e1000e_read_phy_reg_mdic(hw, HV_MCC_UPPER & MAX_PHY_REG_ADDRESS,
3827                                  &phy_data);
3828         ret_val = e1000e_read_phy_reg_mdic(hw,
3829                                            HV_MCC_LOWER & MAX_PHY_REG_ADDRESS,
3830                                            &phy_data);
3831         if (!ret_val)
3832                 adapter->stats.mcc += phy_data;
3833
3834         /* Late Collision Count */
3835         e1000e_read_phy_reg_mdic(hw, HV_LATECOL_UPPER & MAX_PHY_REG_ADDRESS,
3836                                  &phy_data);
3837         ret_val = e1000e_read_phy_reg_mdic(hw,
3838                                            HV_LATECOL_LOWER &
3839                                            MAX_PHY_REG_ADDRESS,
3840                                            &phy_data);
3841         if (!ret_val)
3842                 adapter->stats.latecol += phy_data;
3843
3844         /* Collision Count - also used for adaptive IFS */
3845         e1000e_read_phy_reg_mdic(hw, HV_COLC_UPPER & MAX_PHY_REG_ADDRESS,
3846                                  &phy_data);
3847         ret_val = e1000e_read_phy_reg_mdic(hw,
3848                                            HV_COLC_LOWER & MAX_PHY_REG_ADDRESS,
3849                                            &phy_data);
3850         if (!ret_val)
3851                 hw->mac.collision_delta = phy_data;
3852
3853         /* Defer Count */
3854         e1000e_read_phy_reg_mdic(hw, HV_DC_UPPER & MAX_PHY_REG_ADDRESS,
3855                                  &phy_data);
3856         ret_val = e1000e_read_phy_reg_mdic(hw,
3857                                            HV_DC_LOWER & MAX_PHY_REG_ADDRESS,
3858                                            &phy_data);
3859         if (!ret_val)
3860                 adapter->stats.dc += phy_data;
3861
3862         /* Transmit with no CRS */
3863         e1000e_read_phy_reg_mdic(hw, HV_TNCRS_UPPER & MAX_PHY_REG_ADDRESS,
3864                                  &phy_data);
3865         ret_val = e1000e_read_phy_reg_mdic(hw,
3866                                            HV_TNCRS_LOWER & MAX_PHY_REG_ADDRESS,
3867                                            &phy_data);
3868         if (!ret_val)
3869                 adapter->stats.tncrs += phy_data;
3870
3871 release:
3872         hw->phy.ops.release(hw);
3873 }
3874
3875 /**
3876  * e1000e_update_stats - Update the board statistics counters
3877  * @adapter: board private structure
3878  **/
3879 void e1000e_update_stats(struct e1000_adapter *adapter)
3880 {
3881         struct net_device *netdev = adapter->netdev;
3882         struct e1000_hw *hw = &adapter->hw;
3883         struct pci_dev *pdev = adapter->pdev;
3884
3885         /*
3886          * Prevent stats update while adapter is being reset, or if the pci
3887          * connection is down.
3888          */
3889         if (adapter->link_speed == 0)
3890                 return;
3891         if (pci_channel_offline(pdev))
3892                 return;
3893
3894         adapter->stats.crcerrs += er32(CRCERRS);
3895         adapter->stats.gprc += er32(GPRC);
3896         adapter->stats.gorc += er32(GORCL);
3897         er32(GORCH); /* Clear gorc */
3898         adapter->stats.bprc += er32(BPRC);
3899         adapter->stats.mprc += er32(MPRC);
3900         adapter->stats.roc += er32(ROC);
3901
3902         adapter->stats.mpc += er32(MPC);
3903
3904         /* Half-duplex statistics */
3905         if (adapter->link_duplex == HALF_DUPLEX) {
3906                 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
3907                         e1000e_update_phy_stats(adapter);
3908                 } else {
3909                         adapter->stats.scc += er32(SCC);
3910                         adapter->stats.ecol += er32(ECOL);
3911                         adapter->stats.mcc += er32(MCC);
3912                         adapter->stats.latecol += er32(LATECOL);
3913                         adapter->stats.dc += er32(DC);
3914
3915                         hw->mac.collision_delta = er32(COLC);
3916
3917                         if ((hw->mac.type != e1000_82574) &&
3918                             (hw->mac.type != e1000_82583))
3919                                 adapter->stats.tncrs += er32(TNCRS);
3920                 }
3921                 adapter->stats.colc += hw->mac.collision_delta;
3922         }
3923
3924         adapter->stats.xonrxc += er32(XONRXC);
3925         adapter->stats.xontxc += er32(XONTXC);
3926         adapter->stats.xoffrxc += er32(XOFFRXC);
3927         adapter->stats.xofftxc += er32(XOFFTXC);
3928         adapter->stats.gptc += er32(GPTC);
3929         adapter->stats.gotc += er32(GOTCL);
3930         er32(GOTCH); /* Clear gotc */
3931         adapter->stats.rnbc += er32(RNBC);
3932         adapter->stats.ruc += er32(RUC);
3933
3934         adapter->stats.mptc += er32(MPTC);
3935         adapter->stats.bptc += er32(BPTC);
3936
3937         /* used for adaptive IFS */
3938
3939         hw->mac.tx_packet_delta = er32(TPT);
3940         adapter->stats.tpt += hw->mac.tx_packet_delta;
3941
3942         adapter->stats.algnerrc += er32(ALGNERRC);
3943         adapter->stats.rxerrc += er32(RXERRC);
3944         adapter->stats.cexterr += er32(CEXTERR);
3945         adapter->stats.tsctc += er32(TSCTC);
3946         adapter->stats.tsctfc += er32(TSCTFC);
3947
3948         /* Fill out the OS statistics structure */
3949         netdev->stats.multicast = adapter->stats.mprc;
3950         netdev->stats.collisions = adapter->stats.colc;
3951
3952         /* Rx Errors */
3953
3954         /*
3955          * RLEC on some newer hardware can be incorrect so build
3956          * our own version based on RUC and ROC
3957          */
3958         netdev->stats.rx_errors = adapter->stats.rxerrc +
3959                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3960                 adapter->stats.ruc + adapter->stats.roc +
3961                 adapter->stats.cexterr;
3962         netdev->stats.rx_length_errors = adapter->stats.ruc +
3963                                               adapter->stats.roc;
3964         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3965         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3966         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3967
3968         /* Tx Errors */
3969         netdev->stats.tx_errors = adapter->stats.ecol +
3970                                        adapter->stats.latecol;
3971         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3972         netdev->stats.tx_window_errors = adapter->stats.latecol;
3973         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3974
3975         /* Tx Dropped needs to be maintained elsewhere */
3976
3977         /* Management Stats */
3978         adapter->stats.mgptc += er32(MGTPTC);
3979         adapter->stats.mgprc += er32(MGTPRC);
3980         adapter->stats.mgpdc += er32(MGTPDC);
3981 }
3982
3983 /**
3984  * e1000_phy_read_status - Update the PHY register status snapshot
3985  * @adapter: board private structure
3986  **/
3987 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3988 {
3989         struct e1000_hw *hw = &adapter->hw;
3990         struct e1000_phy_regs *phy = &adapter->phy_regs;
3991         int ret_val;
3992
3993         if ((er32(STATUS) & E1000_STATUS_LU) &&
3994             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3995                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3996                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3997                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3998                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3999                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
4000                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
4001                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
4002                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
4003                 if (ret_val)
4004                         e_warn("Error reading PHY register\n");
4005         } else {
4006                 /*
4007                  * Do not read PHY registers if link is not up
4008                  * Set values to typical power-on defaults
4009                  */
4010                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4011                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4012                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4013                              BMSR_ERCAP);
4014                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4015                                   ADVERTISE_ALL | ADVERTISE_CSMA);
4016                 phy->lpa = 0;
4017                 phy->expansion = EXPANSION_ENABLENPAGE;
4018                 phy->ctrl1000 = ADVERTISE_1000FULL;
4019                 phy->stat1000 = 0;
4020                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4021         }
4022 }
4023
4024 static void e1000_print_link_info(struct e1000_adapter *adapter)
4025 {
4026         struct e1000_hw *hw = &adapter->hw;
4027         u32 ctrl = er32(CTRL);
4028
4029         /* Link status message must follow this format for user tools */
4030         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
4031                "Flow Control: %s\n",
4032                adapter->netdev->name,
4033                adapter->link_speed,
4034                (adapter->link_duplex == FULL_DUPLEX) ?
