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_VERSION "1.0.2-k4"
  56 char e1000e_driver_name[] = "e1000e";
  57 const char e1000e_driver_version[] = DRV_VERSION;
  58
  59 static const struct e1000_info *e1000_info_tbl[] = {
  60         [board_82571]           = &e1000_82571_info,
  61         [board_82572]           = &e1000_82572_info,
  62         [board_82573]           = &e1000_82573_info,
  63         [board_82574]           = &e1000_82574_info,
  64         [board_82583]           = &e1000_82583_info,
  65         [board_80003es2lan]     = &e1000_es2_info,
  66         [board_ich8lan]         = &e1000_ich8_info,
  67         [board_ich9lan]         = &e1000_ich9_info,
  68         [board_ich10lan]        = &e1000_ich10_info,
  69         [board_pchlan]          = &e1000_pch_info,
  70         [board_pch2lan]         = &e1000_pch2_info,
  71 };
  72
  73 struct e1000_reg_info {
  74         u32 ofs;
  75         char *name;
  76 };
  77
  78 #define E1000_RDFH      0x02410 /* Rx Data FIFO Head - RW */
  79 #define E1000_RDFT      0x02418 /* Rx Data FIFO Tail - RW */
  80 #define E1000_RDFHS     0x02420 /* Rx Data FIFO Head Saved - RW */
  81 #define E1000_RDFTS     0x02428 /* Rx Data FIFO Tail Saved - RW */
  82 #define E1000_RDFPC     0x02430 /* Rx Data FIFO Packet Count - RW */
  83
  84 #define E1000_TDFH      0x03410 /* Tx Data FIFO Head - RW */
  85 #define E1000_TDFT      0x03418 /* Tx Data FIFO Tail - RW */
  86 #define E1000_TDFHS     0x03420 /* Tx Data FIFO Head Saved - RW */
  87 #define E1000_TDFTS     0x03428 /* Tx Data FIFO Tail Saved - RW */
  88 #define E1000_TDFPC     0x03430 /* Tx Data FIFO Packet Count - RW */
  89
  90 static const struct e1000_reg_info e1000_reg_info_tbl[] = {
  91
  92         /* General Registers */
  93         {E1000_CTRL, "CTRL"},
  94         {E1000_STATUS, "STATUS"},
  95         {E1000_CTRL_EXT, "CTRL_EXT"},
  96
  97         /* Interrupt Registers */
  98         {E1000_ICR, "ICR"},
  99
 100         /* RX Registers */
 101         {E1000_RCTL, "RCTL"},
 102         {E1000_RDLEN, "RDLEN"},
 103         {E1000_RDH, "RDH"},
 104         {E1000_RDT, "RDT"},
 105         {E1000_RDTR, "RDTR"},
 106         {E1000_RXDCTL(0), "RXDCTL"},
 107         {E1000_ERT, "ERT"},
 108         {E1000_RDBAL, "RDBAL"},
 109         {E1000_RDBAH, "RDBAH"},
 110         {E1000_RDFH, "RDFH"},
 111         {E1000_RDFT, "RDFT"},
 112         {E1000_RDFHS, "RDFHS"},
 113         {E1000_RDFTS, "RDFTS"},
 114         {E1000_RDFPC, "RDFPC"},
 115
 116         /* TX Registers */
 117         {E1000_TCTL, "TCTL"},
 118         {E1000_TDBAL, "TDBAL"},
 119         {E1000_TDBAH, "TDBAH"},
 120         {E1000_TDLEN, "TDLEN"},
 121         {E1000_TDH, "TDH"},
 122         {E1000_TDT, "TDT"},
 123         {E1000_TIDV, "TIDV"},
 124         {E1000_TXDCTL(0), "TXDCTL"},
 125         {E1000_TADV, "TADV"},
 126         {E1000_TARC(0), "TARC"},
 127         {E1000_TDFH, "TDFH"},
 128         {E1000_TDFT, "TDFT"},
 129         {E1000_TDFHS, "TDFHS"},
 130         {E1000_TDFTS, "TDFTS"},
 131         {E1000_TDFPC, "TDFPC"},
 132
 133         /* List Terminator */
 134         {}
 135 };
 136
 137 /*
 138  * e1000_regdump - register printout routine
 139  */
 140 static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
 141 {
 142         int n = 0;
 143         char rname[16];
 144         u32 regs[8];
 145
 146         switch (reginfo->ofs) {
 147         case E1000_RXDCTL(0):
 148                 for (n = 0; n < 2; n++)
 149                         regs[n] = __er32(hw, E1000_RXDCTL(n));
 150                 break;
 151         case E1000_TXDCTL(0):
 152                 for (n = 0; n < 2; n++)
 153                         regs[n] = __er32(hw, E1000_TXDCTL(n));
 154                 break;
 155         case E1000_TARC(0):
 156                 for (n = 0; n < 2; n++)
 157                         regs[n] = __er32(hw, E1000_TARC(n));
 158                 break;
 159         default:
 160                 printk(KERN_INFO "%-15s %08x\n",
 161                         reginfo->name, __er32(hw, reginfo->ofs));
 162                 return;
 163         }
 164
 165         snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
 166         printk(KERN_INFO "%-15s ", rname);
 167         for (n = 0; n < 2; n++)
 168                 printk(KERN_CONT "%08x ", regs[n]);
 169         printk(KERN_CONT "\n");
 170 }
 171
 172
 173 /*
 174  * e1000e_dump - Print registers, tx-ring and rx-ring
 175  */
 176 static void e1000e_dump(struct e1000_adapter *adapter)
 177 {
 178         struct net_device *netdev = adapter->netdev;
 179         struct e1000_hw *hw = &adapter->hw;
 180         struct e1000_reg_info *reginfo;
 181         struct e1000_ring *tx_ring = adapter->tx_ring;
 182         struct e1000_tx_desc *tx_desc;
 183         struct my_u0 { u64 a; u64 b; } *u0;
 184         struct e1000_buffer *buffer_info;
 185         struct e1000_ring *rx_ring = adapter->rx_ring;
 186         union e1000_rx_desc_packet_split *rx_desc_ps;
 187         struct e1000_rx_desc *rx_desc;
 188         struct my_u1 { u64 a; u64 b; u64 c; u64 d; } *u1;
 189         u32 staterr;
 190         int i = 0;
 191
 192         if (!netif_msg_hw(adapter))
 193                 return;
 194
 195         /* Print netdevice Info */
 196         if (netdev) {
 197                 dev_info(&adapter->pdev->dev, "Net device Info\n");
 198                 printk(KERN_INFO "Device Name     state            "
 199                         "trans_start      last_rx\n");
 200                 printk(KERN_INFO "%-15s %016lX %016lX %016lX\n",
 201                         netdev->name,
 202                         netdev->state,
 203                         netdev->trans_start,
 204                         netdev->last_rx);
 205         }
 206
 207         /* Print Registers */
 208         dev_info(&adapter->pdev->dev, "Register Dump\n");
 209         printk(KERN_INFO " Register Name   Value\n");
 210         for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
 211              reginfo->name; reginfo++) {
 212                 e1000_regdump(hw, reginfo);
 213         }
 214
 215         /* Print TX Ring Summary */
 216         if (!netdev || !netif_running(netdev))
 217                 goto exit;
 218
 219         dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
 220         printk(KERN_INFO "Queue [NTU] [NTC] [bi(ntc)->dma  ]"
 221                 " leng ntw timestamp\n");
 222         buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
 223         printk(KERN_INFO " %5d %5X %5X %016llX %04X %3X %016llX\n",
 224                 0, tx_ring->next_to_use, tx_ring->next_to_clean,
 225                 (u64)buffer_info->dma,
 226                 buffer_info->length,
 227                 buffer_info->next_to_watch,
 228                 (u64)buffer_info->time_stamp);
 229
 230         /* Print TX Rings */
 231         if (!netif_msg_tx_done(adapter))
 232                 goto rx_ring_summary;
 233
 234         dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
 235
 236         /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
 237          *
 238          * Legacy Transmit Descriptor
 239          *   +--------------------------------------------------------------+
 240          * 0 |         Buffer Address [63:0] (Reserved on Write Back)       |
 241          *   +--------------------------------------------------------------+
 242          * 8 | Special  |    CSS     | Status |  CMD    |  CSO   |  Length  |
 243          *   +--------------------------------------------------------------+
 244          *   63       48 47        36 35    32 31     24 23    16 15        0
 245          *
 246          * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
 247          *   63      48 47    40 39       32 31             16 15    8 7      0
 248          *   +----------------------------------------------------------------+
 249          * 0 |  TUCSE  | TUCS0  |   TUCSS   |     IPCSE       | IPCS0 | IPCSS |
 250          *   +----------------------------------------------------------------+
 251          * 8 |   MSS   | HDRLEN | RSV | STA | TUCMD | DTYP |      PAYLEN      |
 252          *   +----------------------------------------------------------------+
 253          *   63      48 47    40 39 36 35 32 31   24 23  20 19                0
 254          *
 255          * Extended Data Descriptor (DTYP=0x1)
 256          *   +----------------------------------------------------------------+
 257          * 0 |                     Buffer Address [63:0]                      |
 258          *   +----------------------------------------------------------------+
 259          * 8 | VLAN tag |  POPTS  | Rsvd | Status | Command | DTYP |  DTALEN  |
 260          *   +----------------------------------------------------------------+
 261          *   63       48 47     40 39  36 35    32 31     24 23  20 19        0
 262          */
 263         printk(KERN_INFO "Tl[desc]     [address 63:0  ] [SpeCssSCmCsLen]"
 264                 " [bi->dma       ] leng  ntw timestamp        bi->skb "
 265                 "<-- Legacy format\n");
 266         printk(KERN_INFO "Tc[desc]     [Ce CoCsIpceCoS] [MssHlRSCm0Plen]"
 267                 " [bi->dma       ] leng  ntw timestamp        bi->skb "
 268                 "<-- Ext Context format\n");
 269         printk(KERN_INFO "Td[desc]     [address 63:0  ] [VlaPoRSCm1Dlen]"
 270                 " [bi->dma       ] leng  ntw timestamp        bi->skb "
 271                 "<-- Ext Data format\n");
 272         for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
 273                 tx_desc = E1000_TX_DESC(*tx_ring, i);
 274                 buffer_info = &tx_ring->buffer_info[i];
 275                 u0 = (struct my_u0 *)tx_desc;
 276                 printk(KERN_INFO "T%c[0x%03X]    %016llX %016llX %016llX "
 277                         "%04X  %3X %016llX %p",
 278                        (!(le64_to_cpu(u0->b) & (1<<29)) ? 'l' :
 279                         ((le64_to_cpu(u0->b) & (1<<20)) ? 'd' : 'c')), i,
 280                        le64_to_cpu(u0->a), le64_to_cpu(u0->b),
 281                        (u64)buffer_info->dma, buffer_info->length,
 282                        buffer_info->next_to_watch, (u64)buffer_info->time_stamp,
 283                        buffer_info->skb);
 284                 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
 285                         printk(KERN_CONT " NTC/U\n");
 286                 else if (i == tx_ring->next_to_use)
 287                         printk(KERN_CONT " NTU\n");
 288                 else if (i == tx_ring->next_to_clean)
 289                         printk(KERN_CONT " NTC\n");
 290                 else
 291                         printk(KERN_CONT "\n");
 292
 293                 if (netif_msg_pktdata(adapter) && buffer_info->dma != 0)
 294                         print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
 295                                         16, 1, phys_to_virt(buffer_info->dma),
 296                                         buffer_info->length, true);
 297         }
 298
 299         /* Print RX Rings Summary */
 300 rx_ring_summary:
 301         dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
 302         printk(KERN_INFO "Queue [NTU] [NTC]\n");
 303         printk(KERN_INFO " %5d %5X %5X\n", 0,
 304                 rx_ring->next_to_use, rx_ring->next_to_clean);
 305
 306         /* Print RX Rings */
 307         if (!netif_msg_rx_status(adapter))
 308                 goto exit;
 309
 310         dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
 311         switch (adapter->rx_ps_pages) {
 312         case 1:
 313         case 2:
 314         case 3:
 315                 /* [Extended] Packet Split Receive Descriptor Format
 316                  *
 317                  *    +-----------------------------------------------------+
 318                  *  0 |                Buffer Address 0 [63:0]              |
 319                  *    +-----------------------------------------------------+
 320                  *  8 |                Buffer Address 1 [63:0]              |
 321                  *    +-----------------------------------------------------+
 322                  * 16 |                Buffer Address 2 [63:0]              |
 323                  *    +-----------------------------------------------------+
 324                  * 24 |                Buffer Address 3 [63:0]              |
 325                  *    +-----------------------------------------------------+
 326                  */
 327                 printk(KERN_INFO "R  [desc]      [buffer 0 63:0 ] "
 328                         "[buffer 1 63:0 ] "
 329                        "[buffer 2 63:0 ] [buffer 3 63:0 ] [bi->dma       ] "
 330                        "[bi->skb] <-- Ext Pkt Split format\n");
 331                 /* [Extended] Receive Descriptor (Write-Back) Format
 332                  *
 333                  *   63       48 47    32 31     13 12    8 7    4 3        0
 334                  *   +------------------------------------------------------+
 335                  * 0 | Packet   | IP     |  Rsvd   | MRQ   | Rsvd | MRQ RSS |
 336                  *   | Checksum | Ident  |         | Queue |      |  Type   |
 337                  *   +------------------------------------------------------+
 338                  * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 339                  *   +------------------------------------------------------+
 340                  *   63       48 47    32 31            20 19               0
 341                  */
 342                 printk(KERN_INFO "RWB[desc]      [ck ipid mrqhsh] "
 343                         "[vl   l0 ee  es] "
 344                        "[ l3  l2  l1 hs] [reserved      ] ---------------- "
 345                        "[bi->skb] <-- Ext Rx Write-Back format\n");
 346                 for (i = 0; i < rx_ring->count; i++) {
 347                         buffer_info = &rx_ring->buffer_info[i];
 348                         rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
 349                         u1 = (struct my_u1 *)rx_desc_ps;
 350                         staterr =
 351                                 le32_to_cpu(rx_desc_ps->wb.middle.status_error);
 352                         if (staterr & E1000_RXD_STAT_DD) {
 353                                 /* Descriptor Done */