4035                                 "Full Duplex" : "Half Duplex",
4036                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
4037                                 "RX/TX" :
4038                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
4039                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
4040 }
4041
4042 static bool e1000e_has_link(struct e1000_adapter *adapter)
4043 {
4044         struct e1000_hw *hw = &adapter->hw;
4045         bool link_active = 0;
4046         s32 ret_val = 0;
4047
4048         /*
4049          * get_link_status is set on LSC (link status) interrupt or
4050          * Rx sequence error interrupt.  get_link_status will stay
4051          * false until the check_for_link establishes link
4052          * for copper adapters ONLY
4053          */
4054         switch (hw->phy.media_type) {
4055         case e1000_media_type_copper:
4056                 if (hw->mac.get_link_status) {
4057                         ret_val = hw->mac.ops.check_for_link(hw);
4058                         link_active = !hw->mac.get_link_status;
4059                 } else {
4060                         link_active = 1;
4061                 }
4062                 break;
4063         case e1000_media_type_fiber:
4064                 ret_val = hw->mac.ops.check_for_link(hw);
4065                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4066                 break;
4067         case e1000_media_type_internal_serdes:
4068                 ret_val = hw->mac.ops.check_for_link(hw);
4069                 link_active = adapter->hw.mac.serdes_has_link;
4070                 break;
4071         default:
4072         case e1000_media_type_unknown:
4073                 break;
4074         }
4075
4076         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4077             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4078                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4079                 e_info("Gigabit has been disabled, downgrading speed\n");
4080         }
4081
4082         return link_active;
4083 }
4084
4085 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4086 {
4087         /* make sure the receive unit is started */
4088         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4089             (adapter->flags & FLAG_RX_RESTART_NOW)) {
4090                 struct e1000_hw *hw = &adapter->hw;
4091                 u32 rctl = er32(RCTL);
4092                 ew32(RCTL, rctl | E1000_RCTL_EN);
4093                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
4094         }
4095 }
4096
4097 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
4098 {
4099         struct e1000_hw *hw = &adapter->hw;
4100
4101         /*
4102          * With 82574 controllers, PHY needs to be checked periodically
4103          * for hung state and reset, if two calls return true
4104          */
4105         if (e1000_check_phy_82574(hw))
4106                 adapter->phy_hang_count++;
4107         else
4108                 adapter->phy_hang_count = 0;
4109
4110         if (adapter->phy_hang_count > 1) {
4111                 adapter->phy_hang_count = 0;
4112                 schedule_work(&adapter->reset_task);
4113         }
4114 }
4115
4116 /**
4117  * e1000_watchdog - Timer Call-back
4118  * @data: pointer to adapter cast into an unsigned long
4119  **/
4120 static void e1000_watchdog(unsigned long data)
4121 {
4122         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4123
4124         /* Do the rest outside of interrupt context */
4125         schedule_work(&adapter->watchdog_task);
4126
4127         /* TODO: make this use queue_delayed_work() */
4128 }
4129
4130 static void e1000_watchdog_task(struct work_struct *work)
4131 {
4132         struct e1000_adapter *adapter = container_of(work,
4133                                         struct e1000_adapter, watchdog_task);
4134         struct net_device *netdev = adapter->netdev;
4135         struct e1000_mac_info *mac = &adapter->hw.mac;
4136         struct e1000_phy_info *phy = &adapter->hw.phy;
4137         struct e1000_ring *tx_ring = adapter->tx_ring;
4138         struct e1000_hw *hw = &adapter->hw;
4139         u32 link, tctl;
4140         int tx_pending = 0;
4141
4142         link = e1000e_has_link(adapter);
4143         if ((netif_carrier_ok(netdev)) && link) {
4144                 /* Cancel scheduled suspend requests. */
4145                 pm_runtime_resume(netdev->dev.parent);
4146
4147                 e1000e_enable_receives(adapter);
4148                 goto link_up;
4149         }
4150
4151         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4152             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4153                 e1000_update_mng_vlan(adapter);
4154
4155         if (link) {
4156                 if (!netif_carrier_ok(netdev)) {
4157                         bool txb2b = 1;
4158
4159                         /* Cancel scheduled suspend requests. */
4160                         pm_runtime_resume(netdev->dev.parent);
4161
4162                         /* update snapshot of PHY registers on LSC */
4163                         e1000_phy_read_status(adapter);
4164                         mac->ops.get_link_up_info(&adapter->hw,
4165                                                    &adapter->link_speed,
4166                                                    &adapter->link_duplex);
4167                         e1000_print_link_info(adapter);
4168                         /*
4169                          * On supported PHYs, check for duplex mismatch only
4170                          * if link has autonegotiated at 10/100 half
4171                          */
4172                         if ((hw->phy.type == e1000_phy_igp_3 ||
4173                              hw->phy.type == e1000_phy_bm) &&
4174                             (hw->mac.autoneg == true) &&
4175                             (adapter->link_speed == SPEED_10 ||
4176                              adapter->link_speed == SPEED_100) &&
4177                             (adapter->link_duplex == HALF_DUPLEX)) {
4178                                 u16 autoneg_exp;
4179
4180                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4181
4182                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4183                                         e_info("Autonegotiated half duplex but"
4184                                                " link partner cannot autoneg. "
4185                                                " Try forcing full duplex if "
4186                                                "link gets many collisions.\n");
4187                         }
4188
4189                         /* adjust timeout factor according to speed/duplex */
4190                         adapter->tx_timeout_factor = 1;
4191                         switch (adapter->link_speed) {
4192                         case SPEED_10:
4193                                 txb2b = 0;
4194                                 adapter->tx_timeout_factor = 16;
4195                                 break;
4196                         case SPEED_100:
4197                                 txb2b = 0;
4198                                 adapter->tx_timeout_factor = 10;
4199                                 break;
4200                         }
4201
4202                         /*
4203                          * workaround: re-program speed mode bit after
4204                          * link-up event
4205                          */
4206                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4207                             !txb2b) {
4208                                 u32 tarc0;
4209                                 tarc0 = er32(TARC(0));
4210                                 tarc0 &= ~SPEED_MODE_BIT;
4211                                 ew32(TARC(0), tarc0);
4212                         }
4213
4214                         /*
4215                          * disable TSO for pcie and 10/100 speeds, to avoid
4216                          * some hardware issues
4217                          */
4218                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
4219                                 switch (adapter->link_speed) {
4220                                 case SPEED_10:
4221                                 case SPEED_100:
4222                                         e_info("10/100 speed: disabling TSO\n");
4223                                         netdev->features &= ~NETIF_F_TSO;
4224                                         netdev->features &= ~NETIF_F_TSO6;
4225                                         break;
4226                                 case SPEED_1000:
4227                                         netdev->features |= NETIF_F_TSO;
4228                                         netdev->features |= NETIF_F_TSO6;
4229                                         break;
4230                                 default:
4231                                         /* oops */
4232                                         break;
4233                                 }
4234                         }
4235
4236                         /*
4237                          * enable transmits in the hardware, need to do this
4238                          * after setting TARC(0)
4239                          */
4240                         tctl = er32(TCTL);
4241                         tctl |= E1000_TCTL_EN;
4242                         ew32(TCTL, tctl);
4243
4244                         /*
4245                          * Perform any post-link-up configuration before
4246                          * reporting link up.
4247                          */
4248                         if (phy->ops.cfg_on_link_up)
4249                                 phy->ops.cfg_on_link_up(hw);
4250
4251                         netif_carrier_on(netdev);
4252
4253                         if (!test_bit(__E1000_DOWN, &adapter->state))
4254                                 mod_timer(&adapter->phy_info_timer,
4255                                           round_jiffies(jiffies + 2 * HZ));
4256                 }
4257         } else {
4258                 if (netif_carrier_ok(netdev)) {
4259                         adapter->link_speed = 0;
4260                         adapter->link_duplex = 0;
4261                         /* Link status message must follow this format */
4262                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4263                                adapter->netdev->name);
4264                         netif_carrier_off(netdev);
4265                         if (!test_bit(__E1000_DOWN, &adapter->state))
4266                                 mod_timer(&adapter->phy_info_timer,
4267                                           round_jiffies(jiffies + 2 * HZ));
4268
4269                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4270                                 schedule_work(&adapter->reset_task);
4271                         else
4272                                 pm_schedule_suspend(netdev->dev.parent,
4273                                                         LINK_TIMEOUT);
4274                 }
4275         }
4276
4277 link_up:
4278         e1000e_update_stats(adapter);
4279
4280         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4281         adapter->tpt_old = adapter->stats.tpt;
4282         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4283         adapter->colc_old = adapter->stats.colc;
4284
4285         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4286         adapter->gorc_old = adapter->stats.gorc;
4287         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4288         adapter->gotc_old = adapter->stats.gotc;
4289
4290         e1000e_update_adaptive(&adapter->hw);
4291
4292         if (!netif_carrier_ok(netdev)) {
4293                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
4294                                tx_ring->count);
4295                 if (tx_pending) {
4296                         /*
4297                          * We've lost link, so the controller stops DMA,
4298                          * but we've got queued Tx work that's never going
4299                          * to get done, so reset controller to flush Tx.
4300                          * (Do the reset outside of interrupt context).
4301                          */
4302                         adapter->tx_timeout_count++;
4303                         schedule_work(&adapter->reset_task);
4304                         /* return immediately since reset is imminent */
4305                         return;
4306                 }
4307         }
4308
4309         /* Simple mode for Interrupt Throttle Rate (ITR) */
4310         if (adapter->itr_setting == 4) {
4311                 /*
4312                  * Symmetric Tx/Rx gets a reduced ITR=2000;
4313                  * Total asymmetrical Tx or Rx gets ITR=8000;
4314                  * everyone else is between 2000-8000.
4315                  */
4316                 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4317                 u32 dif = (adapter->gotc > adapter->gorc ?
4318                             adapter->gotc - adapter->gorc :
4319                             adapter->gorc - adapter->gotc) / 10000;
4320                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4321
4322                 ew32(ITR, 1000000000 / (itr * 256));
4323         }
4324
4325         /* Cause software interrupt to ensure Rx ring is cleaned */
4326         if (adapter->msix_entries)
4327                 ew32(ICS, adapter->rx_ring->ims_val);
4328         else
4329                 ew32(ICS, E1000_ICS_RXDMT0);
4330
4331         /* Force detection of hung controller every watchdog period */
4332         adapter->detect_tx_hung = 1;
4333
4334         /* flush partial descriptors to memory before detecting tx hang */
4335         if (adapter->flags2 & FLAG2_DMA_BURST) {
4336                 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
4337                 ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
4338                 /*
4339                  * no need to flush the writes because the timeout code does
4340                  * an er32 first thing
4341                  */
4342         }
4343
4344         /*
4345          * With 82571 controllers, LAA may be overwritten due to controller
4346          * reset from the other port. Set the appropriate LAA in RAR[0]
4347          */
4348         if (e1000e_get_laa_state_82571(hw))
4349                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4350
4351         if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
4352                 e1000e_check_82574_phy_workaround(adapter);
4353
4354         /* Reset the timer */
4355         if (!test_bit(__E1000_DOWN, &adapter->state))
4356                 mod_timer(&adapter->watchdog_timer,
4357                           round_jiffies(jiffies + 2 * HZ));
4358 }
4359
4360 #define E1000_TX_FLAGS_CSUM             0x00000001
4361 #define E1000_TX_FLAGS_VLAN             0x00000002
4362 #define E1000_TX_FLAGS_TSO              0x00000004
4363 #define E1000_TX_FLAGS_IPV4             0x00000008
4364 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
4365 #define E1000_TX_FLAGS_VLAN_SHIFT       16
4366
4367 static int e1000_tso(struct e1000_adapter *adapter,
4368                      struct sk_buff *skb)
4369 {
4370         struct e1000_ring *tx_ring = adapter->tx_ring;
4371         struct e1000_context_desc *context_desc;
4372         struct e1000_buffer *buffer_info;
4373         unsigned int i;
4374         u32 cmd_length = 0;
4375         u16 ipcse = 0, tucse, mss;
4376         u8 ipcss, ipcso, tucss, tucso, hdr_len;
4377         int err;
4378
4379         if (!skb_is_gso(skb))
4380                 return 0;
4381
4382         if (skb_header_cloned(skb)) {
4383                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4384                 if (err)
4385                         return err;
4386         }
4387
4388         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4389         mss = skb_shinfo(skb)->gso_size;
4390         if (skb->protocol == htons(ETH_P_IP)) {
4391                 struct iphdr *iph = ip_hdr(skb);
4392                 iph->tot_len = 0;
4393                 iph->check = 0;
4394                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4395                                                          0, IPPROTO_TCP, 0);
4396                 cmd_length = E1000_TXD_CMD_IP;
4397                 ipcse = skb_transport_offset(skb) - 1;
4398         } else if (skb_is_gso_v6(skb)) {
4399                 ipv6_hdr(skb)->payload_len = 0;
4400                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4401                                                        &ipv6_hdr(skb)->daddr,
4402                                                        0, IPPROTO_TCP, 0);
4403                 ipcse = 0;
4404         }
4405         ipcss = skb_network_offset(skb);
4406         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4407         tucss = skb_transport_offset(skb);
4408         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4409         tucse = 0;
4410
4411         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4412                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4413
4414         i = tx_ring->next_to_use;
4415         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4416         buffer_info = &tx_ring->buffer_info[i];
4417
4418         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4419         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4420         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4421         context_desc->upper_setup.tcp_fields.tucss = tucss;
4422         context_desc->upper_setup.tcp_fields.tucso = tucso;
4423         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4424         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4425         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4426         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4427
4428         buffer_info->time_stamp = jiffies;
4429         buffer_info->next_to_watch = i;
4430
4431         i++;
4432         if (i == tx_ring->count)
4433                 i = 0;
4434         tx_ring->next_to_use = i;
4435
4436         return 1;
4437 }
4438
4439 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
4440 {
4441         struct e1000_ring *tx_ring = adapter->tx_ring;
4442         struct e1000_context_desc *context_desc;
4443         struct e1000_buffer *buffer_info;
4444         unsigned int i;
4445         u8 css;
4446         u32 cmd_len = E1000_TXD_CMD_DEXT;
4447         __be16 protocol;
4448
4449         if (skb->ip_summed != CHECKSUM_PARTIAL)
4450                 return 0;
4451
4452         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4453                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4454         else
4455                 protocol = skb->protocol;
4456
4457         switch (protocol) {
4458         case cpu_to_be16(ETH_P_IP):
4459                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4460                         cmd_len |= E1000_TXD_CMD_TCP;
4461                 break;
4462         case cpu_to_be16(ETH_P_IPV6):
4463                 /* XXX not handling all IPV6 headers */
4464                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4465                         cmd_len |= E1000_TXD_CMD_TCP;
4466                 break;
4467         default:
4468                 if (unlikely(net_ratelimit()))
4469                         e_warn("checksum_partial proto=%x!\n",
4470                                be16_to_cpu(protocol));
4471                 break;
4472         }
4473
4474         css = skb_checksum_start_offset(skb);
4475
4476         i = tx_ring->next_to_use;
4477         buffer_info = &tx_ring->buffer_info[i];
4478         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4479
4480         context_desc->lower_setup.ip_config = 0;
4481         context_desc->upper_setup.tcp_fields.tucss = css;
4482         context_desc->upper_setup.tcp_fields.tucso =
4483                                 css + skb->csum_offset;
4484         context_desc->upper_setup.tcp_fields.tucse = 0;
4485         context_desc->tcp_seg_setup.data = 0;
4486         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4487
4488         buffer_info->time_stamp = jiffies;
4489         buffer_info->next_to_watch = i;
4490
4491         i++;
4492         if (i == tx_ring->count)
4493                 i = 0;
4494         tx_ring->next_to_use = i;
4495
4496         return 1;
4497 }
4498
4499 #define E1000_MAX_PER_TXD       8192
4500 #define E1000_MAX_TXD_PWR       12
4501
4502 static int e1000_tx_map(struct e1000_adapter *adapter,
4503                         struct sk_buff *skb, unsigned int first,
4504                         unsigned int max_per_txd, unsigned int nr_frags,
4505                         unsigned int mss)
4506 {
4507         struct e1000_ring *tx_ring = adapter->tx_ring;
4508         struct pci_dev *pdev = adapter->pdev;
4509         struct e1000_buffer *buffer_info;
4510         unsigned int len = skb_headlen(skb);
4511         unsigned int offset = 0, size, count = 0, i;
4512         unsigned int f, bytecount, segs;
4513
4514         i = tx_ring->next_to_use;
4515
4516         while (len) {
4517                 buffer_info = &tx_ring->buffer_info[i];
4518                 size = min(len, max_per_txd);
4519
4520                 buffer_info->length = size;
4521                 buffer_info->time_stamp = jiffies;
4522                 buffer_info->next_to_watch = i;
4523                 buffer_info->dma = dma_map_single(&pdev->dev,
4524                                                   skb->data + offset,
4525                                                   size, DMA_TO_DEVICE);
4526                 buffer_info->mapped_as_page = false;
4527                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4528                         goto dma_error;
4529
4530                 len -= size;
4531                 offset += size;
4532                 count++;
4533
4534                 if (len) {
4535                         i++;
4536                         if (i == tx_ring->count)
4537                                 i = 0;
4538                 }
4539         }
4540
4541         for (f = 0; f < nr_frags; f++) {
4542                 struct skb_frag_struct *frag;
4543
4544                 frag = &skb_shinfo(skb)->frags[f];
4545                 len = frag->size;
4546                 offset = frag->page_offset;
4547
4548                 while (len) {
4549                         i++;
4550                         if (i == tx_ring->count)
4551                                 i = 0;
4552
4553                         buffer_info = &tx_ring->buffer_info[i];
4554                         size = min(len, max_per_txd);
4555
4556                         buffer_info->length = size;
4557                         buffer_info->time_stamp = jiffies;
4558                         buffer_info->next_to_watch = i;
4559                         buffer_info->dma = dma_map_page(&pdev->dev, frag->page,
4560                                                         offset, size,
4561                                                         DMA_TO_DEVICE);
4562                         buffer_info->mapped_as_page = true;
4563                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4564                                 goto dma_error;
4565
4566                         len -= size;
4567                         offset += size;
4568                         count++;
4569                 }
4570         }
4571
4572         segs = skb_shinfo(skb)->gso_segs ?: 1;
4573         /* multiply data chunks by size of headers */
4574         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4575
4576         tx_ring->buffer_info[i].skb = skb;
4577         tx_ring->buffer_info[i].segs = segs;
4578         tx_ring->buffer_info[i].bytecount = bytecount;
4579         tx_ring->buffer_info[first].next_to_watch = i;
4580
4581         return count;
4582
4583 dma_error:
4584         dev_err(&pdev->dev, "TX DMA map failed\n");
4585         buffer_info->dma = 0;
4586         if (count)
4587                 count--;
4588
4589         while (count--) {
4590                 if (i==0)
4591                         i += tx_ring->count;
4592                 i--;
4593                 buffer_info = &tx_ring->buffer_info[i];
4594                 e1000_put_txbuf(adapter, buffer_info);
4595         }
4596
4597         return 0;
4598 }
4599
4600 static void e1000_tx_queue(struct e1000_adapter *adapter,
4601                            int tx_flags, int count)
4602 {
4603         struct e1000_ring *tx_ring = adapter->tx_ring;
4604         struct e1000_tx_desc *tx_desc = NULL;
4605         struct e1000_buffer *buffer_info;
4606         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4607         unsigned int i;
4608
4609         if (tx_flags & E1000_TX_FLAGS_TSO) {
4610                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4611                              E1000_TXD_CMD_TSE;
4612                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4613
4614                 if (tx_flags & E1000_TX_FLAGS_IPV4)
4615                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4616         }
4617
4618         if (tx_flags & E1000_TX_FLAGS_CSUM) {
4619                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4620                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4621         }
4622
4623         if (tx_flags & E1000_TX_FLAGS_VLAN) {
4624                 txd_lower |= E1000_TXD_CMD_VLE;
4625                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4626         }
4627
4628         i = tx_ring->next_to_use;
4629
4630         do {
4631                 buffer_info = &tx_ring->buffer_info[i];
4632                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4633                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4634                 tx_desc->lower.data =
4635                         cpu_to_le32(txd_lower | buffer_info->length);
4636                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4637
4638                 i++;
4639                 if (i == tx_ring->count)
4640                         i = 0;
4641         } while (--count > 0);
4642
4643         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4644
4645         /*
4646          * Force memory writes to complete before letting h/w
4647          * know there are new descriptors to fetch.  (Only
4648          * applicable for weak-ordered memory model archs,
4649          * such as IA-64).
4650          */
4651         wmb();
4652
4653         tx_ring->next_to_use = i;
4654         writel(i, adapter->hw.hw_addr + tx_ring->tail);
4655         /*
4656          * we need this if more than one processor can write to our tail
4657          * at a time, it synchronizes IO on IA64/Altix systems
4658          */
4659         mmiowb();
4660 }
4661
4662 #define MINIMUM_DHCP_PACKET_SIZE 282
4663 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4664                                     struct sk_buff *skb)
4665 {
4666         struct e1000_hw *hw =  &adapter->hw;
4667         u16 length, offset;
4668
4669         if (vlan_tx_tag_present(skb)) {
4670                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4671                     (adapter->hw.mng_cookie.status &
4672                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4673                         return 0;
4674         }
4675
4676         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4677                 return 0;
4678
4679         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4680                 return 0;
4681
4682         {
4683                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4684                 struct udphdr *udp;
4685
4686                 if (ip->protocol != IPPROTO_UDP)
4687                         return 0;
4688
4689                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4690                 if (ntohs(udp->dest) != 67)
4691                         return 0;
4692
4693                 offset = (u8 *)udp + 8 - skb->data;
4694                 length = skb->len - offset;
4695                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4696         }
4697
4698         return 0;
4699 }
4700
4701 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4702 {
4703         struct e1000_adapter *adapter = netdev_priv(netdev);
4704
4705         netif_stop_queue(netdev);
4706         /*
4707          * Herbert's original patch had:
4708          *  smp_mb__after_netif_stop_queue();
4709          * but since that doesn't exist yet, just open code it.