 354                                 printk(KERN_INFO "RWB[0x%03X]     %016llX "
 355                                         "%016llX %016llX %016llX "
 356                                         "---------------- %p", i,
 357                                         le64_to_cpu(u1->a),
 358                                         le64_to_cpu(u1->b),
 359                                         le64_to_cpu(u1->c),
 360                                         le64_to_cpu(u1->d),
 361                                         buffer_info->skb);
 362                         } else {
 363                                 printk(KERN_INFO "R  [0x%03X]     %016llX "
 364                                         "%016llX %016llX %016llX %016llX %p", i,
 365                                         le64_to_cpu(u1->a),
 366                                         le64_to_cpu(u1->b),
 367                                         le64_to_cpu(u1->c),
 368                                         le64_to_cpu(u1->d),
 369                                         (u64)buffer_info->dma,
 370                                         buffer_info->skb);
 371
 372                                 if (netif_msg_pktdata(adapter))
 373                                         print_hex_dump(KERN_INFO, "",
 374                                                 DUMP_PREFIX_ADDRESS, 16, 1,
 375                                                 phys_to_virt(buffer_info->dma),
 376                                                 adapter->rx_ps_bsize0, true);
 377                         }
 378
 379                         if (i == rx_ring->next_to_use)
 380                                 printk(KERN_CONT " NTU\n");
 381                         else if (i == rx_ring->next_to_clean)
 382                                 printk(KERN_CONT " NTC\n");
 383                         else
 384                                 printk(KERN_CONT "\n");
 385                 }
 386                 break;
 387         default:
 388         case 0:
 389                 /* Legacy Receive Descriptor Format
 390                  *
 391                  * +-----------------------------------------------------+
 392                  * |                Buffer Address [63:0]                |
 393                  * +-----------------------------------------------------+
 394                  * | VLAN Tag | Errors | Status 0 | Packet csum | Length |
 395                  * +-----------------------------------------------------+
 396                  * 63       48 47    40 39      32 31         16 15      0
 397                  */
 398                 printk(KERN_INFO "Rl[desc]     [address 63:0  ] "
 399                         "[vl er S cks ln] [bi->dma       ] [bi->skb] "
 400                         "<-- Legacy format\n");
 401                 for (i = 0; rx_ring->desc && (i < rx_ring->count); i++) {
 402                         rx_desc = E1000_RX_DESC(*rx_ring, i);
 403                         buffer_info = &rx_ring->buffer_info[i];
 404                         u0 = (struct my_u0 *)rx_desc;
 405                         printk(KERN_INFO "Rl[0x%03X]    %016llX %016llX "
 406                                 "%016llX %p",
 407                                 i, le64_to_cpu(u0->a), le64_to_cpu(u0->b),
 408                                 (u64)buffer_info->dma, buffer_info->skb);
 409                         if (i == rx_ring->next_to_use)
 410                                 printk(KERN_CONT " NTU\n");
 411                         else if (i == rx_ring->next_to_clean)
 412                                 printk(KERN_CONT " NTC\n");
 413                         else
 414                                 printk(KERN_CONT "\n");
 415
 416                         if (netif_msg_pktdata(adapter))
 417                                 print_hex_dump(KERN_INFO, "",
 418                                         DUMP_PREFIX_ADDRESS,
 419                                         16, 1, phys_to_virt(buffer_info->dma),
 420                                         adapter->rx_buffer_len, true);
 421                 }
 422         }
 423
 424 exit:
 425         return;
 426 }
 427
 428 /**
 429  * e1000_desc_unused - calculate if we have unused descriptors
 430  **/
 431 static int e1000_desc_unused(struct e1000_ring *ring)
 432 {
 433         if (ring->next_to_clean > ring->next_to_use)
 434                 return ring->next_to_clean - ring->next_to_use - 1;
 435
 436         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
 437 }
 438
 439 /**
 440  * e1000_receive_skb - helper function to handle Rx indications
 441  * @adapter: board private structure
 442  * @status: descriptor status field as written by hardware
 443  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
 444  * @skb: pointer to sk_buff to be indicated to stack
 445  **/
 446 static void e1000_receive_skb(struct e1000_adapter *adapter,
 447                               struct net_device *netdev,
 448                               struct sk_buff *skb,
 449                               u8 status, __le16 vlan)
 450 {
 451         skb->protocol = eth_type_trans(skb, netdev);
 452
 453         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
 454                 vlan_gro_receive(&adapter->napi, adapter->vlgrp,
 455                                  le16_to_cpu(vlan), skb);
 456         else
 457                 napi_gro_receive(&adapter->napi, skb);
 458 }
 459
 460 /**
 461  * e1000_rx_checksum - Receive Checksum Offload for 82543
 462  * @adapter:     board private structure
 463  * @status_err:  receive descriptor status and error fields
 464  * @csum:       receive descriptor csum field
 465  * @sk_buff:     socket buffer with received data
 466  **/
 467 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
 468                               u32 csum, struct sk_buff *skb)
 469 {
 470         u16 status = (u16)status_err;
 471         u8 errors = (u8)(status_err >> 24);
 472         skb->ip_summed = CHECKSUM_NONE;
 473
 474         /* Ignore Checksum bit is set */
 475         if (status & E1000_RXD_STAT_IXSM)
 476                 return;
 477         /* TCP/UDP checksum error bit is set */
 478         if (errors & E1000_RXD_ERR_TCPE) {
 479                 /* let the stack verify checksum errors */
 480                 adapter->hw_csum_err++;
 481                 return;
 482         }
 483
 484         /* TCP/UDP Checksum has not been calculated */
 485         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
 486                 return;
 487
 488         /* It must be a TCP or UDP packet with a valid checksum */
 489         if (status & E1000_RXD_STAT_TCPCS) {
 490                 /* TCP checksum is good */
 491                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 492         } else {
 493                 /*
 494                  * IP fragment with UDP payload
 495                  * Hardware complements the payload checksum, so we undo it
 496                  * and then put the value in host order for further stack use.
 497                  */
 498                 __sum16 sum = (__force __sum16)htons(csum);
 499                 skb->csum = csum_unfold(~sum);
 500                 skb->ip_summed = CHECKSUM_COMPLETE;
 501         }
 502         adapter->hw_csum_good++;
 503 }
 504
 505 /**
 506  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
 507  * @adapter: address of board private structure
 508  **/
 509 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
 510                                    int cleaned_count)
 511 {
 512         struct net_device *netdev = adapter->netdev;
 513         struct pci_dev *pdev = adapter->pdev;
 514         struct e1000_ring *rx_ring = adapter->rx_ring;
 515         struct e1000_rx_desc *rx_desc;
 516         struct e1000_buffer *buffer_info;
 517         struct sk_buff *skb;
 518         unsigned int i;
 519         unsigned int bufsz = adapter->rx_buffer_len;
 520
 521         i = rx_ring->next_to_use;
 522         buffer_info = &rx_ring->buffer_info[i];
 523
 524         while (cleaned_count--) {
 525                 skb = buffer_info->skb;
 526                 if (skb) {
 527                         skb_trim(skb, 0);
 528                         goto map_skb;
 529                 }
 530
 531                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 532                 if (!skb) {
 533                         /* Better luck next round */
 534                         adapter->alloc_rx_buff_failed++;
 535                         break;
 536                 }
 537
 538                 buffer_info->skb = skb;
 539 map_skb:
 540                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 541                                                   adapter->rx_buffer_len,
 542                                                   DMA_FROM_DEVICE);
 543                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 544                         dev_err(&pdev->dev, "RX DMA map failed\n");
 545                         adapter->rx_dma_failed++;
 546                         break;
 547                 }
 548
 549                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 550                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 551
 552                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 553                         /*
 554                          * Force memory writes to complete before letting h/w
 555                          * know there are new descriptors to fetch.  (Only
 556                          * applicable for weak-ordered memory model archs,
 557                          * such as IA-64).
 558                          */
 559                         wmb();
 560                         writel(i, adapter->hw.hw_addr + rx_ring->tail);
 561                 }
 562                 i++;
 563                 if (i == rx_ring->count)
 564                         i = 0;
 565                 buffer_info = &rx_ring->buffer_info[i];
 566         }
 567
 568         rx_ring->next_to_use = i;
 569 }
 570
 571 /**
 572  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
 573  * @adapter: address of board private structure
 574  **/
 575 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 576                                       int cleaned_count)
 577 {
 578         struct net_device *netdev = adapter->netdev;
 579         struct pci_dev *pdev = adapter->pdev;
 580         union e1000_rx_desc_packet_split *rx_desc;
 581         struct e1000_ring *rx_ring = adapter->rx_ring;
 582         struct e1000_buffer *buffer_info;
 583         struct e1000_ps_page *ps_page;
 584         struct sk_buff *skb;
 585         unsigned int i, j;
 586
 587         i = rx_ring->next_to_use;
 588         buffer_info = &rx_ring->buffer_info[i];
 589
 590         while (cleaned_count--) {
 591                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 592
 593                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 594                         ps_page = &buffer_info->ps_pages[j];
 595                         if (j >= adapter->rx_ps_pages) {
 596                                 /* all unused desc entries get hw null ptr */
 597                                 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
 598                                 continue;
 599                         }
 600                         if (!ps_page->page) {
 601                                 ps_page->page = alloc_page(GFP_ATOMIC);
 602                                 if (!ps_page->page) {
 603                                         adapter->alloc_rx_buff_failed++;
 604                                         goto no_buffers;
 605                                 }
 606                                 ps_page->dma = dma_map_page(&pdev->dev,
 607                                                             ps_page->page,
 608                                                             0, PAGE_SIZE,
 609                                                             DMA_FROM_DEVICE);
 610                                 if (dma_mapping_error(&pdev->dev,
 611                                                       ps_page->dma)) {
 612                                         dev_err(&adapter->pdev->dev,
 613                                           "RX DMA page map failed\n");
 614                                         adapter->rx_dma_failed++;
 615                                         goto no_buffers;
 616                                 }
 617                         }
 618                         /*
 619                          * Refresh the desc even if buffer_addrs
 620                          * didn't change because each write-back
 621                          * erases this info.
 622                          */
 623                         rx_desc->read.buffer_addr[j+1] =
 624                              cpu_to_le64(ps_page->dma);
 625                 }
 626
 627                 skb = netdev_alloc_skb_ip_align(netdev,
 628                                                 adapter->rx_ps_bsize0);
 629
 630                 if (!skb) {
 631                         adapter->alloc_rx_buff_failed++;
 632                         break;
 633                 }
 634
 635                 buffer_info->skb = skb;
 636                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 637                                                   adapter->rx_ps_bsize0,
 638                                                   DMA_FROM_DEVICE);
 639                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 640                         dev_err(&pdev->dev, "RX DMA map failed\n");
 641                         adapter->rx_dma_failed++;
 642                         /* cleanup skb */
 643                         dev_kfree_skb_any(skb);
 644                         buffer_info->skb = NULL;
 645                         break;
 646                 }
 647
 648                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 649
 650                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 651                         /*
 652                          * Force memory writes to complete before letting h/w
 653                          * know there are new descriptors to fetch.  (Only
 654                          * applicable for weak-ordered memory model archs,
 655                          * such as IA-64).