4710          */
4711         smp_mb();
4712
4713         /*
4714          * We need to check again in a case another CPU has just
4715          * made room available.
4716          */
4717         if (e1000_desc_unused(adapter->tx_ring) < size)
4718                 return -EBUSY;
4719
4720         /* A reprieve! */
4721         netif_start_queue(netdev);
4722         ++adapter->restart_queue;
4723         return 0;
4724 }
4725
4726 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4727 {
4728         struct e1000_adapter *adapter = netdev_priv(netdev);
4729
4730         if (e1000_desc_unused(adapter->tx_ring) >= size)
4731                 return 0;
4732         return __e1000_maybe_stop_tx(netdev, size);
4733 }
4734
4735 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4736 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4737                                     struct net_device *netdev)
4738 {
4739         struct e1000_adapter *adapter = netdev_priv(netdev);
4740         struct e1000_ring *tx_ring = adapter->tx_ring;
4741         unsigned int first;
4742         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4743         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4744         unsigned int tx_flags = 0;
4745         unsigned int len = skb_headlen(skb);
4746         unsigned int nr_frags;
4747         unsigned int mss;
4748         int count = 0;
4749         int tso;
4750         unsigned int f;
4751
4752         if (test_bit(__E1000_DOWN, &adapter->state)) {
4753                 dev_kfree_skb_any(skb);
4754                 return NETDEV_TX_OK;
4755         }
4756
4757         if (skb->len <= 0) {
4758                 dev_kfree_skb_any(skb);
4759                 return NETDEV_TX_OK;
4760         }
4761
4762         mss = skb_shinfo(skb)->gso_size;
4763         /*
4764          * The controller does a simple calculation to
4765          * make sure there is enough room in the FIFO before
4766          * initiating the DMA for each buffer.  The calc is:
4767          * 4 = ceil(buffer len/mss).  To make sure we don't
4768          * overrun the FIFO, adjust the max buffer len if mss
4769          * drops.
4770          */
4771         if (mss) {
4772                 u8 hdr_len;
4773                 max_per_txd = min(mss << 2, max_per_txd);
4774                 max_txd_pwr = fls(max_per_txd) - 1;
4775
4776                 /*
4777                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4778                  * points to just header, pull a few bytes of payload from
4779                  * frags into skb->data
4780                  */
4781                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4782                 /*
4783                  * we do this workaround for ES2LAN, but it is un-necessary,
4784                  * avoiding it could save a lot of cycles
4785                  */
4786                 if (skb->data_len && (hdr_len == len)) {
4787                         unsigned int pull_size;
4788
4789                         pull_size = min((unsigned int)4, skb->data_len);
4790                         if (!__pskb_pull_tail(skb, pull_size)) {
4791                                 e_err("__pskb_pull_tail failed.\n");
4792                                 dev_kfree_skb_any(skb);
4793                                 return NETDEV_TX_OK;
4794                         }
4795                         len = skb_headlen(skb);
4796                 }
4797         }
4798
4799         /* reserve a descriptor for the offload context */
4800         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4801                 count++;
4802         count++;
4803
4804         count += TXD_USE_COUNT(len, max_txd_pwr);
4805
4806         nr_frags = skb_shinfo(skb)->nr_frags;
4807         for (f = 0; f < nr_frags; f++)
4808                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4809                                        max_txd_pwr);
4810
4811         if (adapter->hw.mac.tx_pkt_filtering)
4812                 e1000_transfer_dhcp_info(adapter, skb);
4813
4814         /*
4815          * need: count + 2 desc gap to keep tail from touching
4816          * head, otherwise try next time
4817          */
4818         if (e1000_maybe_stop_tx(netdev, count + 2))
4819                 return NETDEV_TX_BUSY;
4820
4821         if (vlan_tx_tag_present(skb)) {
4822                 tx_flags |= E1000_TX_FLAGS_VLAN;
4823                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4824         }
4825
4826         first = tx_ring->next_to_use;
4827
4828         tso = e1000_tso(adapter, skb);
4829         if (tso < 0) {
4830                 dev_kfree_skb_any(skb);
4831                 return NETDEV_TX_OK;
4832         }
4833
4834         if (tso)
4835                 tx_flags |= E1000_TX_FLAGS_TSO;
4836         else if (e1000_tx_csum(adapter, skb))
4837                 tx_flags |= E1000_TX_FLAGS_CSUM;
4838
4839         /*
4840          * Old method was to assume IPv4 packet by default if TSO was enabled.
4841          * 82571 hardware supports TSO capabilities for IPv6 as well...
4842          * no longer assume, we must.
4843          */
4844         if (skb->protocol == htons(ETH_P_IP))
4845                 tx_flags |= E1000_TX_FLAGS_IPV4;
4846
4847         /* if count is 0 then mapping error has occured */
4848         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4849         if (count) {
4850                 e1000_tx_queue(adapter, tx_flags, count);
4851                 /* Make sure there is space in the ring for the next send. */
4852                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4853
4854         } else {
4855                 dev_kfree_skb_any(skb);
4856                 tx_ring->buffer_info[first].time_stamp = 0;
4857                 tx_ring->next_to_use = first;
4858         }
4859
4860         return NETDEV_TX_OK;
4861 }
4862
4863 /**
4864  * e1000_tx_timeout - Respond to a Tx Hang
4865  * @netdev: network interface device structure
4866  **/
4867 static void e1000_tx_timeout(struct net_device *netdev)
4868 {
4869         struct e1000_adapter *adapter = netdev_priv(netdev);
4870
4871         /* Do the reset outside of interrupt context */
4872         adapter->tx_timeout_count++;
4873         schedule_work(&adapter->reset_task);
4874 }
4875
4876 static void e1000_reset_task(struct work_struct *work)
4877 {
4878         struct e1000_adapter *adapter;
4879         adapter = container_of(work, struct e1000_adapter, reset_task);
4880
4881         if (!((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4882               (adapter->flags & FLAG_RX_RESTART_NOW))) {
4883                 e1000e_dump(adapter);
4884                 e_err("Reset adapter\n");
4885         }
4886         e1000e_reinit_locked(adapter);
4887 }
4888
4889 /**
4890  * e1000_get_stats - Get System Network Statistics
4891  * @netdev: network interface device structure
4892  *
4893  * Returns the address of the device statistics structure.
4894  * The statistics are actually updated from the timer callback.
4895  **/
4896 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4897 {
4898         /* only return the current stats */
4899         return &netdev->stats;
4900 }
4901
4902 /**
4903  * e1000_change_mtu - Change the Maximum Transfer Unit
4904  * @netdev: network interface device structure
4905  * @new_mtu: new value for maximum frame size
4906  *
4907  * Returns 0 on success, negative on failure
4908  **/
4909 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4910 {
4911         struct e1000_adapter *adapter = netdev_priv(netdev);
4912         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4913
4914         /* Jumbo frame support */
4915         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
4916             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4917                 e_err("Jumbo Frames not supported.\n");
4918                 return -EINVAL;
4919         }
4920
4921         /* Supported frame sizes */
4922         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4923             (max_frame > adapter->max_hw_frame_size)) {
4924                 e_err("Unsupported MTU setting\n");
4925                 return -EINVAL;
4926         }
4927
4928         /* Jumbo frame workaround on 82579 requires CRC be stripped */
4929         if ((adapter->hw.mac.type == e1000_pch2lan) &&
4930             !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
4931             (new_mtu > ETH_DATA_LEN)) {
4932                 e_err("Jumbo Frames not supported on 82579 when CRC "
4933                       "stripping is disabled.\n");
4934                 return -EINVAL;
4935         }
4936
4937         /* 82573 Errata 17 */
4938         if (((adapter->hw.mac.type == e1000_82573) ||
4939              (adapter->hw.mac.type == e1000_82574)) &&
4940             (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
4941                 adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
4942                 e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
4943         }
4944
4945         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4946                 msleep(1);
4947         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
4948         adapter->max_frame_size = max_frame;
4949         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4950         netdev->mtu = new_mtu;
4951         if (netif_running(netdev))
4952                 e1000e_down(adapter);
4953
4954         /*
4955          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4956          * means we reserve 2 more, this pushes us to allocate from the next
4957          * larger slab size.
4958          * i.e. RXBUFFER_2048 --> size-4096 slab
4959          * However with the new *_jumbo_rx* routines, jumbo receives will use
4960          * fragmented skbs
4961          */
4962
4963         if (max_frame <= 2048)
4964                 adapter->rx_buffer_len = 2048;
4965         else
4966                 adapter->rx_buffer_len = 4096;
4967
4968         /* adjust allocation if LPE protects us, and we aren't using SBP */
4969         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4970              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4971                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4972                                          + ETH_FCS_LEN;
4973
4974         if (netif_running(netdev))
4975                 e1000e_up(adapter);
4976         else
4977                 e1000e_reset(adapter);
4978
4979         clear_bit(__E1000_RESETTING, &adapter->state);
4980
4981         return 0;
4982 }
4983
4984 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4985                            int cmd)
4986 {
4987         struct e1000_adapter *adapter = netdev_priv(netdev);
4988         struct mii_ioctl_data *data = if_mii(ifr);
4989
4990         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4991                 return -EOPNOTSUPP;
4992
4993         switch (cmd) {
4994         case SIOCGMIIPHY:
4995                 data->phy_id = adapter->hw.phy.addr;
4996                 break;
4997         case SIOCGMIIREG:
4998                 e1000_phy_read_status(adapter);
4999
5000                 switch (data->reg_num & 0x1F) {
5001                 case MII_BMCR:
5002                         data->val_out = adapter->phy_regs.bmcr;
5003                         break;
5004                 case MII_BMSR:
5005                         data->val_out = adapter->phy_regs.bmsr;
5006                         break;
5007                 case MII_PHYSID1:
5008                         data->val_out = (adapter->hw.phy.id >> 16);
5009                         break;
5010                 case MII_PHYSID2:
5011                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
5012                         break;
5013                 case MII_ADVERTISE:
5014                         data->val_out = adapter->phy_regs.advertise;
5015                         break;
5016                 case MII_LPA:
5017                         data->val_out = adapter->phy_regs.lpa;
5018                         break;
5019                 case MII_EXPANSION:
5020                         data->val_out = adapter->phy_regs.expansion;
5021                         break;
5022                 case MII_CTRL1000:
5023                         data->val_out = adapter->phy_regs.ctrl1000;
5024                         break;
5025                 case MII_STAT1000:
5026                         data->val_out = adapter->phy_regs.stat1000;
5027                         break;
5028                 case MII_ESTATUS:
5029                         data->val_out = adapter->phy_regs.estatus;
5030                         break;
5031                 default:
5032                         return -EIO;
5033                 }
5034                 break;
5035         case SIOCSMIIREG:
5036         default:
5037                 return -EOPNOTSUPP;
5038         }
5039         return 0;
5040 }
5041
5042 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5043 {
5044         switch (cmd) {
5045         case SIOCGMIIPHY:
5046         case SIOCGMIIREG:
5047         case SIOCSMIIREG:
5048                 return e1000_mii_ioctl(netdev, ifr, cmd);
5049         default:
5050                 return -EOPNOTSUPP;
5051         }
5052 }
5053
5054 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
5055 {
5056         struct e1000_hw *hw = &adapter->hw;
5057         u32 i, mac_reg;
5058         u16 phy_reg;
5059         int retval = 0;
5060
5061         /* copy MAC RARs to PHY RARs */
5062         e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5063
5064         /* copy MAC MTA to PHY MTA */
5065         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5066                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5067                 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
5068                 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
5069         }
5070
5071         /* configure PHY Rx Control register */
5072         e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
5073         mac_reg = er32(RCTL);
5074         if (mac_reg & E1000_RCTL_UPE)
5075                 phy_reg |= BM_RCTL_UPE;
5076         if (mac_reg & E1000_RCTL_MPE)
5077                 phy_reg |= BM_RCTL_MPE;
5078         phy_reg &= ~(BM_RCTL_MO_MASK);
5079         if (mac_reg & E1000_RCTL_MO_3)
5080                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5081                                 << BM_RCTL_MO_SHIFT);
5082         if (mac_reg & E1000_RCTL_BAM)
5083                 phy_reg |= BM_RCTL_BAM;
5084         if (mac_reg & E1000_RCTL_PMCF)
5085                 phy_reg |= BM_RCTL_PMCF;
5086         mac_reg = er32(CTRL);
5087         if (mac_reg & E1000_CTRL_RFCE)
5088                 phy_reg |= BM_RCTL_RFCE;
5089         e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
5090
5091         /* enable PHY wakeup in MAC register */
5092         ew32(WUFC, wufc);
5093         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
5094
5095         /* configure and enable PHY wakeup in PHY registers */
5096         e1e_wphy(&adapter->hw, BM_WUFC, wufc);
5097         e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
5098
5099         /* activate PHY wakeup */
5100         retval = hw->phy.ops.acquire(hw);
5101         if (retval) {
5102                 e_err("Could not acquire PHY\n");
5103                 return retval;
5104         }
5105         e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
5106                                  (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
5107         retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
5108         if (retval) {
5109                 e_err("Could not read PHY page 769\n");
5110                 goto out;
5111         }
5112         phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5113         retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
5114         if (retval)
5115                 e_err("Could not set PHY Host Wakeup bit\n");
5116 out:
5117         hw->phy.ops.release(hw);
5118
5119         return retval;
5120 }
5121
5122 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5123                             bool runtime)
5124 {
5125         struct net_device *netdev = pci_get_drvdata(pdev);
5126         struct e1000_adapter *adapter = netdev_priv(netdev);
5127         struct e1000_hw *hw = &adapter->hw;
5128         u32 ctrl, ctrl_ext, rctl, status;
5129         /* Runtime suspend should only enable wakeup for link changes */
5130         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5131         int retval = 0;
5132
5133         netif_device_detach(netdev);
5134
5135         if (netif_running(netdev)) {
5136                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5137                 e1000e_down(adapter);
5138                 e1000_free_irq(adapter);
5139         }
5140         e1000e_reset_interrupt_capability(adapter);
5141
5142         retval = pci_save_state(pdev);
5143         if (retval)
5144                 return retval;
5145
5146         status = er32(STATUS);
5147         if (status & E1000_STATUS_LU)
5148                 wufc &= ~E1000_WUFC_LNKC;
5149
5150         if (wufc) {
5151                 e1000_setup_rctl(adapter);
5152                 e1000_set_multi(netdev);
5153
5154                 /* turn on all-multi mode if wake on multicast is enabled */
5155                 if (wufc & E1000_WUFC_MC) {
5156                         rctl = er32(RCTL);
5157                         rctl |= E1000_RCTL_MPE;
5158                         ew32(RCTL, rctl);
5159                 }
5160
5161                 ctrl = er32(CTRL);
5162                 /* advertise wake from D3Cold */
5163                 #define E1000_CTRL_ADVD3WUC 0x00100000
5164                 /* phy power management enable */
5165                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5166                 ctrl |= E1000_CTRL_ADVD3WUC;
5167                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5168                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5169                 ew32(CTRL, ctrl);
5170
5171                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5172                     adapter->hw.phy.media_type ==
5173                     e1000_media_type_internal_serdes) {
5174                         /* keep the laser running in D3 */
5175                         ctrl_ext = er32(CTRL_EXT);
5176                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5177                         ew32(CTRL_EXT, ctrl_ext);
5178                 }
5179
5180                 if (adapter->flags & FLAG_IS_ICH)
5181                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
5182
5183                 /* Allow time for pending master requests to run */
5184                 e1000e_disable_pcie_master(&adapter->hw);
5185
5186                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5187                         /* enable wakeup by the PHY */
5188                         retval = e1000_init_phy_wakeup(adapter, wufc);
5189                         if (retval)
5190                                 return retval;
5191                 } else {
5192                         /* enable wakeup by the MAC */
5193                         ew32(WUFC, wufc);
5194                         ew32(WUC, E1000_WUC_PME_EN);
5195                 }
5196         } else {
5197                 ew32(WUC, 0);
5198                 ew32(WUFC, 0);
5199         }
5200
5201         *enable_wake = !!wufc;
5202
5203         /* make sure adapter isn't asleep if manageability is enabled */
5204         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5205             (hw->mac.ops.check_mng_mode(hw)))
5206                 *enable_wake = true;
5207
5208         if (adapter->hw.phy.type == e1000_phy_igp_3)
5209                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5210
5211         /*
5212          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5213          * would have already happened in close and is redundant.
5214          */
5215         e1000_release_hw_control(adapter);
5216
5217         pci_disable_device(pdev);
5218
5219         return 0;
5220 }
5221
5222 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5223 {
5224         if (sleep && wake) {
5225                 pci_prepare_to_sleep(pdev);
5226                 return;
5227         }
5228
5229         pci_wake_from_d3(pdev, wake);
5230         pci_set_power_state(pdev, PCI_D3hot);
5231 }
5232
5233 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5234                                     bool wake)
5235 {
5236         struct net_device *netdev = pci_get_drvdata(pdev);
5237         struct e1000_adapter *adapter = netdev_priv(netdev);
5238
5239         /*
5240          * The pci-e switch on some quad port adapters will report a
5241          * correctable error when the MAC transitions from D0 to D3.  To
5242          * prevent this we need to mask off the correctable errors on the
5243          * downstream port of the pci-e switch.
5244          */
5245         if (adapter->flags & FLAG_IS_QUAD_PORT) {
5246                 struct pci_dev *us_dev = pdev->bus->self;
5247                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
5248                 u16 devctl;
5249
5250                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5251                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5252                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
5253
5254                 e1000_power_off(pdev, sleep, wake);
5255
5256                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5257         } else {
5258                 e1000_power_off(pdev, sleep, wake);
5259         }
5260 }
5261
5262 #ifdef CONFIG_PCIEASPM
5263 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5264 {
5265         pci_disable_link_state(pdev, state);
5266 }
5267 #else
5268 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5269 {
5270         int pos;
5271         u16 reg16;
5272
5273         /*
5274          * Both device and parent should have the same ASPM setting.
5275          * Disable ASPM in downstream component first and then upstream.