 656                          */
 657                         wmb();
 658                         writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
 659                 }
 660
 661                 i++;
 662                 if (i == rx_ring->count)
 663                         i = 0;
 664                 buffer_info = &rx_ring->buffer_info[i];
 665         }
 666
 667 no_buffers:
 668         rx_ring->next_to_use = i;
 669 }
 670
 671 /**
 672  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
 673  * @adapter: address of board private structure
 674  * @cleaned_count: number of buffers to allocate this pass
 675  **/
 676
 677 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
 678                                          int cleaned_count)
 679 {
 680         struct net_device *netdev = adapter->netdev;
 681         struct pci_dev *pdev = adapter->pdev;
 682         struct e1000_rx_desc *rx_desc;
 683         struct e1000_ring *rx_ring = adapter->rx_ring;
 684         struct e1000_buffer *buffer_info;
 685         struct sk_buff *skb;
 686         unsigned int i;
 687         unsigned int bufsz = 256 - 16 /* for skb_reserve */;
 688
 689         i = rx_ring->next_to_use;
 690         buffer_info = &rx_ring->buffer_info[i];
 691
 692         while (cleaned_count--) {
 693                 skb = buffer_info->skb;
 694                 if (skb) {
 695                         skb_trim(skb, 0);
 696                         goto check_page;
 697                 }
 698
 699                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
 700                 if (unlikely(!skb)) {
 701                         /* Better luck next round */
 702                         adapter->alloc_rx_buff_failed++;
 703                         break;
 704                 }
 705
 706                 buffer_info->skb = skb;
 707 check_page:
 708                 /* allocate a new page if necessary */
 709                 if (!buffer_info->page) {
 710                         buffer_info->page = alloc_page(GFP_ATOMIC);
 711                         if (unlikely(!buffer_info->page)) {
 712                                 adapter->alloc_rx_buff_failed++;
 713                                 break;
 714                         }
 715                 }
 716
 717                 if (!buffer_info->dma)
 718                         buffer_info->dma = dma_map_page(&pdev->dev,
 719                                                         buffer_info->page, 0,
 720                                                         PAGE_SIZE,
 721                                                         DMA_FROM_DEVICE);
 722
 723                 rx_desc = E1000_RX_DESC(*rx_ring, i);
 724                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
 725
 726                 if (unlikely(++i == rx_ring->count))
 727                         i = 0;
 728                 buffer_info = &rx_ring->buffer_info[i];
 729         }
 730
 731         if (likely(rx_ring->next_to_use != i)) {
 732                 rx_ring->next_to_use = i;
 733                 if (unlikely(i-- == 0))
 734                         i = (rx_ring->count - 1);
 735
 736                 /* Force memory writes to complete before letting h/w
 737                  * know there are new descriptors to fetch.  (Only
 738                  * applicable for weak-ordered memory model archs,
 739                  * such as IA-64). */
 740                 wmb();
 741                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 742         }
 743 }
 744
 745 /**
 746  * e1000_clean_rx_irq - Send received data up the network stack; legacy
 747  * @adapter: board private structure
 748  *
 749  * the return value indicates whether actual cleaning was done, there
 750  * is no guarantee that everything was cleaned
 751  **/
 752 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
 753                                int *work_done, int work_to_do)
 754 {
 755         struct net_device *netdev = adapter->netdev;
 756         struct pci_dev *pdev = adapter->pdev;
 757         struct e1000_hw *hw = &adapter->hw;
 758         struct e1000_ring *rx_ring = adapter->rx_ring;
 759         struct e1000_rx_desc *rx_desc, *next_rxd;
 760         struct e1000_buffer *buffer_info, *next_buffer;
 761         u32 length;
 762         unsigned int i;
 763         int cleaned_count = 0;
 764         bool cleaned = 0;
 765         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 766
 767         i = rx_ring->next_to_clean;
 768         rx_desc = E1000_RX_DESC(*rx_ring, i);
 769         buffer_info = &rx_ring->buffer_info[i];
 770
 771         while (rx_desc->status & E1000_RXD_STAT_DD) {
 772                 struct sk_buff *skb;
 773                 u8 status;
 774
 775                 if (*work_done >= work_to_do)
 776                         break;
 777                 (*work_done)++;
 778
 779                 status = rx_desc->status;
 780                 skb = buffer_info->skb;
 781                 buffer_info->skb = NULL;
 782
 783                 prefetch(skb->data - NET_IP_ALIGN);
 784
 785                 i++;
 786                 if (i == rx_ring->count)
 787                         i = 0;
 788                 next_rxd = E1000_RX_DESC(*rx_ring, i);
 789                 prefetch(next_rxd);
 790
 791                 next_buffer = &rx_ring->buffer_info[i];
 792
 793                 cleaned = 1;
 794                 cleaned_count++;
 795                 dma_unmap_single(&pdev->dev,
 796                                  buffer_info->dma,
 797                                  adapter->rx_buffer_len,
 798                                  DMA_FROM_DEVICE);
 799                 buffer_info->dma = 0;
 800
 801                 length = le16_to_cpu(rx_desc->length);
 802
 803                 /*
 804                  * !EOP means multiple descriptors were used to store a single
 805                  * packet, if that's the case we need to toss it.  In fact, we
 806                  * need to toss every packet with the EOP bit clear and the
 807                  * next frame that _does_ have the EOP bit set, as it is by
 808                  * definition only a frame fragment
 809                  */
 810                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
 811                         adapter->flags2 |= FLAG2_IS_DISCARDING;
 812
 813                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
 814                         /* All receives must fit into a single buffer */
 815                         e_dbg("Receive packet consumed multiple buffers\n");
 816                         /* recycle */
 817                         buffer_info->skb = skb;
 818                         if (status & E1000_RXD_STAT_EOP)
 819                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
 820                         goto next_desc;
 821                 }
 822
 823                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
 824                         /* recycle */
 825                         buffer_info->skb = skb;
 826                         goto next_desc;
 827                 }
 828
 829                 /* adjust length to remove Ethernet CRC */
 830                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 831                         length -= 4;
 832
 833                 total_rx_bytes += length;
 834                 total_rx_packets++;
 835
 836                 /*
 837                  * code added for copybreak, this should improve
 838                  * performance for small packets with large amounts
 839                  * of reassembly being done in the stack
 840                  */
 841                 if (length < copybreak) {
 842                         struct sk_buff *new_skb =
 843                             netdev_alloc_skb_ip_align(netdev, length);
 844                         if (new_skb) {
 845                                 skb_copy_to_linear_data_offset(new_skb,
 846                                                                -NET_IP_ALIGN,
 847                                                                (skb->data -
 848                                                                 NET_IP_ALIGN),
 849                                                                (length +
 850                                                                 NET_IP_ALIGN));
 851                                 /* save the skb in buffer_info as good */
 852                                 buffer_info->skb = skb;
 853                                 skb = new_skb;
 854                         }
 855                         /* else just continue with the old one */
 856                 }
 857                 /* end copybreak code */
 858                 skb_put(skb, length);
 859
 860                 /* Receive Checksum Offload */
 861                 e1000_rx_checksum(adapter,
 862                                   (u32)(status) |
 863                                   ((u32)(rx_desc->errors) << 24),
 864                                   le16_to_cpu(rx_desc->csum), skb);
 865
 866                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
 867
 868 next_desc:
 869                 rx_desc->status = 0;
 870
 871                 /* return some buffers to hardware, one at a time is too slow */
 872                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 873                         adapter->alloc_rx_buf(adapter, cleaned_count);
 874                         cleaned_count = 0;
 875                 }
 876
 877                 /* use prefetched values */
 878                 rx_desc = next_rxd;
 879                 buffer_info = next_buffer;
 880         }
 881         rx_ring->next_to_clean = i;
 882
 883         cleaned_count = e1000_desc_unused(rx_ring);
 884         if (cleaned_count)
 885                 adapter->alloc_rx_buf(adapter, cleaned_count);
 886
 887         adapter->total_rx_bytes += total_rx_bytes;
 888         adapter->total_rx_packets += total_rx_packets;
 889         netdev->stats.rx_bytes += total_rx_bytes;
 890         netdev->stats.rx_packets += total_rx_packets;
 891         return cleaned;
 892 }
 893
 894 static void e1000_put_txbuf(struct e1000_adapter *adapter,
 895                              struct e1000_buffer *buffer_info)
 896 {
 897         if (buffer_info->dma) {
 898                 if (buffer_info->mapped_as_page)
 899                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
 900                                        buffer_info->length, DMA_TO_DEVICE);
 901                 else
 902                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
 903                                          buffer_info->length, DMA_TO_DEVICE);
 904                 buffer_info->dma = 0;
 905         }
 906         if (buffer_info->skb) {
 907                 dev_kfree_skb_any(buffer_info->skb);
 908                 buffer_info->skb = NULL;
 909         }
 910         buffer_info->time_stamp = 0;
 911 }
 912
 913 static void e1000_print_hw_hang(struct work_struct *work)
 914 {
 915         struct e1000_adapter *adapter = container_of(work,
 916                                                      struct e1000_adapter,
 917                                                      print_hang_task);
 918         struct e1000_ring *tx_ring = adapter->tx_ring;
 919         unsigned int i = tx_ring->next_to_clean;
 920         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
 921         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
 922         struct e1000_hw *hw = &adapter->hw;
 923         u16 phy_status, phy_1000t_status, phy_ext_status;
 924         u16 pci_status;
 925
 926         e1e_rphy(hw, PHY_STATUS, &phy_status);
 927         e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
 928         e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
 929
 930         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
 931
 932         /* detected Hardware unit hang */
 933         e_err("Detected Hardware Unit Hang:\n"
 934               "  TDH                  <%x>\n"
 935               "  TDT                  <%x>\n"
 936               "  next_to_use          <%x>\n"
 937               "  next_to_clean        <%x>\n"
 938               "buffer_info[next_to_clean]:\n"
 939               "  time_stamp           <%lx>\n"
 940               "  next_to_watch        <%x>\n"
 941               "  jiffies              <%lx>\n"
 942               "  next_to_watch.status <%x>\n"
 943               "MAC Status             <%x>\n"
 944               "PHY Status             <%x>\n"
 945               "PHY 1000BASE-T Status  <%x>\n"
 946               "PHY Extended Status    <%x>\n"
 947               "PCI Status             <%x>\n",
 948               readl(adapter->hw.hw_addr + tx_ring->head),
 949               readl(adapter->hw.hw_addr + tx_ring->tail),
 950               tx_ring->next_to_use,
 951               tx_ring->next_to_clean,
 952               tx_ring->buffer_info[eop].time_stamp,
 953               eop,
 954               jiffies,
 955               eop_desc->upper.fields.status,
 956               er32(STATUS),
 957               phy_status,
 958               phy_1000t_status,
 959               phy_ext_status,
 960               pci_status);
 961 }
 962
 963 /**
 964  * e1000_clean_tx_irq - Reclaim resources after transmit completes
 965  * @adapter: board private structure
 966  *
 967  * the return value indicates whether actual cleaning was done, there
 968  * is no guarantee that everything was cleaned
 969  **/
 970 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
 971 {
 972         struct net_device *netdev = adapter->netdev;
 973         struct e1000_hw *hw = &adapter->hw;
 974         struct e1000_ring *tx_ring = adapter->tx_ring;
 975         struct e1000_tx_desc *tx_desc, *eop_desc;
 976         struct e1000_buffer *buffer_info;
 977         unsigned int i, eop;
 978         unsigned int count = 0;
 979         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
 980
 981         i = tx_ring->next_to_clean;
 982         eop = tx_ring->buffer_info[i].next_to_watch;
 983         eop_desc = E1000_TX_DESC(*tx_ring, eop);
 984
 985         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
 986                (count < tx_ring->count)) {
 987                 bool cleaned = false;
 988                 for (; !cleaned; count++) {
 989                         tx_desc = E1000_TX_DESC(*tx_ring, i);
 990                         buffer_info = &tx_ring->buffer_info[i];
 991                         cleaned = (i == eop);
 992
 993                         if (cleaned) {
 994                                 total_tx_packets += buffer_info->segs;
 995                                 total_tx_bytes += buffer_info->bytecount;
 996                         }
 997
 998                         e1000_put_txbuf(adapter, buffer_info);
 999                         tx_desc->upper.data = 0;
1000
1001                         i++;
1002                         if (i == tx_ring->count)
1003                                 i = 0;
1004                 }
1005
1006                 if (i == tx_ring->next_to_use)
1007                         break;
1008                 eop = tx_ring->buffer_info[i].next_to_watch;
1009                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1010         }
1011
1012         tx_ring->next_to_clean = i;
1013
1014 #define TX_WAKE_THRESHOLD 32
1015         if (count && netif_carrier_ok(netdev) &&
1016             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1017                 /* Make sure that anybody stopping the queue after this
1018                  * sees the new next_to_clean.
1019                  */
1020                 smp_mb();
1021
1022                 if (netif_queue_stopped(netdev) &&
1023                     !(test_bit(__E1000_DOWN, &adapter->state))) {
1024                         netif_wake_queue(netdev);
1025                         ++adapter->restart_queue;
1026                 }
1027         }
1028
1029         if (adapter->detect_tx_hung) {
1030                 /*
1031                  * Detect a transmit hang in hardware, this serializes the
1032                  * check with the clearing of time_stamp and movement of i
1033                  */
1034                 adapter->detect_tx_hung = 0;
1035                 if (tx_ring->buffer_info[i].time_stamp &&
1036                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1037                                + (adapter->tx_timeout_factor * HZ)) &&
1038                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
1039                         schedule_work(&adapter->print_hang_task);
1040                         netif_stop_queue(netdev);
1041                 }
1042         }
1043         adapter->total_tx_bytes += total_tx_bytes;
1044         adapter->total_tx_packets += total_tx_packets;
1045         netdev->stats.tx_bytes += total_tx_bytes;
1046         netdev->stats.tx_packets += total_tx_packets;
1047         return (count < tx_ring->count);
1048 }
1049
1050 /**
1051  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1052  * @adapter: board private structure
1053  *
1054  * the return value indicates whether actual cleaning was done, there
1055  * is no guarantee that everything was cleaned
1056  **/
1057 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
1058                                   int *work_done, int work_to_do)
1059 {
1060         struct e1000_hw *hw = &adapter->hw;
1061         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1062         struct net_device *netdev = adapter->netdev;
1063         struct pci_dev *pdev = adapter->pdev;
1064         struct e1000_ring *rx_ring = adapter->rx_ring;
1065         struct e1000_buffer *buffer_info, *next_buffer;
1066         struct e1000_ps_page *ps_page;
1067         struct sk_buff *skb;
1068         unsigned int i, j;
1069         u32 length, staterr;
1070         int cleaned_count = 0;
1071         bool cleaned = 0;
1072         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1073
1074         i = rx_ring->next_to_clean;
1075         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1076         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1077         buffer_info = &rx_ring->buffer_info[i];
1078
1079         while (staterr & E1000_RXD_STAT_DD) {
1080                 if (*work_done >= work_to_do)
1081                         break;
1082                 (*work_done)++;
1083                 skb = buffer_info->skb;
1084
1085                 /* in the packet split case this is header only */
1086                 prefetch(skb->data - NET_IP_ALIGN);
1087
1088                 i++;
1089                 if (i == rx_ring->count)
1090                         i = 0;
1091                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1092                 prefetch(next_rxd);
1093
1094                 next_buffer = &rx_ring->buffer_info[i];
1095
1096                 cleaned = 1;
1097                 cleaned_count++;
1098                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1099                                  adapter->rx_ps_bsize0,
1100                                  DMA_FROM_DEVICE);
1101                 buffer_info->dma = 0;
1102
1103                 /* see !EOP comment in other rx routine */
1104                 if (!(staterr & E1000_RXD_STAT_EOP))
1105                         adapter->flags2 |= FLAG2_IS_DISCARDING;
1106
1107                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1108                         e_dbg("Packet Split buffers didn't pick up the full "
1109                               "packet\n");
1110                         dev_kfree_skb_irq(skb);
1111                         if (staterr & E1000_RXD_STAT_EOP)
1112                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1113                         goto next_desc;
1114                 }
1115
1116                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
1117                         dev_kfree_skb_irq(skb);
1118                         goto next_desc;
1119                 }
1120
1121                 length = le16_to_cpu(rx_desc->wb.middle.length0);
1122
1123                 if (!length) {
1124                         e_dbg("Last part of the packet spanning multiple "
1125                               "descriptors\n");
1126                         dev_kfree_skb_irq(skb);
1127                         goto next_desc;
1128                 }
1129
1130                 /* Good Receive */
1131                 skb_put(skb, length);
1132
1133                 {
1134                 /*
1135                  * this looks ugly, but it seems compiler issues make it
1136                  * more efficient than reusing j
1137                  */
1138                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1139
1140                 /*
1141                  * page alloc/put takes too long and effects small packet
1142                  * throughput, so unsplit small packets and save the alloc/put
1143                  * only valid in softirq (napi) context to call kmap_*
1144                  */
1145                 if (l1 && (l1 <= copybreak) &&
1146                     ((length + l1) <= adapter->rx_ps_bsize0)) {
1147                         u8 *vaddr;
1148
1149                         ps_page = &buffer_info->ps_pages[0];
1150
1151                         /*
1152                          * there is no documentation about how to call
1153                          * kmap_atomic, so we can't hold the mapping
1154                          * very long
1155                          */
1156                         dma_sync_single_for_cpu(&pdev->dev, ps_page->dma,
1157                                                 PAGE_SIZE, DMA_FROM_DEVICE);
1158                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
1159                         memcpy(skb_tail_pointer(skb), vaddr, l1);
1160                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
1161                         dma_sync_single_for_device(&pdev->dev, ps_page->dma,
1162                                                    PAGE_SIZE, DMA_FROM_DEVICE);
1163
1164                         /* remove the CRC */