5276          */
5277         pos = pci_pcie_cap(pdev);
5278         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5279         reg16 &= ~state;
5280         pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5281
5282         if (!pdev->bus->self)
5283                 return;
5284
5285         pos = pci_pcie_cap(pdev->bus->self);
5286         pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5287         reg16 &= ~state;
5288         pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5289 }
5290 #endif
5291 void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5292 {
5293         dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5294                  (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5295                  (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5296
5297         __e1000e_disable_aspm(pdev, state);
5298 }
5299
5300 #ifdef CONFIG_PM_OPS
5301 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5302 {
5303         return !!adapter->tx_ring->buffer_info;
5304 }
5305
5306 static int __e1000_resume(struct pci_dev *pdev)
5307 {
5308         struct net_device *netdev = pci_get_drvdata(pdev);
5309         struct e1000_adapter *adapter = netdev_priv(netdev);
5310         struct e1000_hw *hw = &adapter->hw;
5311         u32 err;
5312
5313         pci_set_power_state(pdev, PCI_D0);
5314         pci_restore_state(pdev);
5315         pci_save_state(pdev);
5316         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5317                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5318
5319         e1000e_set_interrupt_capability(adapter);
5320         if (netif_running(netdev)) {
5321                 err = e1000_request_irq(adapter);
5322                 if (err)
5323                         return err;
5324         }
5325
5326         e1000e_power_up_phy(adapter);
5327
5328         /* report the system wakeup cause from S3/S4 */
5329         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5330                 u16 phy_data;
5331
5332                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5333                 if (phy_data) {
5334                         e_info("PHY Wakeup cause - %s\n",
5335                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5336                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5337                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5338                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5339                                 phy_data & E1000_WUS_LNKC ? "Link Status "
5340                                 " Change" : "other");
5341                 }
5342                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5343         } else {
5344                 u32 wus = er32(WUS);
5345                 if (wus) {
5346                         e_info("MAC Wakeup cause - %s\n",
5347                                 wus & E1000_WUS_EX ? "Unicast Packet" :
5348                                 wus & E1000_WUS_MC ? "Multicast Packet" :
5349                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
5350                                 wus & E1000_WUS_MAG ? "Magic Packet" :
5351                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
5352                                 "other");
5353                 }
5354                 ew32(WUS, ~0);
5355         }
5356
5357         e1000e_reset(adapter);
5358
5359         e1000_init_manageability_pt(adapter);
5360
5361         if (netif_running(netdev))
5362                 e1000e_up(adapter);
5363
5364         netif_device_attach(netdev);
5365
5366         /*
5367          * If the controller has AMT, do not set DRV_LOAD until the interface
5368          * is up.  For all other cases, let the f/w know that the h/w is now
5369          * under the control of the driver.
5370          */
5371         if (!(adapter->flags & FLAG_HAS_AMT))
5372                 e1000_get_hw_control(adapter);
5373
5374         return 0;
5375 }
5376
5377 #ifdef CONFIG_PM_SLEEP
5378 static int e1000_suspend(struct device *dev)
5379 {
5380         struct pci_dev *pdev = to_pci_dev(dev);
5381         int retval;
5382         bool wake;
5383
5384         retval = __e1000_shutdown(pdev, &wake, false);
5385         if (!retval)
5386                 e1000_complete_shutdown(pdev, true, wake);
5387
5388         return retval;
5389 }
5390
5391 static int e1000_resume(struct device *dev)
5392 {
5393         struct pci_dev *pdev = to_pci_dev(dev);
5394         struct net_device *netdev = pci_get_drvdata(pdev);
5395         struct e1000_adapter *adapter = netdev_priv(netdev);
5396
5397         if (e1000e_pm_ready(adapter))
5398                 adapter->idle_check = true;
5399
5400         return __e1000_resume(pdev);
5401 }
5402 #endif /* CONFIG_PM_SLEEP */
5403
5404 #ifdef CONFIG_PM_RUNTIME
5405 static int e1000_runtime_suspend(struct device *dev)
5406 {
5407         struct pci_dev *pdev = to_pci_dev(dev);
5408         struct net_device *netdev = pci_get_drvdata(pdev);
5409         struct e1000_adapter *adapter = netdev_priv(netdev);
5410
5411         if (e1000e_pm_ready(adapter)) {
5412                 bool wake;
5413
5414                 __e1000_shutdown(pdev, &wake, true);
5415         }
5416
5417         return 0;
5418 }
5419
5420 static int e1000_idle(struct device *dev)
5421 {
5422         struct pci_dev *pdev = to_pci_dev(dev);
5423         struct net_device *netdev = pci_get_drvdata(pdev);
5424         struct e1000_adapter *adapter = netdev_priv(netdev);
5425
5426         if (!e1000e_pm_ready(adapter))
5427                 return 0;
5428
5429         if (adapter->idle_check) {
5430                 adapter->idle_check = false;
5431                 if (!e1000e_has_link(adapter))
5432                         pm_schedule_suspend(dev, MSEC_PER_SEC);
5433         }
5434
5435         return -EBUSY;
5436 }
5437
5438 static int e1000_runtime_resume(struct device *dev)
5439 {
5440         struct pci_dev *pdev = to_pci_dev(dev);
5441         struct net_device *netdev = pci_get_drvdata(pdev);
5442         struct e1000_adapter *adapter = netdev_priv(netdev);
5443
5444         if (!e1000e_pm_ready(adapter))
5445                 return 0;
5446
5447         adapter->idle_check = !dev->power.runtime_auto;
5448         return __e1000_resume(pdev);
5449 }
5450 #endif /* CONFIG_PM_RUNTIME */
5451 #endif /* CONFIG_PM_OPS */
5452
5453 static void e1000_shutdown(struct pci_dev *pdev)
5454 {
5455         bool wake = false;
5456
5457         __e1000_shutdown(pdev, &wake, false);
5458
5459         if (system_state == SYSTEM_POWER_OFF)
5460                 e1000_complete_shutdown(pdev, false, wake);
5461 }
5462
5463 #ifdef CONFIG_NET_POLL_CONTROLLER
5464
5465 static irqreturn_t e1000_intr_msix(int irq, void *data)
5466 {
5467         struct net_device *netdev = data;
5468         struct e1000_adapter *adapter = netdev_priv(netdev);
5469         int vector, msix_irq;
5470
5471         if (adapter->msix_entries) {
5472                 vector = 0;
5473                 msix_irq = adapter->msix_entries[vector].vector;
5474                 disable_irq(msix_irq);
5475                 e1000_intr_msix_rx(msix_irq, netdev);
5476                 enable_irq(msix_irq);
5477
5478                 vector++;
5479                 msix_irq = adapter->msix_entries[vector].vector;
5480                 disable_irq(msix_irq);
5481                 e1000_intr_msix_tx(msix_irq, netdev);
5482                 enable_irq(msix_irq);
5483
5484                 vector++;
5485                 msix_irq = adapter->msix_entries[vector].vector;
5486                 disable_irq(msix_irq);
5487                 e1000_msix_other(msix_irq, netdev);
5488                 enable_irq(msix_irq);
5489         }
5490
5491         return IRQ_HANDLED;
5492 }
5493
5494 /*
5495  * Polling 'interrupt' - used by things like netconsole to send skbs
5496  * without having to re-enable interrupts. It's not called while
5497  * the interrupt routine is executing.
5498  */
5499 static void e1000_netpoll(struct net_device *netdev)
5500 {
5501         struct e1000_adapter *adapter = netdev_priv(netdev);
5502
5503         switch (adapter->int_mode) {
5504         case E1000E_INT_MODE_MSIX:
5505                 e1000_intr_msix(adapter->pdev->irq, netdev);
5506                 break;
5507         case E1000E_INT_MODE_MSI:
5508                 disable_irq(adapter->pdev->irq);
5509                 e1000_intr_msi(adapter->pdev->irq, netdev);
5510                 enable_irq(adapter->pdev->irq);
5511                 break;
5512         default: /* E1000E_INT_MODE_LEGACY */
5513                 disable_irq(adapter->pdev->irq);
5514                 e1000_intr(adapter->pdev->irq, netdev);
5515                 enable_irq(adapter->pdev->irq);
5516                 break;
5517         }
5518 }
5519 #endif
5520
5521 /**
5522  * e1000_io_error_detected - called when PCI error is detected
5523  * @pdev: Pointer to PCI device
5524  * @state: The current pci connection state
5525  *
5526  * This function is called after a PCI bus error affecting
5527  * this device has been detected.
5528  */
5529 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5530                                                 pci_channel_state_t state)
5531 {
5532         struct net_device *netdev = pci_get_drvdata(pdev);
5533         struct e1000_adapter *adapter = netdev_priv(netdev);
5534
5535         netif_device_detach(netdev);
5536
5537         if (state == pci_channel_io_perm_failure)
5538                 return PCI_ERS_RESULT_DISCONNECT;
5539
5540         if (netif_running(netdev))
5541                 e1000e_down(adapter);
5542         pci_disable_device(pdev);
5543
5544         /* Request a slot slot reset. */
5545         return PCI_ERS_RESULT_NEED_RESET;
5546 }
5547
5548 /**
5549  * e1000_io_slot_reset - called after the pci bus has been reset.
5550  * @pdev: Pointer to PCI device
5551  *
5552  * Restart the card from scratch, as if from a cold-boot. Implementation
5553  * resembles the first-half of the e1000_resume routine.
5554  */
5555 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5556 {
5557         struct net_device *netdev = pci_get_drvdata(pdev);
5558         struct e1000_adapter *adapter = netdev_priv(netdev);
5559         struct e1000_hw *hw = &adapter->hw;
5560         int err;
5561         pci_ers_result_t result;
5562
5563         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5564                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5565         err = pci_enable_device_mem(pdev);
5566         if (err) {
5567                 dev_err(&pdev->dev,
5568                         "Cannot re-enable PCI device after reset.\n");
5569                 result = PCI_ERS_RESULT_DISCONNECT;
5570         } else {
5571                 pci_set_master(pdev);
5572                 pdev->state_saved = true;
5573                 pci_restore_state(pdev);
5574
5575                 pci_enable_wake(pdev, PCI_D3hot, 0);
5576                 pci_enable_wake(pdev, PCI_D3cold, 0);
5577
5578                 e1000e_reset(adapter);
5579                 ew32(WUS, ~0);
5580                 result = PCI_ERS_RESULT_RECOVERED;
5581         }
5582
5583         pci_cleanup_aer_uncorrect_error_status(pdev);
5584
5585         return result;
5586 }
5587
5588 /**
5589  * e1000_io_resume - called when traffic can start flowing again.
5590  * @pdev: Pointer to PCI device
5591  *
5592  * This callback is called when the error recovery driver tells us that
5593  * its OK to resume normal operation. Implementation resembles the
5594  * second-half of the e1000_resume routine.