1165                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1166                                 l1 -= 4;
1167
1168                         skb_put(skb, l1);
1169                         goto copydone;
1170                 } /* if */
1171                 }
1172
1173                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1174                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1175                         if (!length)
1176                                 break;
1177
1178                         ps_page = &buffer_info->ps_pages[j];
1179                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1180                                        DMA_FROM_DEVICE);
1181                         ps_page->dma = 0;
1182                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1183                         ps_page->page = NULL;
1184                         skb->len += length;
1185                         skb->data_len += length;
1186                         skb->truesize += length;
1187                 }
1188
1189                 /* strip the ethernet crc, problem is we're using pages now so
1190                  * this whole operation can get a little cpu intensive
1191                  */
1192                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1193                         pskb_trim(skb, skb->len - 4);
1194
1195 copydone:
1196                 total_rx_bytes += skb->len;
1197                 total_rx_packets++;
1198
1199                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
1200                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
1201
1202                 if (rx_desc->wb.upper.header_status &
1203                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1204                         adapter->rx_hdr_split++;
1205
1206                 e1000_receive_skb(adapter, netdev, skb,
1207                                   staterr, rx_desc->wb.middle.vlan);
1208
1209 next_desc:
1210                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1211                 buffer_info->skb = NULL;
1212
1213                 /* return some buffers to hardware, one at a time is too slow */
1214                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1215                         adapter->alloc_rx_buf(adapter, cleaned_count);
1216                         cleaned_count = 0;
1217                 }
1218
1219                 /* use prefetched values */
1220                 rx_desc = next_rxd;
1221                 buffer_info = next_buffer;
1222
1223                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1224         }
1225         rx_ring->next_to_clean = i;
1226
1227         cleaned_count = e1000_desc_unused(rx_ring);
1228         if (cleaned_count)
1229                 adapter->alloc_rx_buf(adapter, cleaned_count);
1230
1231         adapter->total_rx_bytes += total_rx_bytes;
1232         adapter->total_rx_packets += total_rx_packets;
1233         netdev->stats.rx_bytes += total_rx_bytes;
1234         netdev->stats.rx_packets += total_rx_packets;
1235         return cleaned;
1236 }
1237
1238 /**
1239  * e1000_consume_page - helper function
1240  **/
1241 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1242                                u16 length)
1243 {
1244         bi->page = NULL;
1245         skb->len += length;
1246         skb->data_len += length;
1247         skb->truesize += length;
1248 }
1249
1250 /**
1251  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1252  * @adapter: board private structure
1253  *
1254  * the return value indicates whether actual cleaning was done, there
1255  * is no guarantee that everything was cleaned
1256  **/
1257
1258 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
1259                                      int *work_done, int work_to_do)
1260 {
1261         struct net_device *netdev = adapter->netdev;
1262         struct pci_dev *pdev = adapter->pdev;
1263         struct e1000_ring *rx_ring = adapter->rx_ring;
1264         struct e1000_rx_desc *rx_desc, *next_rxd;
1265         struct e1000_buffer *buffer_info, *next_buffer;
1266         u32 length;
1267         unsigned int i;
1268         int cleaned_count = 0;
1269         bool cleaned = false;
1270         unsigned int total_rx_bytes=0, total_rx_packets=0;
1271
1272         i = rx_ring->next_to_clean;
1273         rx_desc = E1000_RX_DESC(*rx_ring, i);
1274         buffer_info = &rx_ring->buffer_info[i];
1275
1276         while (rx_desc->status & E1000_RXD_STAT_DD) {
1277                 struct sk_buff *skb;
1278                 u8 status;
1279
1280                 if (*work_done >= work_to_do)
1281                         break;
1282                 (*work_done)++;
1283
1284                 status = rx_desc->status;
1285                 skb = buffer_info->skb;
1286                 buffer_info->skb = NULL;
1287
1288                 ++i;
1289                 if (i == rx_ring->count)
1290                         i = 0;
1291                 next_rxd = E1000_RX_DESC(*rx_ring, i);
1292                 prefetch(next_rxd);
1293
1294                 next_buffer = &rx_ring->buffer_info[i];
1295
1296                 cleaned = true;
1297                 cleaned_count++;
1298                 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1299                                DMA_FROM_DEVICE);
1300                 buffer_info->dma = 0;
1301
1302                 length = le16_to_cpu(rx_desc->length);
1303
1304                 /* errors is only valid for DD + EOP descriptors */
1305                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
1306                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
1307                                 /* recycle both page and skb */
1308                                 buffer_info->skb = skb;
1309                                 /* an error means any chain goes out the window
1310                                  * too */
1311                                 if (rx_ring->rx_skb_top)
1312                                         dev_kfree_skb(rx_ring->rx_skb_top);
1313                                 rx_ring->rx_skb_top = NULL;
1314                                 goto next_desc;
1315                 }
1316
1317 #define rxtop rx_ring->rx_skb_top
1318                 if (!(status & E1000_RXD_STAT_EOP)) {
1319                         /* this descriptor is only the beginning (or middle) */
1320                         if (!rxtop) {
1321                                 /* this is the beginning of a chain */
1322                                 rxtop = skb;
1323                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1324                                                    0, length);
1325                         } else {
1326                                 /* this is the middle of a chain */
1327                                 skb_fill_page_desc(rxtop,
1328                                     skb_shinfo(rxtop)->nr_frags,
1329                                     buffer_info->page, 0, length);
1330                                 /* re-use the skb, only consumed the page */
1331                                 buffer_info->skb = skb;
1332                         }
1333                         e1000_consume_page(buffer_info, rxtop, length);
1334                         goto next_desc;
1335                 } else {
1336                         if (rxtop) {
1337                                 /* end of the chain */
1338                                 skb_fill_page_desc(rxtop,
1339                                     skb_shinfo(rxtop)->nr_frags,
1340                                     buffer_info->page, 0, length);
1341                                 /* re-use the current skb, we only consumed the
1342                                  * page */
1343                                 buffer_info->skb = skb;
1344                                 skb = rxtop;
1345                                 rxtop = NULL;
1346                                 e1000_consume_page(buffer_info, skb, length);
1347                         } else {
1348                                 /* no chain, got EOP, this buf is the packet
1349                                  * copybreak to save the put_page/alloc_page */
1350                                 if (length <= copybreak &&
1351                                     skb_tailroom(skb) >= length) {
1352                                         u8 *vaddr;
1353                                         vaddr = kmap_atomic(buffer_info->page,
1354                                                            KM_SKB_DATA_SOFTIRQ);
1355                                         memcpy(skb_tail_pointer(skb), vaddr,
1356                                                length);
1357                                         kunmap_atomic(vaddr,
1358                                                       KM_SKB_DATA_SOFTIRQ);
1359                                         /* re-use the page, so don't erase
1360                                          * buffer_info->page */
1361                                         skb_put(skb, length);
1362                                 } else {
1363                                         skb_fill_page_desc(skb, 0,
1364                                                            buffer_info->page, 0,
1365                                                            length);
1366                                         e1000_consume_page(buffer_info, skb,
1367                                                            length);
1368                                 }
1369                         }
1370                 }
1371
1372                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1373                 e1000_rx_checksum(adapter,
1374                                   (u32)(status) |
1375                                   ((u32)(rx_desc->errors) << 24),
1376                                   le16_to_cpu(rx_desc->csum), skb);
1377
1378                 /* probably a little skewed due to removing CRC */
1379                 total_rx_bytes += skb->len;
1380                 total_rx_packets++;
1381
1382                 /* eth type trans needs skb->data to point to something */
1383                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1384                         e_err("pskb_may_pull failed.\n");
1385                         dev_kfree_skb(skb);
1386                         goto next_desc;
1387                 }
1388
1389                 e1000_receive_skb(adapter, netdev, skb, status,
1390                                   rx_desc->special);
1391
1392 next_desc:
1393                 rx_desc->status = 0;
1394
1395                 /* return some buffers to hardware, one at a time is too slow */
1396                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1397                         adapter->alloc_rx_buf(adapter, cleaned_count);
1398                         cleaned_count = 0;
1399                 }
1400
1401                 /* use prefetched values */
1402                 rx_desc = next_rxd;
1403                 buffer_info = next_buffer;
1404         }
1405         rx_ring->next_to_clean = i;
1406
1407         cleaned_count = e1000_desc_unused(rx_ring);
1408         if (cleaned_count)
1409                 adapter->alloc_rx_buf(adapter, cleaned_count);
1410
1411         adapter->total_rx_bytes += total_rx_bytes;
1412         adapter->total_rx_packets += total_rx_packets;
1413         netdev->stats.rx_bytes += total_rx_bytes;
1414         netdev->stats.rx_packets += total_rx_packets;
1415         return cleaned;
1416 }
1417
1418 /**
1419  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1420  * @adapter: board private structure
1421  **/
1422 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1423 {
1424         struct e1000_ring *rx_ring = adapter->rx_ring;
1425         struct e1000_buffer *buffer_info;
1426         struct e1000_ps_page *ps_page;
1427         struct pci_dev *pdev = adapter->pdev;
1428         unsigned int i, j;
1429
1430         /* Free all the Rx ring sk_buffs */
1431         for (i = 0; i < rx_ring->count; i++) {
1432                 buffer_info = &rx_ring->buffer_info[i];
1433                 if (buffer_info->dma) {
1434                         if (adapter->clean_rx == e1000_clean_rx_irq)
1435                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1436                                                  adapter->rx_buffer_len,
1437                                                  DMA_FROM_DEVICE);
1438                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1439                                 dma_unmap_page(&pdev->dev, buffer_info->dma,
1440                                                PAGE_SIZE,
1441                                                DMA_FROM_DEVICE);
1442                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1443                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1444                                                  adapter->rx_ps_bsize0,
1445                                                  DMA_FROM_DEVICE);
1446                         buffer_info->dma = 0;
1447                 }
1448
1449                 if (buffer_info->page) {
1450                         put_page(buffer_info->page);
1451                         buffer_info->page = NULL;
1452                 }
1453
1454                 if (buffer_info->skb) {
1455                         dev_kfree_skb(buffer_info->skb);
1456                         buffer_info->skb = NULL;
1457                 }
1458
1459                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1460                         ps_page = &buffer_info->ps_pages[j];
1461                         if (!ps_page->page)
1462                                 break;
1463                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1464                                        DMA_FROM_DEVICE);
1465                         ps_page->dma = 0;
1466                         put_page(ps_page->page);
1467                         ps_page->page = NULL;
1468                 }
1469         }
1470
1471         /* there also may be some cached data from a chained receive */
1472         if (rx_ring->rx_skb_top) {
1473                 dev_kfree_skb(rx_ring->rx_skb_top);
1474                 rx_ring->rx_skb_top = NULL;
1475         }
1476
1477         /* Zero out the descriptor ring */
1478         memset(rx_ring->desc, 0, rx_ring->size);
1479
1480         rx_ring->next_to_clean = 0;
1481         rx_ring->next_to_use = 0;
1482         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1483
1484         writel(0, adapter->hw.hw_addr + rx_ring->head);
1485         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1486 }
1487
1488 static void e1000e_downshift_workaround(struct work_struct *work)
1489 {
1490         struct e1000_adapter *adapter = container_of(work,
1491                                         struct e1000_adapter, downshift_task);
1492
1493         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1494 }
1495
1496 /**
1497  * e1000_intr_msi - Interrupt Handler
1498  * @irq: interrupt number
1499  * @data: pointer to a network interface device structure
1500  **/
1501 static irqreturn_t e1000_intr_msi(int irq, void *data)
1502 {
1503         struct net_device *netdev = data;
1504         struct e1000_adapter *adapter = netdev_priv(netdev);
1505         struct e1000_hw *hw = &adapter->hw;
1506         u32 icr = er32(ICR);
1507
1508         /*
1509          * read ICR disables interrupts using IAM
1510          */
1511
1512         if (icr & E1000_ICR_LSC) {
1513                 hw->mac.get_link_status = 1;
1514                 /*
1515                  * ICH8 workaround-- Call gig speed drop workaround on cable
1516                  * disconnect (LSC) before accessing any PHY registers
1517                  */
1518                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1519                     (!(er32(STATUS) & E1000_STATUS_LU)))
1520                         schedule_work(&adapter->downshift_task);
1521
1522                 /*
1523                  * 80003ES2LAN workaround-- For packet buffer work-around on
1524                  * link down event; disable receives here in the ISR and reset
1525                  * adapter in watchdog
1526                  */
1527                 if (netif_carrier_ok(netdev) &&
1528                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1529                         /* disable receives */
1530                         u32 rctl = er32(RCTL);
1531                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1532                         adapter->flags |= FLAG_RX_RESTART_NOW;
1533                 }
1534                 /* guard against interrupt when we're going down */
1535                 if (!test_bit(__E1000_DOWN, &adapter->state))
1536                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1537         }
1538
1539         if (napi_schedule_prep(&adapter->napi)) {
1540                 adapter->total_tx_bytes = 0;
1541                 adapter->total_tx_packets = 0;
1542                 adapter->total_rx_bytes = 0;
1543                 adapter->total_rx_packets = 0;
1544                 __napi_schedule(&adapter->napi);
1545         }
1546
1547         return IRQ_HANDLED;
1548 }
1549
1550 /**
1551  * e1000_intr - Interrupt Handler
1552  * @irq: interrupt number
1553  * @data: pointer to a network interface device structure
1554  **/
1555 static irqreturn_t e1000_intr(int irq, void *data)
1556 {
1557         struct net_device *netdev = data;
1558         struct e1000_adapter *adapter = netdev_priv(netdev);
1559         struct e1000_hw *hw = &adapter->hw;
1560         u32 rctl, icr = er32(ICR);
1561
1562         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1563                 return IRQ_NONE;  /* Not our interrupt */
1564
1565         /*
1566          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1567          * not set, then the adapter didn't send an interrupt
1568          */
1569         if (!(icr & E1000_ICR_INT_ASSERTED))
1570                 return IRQ_NONE;
1571
1572         /*
1573          * Interrupt Auto-Mask...upon reading ICR,
1574          * interrupts are masked.  No need for the
1575          * IMC write
1576          */
1577
1578         if (icr & E1000_ICR_LSC) {
1579                 hw->mac.get_link_status = 1;
1580                 /*
1581                  * ICH8 workaround-- Call gig speed drop workaround on cable
1582                  * disconnect (LSC) before accessing any PHY registers
1583                  */
1584                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1585                     (!(er32(STATUS) & E1000_STATUS_LU)))
1586                         schedule_work(&adapter->downshift_task);
1587
1588                 /*
1589                  * 80003ES2LAN workaround--
1590                  * For packet buffer work-around on link down event;
1591                  * disable receives here in the ISR and
1592                  * reset adapter in watchdog
1593                  */
1594                 if (netif_carrier_ok(netdev) &&
1595                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1596                         /* disable receives */
1597                         rctl = er32(RCTL);
1598                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1599                         adapter->flags |= FLAG_RX_RESTART_NOW;
1600                 }
1601                 /* guard against interrupt when we're going down */
1602                 if (!test_bit(__E1000_DOWN, &adapter->state))
1603                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1604         }
1605
1606         if (napi_schedule_prep(&adapter->napi)) {
1607                 adapter->total_tx_bytes = 0;
1608                 adapter->total_tx_packets = 0;
1609                 adapter->total_rx_bytes = 0;
1610                 adapter->total_rx_packets = 0;
1611                 __napi_schedule(&adapter->napi);
1612         }
1613
1614         return IRQ_HANDLED;
1615 }
1616
1617 static irqreturn_t e1000_msix_other(int irq, void *data)
1618 {
1619         struct net_device *netdev = data;
1620         struct e1000_adapter *adapter = netdev_priv(netdev);
1621         struct e1000_hw *hw = &adapter->hw;
1622         u32 icr = er32(ICR);
1623
1624         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1625                 if (!test_bit(__E1000_DOWN, &adapter->state))
1626                         ew32(IMS, E1000_IMS_OTHER);
1627                 return IRQ_NONE;
1628         }
1629
1630         if (icr & adapter->eiac_mask)
1631                 ew32(ICS, (icr & adapter->eiac_mask));
1632
1633         if (icr & E1000_ICR_OTHER) {
1634                 if (!(icr & E1000_ICR_LSC))
1635                         goto no_link_interrupt;
1636                 hw->mac.get_link_status = 1;
1637                 /* guard against interrupt when we're going down */
1638                 if (!test_bit(__E1000_DOWN, &adapter->state))
1639                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1640         }
1641
1642 no_link_interrupt:
1643         if (!test_bit(__E1000_DOWN, &adapter->state))
1644                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1645
1646         return IRQ_HANDLED;
1647 }
1648
1649
1650 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1651 {
1652         struct net_device *netdev = data;
1653         struct e1000_adapter *adapter = netdev_priv(netdev);
1654         struct e1000_hw *hw = &adapter->hw;
1655         struct e1000_ring *tx_ring = adapter->tx_ring;
1656
1657
1658         adapter->total_tx_bytes = 0;
1659         adapter->total_tx_packets = 0;
1660
1661         if (!e1000_clean_tx_irq(adapter))
1662                 /* Ring was not completely cleaned, so fire another interrupt */
1663                 ew32(ICS, tx_ring->ims_val);
1664
1665         return IRQ_HANDLED;
1666 }
1667
1668 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1669 {
1670         struct net_device *netdev = data;
1671         struct e1000_adapter *adapter = netdev_priv(netdev);
1672
1673         /* Write the ITR value calculated at the end of the
1674          * previous interrupt.
1675          */
1676         if (adapter->rx_ring->set_itr) {
1677                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1678                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1679                 adapter->rx_ring->set_itr = 0;
1680         }
1681
1682         if (napi_schedule_prep(&adapter->napi)) {
1683                 adapter->total_rx_bytes = 0;
1684                 adapter->total_rx_packets = 0;
1685                 __napi_schedule(&adapter->napi);
1686         }
1687         return IRQ_HANDLED;
1688 }
1689
1690 /**
1691  * e1000_configure_msix - Configure MSI-X hardware
1692  *
1693  * e1000_configure_msix sets up the hardware to properly
1694  * generate MSI-X interrupts.
1695  **/
1696 static void e1000_configure_msix(struct e1000_adapter *adapter)
1697 {
1698         struct e1000_hw *hw = &adapter->hw;
1699         struct e1000_ring *rx_ring = adapter->rx_ring;
1700         struct e1000_ring *tx_ring = adapter->tx_ring;
1701         int vector = 0;
1702         u32 ctrl_ext, ivar = 0;
1703
1704         adapter->eiac_mask = 0;
1705
1706         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1707         if (hw->mac.type == e1000_82574) {
1708                 u32 rfctl = er32(RFCTL);
1709                 rfctl |= E1000_RFCTL_ACK_DIS;
1710                 ew32(RFCTL, rfctl);
1711         }
1712
1713 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1714         /* Configure Rx vector */
1715         rx_ring->ims_val = E1000_IMS_RXQ0;
1716         adapter->eiac_mask |= rx_ring->ims_val;
1717         if (rx_ring->itr_val)
1718                 writel(1000000000 / (rx_ring->itr_val * 256),
1719                        hw->hw_addr + rx_ring->itr_register);
1720         else
1721                 writel(1, hw->hw_addr + rx_ring->itr_register);
1722         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1723
1724         /* Configure Tx vector */
1725         tx_ring->ims_val = E1000_IMS_TXQ0;
1726         vector++;
1727         if (tx_ring->itr_val)
1728                 writel(1000000000 / (tx_ring->itr_val * 256),
1729                        hw->hw_addr + tx_ring->itr_register);
1730         else
1731                 writel(1, hw->hw_addr + tx_ring->itr_register);
1732         adapter->eiac_mask |= tx_ring->ims_val;
1733         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1734
1735         /* set vector for Other Causes, e.g. link changes */
1736         vector++;
1737         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1738         if (rx_ring->itr_val)
1739                 writel(1000000000 / (rx_ring->itr_val * 256),
1740                        hw->hw_addr + E1000_EITR_82574(vector));
1741         else
1742                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1743
1744         /* Cause Tx interrupts on every write back */
1745         ivar |= (1 << 31);
1746
1747         ew32(IVAR, ivar);
1748
1749         /* enable MSI-X PBA support */
1750         ctrl_ext = er32(CTRL_EXT);
1751         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1752
1753         /* Auto-Mask Other interrupts upon ICR read */
1754 #define E1000_EIAC_MASK_82574   0x01F00000
1755         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1756         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1757         ew32(CTRL_EXT, ctrl_ext);
1758         e1e_flush();
1759 }
1760
1761 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1762 {
1763         if (adapter->msix_entries) {
1764                 pci_disable_msix(adapter->pdev);
1765                 kfree(adapter->msix_entries);
1766                 adapter->msix_entries = NULL;
1767         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1768                 pci_disable_msi(adapter->pdev);
1769                 adapter->flags &= ~FLAG_MSI_ENABLED;
1770         }
1771 }
1772
1773 /**
1774  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1775  *
1776  * Attempt to configure interrupts using the best available
1777  * capabilities of the hardware and kernel.
1778  **/
1779 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1780 {
1781         int err;
1782         int numvecs, i;
1783
1784
1785         switch (adapter->int_mode) {
1786         case E1000E_INT_MODE_MSIX:
1787                 if (adapter->flags & FLAG_HAS_MSIX) {
1788                         numvecs = 3; /* RxQ0, TxQ0 and other */
1789                         adapter->msix_entries = kcalloc(numvecs,
1790                                                       sizeof(struct msix_entry),
1791                                                       GFP_KERNEL);
1792                         if (adapter->msix_entries) {
1793                                 for (i = 0; i < numvecs; i++)
1794                                         adapter->msix_entries[i].entry = i;
1795
1796                                 err = pci_enable_msix(adapter->pdev,
1797                                                       adapter->msix_entries,
1798                                                       numvecs);
1799                                 if (err == 0)
1800                                         return;
1801                         }
1802                         /* MSI-X failed, so fall through and try MSI */
1803                         e_err("Failed to initialize MSI-X interrupts.  "
1804                               "Falling back to MSI interrupts.\n");
1805                         e1000e_reset_interrupt_capability(adapter);
1806                 }
1807                 adapter->int_mode = E1000E_INT_MODE_MSI;
1808                 /* Fall through */
1809         case E1000E_INT_MODE_MSI:
1810                 if (!pci_enable_msi(adapter->pdev)) {
1811                         adapter->flags |= FLAG_MSI_ENABLED;
1812                 } else {
1813                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1814                         e_err("Failed to initialize MSI interrupts.  Falling "
1815                               "back to legacy interrupts.\n");
1816                 }
1817                 /* Fall through */
1818         case E1000E_INT_MODE_LEGACY:
1819                 /* Don't do anything; this is the system default */
1820                 break;
1821         }
1822 }
1823
1824 /**
1825  * e1000_request_msix - Initialize MSI-X interrupts
1826  *
1827  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1828  * kernel.
1829  **/
1830 static int e1000_request_msix(struct e1000_adapter *adapter)
1831 {
1832         struct net_device *netdev = adapter->netdev;
1833         int err = 0, vector = 0;
1834
1835         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1836                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1837         else
1838                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1839         err = request_irq(adapter->msix_entries[vector].vector,
1840                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1841                           netdev);
1842         if (err)
1843                 goto out;
1844         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1845         adapter->rx_ring->itr_val = adapter->itr;
1846         vector++;
1847
1848         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1849                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1850         else
1851                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1852         err = request_irq(adapter->msix_entries[vector].vector,
1853                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1854                           netdev);
1855         if (err)
1856                 goto out;
1857         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1858         adapter->tx_ring->itr_val = adapter->itr;
1859         vector++;
1860
1861         err = request_irq(adapter->msix_entries[vector].vector,
1862                           e1000_msix_other, 0, netdev->name, netdev);
1863         if (err)
1864                 goto out;
1865
1866         e1000_configure_msix(adapter);
1867         return 0;
1868 out:
1869         return err;
1870 }
1871
1872 /**
1873  * e1000_request_irq - initialize interrupts
1874  *
1875  * Attempts to configure interrupts using the best available
1876  * capabilities of the hardware and kernel.
1877  **/
1878 static int e1000_request_irq(struct e1000_adapter *adapter)
1879 {
1880         struct net_device *netdev = adapter->netdev;
1881         int err;
1882
1883         if (adapter->msix_entries) {
1884                 err = e1000_request_msix(adapter);
1885                 if (!err)
1886                         return err;
1887                 /* fall back to MSI */
1888                 e1000e_reset_interrupt_capability(adapter);
1889                 adapter->int_mode = E1000E_INT_MODE_MSI;
1890                 e1000e_set_interrupt_capability(adapter);
1891         }
1892         if (adapter->flags & FLAG_MSI_ENABLED) {
1893                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1894                                   netdev->name, netdev);
1895                 if (!err)
1896                         return err;
1897
1898                 /* fall back to legacy interrupt */
1899                 e1000e_reset_interrupt_capability(adapter);
1900                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1901         }
1902
1903         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1904                           netdev->name, netdev);
1905         if (err)
1906                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1907
1908         return err;
1909 }
1910
1911 static void e1000_free_irq(struct e1000_adapter *adapter)
1912 {
1913         struct net_device *netdev = adapter->netdev;
1914
1915         if (adapter->msix_entries) {
1916                 int vector = 0;
1917
1918                 free_irq(adapter->msix_entries[vector].vector, netdev);
1919                 vector++;
1920
1921                 free_irq(adapter->msix_entries[vector].vector, netdev);
1922                 vector++;
1923
1924                 /* Other Causes interrupt vector */
1925                 free_irq(adapter->msix_entries[vector].vector, netdev);
1926                 return;
1927         }
1928
1929         free_irq(adapter->pdev->irq, netdev);
1930 }
1931
1932 /**
1933  * e1000_irq_disable - Mask off interrupt generation on the NIC
1934  **/
1935 static void e1000_irq_disable(struct e1000_adapter *adapter)
1936 {
1937         struct e1000_hw *hw = &adapter->hw;
1938
1939         ew32(IMC, ~0);
1940         if (adapter->msix_entries)
1941                 ew32(EIAC_82574, 0);
1942         e1e_flush();
1943         synchronize_irq(adapter->pdev->irq);
1944 }
1945
1946 /**
1947  * e1000_irq_enable - Enable default interrupt generation settings
1948  **/
1949 static void e1000_irq_enable(struct e1000_adapter *adapter)
1950 {
1951         struct e1000_hw *hw = &adapter->hw;
1952
1953         if (adapter->msix_entries) {
1954                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1955                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1956         } else {
1957                 ew32(IMS, IMS_ENABLE_MASK);
1958         }
1959         e1e_flush();
1960 }
1961
1962 /**
1963  * e1000_get_hw_control - get control of the h/w from f/w
1964  * @adapter: address of board private structure
1965  *
1966  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1967  * For ASF and Pass Through versions of f/w this means that
1968  * the driver is loaded. For AMT version (only with 82573)
1969  * of the f/w this means that the network i/f is open.
1970  **/
1971 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1972 {
1973         struct e1000_hw *hw = &adapter->hw;
1974         u32 ctrl_ext;
1975         u32 swsm;
1976
1977         /* Let firmware know the driver has taken over */
1978         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1979                 swsm = er32(SWSM);
1980                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1981         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1982                 ctrl_ext = er32(CTRL_EXT);
1983                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1984         }
1985 }
1986
1987 /**
1988  * e1000_release_hw_control - release control of the h/w to f/w
1989  * @adapter: address of board private structure
1990  *
1991  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1992  * For ASF and Pass Through versions of f/w this means that the
1993  * driver is no longer loaded. For AMT version (only with 82573) i
1994  * of the f/w this means that the network i/f is closed.