5595  */
5596 static void e1000_io_resume(struct pci_dev *pdev)
5597 {
5598         struct net_device *netdev = pci_get_drvdata(pdev);
5599         struct e1000_adapter *adapter = netdev_priv(netdev);
5600
5601         e1000_init_manageability_pt(adapter);
5602
5603         if (netif_running(netdev)) {
5604                 if (e1000e_up(adapter)) {
5605                         dev_err(&pdev->dev,
5606                                 "can't bring device back up after reset\n");
5607                         return;
5608                 }
5609         }
5610
5611         netif_device_attach(netdev);
5612
5613         /*
5614          * If the controller has AMT, do not set DRV_LOAD until the interface
5615          * is up.  For all other cases, let the f/w know that the h/w is now
5616          * under the control of the driver.
5617          */
5618         if (!(adapter->flags & FLAG_HAS_AMT))
5619                 e1000_get_hw_control(adapter);
5620
5621 }
5622
5623 static void e1000_print_device_info(struct e1000_adapter *adapter)
5624 {
5625         struct e1000_hw *hw = &adapter->hw;
5626         struct net_device *netdev = adapter->netdev;
5627         u32 ret_val;
5628         u8 pba_str[E1000_PBANUM_LENGTH];
5629
5630         /* print bus type/speed/width info */
5631         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
5632                /* bus width */
5633                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5634                 "Width x1"),
5635                /* MAC address */
5636                netdev->dev_addr);
5637         e_info("Intel(R) PRO/%s Network Connection\n",
5638                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5639         ret_val = e1000_read_pba_string_generic(hw, pba_str,
5640                                                 E1000_PBANUM_LENGTH);
5641         if (ret_val)
5642                 strncpy((char *)pba_str, "Unknown", sizeof(pba_str) - 1);
5643         e_info("MAC: %d, PHY: %d, PBA No: %s\n",
5644                hw->mac.type, hw->phy.type, pba_str);
5645 }
5646
5647 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5648 {
5649         struct e1000_hw *hw = &adapter->hw;
5650         int ret_val;
5651         u16 buf = 0;
5652
5653         if (hw->mac.type != e1000_82573)
5654                 return;
5655
5656         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
5657         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
5658                 /* Deep Smart Power Down (DSPD) */
5659                 dev_warn(&adapter->pdev->dev,
5660                          "Warning: detected DSPD enabled in EEPROM\n");
5661         }
5662 }
5663
5664 static const struct net_device_ops e1000e_netdev_ops = {
5665         .ndo_open               = e1000_open,
5666         .ndo_stop               = e1000_close,
5667         .ndo_start_xmit         = e1000_xmit_frame,
5668         .ndo_get_stats          = e1000_get_stats,
5669         .ndo_set_multicast_list = e1000_set_multi,
5670         .ndo_set_mac_address    = e1000_set_mac,
5671         .ndo_change_mtu         = e1000_change_mtu,
5672         .ndo_do_ioctl           = e1000_ioctl,
5673         .ndo_tx_timeout         = e1000_tx_timeout,
5674         .ndo_validate_addr      = eth_validate_addr,
5675
5676         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
5677         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
5678         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
5679 #ifdef CONFIG_NET_POLL_CONTROLLER
5680         .ndo_poll_controller    = e1000_netpoll,
5681 #endif
5682 };
5683
5684 /**
5685  * e1000_probe - Device Initialization Routine
5686  * @pdev: PCI device information struct
5687  * @ent: entry in e1000_pci_tbl
5688  *
5689  * Returns 0 on success, negative on failure
5690  *
5691  * e1000_probe initializes an adapter identified by a pci_dev structure.
5692  * The OS initialization, configuring of the adapter private structure,
5693  * and a hardware reset occur.
5694  **/
5695 static int __devinit e1000_probe(struct pci_dev *pdev,
5696                                  const struct pci_device_id *ent)
5697 {
5698         struct net_device *netdev;
5699         struct e1000_adapter *adapter;
5700         struct e1000_hw *hw;
5701         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
5702         resource_size_t mmio_start, mmio_len;
5703         resource_size_t flash_start, flash_len;
5704
5705         static int cards_found;
5706         int i, err, pci_using_dac;
5707         u16 eeprom_data = 0;
5708         u16 eeprom_apme_mask = E1000_EEPROM_APME;
5709
5710         if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
5711                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5712
5713         err = pci_enable_device_mem(pdev);
5714         if (err)
5715                 return err;
5716
5717         pci_using_dac = 0;
5718         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
5719         if (!err) {
5720                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
5721                 if (!err)
5722                         pci_using_dac = 1;
5723         } else {
5724                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
5725                 if (err) {
5726                         err = dma_set_coherent_mask(&pdev->dev,
5727                                                     DMA_BIT_MASK(32));
5728                         if (err) {
5729                                 dev_err(&pdev->dev, "No usable DMA "
5730                                         "configuration, aborting\n");
5731                                 goto err_dma;
5732                         }
5733                 }
5734         }
5735
5736         err = pci_request_selected_regions_exclusive(pdev,
5737                                           pci_select_bars(pdev, IORESOURCE_MEM),
5738                                           e1000e_driver_name);
5739         if (err)
5740                 goto err_pci_reg;
5741
5742         /* AER (Advanced Error Reporting) hooks */
5743         pci_enable_pcie_error_reporting(pdev);
5744
5745         pci_set_master(pdev);
5746         /* PCI config space info */
5747         err = pci_save_state(pdev);
5748         if (err)
5749                 goto err_alloc_etherdev;
5750
5751         err = -ENOMEM;
5752         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
5753         if (!netdev)
5754                 goto err_alloc_etherdev;
5755
5756         SET_NETDEV_DEV(netdev, &pdev->dev);
5757
5758         netdev->irq = pdev->irq;
5759
5760         pci_set_drvdata(pdev, netdev);
5761         adapter = netdev_priv(netdev);
5762         hw = &adapter->hw;
5763         adapter->netdev = netdev;
5764         adapter->pdev = pdev;
5765         adapter->ei = ei;
5766         adapter->pba = ei->pba;
5767         adapter->flags = ei->flags;
5768         adapter->flags2 = ei->flags2;
5769         adapter->hw.adapter = adapter;
5770         adapter->hw.mac.type = ei->mac;
5771         adapter->max_hw_frame_size = ei->max_hw_frame_size;
5772         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5773
5774         mmio_start = pci_resource_start(pdev, 0);
5775         mmio_len = pci_resource_len(pdev, 0);
5776
5777         err = -EIO;
5778         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5779         if (!adapter->hw.hw_addr)
5780                 goto err_ioremap;
5781
5782         if ((adapter->flags & FLAG_HAS_FLASH) &&
5783             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5784                 flash_start = pci_resource_start(pdev, 1);
5785                 flash_len = pci_resource_len(pdev, 1);
5786                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5787                 if (!adapter->hw.flash_address)
5788                         goto err_flashmap;
5789         }
5790
5791         /* construct the net_device struct */
5792         netdev->netdev_ops              = &e1000e_netdev_ops;
5793         e1000e_set_ethtool_ops(netdev);
5794         netdev->watchdog_timeo          = 5 * HZ;
5795         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5796         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5797
5798         netdev->mem_start = mmio_start;
5799         netdev->mem_end = mmio_start + mmio_len;
5800
5801         adapter->bd_number = cards_found++;
5802
5803         e1000e_check_options(adapter);
5804
5805         /* setup adapter struct */
5806         err = e1000_sw_init(adapter);
5807         if (err)
5808                 goto err_sw_init;
5809
5810         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5811         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5812         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5813
5814         err = ei->get_variants(adapter);
5815         if (err)
5816                 goto err_hw_init;
5817
5818         if ((adapter->flags & FLAG_IS_ICH) &&
5819             (adapter->flags & FLAG_READ_ONLY_NVM))
5820                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5821
5822         hw->mac.ops.get_bus_info(&adapter->hw);
5823
5824         adapter->hw.phy.autoneg_wait_to_complete = 0;
5825
5826         /* Copper options */
5827         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5828                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5829                 adapter->hw.phy.disable_polarity_correction = 0;
5830                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5831         }
5832
5833         if (e1000_check_reset_block(&adapter->hw))
5834                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5835
5836         netdev->features = NETIF_F_SG |
5837                            NETIF_F_HW_CSUM |
5838                            NETIF_F_HW_VLAN_TX |
5839                            NETIF_F_HW_VLAN_RX;
5840
5841         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5842                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5843
5844         netdev->features |= NETIF_F_TSO;
5845         netdev->features |= NETIF_F_TSO6;
5846
5847         netdev->vlan_features |= NETIF_F_TSO;
5848         netdev->vlan_features |= NETIF_F_TSO6;
5849         netdev->vlan_features |= NETIF_F_HW_CSUM;
5850         netdev->vlan_features |= NETIF_F_SG;
5851
5852         if (pci_using_dac) {
5853                 netdev->features |= NETIF_F_HIGHDMA;
5854                 netdev->vlan_features |= NETIF_F_HIGHDMA;
5855         }
5856
5857         if (e1000e_enable_mng_pass_thru(&adapter->hw))
5858                 adapter->flags |= FLAG_MNG_PT_ENABLED;
5859
5860         /*
5861          * before reading the NVM, reset the controller to
5862          * put the device in a known good starting state
5863          */
5864         adapter->hw.mac.ops.reset_hw(&adapter->hw);
5865
5866         /*
5867          * systems with ASPM and others may see the checksum fail on the first
5868          * attempt. Let's give it a few tries
5869          */
5870         for (i = 0;; i++) {
5871                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5872                         break;
5873                 if (i == 2) {
5874                         e_err("The NVM Checksum Is Not Valid\n");
5875                         err = -EIO;
5876                         goto err_eeprom;
5877                 }
5878         }
5879
5880         e1000_eeprom_checks(adapter);
5881
5882         /* copy the MAC address */
5883         if (e1000e_read_mac_addr(&adapter->hw))
5884                 e_err("NVM Read Error while reading MAC address\n");
5885
5886         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5887         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5888
5889         if (!is_valid_ether_addr(netdev->perm_addr)) {
5890                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5891                 err = -EIO;
5892                 goto err_eeprom;
5893         }
5894
5895         init_timer(&adapter->watchdog_timer);
5896         adapter->watchdog_timer.function = e1000_watchdog;
5897         adapter->watchdog_timer.data = (unsigned long) adapter;
5898
5899         init_timer(&adapter->phy_info_timer);
5900         adapter->phy_info_timer.function = e1000_update_phy_info;
5901         adapter->phy_info_timer.data = (unsigned long) adapter;
5902
5903         INIT_WORK(&adapter->reset_task, e1000_reset_task);
5904         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
5905         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
5906         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
5907         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
5908         INIT_WORK(&adapter->led_blink_task, e1000e_led_blink_task);
5909
5910         /* Initialize link parameters. User can change them with ethtool */
5911         adapter->hw.mac.autoneg = 1;
5912         adapter->fc_autoneg = 1;
5913         adapter->hw.fc.requested_mode = e1000_fc_default;
5914         adapter->hw.fc.current_mode = e1000_fc_default;
5915         adapter->hw.phy.autoneg_advertised = 0x2f;
5916
5917         /* ring size defaults */
5918         adapter->rx_ring->count = 256;
5919         adapter->tx_ring->count = 256;
5920
5921         /*
5922          * Initial Wake on LAN setting - If APM wake is enabled in
5923          * the EEPROM, enable the ACPI Magic Packet filter
5924          */
5925         if (adapter->flags & FLAG_APME_IN_WUC) {
5926                 /* APME bit in EEPROM is mapped to WUC.APME */
5927                 eeprom_data = er32(WUC);
5928                 eeprom_apme_mask = E1000_WUC_APME;
5929                 if (eeprom_data & E1000_WUC_PHY_WAKE)
5930                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
5931         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5932                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5933                     (adapter->hw.bus.func == 1))
5934                         e1000_read_nvm(&adapter->hw,
5935                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5936                 else
5937                         e1000_read_nvm(&adapter->hw,
5938                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5939         }
5940
5941         /* fetch WoL from EEPROM */
5942         if (eeprom_data & eeprom_apme_mask)
5943                 adapter->eeprom_wol |= E1000_WUFC_MAG;
5944
5945         /*
5946          * now that we have the eeprom settings, apply the special cases
5947          * where the eeprom may be wrong or the board simply won't support
5948          * wake on lan on a particular port
5949          */
5950         if (!(adapter->flags & FLAG_HAS_WOL))
5951                 adapter->eeprom_wol = 0;
5952
5953         /* initialize the wol settings based on the eeprom settings */
5954         adapter->wol = adapter->eeprom_wol;
5955         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5956
5957         /* save off EEPROM version number */
5958         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
5959
5960         /* reset the hardware with the new settings */
5961         e1000e_reset(adapter);
5962
5963         /*
5964          * If the controller has AMT, do not set DRV_LOAD until the interface
5965          * is up.  For all other cases, let the f/w know that the h/w is now
5966          * under the control of the driver.