1995  *
1996  **/
1997 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1998 {
1999         struct e1000_hw *hw = &adapter->hw;
2000         u32 ctrl_ext;
2001         u32 swsm;
2002
2003         /* Let firmware taken over control of h/w */
2004         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2005                 swsm = er32(SWSM);
2006                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2007         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2008                 ctrl_ext = er32(CTRL_EXT);
2009                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2010         }
2011 }
2012
2013 /**
2014  * @e1000_alloc_ring - allocate memory for a ring structure
2015  **/
2016 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2017                                 struct e1000_ring *ring)
2018 {
2019         struct pci_dev *pdev = adapter->pdev;
2020
2021         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2022                                         GFP_KERNEL);
2023         if (!ring->desc)
2024                 return -ENOMEM;
2025
2026         return 0;
2027 }
2028
2029 /**
2030  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2031  * @adapter: board private structure
2032  *
2033  * Return 0 on success, negative on failure
2034  **/
2035 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
2036 {
2037         struct e1000_ring *tx_ring = adapter->tx_ring;
2038         int err = -ENOMEM, size;
2039
2040         size = sizeof(struct e1000_buffer) * tx_ring->count;
2041         tx_ring->buffer_info = vmalloc(size);
2042         if (!tx_ring->buffer_info)
2043                 goto err;
2044         memset(tx_ring->buffer_info, 0, size);
2045
2046         /* round up to nearest 4K */
2047         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2048         tx_ring->size = ALIGN(tx_ring->size, 4096);
2049
2050         err = e1000_alloc_ring_dma(adapter, tx_ring);
2051         if (err)
2052                 goto err;
2053
2054         tx_ring->next_to_use = 0;
2055         tx_ring->next_to_clean = 0;
2056
2057         return 0;
2058 err:
2059         vfree(tx_ring->buffer_info);
2060         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2061         return err;
2062 }
2063
2064 /**
2065  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2066  * @adapter: board private structure
2067  *
2068  * Returns 0 on success, negative on failure
2069  **/
2070 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
2071 {
2072         struct e1000_ring *rx_ring = adapter->rx_ring;
2073         struct e1000_buffer *buffer_info;
2074         int i, size, desc_len, err = -ENOMEM;
2075
2076         size = sizeof(struct e1000_buffer) * rx_ring->count;
2077         rx_ring->buffer_info = vmalloc(size);
2078         if (!rx_ring->buffer_info)
2079                 goto err;
2080         memset(rx_ring->buffer_info, 0, size);
2081
2082         for (i = 0; i < rx_ring->count; i++) {
2083                 buffer_info = &rx_ring->buffer_info[i];
2084                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2085                                                 sizeof(struct e1000_ps_page),
2086                                                 GFP_KERNEL);
2087                 if (!buffer_info->ps_pages)
2088                         goto err_pages;
2089         }
2090
2091         desc_len = sizeof(union e1000_rx_desc_packet_split);
2092
2093         /* Round up to nearest 4K */
2094         rx_ring->size = rx_ring->count * desc_len;
2095         rx_ring->size = ALIGN(rx_ring->size, 4096);
2096
2097         err = e1000_alloc_ring_dma(adapter, rx_ring);
2098         if (err)
2099                 goto err_pages;
2100
2101         rx_ring->next_to_clean = 0;
2102         rx_ring->next_to_use = 0;
2103         rx_ring->rx_skb_top = NULL;
2104
2105         return 0;
2106
2107 err_pages:
2108         for (i = 0; i < rx_ring->count; i++) {
2109                 buffer_info = &rx_ring->buffer_info[i];
2110                 kfree(buffer_info->ps_pages);
2111         }
2112 err:
2113         vfree(rx_ring->buffer_info);
2114         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2115         return err;
2116 }
2117
2118 /**
2119  * e1000_clean_tx_ring - Free Tx Buffers
2120  * @adapter: board private structure
2121  **/
2122 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
2123 {
2124         struct e1000_ring *tx_ring = adapter->tx_ring;
2125         struct e1000_buffer *buffer_info;
2126         unsigned long size;
2127         unsigned int i;
2128
2129         for (i = 0; i < tx_ring->count; i++) {
2130                 buffer_info = &tx_ring->buffer_info[i];
2131                 e1000_put_txbuf(adapter, buffer_info);
2132         }
2133
2134         size = sizeof(struct e1000_buffer) * tx_ring->count;
2135         memset(tx_ring->buffer_info, 0, size);
2136
2137         memset(tx_ring->desc, 0, tx_ring->size);
2138
2139         tx_ring->next_to_use = 0;
2140         tx_ring->next_to_clean = 0;
2141
2142         writel(0, adapter->hw.hw_addr + tx_ring->head);
2143         writel(0, adapter->hw.hw_addr + tx_ring->tail);
2144 }
2145
2146 /**
2147  * e1000e_free_tx_resources - Free Tx Resources per Queue
2148  * @adapter: board private structure
2149  *
2150  * Free all transmit software resources
2151  **/
2152 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
2153 {
2154         struct pci_dev *pdev = adapter->pdev;
2155         struct e1000_ring *tx_ring = adapter->tx_ring;
2156
2157         e1000_clean_tx_ring(adapter);
2158
2159         vfree(tx_ring->buffer_info);
2160         tx_ring->buffer_info = NULL;
2161
2162         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2163                           tx_ring->dma);
2164         tx_ring->desc = NULL;
2165 }
2166
2167 /**
2168  * e1000e_free_rx_resources - Free Rx Resources
2169  * @adapter: board private structure
2170  *
2171  * Free all receive software resources
2172  **/
2173
2174 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
2175 {
2176         struct pci_dev *pdev = adapter->pdev;
2177         struct e1000_ring *rx_ring = adapter->rx_ring;
2178         int i;
2179
2180         e1000_clean_rx_ring(adapter);
2181
2182         for (i = 0; i < rx_ring->count; i++) {
2183                 kfree(rx_ring->buffer_info[i].ps_pages);
2184         }
2185
2186         vfree(rx_ring->buffer_info);
2187         rx_ring->buffer_info = NULL;
2188
2189         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2190                           rx_ring->dma);
2191         rx_ring->desc = NULL;
2192 }
2193
2194 /**
2195  * e1000_update_itr - update the dynamic ITR value based on statistics
2196  * @adapter: pointer to adapter
2197  * @itr_setting: current adapter->itr
2198  * @packets: the number of packets during this measurement interval
2199  * @bytes: the number of bytes during this measurement interval
2200  *
2201  *      Stores a new ITR value based on packets and byte
2202  *      counts during the last interrupt.  The advantage of per interrupt
2203  *      computation is faster updates and more accurate ITR for the current
2204  *      traffic pattern.  Constants in this function were computed
2205  *      based on theoretical maximum wire speed and thresholds were set based
2206  *      on testing data as well as attempting to minimize response time
2207  *      while increasing bulk throughput.  This functionality is controlled
2208  *      by the InterruptThrottleRate module parameter.
2209  **/
2210 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2211                                      u16 itr_setting, int packets,
2212                                      int bytes)
2213 {
2214         unsigned int retval = itr_setting;
2215
2216         if (packets == 0)
2217                 goto update_itr_done;
2218
2219         switch (itr_setting) {
2220         case lowest_latency:
2221                 /* handle TSO and jumbo frames */
2222                 if (bytes/packets > 8000)
2223                         retval = bulk_latency;
2224                 else if ((packets < 5) && (bytes > 512)) {
2225                         retval = low_latency;
2226                 }
2227                 break;
2228         case low_latency:  /* 50 usec aka 20000 ints/s */
2229                 if (bytes > 10000) {
2230                         /* this if handles the TSO accounting */
2231                         if (bytes/packets > 8000) {
2232                                 retval = bulk_latency;
2233                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
2234                                 retval = bulk_latency;
2235                         } else if ((packets > 35)) {
2236                                 retval = lowest_latency;
2237                         }
2238                 } else if (bytes/packets > 2000) {
2239                         retval = bulk_latency;
2240                 } else if (packets <= 2 && bytes < 512) {
2241                         retval = lowest_latency;
2242                 }
2243                 break;
2244         case bulk_latency: /* 250 usec aka 4000 ints/s */
2245                 if (bytes > 25000) {
2246                         if (packets > 35) {
2247                                 retval = low_latency;
2248                         }
2249                 } else if (bytes < 6000) {
2250                         retval = low_latency;
2251                 }
2252                 break;
2253         }
2254
2255 update_itr_done:
2256         return retval;
2257 }
2258
2259 static void e1000_set_itr(struct e1000_adapter *adapter)
2260 {
2261         struct e1000_hw *hw = &adapter->hw;
2262         u16 current_itr;
2263         u32 new_itr = adapter->itr;
2264
2265         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2266         if (adapter->link_speed != SPEED_1000) {
2267                 current_itr = 0;
2268                 new_itr = 4000;
2269                 goto set_itr_now;
2270         }
2271
2272         adapter->tx_itr = e1000_update_itr(adapter,
2273                                     adapter->tx_itr,
2274                                     adapter->total_tx_packets,
2275                                     adapter->total_tx_bytes);
2276         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2277         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2278                 adapter->tx_itr = low_latency;
2279
2280         adapter->rx_itr = e1000_update_itr(adapter,
2281                                     adapter->rx_itr,
2282                                     adapter->total_rx_packets,
2283                                     adapter->total_rx_bytes);
2284         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2285         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2286                 adapter->rx_itr = low_latency;
2287
2288         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2289
2290         switch (current_itr) {
2291         /* counts and packets in update_itr are dependent on these numbers */
2292         case lowest_latency:
2293                 new_itr = 70000;
2294                 break;
2295         case low_latency:
2296                 new_itr = 20000; /* aka hwitr = ~200 */
2297                 break;
2298         case bulk_latency:
2299                 new_itr = 4000;
2300                 break;
2301         default:
2302                 break;
2303         }
2304
2305 set_itr_now:
2306         if (new_itr != adapter->itr) {
2307                 /*
2308                  * this attempts to bias the interrupt rate towards Bulk
2309                  * by adding intermediate steps when interrupt rate is
2310                  * increasing
2311                  */
2312                 new_itr = new_itr > adapter->itr ?
2313                              min(adapter->itr + (new_itr >> 2), new_itr) :
2314                              new_itr;
2315                 adapter->itr = new_itr;
2316                 adapter->rx_ring->itr_val = new_itr;
2317                 if (adapter->msix_entries)
2318                         adapter->rx_ring->set_itr = 1;
2319                 else
2320                         ew32(ITR, 1000000000 / (new_itr * 256));
2321         }
2322 }
2323
2324 /**
2325  * e1000_alloc_queues - Allocate memory for all rings
2326  * @adapter: board private structure to initialize
2327  **/
2328 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2329 {
2330         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2331         if (!adapter->tx_ring)
2332                 goto err;
2333
2334         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2335         if (!adapter->rx_ring)
2336                 goto err;
2337
2338         return 0;
2339 err:
2340         e_err("Unable to allocate memory for queues\n");
2341         kfree(adapter->rx_ring);
2342         kfree(adapter->tx_ring);
2343         return -ENOMEM;
2344 }
2345
2346 /**
2347  * e1000_clean - NAPI Rx polling callback
2348  * @napi: struct associated with this polling callback
2349  * @budget: amount of packets driver is allowed to process this poll
2350  **/
2351 static int e1000_clean(struct napi_struct *napi, int budget)
2352 {
2353         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2354         struct e1000_hw *hw = &adapter->hw;
2355         struct net_device *poll_dev = adapter->netdev;
2356         int tx_cleaned = 1, work_done = 0;
2357
2358         adapter = netdev_priv(poll_dev);
2359
2360         if (adapter->msix_entries &&
2361             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2362                 goto clean_rx;
2363
2364         tx_cleaned = e1000_clean_tx_irq(adapter);
2365
2366 clean_rx:
2367         adapter->clean_rx(adapter, &work_done, budget);
2368
2369         if (!tx_cleaned)
2370                 work_done = budget;
2371
2372         /* If budget not fully consumed, exit the polling mode */
2373         if (work_done < budget) {
2374                 if (adapter->itr_setting & 3)
2375                         e1000_set_itr(adapter);
2376                 napi_complete(napi);
2377                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2378                         if (adapter->msix_entries)
2379                                 ew32(IMS, adapter->rx_ring->ims_val);
2380                         else
2381                                 e1000_irq_enable(adapter);
2382                 }
2383         }
2384
2385         return work_done;
2386 }
2387
2388 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2389 {
2390         struct e1000_adapter *adapter = netdev_priv(netdev);
2391         struct e1000_hw *hw = &adapter->hw;
2392         u32 vfta, index;
2393
2394         /* don't update vlan cookie if already programmed */
2395         if ((adapter->hw.mng_cookie.status &
2396              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2397             (vid == adapter->mng_vlan_id))
2398                 return;
2399
2400         /* add VID to filter table */
2401         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2402                 index = (vid >> 5) & 0x7F;
2403                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2404                 vfta |= (1 << (vid & 0x1F));
2405                 hw->mac.ops.write_vfta(hw, index, vfta);
2406         }
2407 }
2408
2409 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2410 {
2411         struct e1000_adapter *adapter = netdev_priv(netdev);
2412         struct e1000_hw *hw = &adapter->hw;
2413         u32 vfta, index;
2414
2415         if (!test_bit(__E1000_DOWN, &adapter->state))
2416                 e1000_irq_disable(adapter);
2417         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2418
2419         if (!test_bit(__E1000_DOWN, &adapter->state))
2420                 e1000_irq_enable(adapter);
2421
2422         if ((adapter->hw.mng_cookie.status &
2423              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2424             (vid == adapter->mng_vlan_id)) {
2425                 /* release control to f/w */
2426                 e1000_release_hw_control(adapter);
2427                 return;
2428         }
2429
2430         /* remove VID from filter table */
2431         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2432                 index = (vid >> 5) & 0x7F;
2433                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2434                 vfta &= ~(1 << (vid & 0x1F));
2435                 hw->mac.ops.write_vfta(hw, index, vfta);
2436         }
2437 }
2438
2439 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2440 {
2441         struct net_device *netdev = adapter->netdev;
2442         u16 vid = adapter->hw.mng_cookie.vlan_id;
2443         u16 old_vid = adapter->mng_vlan_id;
2444
2445         if (!adapter->vlgrp)
2446                 return;
2447
2448         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2449                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2450                 if (adapter->hw.mng_cookie.status &
2451                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2452                         e1000_vlan_rx_add_vid(netdev, vid);
2453                         adapter->mng_vlan_id = vid;
2454                 }
2455
2456                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2457                                 (vid != old_vid) &&
2458                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2459                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2460         } else {
2461                 adapter->mng_vlan_id = vid;
2462         }
2463 }
2464
2465
2466 static void e1000_vlan_rx_register(struct net_device *netdev,
2467                                    struct vlan_group *grp)
2468 {
2469         struct e1000_adapter *adapter = netdev_priv(netdev);
2470         struct e1000_hw *hw = &adapter->hw;
2471         u32 ctrl, rctl;
2472
2473         if (!test_bit(__E1000_DOWN, &adapter->state))
2474                 e1000_irq_disable(adapter);
2475         adapter->vlgrp = grp;
2476
2477         if (grp) {
2478                 /* enable VLAN tag insert/strip */
2479                 ctrl = er32(CTRL);
2480                 ctrl |= E1000_CTRL_VME;
2481                 ew32(CTRL, ctrl);
2482
2483                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2484                         /* enable VLAN receive filtering */
2485                         rctl = er32(RCTL);
2486                         rctl &= ~E1000_RCTL_CFIEN;
2487                         ew32(RCTL, rctl);
2488                         e1000_update_mng_vlan(adapter);
2489                 }
2490         } else {
2491                 /* disable VLAN tag insert/strip */
2492                 ctrl = er32(CTRL);
2493                 ctrl &= ~E1000_CTRL_VME;
2494                 ew32(CTRL, ctrl);
2495
2496                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2497                         if (adapter->mng_vlan_id !=
2498                             (u16)E1000_MNG_VLAN_NONE) {
2499                                 e1000_vlan_rx_kill_vid(netdev,
2500                                                        adapter->mng_vlan_id);
2501                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2502                         }
2503                 }
2504         }
2505
2506         if (!test_bit(__E1000_DOWN, &adapter->state))
2507                 e1000_irq_enable(adapter);
2508 }
2509
2510 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2511 {
2512         u16 vid;
2513
2514         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2515
2516         if (!adapter->vlgrp)
2517                 return;
2518
2519         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2520                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2521                         continue;
2522                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2523         }
2524 }
2525
2526 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2527 {
2528         struct e1000_hw *hw = &adapter->hw;
2529         u32 manc, manc2h, mdef, i, j;
2530
2531         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2532                 return;
2533
2534         manc = er32(MANC);
2535
2536         /*
2537          * enable receiving management packets to the host. this will probably
2538          * generate destination unreachable messages from the host OS, but
2539          * the packets will be handled on SMBUS
2540          */
2541         manc |= E1000_MANC_EN_MNG2HOST;
2542         manc2h = er32(MANC2H);
2543
2544         switch (hw->mac.type) {
2545         default:
2546                 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2547                 break;
2548         case e1000_82574:
2549         case e1000_82583:
2550                 /*
2551                  * Check if IPMI pass-through decision filter already exists;
2552                  * if so, enable it.
2553                  */
2554                 for (i = 0, j = 0; i < 8; i++) {
2555                         mdef = er32(MDEF(i));
2556
2557                         /* Ignore filters with anything other than IPMI ports */
2558                         if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2559                                 continue;
2560
2561                         /* Enable this decision filter in MANC2H */
2562                         if (mdef)
2563                                 manc2h |= (1 << i);
2564
2565                         j |= mdef;
2566                 }
2567
2568                 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2569                         break;
2570
2571                 /* Create new decision filter in an empty filter */
2572                 for (i = 0, j = 0; i < 8; i++)
2573                         if (er32(MDEF(i)) == 0) {
2574                                 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2575                                                E1000_MDEF_PORT_664));
2576                                 manc2h |= (1 << 1);
2577                                 j++;
2578                                 break;
2579                         }
2580
2581                 if (!j)
2582                         e_warn("Unable to create IPMI pass-through filter\n");
2583                 break;
2584         }
2585
2586         ew32(MANC2H, manc2h);
2587         ew32(MANC, manc);
2588 }
2589
2590 /**
2591  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2592  * @adapter: board private structure
2593  *
2594  * Configure the Tx unit of the MAC after a reset.