5967          */
5968         if (!(adapter->flags & FLAG_HAS_AMT))
5969                 e1000_get_hw_control(adapter);
5970
5971         strncpy(netdev->name, "eth%d", sizeof(netdev->name) - 1);
5972         err = register_netdev(netdev);
5973         if (err)
5974                 goto err_register;
5975
5976         /* carrier off reporting is important to ethtool even BEFORE open */
5977         netif_carrier_off(netdev);
5978
5979         e1000_print_device_info(adapter);
5980
5981         if (pci_dev_run_wake(pdev))
5982                 pm_runtime_put_noidle(&pdev->dev);
5983
5984         return 0;
5985
5986 err_register:
5987         if (!(adapter->flags & FLAG_HAS_AMT))
5988                 e1000_release_hw_control(adapter);
5989 err_eeprom:
5990         if (!e1000_check_reset_block(&adapter->hw))
5991                 e1000_phy_hw_reset(&adapter->hw);
5992 err_hw_init:
5993
5994         kfree(adapter->tx_ring);
5995         kfree(adapter->rx_ring);
5996 err_sw_init:
5997         if (adapter->hw.flash_address)
5998                 iounmap(adapter->hw.flash_address);
5999         e1000e_reset_interrupt_capability(adapter);
6000 err_flashmap:
6001         iounmap(adapter->hw.hw_addr);
6002 err_ioremap:
6003         free_netdev(netdev);
6004 err_alloc_etherdev:
6005         pci_release_selected_regions(pdev,
6006                                      pci_select_bars(pdev, IORESOURCE_MEM));
6007 err_pci_reg:
6008 err_dma:
6009         pci_disable_device(pdev);
6010         return err;
6011 }
6012
6013 /**
6014  * e1000_remove - Device Removal Routine
6015  * @pdev: PCI device information struct
6016  *
6017  * e1000_remove is called by the PCI subsystem to alert the driver
6018  * that it should release a PCI device.  The could be caused by a
6019  * Hot-Plug event, or because the driver is going to be removed from
6020  * memory.
6021  **/
6022 static void __devexit e1000_remove(struct pci_dev *pdev)
6023 {
6024         struct net_device *netdev = pci_get_drvdata(pdev);
6025         struct e1000_adapter *adapter = netdev_priv(netdev);
6026         bool down = test_bit(__E1000_DOWN, &adapter->state);
6027
6028         /*
6029          * The timers may be rescheduled, so explicitly disable them
6030          * from being rescheduled.
6031          */
6032         if (!down)
6033                 set_bit(__E1000_DOWN, &adapter->state);
6034         del_timer_sync(&adapter->watchdog_timer);
6035         del_timer_sync(&adapter->phy_info_timer);
6036
6037         cancel_work_sync(&adapter->reset_task);
6038         cancel_work_sync(&adapter->watchdog_task);
6039         cancel_work_sync(&adapter->downshift_task);
6040         cancel_work_sync(&adapter->update_phy_task);
6041         cancel_work_sync(&adapter->led_blink_task);
6042         cancel_work_sync(&adapter->print_hang_task);
6043
6044         if (!(netdev->flags & IFF_UP))
6045                 e1000_power_down_phy(adapter);
6046
6047         /* Don't lie to e1000_close() down the road. */
6048         if (!down)
6049                 clear_bit(__E1000_DOWN, &adapter->state);
6050         unregister_netdev(netdev);
6051
6052         if (pci_dev_run_wake(pdev))
6053                 pm_runtime_get_noresume(&pdev->dev);
6054
6055         /*
6056          * Release control of h/w to f/w.  If f/w is AMT enabled, this
6057          * would have already happened in close and is redundant.
6058          */
6059         e1000_release_hw_control(adapter);
6060
6061         e1000e_reset_interrupt_capability(adapter);
6062         kfree(adapter->tx_ring);
6063         kfree(adapter->rx_ring);
6064
6065         iounmap(adapter->hw.hw_addr);
6066         if (adapter->hw.flash_address)
6067                 iounmap(adapter->hw.flash_address);
6068         pci_release_selected_regions(pdev,
6069                                      pci_select_bars(pdev, IORESOURCE_MEM));
6070
6071         free_netdev(netdev);
6072
6073         /* AER disable */
6074         pci_disable_pcie_error_reporting(pdev);
6075
6076         pci_disable_device(pdev);
6077 }
6078
6079 /* PCI Error Recovery (ERS) */
6080 static struct pci_error_handlers e1000_err_handler = {
6081         .error_detected = e1000_io_error_detected,
6082         .slot_reset = e1000_io_slot_reset,
6083         .resume = e1000_io_resume,
6084 };
6085
6086 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
6087         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
6088         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
6089         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
6090         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
6091         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
6092         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
6093         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
6094         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
6095         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
6096
6097         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
6098         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
6099         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
6100         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
6101
6102         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
6103         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
6104         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
6105
6106         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
6107         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
6108         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
6109
6110         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
6111           board_80003es2lan },
6112         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
6113           board_80003es2lan },
6114         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
6115           board_80003es2lan },
6116         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
6117           board_80003es2lan },
6118
6119         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
6120         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
6121         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
6122         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
6123         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
6124         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
6125         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
6126         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
6127
6128         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
6129         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
6130         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
6131         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
6132         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
6133         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
6134         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
6135         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
6136         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
6137
6138         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
6139         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
6140         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
6141
6142         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
6143         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
6144         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
6145
6146         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
6147         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
6148         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
6149         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
6150
6151         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
6152         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
6153
6154         { }     /* terminate list */
6155 };
6156 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
6157
6158 #ifdef CONFIG_PM_OPS
6159 static const struct dev_pm_ops e1000_pm_ops = {
6160         SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
6161         SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
6162                                 e1000_runtime_resume, e1000_idle)
6163 };
6164 #endif
6165
6166 /* PCI Device API Driver */
6167 static struct pci_driver e1000_driver = {
6168         .name     = e1000e_driver_name,
6169         .id_table = e1000_pci_tbl,
6170         .probe    = e1000_probe,
6171         .remove   = __devexit_p(e1000_remove),
6172 #ifdef CONFIG_PM_OPS
6173         .driver.pm = &e1000_pm_ops,
6174 #endif
6175         .shutdown = e1000_shutdown,
6176         .err_handler = &e1000_err_handler
6177 };
6178
6179 /**
6180  * e1000_init_module - Driver Registration Routine
6181  *
6182  * e1000_init_module is the first routine called when the driver is
6183  * loaded. All it does is register with the PCI subsystem.
6184  **/
6185 static int __init e1000_init_module(void)
6186 {
6187         int ret;
6188         pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
6189                 e1000e_driver_version);
6190         pr_info("Copyright (c) 1999 - 2010 Intel Corporation.\n");
6191         ret = pci_register_driver(&e1000_driver);
6192
6193         return ret;
6194 }
6195 module_init(e1000_init_module);
6196
6197 /**
6198  * e1000_exit_module - Driver Exit Cleanup Routine
6199  *
6200  * e1000_exit_module is called just before the driver is removed
6201  * from memory.
6202  **/
6203 static void __exit e1000_exit_module(void)
6204 {
6205         pci_unregister_driver(&e1000_driver);
6206 }
6207 module_exit(e1000_exit_module);
6208
6209
6210 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
6211 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
6212 MODULE_LICENSE("GPL");
6213 MODULE_VERSION(DRV_VERSION);
6214
6215 /* e1000_main.c */