2595  **/
2596 static void e1000_configure_tx(struct e1000_adapter *adapter)
2597 {
2598         struct e1000_hw *hw = &adapter->hw;
2599         struct e1000_ring *tx_ring = adapter->tx_ring;
2600         u64 tdba;
2601         u32 tdlen, tctl, tipg, tarc;
2602         u32 ipgr1, ipgr2;
2603
2604         /* Setup the HW Tx Head and Tail descriptor pointers */
2605         tdba = tx_ring->dma;
2606         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2607         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2608         ew32(TDBAH, (tdba >> 32));
2609         ew32(TDLEN, tdlen);
2610         ew32(TDH, 0);
2611         ew32(TDT, 0);
2612         tx_ring->head = E1000_TDH;
2613         tx_ring->tail = E1000_TDT;
2614
2615         /* Set the default values for the Tx Inter Packet Gap timer */
2616         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2617         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2618         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2619
2620         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2621                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2622
2623         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2624         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2625         ew32(TIPG, tipg);
2626
2627         /* Set the Tx Interrupt Delay register */
2628         ew32(TIDV, adapter->tx_int_delay);
2629         /* Tx irq moderation */
2630         ew32(TADV, adapter->tx_abs_int_delay);
2631
2632         /* Program the Transmit Control Register */
2633         tctl = er32(TCTL);
2634         tctl &= ~E1000_TCTL_CT;
2635         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2636                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2637
2638         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2639                 tarc = er32(TARC(0));
2640                 /*
2641                  * set the speed mode bit, we'll clear it if we're not at
2642                  * gigabit link later
2643                  */
2644 #define SPEED_MODE_BIT (1 << 21)
2645                 tarc |= SPEED_MODE_BIT;
2646                 ew32(TARC(0), tarc);
2647         }
2648
2649         /* errata: program both queues to unweighted RR */
2650         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2651                 tarc = er32(TARC(0));
2652                 tarc |= 1;
2653                 ew32(TARC(0), tarc);
2654                 tarc = er32(TARC(1));
2655                 tarc |= 1;
2656                 ew32(TARC(1), tarc);
2657         }
2658
2659         /* Setup Transmit Descriptor Settings for eop descriptor */
2660         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2661
2662         /* only set IDE if we are delaying interrupts using the timers */
2663         if (adapter->tx_int_delay)
2664                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2665
2666         /* enable Report Status bit */
2667         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2668
2669         ew32(TCTL, tctl);
2670
2671         e1000e_config_collision_dist(hw);
2672 }
2673
2674 /**
2675  * e1000_setup_rctl - configure the receive control registers
2676  * @adapter: Board private structure
2677  **/
2678 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2679                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2680 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2681 {
2682         struct e1000_hw *hw = &adapter->hw;
2683         u32 rctl, rfctl;
2684         u32 psrctl = 0;
2685         u32 pages = 0;
2686
2687         /* Program MC offset vector base */
2688         rctl = er32(RCTL);
2689         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2690         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2691                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2692                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2693
2694         /* Do not Store bad packets */
2695         rctl &= ~E1000_RCTL_SBP;
2696
2697         /* Enable Long Packet receive */
2698         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2699                 rctl &= ~E1000_RCTL_LPE;
2700         else
2701                 rctl |= E1000_RCTL_LPE;
2702
2703         /* Some systems expect that the CRC is included in SMBUS traffic. The
2704          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2705          * host memory when this is enabled
2706          */
2707         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2708                 rctl |= E1000_RCTL_SECRC;
2709
2710         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2711         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2712                 u16 phy_data;
2713
2714                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2715                 phy_data &= 0xfff8;
2716                 phy_data |= (1 << 2);
2717                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2718
2719                 e1e_rphy(hw, 22, &phy_data);
2720                 phy_data &= 0x0fff;
2721                 phy_data |= (1 << 14);
2722                 e1e_wphy(hw, 0x10, 0x2823);
2723                 e1e_wphy(hw, 0x11, 0x0003);
2724                 e1e_wphy(hw, 22, phy_data);
2725         }
2726
2727         /* Workaround Si errata on 82579 - configure jumbo frame flow */
2728         if (hw->mac.type == e1000_pch2lan) {
2729                 s32 ret_val;
2730
2731                 if (rctl & E1000_RCTL_LPE)
2732                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2733                 else
2734                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2735         }
2736
2737         /* Setup buffer sizes */
2738         rctl &= ~E1000_RCTL_SZ_4096;
2739         rctl |= E1000_RCTL_BSEX;
2740         switch (adapter->rx_buffer_len) {
2741         case 2048:
2742         default:
2743                 rctl |= E1000_RCTL_SZ_2048;
2744                 rctl &= ~E1000_RCTL_BSEX;
2745                 break;
2746         case 4096:
2747                 rctl |= E1000_RCTL_SZ_4096;
2748                 break;
2749         case 8192:
2750                 rctl |= E1000_RCTL_SZ_8192;
2751                 break;
2752         case 16384:
2753                 rctl |= E1000_RCTL_SZ_16384;
2754                 break;
2755         }
2756
2757         /*
2758          * 82571 and greater support packet-split where the protocol
2759          * header is placed in skb->data and the packet data is
2760          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2761          * In the case of a non-split, skb->data is linearly filled,
2762          * followed by the page buffers.  Therefore, skb->data is
2763          * sized to hold the largest protocol header.
2764          *
2765          * allocations using alloc_page take too long for regular MTU
2766          * so only enable packet split for jumbo frames
2767          *
2768          * Using pages when the page size is greater than 16k wastes
2769          * a lot of memory, since we allocate 3 pages at all times
2770          * per packet.
2771          */
2772         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2773         if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2774             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2775                 adapter->rx_ps_pages = pages;
2776         else
2777                 adapter->rx_ps_pages = 0;
2778
2779         if (adapter->rx_ps_pages) {
2780                 /* Configure extra packet-split registers */
2781                 rfctl = er32(RFCTL);
2782                 rfctl |= E1000_RFCTL_EXTEN;
2783                 /*
2784                  * disable packet split support for IPv6 extension headers,
2785                  * because some malformed IPv6 headers can hang the Rx
2786                  */
2787                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2788                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2789
2790                 ew32(RFCTL, rfctl);
2791
2792                 /* Enable Packet split descriptors */
2793                 rctl |= E1000_RCTL_DTYP_PS;
2794
2795                 psrctl |= adapter->rx_ps_bsize0 >>
2796                         E1000_PSRCTL_BSIZE0_SHIFT;
2797
2798                 switch (adapter->rx_ps_pages) {
2799                 case 3:
2800                         psrctl |= PAGE_SIZE <<
2801                                 E1000_PSRCTL_BSIZE3_SHIFT;
2802                 case 2:
2803                         psrctl |= PAGE_SIZE <<
2804                                 E1000_PSRCTL_BSIZE2_SHIFT;
2805                 case 1:
2806                         psrctl |= PAGE_SIZE >>
2807                                 E1000_PSRCTL_BSIZE1_SHIFT;
2808                         break;
2809                 }
2810
2811                 ew32(PSRCTL, psrctl);
2812         }
2813
2814         ew32(RCTL, rctl);
2815         /* just started the receive unit, no need to restart */
2816         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2817 }
2818
2819 /**
2820  * e1000_configure_rx - Configure Receive Unit after Reset
2821  * @adapter: board private structure
2822  *
2823  * Configure the Rx unit of the MAC after a reset.
2824  **/
2825 static void e1000_configure_rx(struct e1000_adapter *adapter)
2826 {
2827         struct e1000_hw *hw = &adapter->hw;
2828         struct e1000_ring *rx_ring = adapter->rx_ring;
2829         u64 rdba;
2830         u32 rdlen, rctl, rxcsum, ctrl_ext;
2831
2832         if (adapter->rx_ps_pages) {
2833                 /* this is a 32 byte descriptor */
2834                 rdlen = rx_ring->count *
2835                         sizeof(union e1000_rx_desc_packet_split);
2836                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2837                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2838         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2839                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2840                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2841                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2842         } else {
2843                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2844                 adapter->clean_rx = e1000_clean_rx_irq;
2845                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2846         }
2847
2848         /* disable receives while setting up the descriptors */
2849         rctl = er32(RCTL);
2850         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2851         e1e_flush();
2852         msleep(10);
2853
2854         /* set the Receive Delay Timer Register */
2855         ew32(RDTR, adapter->rx_int_delay);
2856
2857         /* irq moderation */
2858         ew32(RADV, adapter->rx_abs_int_delay);
2859         if (adapter->itr_setting != 0)
2860                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2861
2862         ctrl_ext = er32(CTRL_EXT);
2863         /* Auto-Mask interrupts upon ICR access */
2864         ctrl_ext |= E1000_CTRL_EXT_IAME;
2865         ew32(IAM, 0xffffffff);
2866         ew32(CTRL_EXT, ctrl_ext);
2867         e1e_flush();
2868
2869         /*
2870          * Setup the HW Rx Head and Tail Descriptor Pointers and
2871          * the Base and Length of the Rx Descriptor Ring
2872          */
2873         rdba = rx_ring->dma;
2874         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2875         ew32(RDBAH, (rdba >> 32));
2876         ew32(RDLEN, rdlen);
2877         ew32(RDH, 0);
2878         ew32(RDT, 0);
2879         rx_ring->head = E1000_RDH;
2880         rx_ring->tail = E1000_RDT;
2881
2882         /* Enable Receive Checksum Offload for TCP and UDP */
2883         rxcsum = er32(RXCSUM);
2884         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2885                 rxcsum |= E1000_RXCSUM_TUOFL;
2886
2887                 /*
2888                  * IPv4 payload checksum for UDP fragments must be
2889                  * used in conjunction with packet-split.
2890                  */
2891                 if (adapter->rx_ps_pages)
2892                         rxcsum |= E1000_RXCSUM_IPPCSE;
2893         } else {
2894                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2895                 /* no need to clear IPPCSE as it defaults to 0 */
2896         }
2897         ew32(RXCSUM, rxcsum);
2898
2899         /*
2900          * Enable early receives on supported devices, only takes effect when
2901          * packet size is equal or larger than the specified value (in 8 byte
2902          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2903          */
2904         if (adapter->flags & FLAG_HAS_ERT) {
2905                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2906                         u32 rxdctl = er32(RXDCTL(0));
2907                         ew32(RXDCTL(0), rxdctl | 0x3);
2908                         ew32(ERT, E1000_ERT_2048 | (1 << 13));
2909                         /*
2910                          * With jumbo frames and early-receive enabled,
2911                          * excessive C-state transition latencies result in
2912                          * dropped transactions.
2913                          */
2914                         pm_qos_update_request(
2915                                 adapter->netdev->pm_qos_req, 55);
2916                 } else {
2917                         pm_qos_update_request(
2918                                 adapter->netdev->pm_qos_req,
2919                                 PM_QOS_DEFAULT_VALUE);
2920                 }
2921         }
2922
2923         /* Enable Receives */
2924         ew32(RCTL, rctl);
2925 }
2926
2927 /**
2928  *  e1000_update_mc_addr_list - Update Multicast addresses
2929  *  @hw: pointer to the HW structure
2930  *  @mc_addr_list: array of multicast addresses to program
2931  *  @mc_addr_count: number of multicast addresses to program
2932  *
2933  *  Updates the Multicast Table Array.
2934  *  The caller must have a packed mc_addr_list of multicast addresses.
2935  **/
2936 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2937                                       u32 mc_addr_count)
2938 {
2939         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
2940 }
2941
2942 /**
2943  * e1000_set_multi - Multicast and Promiscuous mode set
2944  * @netdev: network interface device structure
2945  *
2946  * The set_multi entry point is called whenever the multicast address
2947  * list or the network interface flags are updated.  This routine is
2948  * responsible for configuring the hardware for proper multicast,
2949  * promiscuous mode, and all-multi behavior.
2950  **/
2951 static void e1000_set_multi(struct net_device *netdev)
2952 {
2953         struct e1000_adapter *adapter = netdev_priv(netdev);
2954         struct e1000_hw *hw = &adapter->hw;
2955         struct netdev_hw_addr *ha;
2956         u8  *mta_list;
2957         u32 rctl;
2958         int i;
2959
2960         /* Check for Promiscuous and All Multicast modes */
2961
2962         rctl = er32(RCTL);
2963
2964         if (netdev->flags & IFF_PROMISC) {
2965                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2966                 rctl &= ~E1000_RCTL_VFE;
2967         } else {
2968                 if (netdev->flags & IFF_ALLMULTI) {
2969                         rctl |= E1000_RCTL_MPE;
2970                         rctl &= ~E1000_RCTL_UPE;
2971                 } else {
2972                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2973                 }
2974                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2975                         rctl |= E1000_RCTL_VFE;
2976         }
2977
2978         ew32(RCTL, rctl);
2979
2980         if (!netdev_mc_empty(netdev)) {
2981                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
2982                 if (!mta_list)
2983                         return;
2984
2985                 /* prepare a packed array of only addresses. */
2986                 i = 0;
2987                 netdev_for_each_mc_addr(ha, netdev)
2988                         memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
2989
2990                 e1000_update_mc_addr_list(hw, mta_list, i);
2991                 kfree(mta_list);
2992         } else {
2993                 /*
2994                  * if we're called from probe, we might not have
2995                  * anything to do here, so clear out the list
2996                  */
2997                 e1000_update_mc_addr_list(hw, NULL, 0);
2998         }
2999 }
3000
3001 /**
3002  * e1000_configure - configure the hardware for Rx and Tx
3003  * @adapter: private board structure
3004  **/
3005 static void e1000_configure(struct e1000_adapter *adapter)
3006 {
3007         e1000_set_multi(adapter->netdev);
3008
3009         e1000_restore_vlan(adapter);
3010         e1000_init_manageability_pt(adapter);
3011
3012         e1000_configure_tx(adapter);
3013         e1000_setup_rctl(adapter);
3014         e1000_configure_rx(adapter);
3015         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
3016 }
3017
3018 /**
3019  * e1000e_power_up_phy - restore link in case the phy was powered down
3020  * @adapter: address of board private structure
3021  *
3022  * The phy may be powered down to save power and turn off link when the
3023  * driver is unloaded and wake on lan is not enabled (among others)
3024  * *** this routine MUST be followed by a call to e1000e_reset ***
3025  **/
3026 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3027 {
3028         if (adapter->hw.phy.ops.power_up)
3029                 adapter->hw.phy.ops.power_up(&adapter->hw);
3030
3031         adapter->hw.mac.ops.setup_link(&adapter->hw);
3032 }
3033
3034 /**
3035  * e1000_power_down_phy - Power down the PHY
3036  *
3037  * Power down the PHY so no link is implied when interface is down.
3038  * The PHY cannot be powered down if management or WoL is active.
3039  */
3040 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3041 {
3042         /* WoL is enabled */
3043         if (adapter->wol)
3044                 return;
3045
3046         if (adapter->hw.phy.ops.power_down)
3047                 adapter->hw.phy.ops.power_down(&adapter->hw);
3048 }
3049
3050 /**
3051  * e1000e_reset - bring the hardware into a known good state
3052  *
3053  * This function boots the hardware and enables some settings that
3054  * require a configuration cycle of the hardware - those cannot be
3055  * set/changed during runtime. After reset the device needs to be
3056  * properly configured for Rx, Tx etc.
3057  */
3058 void e1000e_reset(struct e1000_adapter *adapter)
3059 {
3060         struct e1000_mac_info *mac = &adapter->hw.mac;
3061         struct e1000_fc_info *fc = &adapter->hw.fc;
3062         struct e1000_hw *hw = &adapter->hw;
3063         u32 tx_space, min_tx_space, min_rx_space;
3064         u32 pba = adapter->pba;
3065         u16 hwm;
3066
3067         /* reset Packet Buffer Allocation to default */
3068         ew32(PBA, pba);
3069
3070         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3071                 /*
3072                  * To maintain wire speed transmits, the Tx FIFO should be
3073                  * large enough to accommodate two full transmit packets,
3074                  * rounded up to the next 1KB and expressed in KB.  Likewise,
3075                  * the Rx FIFO should be large enough to accommodate at least
3076                  * one full receive packet and is similarly rounded up and
3077                  * expressed in KB.
3078                  */
3079                 pba = er32(PBA);
3080                 /* upper 16 bits has Tx packet buffer allocation size in KB */
3081                 tx_space = pba >> 16;
3082                 /* lower 16 bits has Rx packet buffer allocation size in KB */
3083                 pba &= 0xffff;
3084                 /*
3085                  * the Tx fifo also stores 16 bytes of information about the tx
3086                  * but don't include ethernet FCS because hardware appends it
3087                  */
3088                 min_tx_space = (adapter->max_frame_size +
3089                                 sizeof(struct e1000_tx_desc) -
3090                                 ETH_FCS_LEN) * 2;
3091                 min_tx_space = ALIGN(min_tx_space, 1024);
3092                 min_tx_space >>= 10;
3093                 /* software strips receive CRC, so leave room for it */
3094                 min_rx_space = adapter->max_frame_size;
3095                 min_rx_space = ALIGN(min_rx_space, 1024);
3096                 min_rx_space >>= 10;
3097
3098                 /*
3099                  * If current Tx allocation is less than the min Tx FIFO size,
3100                  * and the min Tx FIFO size is less than the current Rx FIFO
3101                  * allocation, take space away from current Rx allocation
3102                  */
3103                 if ((tx_space < min_tx_space) &&
3104                     ((min_tx_space - tx_space) < pba)) {
3105                         pba -= min_tx_space - tx_space;
3106
3107                         /*
3108                          * if short on Rx space, Rx wins and must trump tx
3109                          * adjustment or use Early Receive if available
3110                          */
3111                         if ((pba < min_rx_space) &&
3112                             (!(adapter->flags & FLAG_HAS_ERT)))
3113                                 /* ERT enabled in e1000_configure_rx */
3114                                 pba = min_rx_space;
3115                 }
3116
3117                 ew32(PBA, pba);
3118         }
3119
3120
3121         /*
3122          * flow control settings
3123          *
3124          * The high water mark must be low enough to fit one full frame
3125          * (or the size used for early receive) above it in the Rx FIFO.
3126          * Set it to the lower of:
3127          * - 90% of the Rx FIFO size, and
3128          * - the full Rx FIFO size minus the early receive size (for parts
3129          *   with ERT support assuming ERT set to E1000_ERT_2048), or
3130          * - the full Rx FIFO size minus one full frame
3131          */
3132         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3133                 fc->pause_time = 0xFFFF;
3134         else
3135                 fc->pause_time = E1000_FC_PAUSE_TIME;
3136         fc->send_xon = 1;
3137         fc->current_mode = fc->requested_mode;
3138
3139         switch (hw->mac.type) {
3140         default:
3141                 if ((adapter->flags & FLAG_HAS_ERT) &&
3142                     (adapter->netdev->mtu > ETH_DATA_LEN))
3143                         hwm = min(((pba << 10) * 9 / 10),
3144                                   ((pba << 10) - (E1000_ERT_2048 << 3)));
3145                 else
3146                         hwm = min(((pba << 10) * 9 / 10),
3147                                   ((pba << 10) - adapter->max_frame_size));
3148
3149                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3150                 fc->low_water = fc->high_water - 8;
3151                 break;
3152         case e1000_pchlan:
3153                 /*
3154                  * Workaround PCH LOM adapter hangs with certain network
3155                  * loads.  If hangs persist, try disabling Tx flow control.
3156                  */
3157                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3158                         fc->high_water = 0x3500;
3159                         fc->low_water  = 0x1500;
3160                 } else {
3161                         fc->high_water = 0x5000;
3162                         fc->low_water  = 0x3000;
3163                 }
3164                 fc->refresh_time = 0x1000;
3165                 break;
3166         case e1000_pch2lan:
3167                 fc->high_water = 0x05C20;
3168                 fc->low_water = 0x05048;
3169                 fc->pause_time = 0x0650;
3170                 fc->refresh_time = 0x0400;
3171                 break;
3172         }
3173
3174         /* Allow time for pending master requests to run */
3175         mac->ops.reset_hw(hw);
3176
3177         /*
3178          * For parts with AMT enabled, let the firmware know
3179          * that the network interface is in control
3180          */
3181         if (adapter->flags & FLAG_HAS_AMT)
3182                 e1000_get_hw_control(adapter);
3183
3184         ew32(WUC, 0);
3185         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP)
3186                 e1e_wphy(&adapter->hw, BM_WUC, 0);
3187
3188         if (mac->ops.init_hw(hw))
3189                 e_err("Hardware Error\n");
3190
3191         e1000_update_mng_vlan(adapter);
3192
3193         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3194         ew32(VET, ETH_P_8021Q);
3195
3196         e1000e_reset_adaptive(hw);
3197         e1000_get_phy_info(hw);
3198
3199         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3200             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3201                 u16 phy_data = 0;
3202                 /*
3203                  * speed up time to link by disabling smart power down, ignore
3204                  * the return value of this function because there is nothing
3205                  * different we would do if it failed
3206                  */
3207                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3208                 phy_data &= ~IGP02E1000_PM_SPD;
3209                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3210         }
3211 }
3212
3213 int e1000e_up(struct e1000_adapter *adapter)
3214 {
3215         struct e1000_hw *hw = &adapter->hw;
3216
3217         /* DMA latency requirement to workaround early-receive/jumbo issue */
3218         if (adapter->flags & FLAG_HAS_ERT)
3219                 adapter->netdev->pm_qos_req =
3220                         pm_qos_add_request(PM_QOS_CPU_DMA_LATENCY,
3221                                        PM_QOS_DEFAULT_VALUE);
3222
3223         /* hardware has been reset, we need to reload some things */
3224         e1000_configure(adapter);
3225
3226         clear_bit(__E1000_DOWN, &adapter->state);
3227
3228         napi_enable(&adapter->napi);
3229         if (adapter->msix_entries)
3230                 e1000_configure_msix(adapter);
3231         e1000_irq_enable(adapter);
3232
3233         netif_wake_queue(adapter->netdev);
3234
3235         /* fire a link change interrupt to start the watchdog */
3236         if (adapter->msix_entries)
3237                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3238         else
3239                 ew32(ICS, E1000_ICS_LSC);
3240
3241         return 0;
3242 }
3243
3244 void e1000e_down(struct e1000_adapter *adapter)
3245 {
3246         struct net_device *netdev = adapter->netdev;
3247         struct e1000_hw *hw = &adapter->hw;
3248         u32 tctl, rctl;
3249
3250         /*
3251          * signal that we're down so the interrupt handler does not
3252          * reschedule our watchdog timer
3253          */
3254         set_bit(__E1000_DOWN, &adapter->state);
3255
3256         /* disable receives in the hardware */
3257         rctl = er32(RCTL);
3258         ew32(RCTL, rctl & ~E1000_RCTL_EN);
3259         /* flush and sleep below */
3260
3261         netif_stop_queue(netdev);
3262
3263         /* disable transmits in the hardware */
3264         tctl = er32(TCTL);
3265         tctl &= ~E1000_TCTL_EN;
3266         ew32(TCTL, tctl);
3267         /* flush both disables and wait for them to finish */
3268         e1e_flush();
3269         msleep(10);
3270
3271         napi_disable(&adapter->napi);
3272         e1000_irq_disable(adapter);
3273
3274         del_timer_sync(&adapter->watchdog_timer);
3275         del_timer_sync(&adapter->phy_info_timer);
3276
3277         netif_carrier_off(netdev);
3278         adapter->link_speed = 0;
3279         adapter->link_duplex = 0;
3280
3281         if (!pci_channel_offline(adapter->pdev))
3282                 e1000e_reset(adapter);
3283         e1000_clean_tx_ring(adapter);
3284         e1000_clean_rx_ring(adapter);
3285
3286         if (adapter->flags & FLAG_HAS_ERT) {
3287                 pm_qos_remove_request(
3288                               adapter->netdev->pm_qos_req);
3289                 adapter->netdev->pm_qos_req = NULL;
3290         }
3291
3292         /*
3293          * TODO: for power management, we could drop the link and
3294          * pci_disable_device here.
3295          */
3296 }
3297
3298 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3299 {
3300         might_sleep();
3301         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3302                 msleep(1);
3303         e1000e_down(adapter);
3304         e1000e_up(adapter);
3305         clear_bit(__E1000_RESETTING, &adapter->state);
3306 }
3307
3308 /**
3309  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3310  * @adapter: board private structure to initialize
3311  *
3312  * e1000_sw_init initializes the Adapter private data structure.
3313  * Fields are initialized based on PCI device information and
3314  * OS network device settings (MTU size).
3315  **/
3316 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3317 {
3318         struct net_device *netdev = adapter->netdev;
3319
3320         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3321         adapter->rx_ps_bsize0 = 128;
3322         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3323         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3324
3325         e1000e_set_interrupt_capability(adapter);
3326
3327         if (e1000_alloc_queues(adapter))
3328                 return -ENOMEM;
3329
3330         /* Explicitly disable IRQ since the NIC can be in any state. */
3331         e1000_irq_disable(adapter);
3332
3333         set_bit(__E1000_DOWN, &adapter->state);
3334         return 0;
3335 }
3336
3337 /**
3338  * e1000_intr_msi_test - Interrupt Handler
3339  * @irq: interrupt number
3340  * @data: pointer to a network interface device structure
3341  **/
3342 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3343 {
3344         struct net_device *netdev = data;
3345         struct e1000_adapter *adapter = netdev_priv(netdev);
3346         struct e1000_hw *hw = &adapter->hw;
3347         u32 icr = er32(ICR);
3348
3349         e_dbg("icr is %08X\n", icr);
3350         if (icr & E1000_ICR_RXSEQ) {
3351                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3352                 wmb();
3353         }
3354
3355         return IRQ_HANDLED;
3356 }
3357
3358 /**
3359  * e1000_test_msi_interrupt - Returns 0 for successful test
3360  * @adapter: board private struct
3361  *
3362  * code flow taken from tg3.c
3363  **/
3364 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3365 {
3366         struct net_device *netdev = adapter->netdev;
3367         struct e1000_hw *hw = &adapter->hw;
3368         int err;
3369
3370         /* poll_enable hasn't been called yet, so don't need disable */
3371         /* clear any pending events */
3372         er32(ICR);
3373
3374         /* free the real vector and request a test handler */
3375         e1000_free_irq(adapter);
3376         e1000e_reset_interrupt_capability(adapter);
3377
3378         /* Assume that the test fails, if it succeeds then the test
3379          * MSI irq handler will unset this flag */
3380         adapter->flags |= FLAG_MSI_TEST_FAILED;
3381
3382         err = pci_enable_msi(adapter->pdev);
3383         if (err)
3384                 goto msi_test_failed;
3385
3386         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3387                           netdev->name, netdev);
3388         if (err) {
3389                 pci_disable_msi(adapter->pdev);
3390                 goto msi_test_failed;
3391         }
3392
3393         wmb();
3394
3395         e1000_irq_enable(adapter);
3396
3397         /* fire an unusual interrupt on the test handler */
3398         ew32(ICS, E1000_ICS_RXSEQ);
3399         e1e_flush();
3400         msleep(50);
3401
3402         e1000_irq_disable(adapter);
3403
3404         rmb();
3405
3406         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3407                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3408                 err = -EIO;
3409                 e_info("MSI interrupt test failed!\n");
3410         }
3411
3412         free_irq(adapter->pdev->irq, netdev);
3413         pci_disable_msi(adapter->pdev);
3414
3415         if (err == -EIO)
3416                 goto msi_test_failed;
3417
3418         /* okay so the test worked, restore settings */
3419         e_dbg("MSI interrupt test succeeded!\n");
3420 msi_test_failed:
3421         e1000e_set_interrupt_capability(adapter);
3422         e1000_request_irq(adapter);
3423         return err;
3424 }
3425
3426 /**
3427  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3428  * @adapter: board private struct
3429  *
3430  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3431  **/
3432 static int e1000_test_msi(struct e1000_adapter *adapter)
3433 {
3434         int err;
3435         u16 pci_cmd;
3436
3437         if (!(adapter->flags & FLAG_MSI_ENABLED))
3438                 return 0;
3439
3440         /* disable SERR in case the MSI write causes a master abort */
3441         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3442         pci_write_config_word(adapter->pdev, PCI_COMMAND,
3443                               pci_cmd & ~PCI_COMMAND_SERR);
3444
3445         err = e1000_test_msi_interrupt(adapter);
3446
3447         /* restore previous setting of command word */
3448         pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3449
3450         /* success ! */
3451         if (!err)
3452                 return 0;
3453
3454         /* EIO means MSI test failed */
3455         if (err != -EIO)
3456                 return err;
3457
3458         /* back to INTx mode */
3459         e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3460
3461         e1000_free_irq(adapter);
3462
3463         err = e1000_request_irq(adapter);
3464
3465         return err;
3466 }
3467
3468 /**
3469  * e1000_open - Called when a network interface is made active
3470  * @netdev: network interface device structure
3471  *
3472  * Returns 0 on success, negative value on failure
3473  *
3474  * The open entry point is called when a network interface is made
3475  * active by the system (IFF_UP).  At this point all resources needed
3476  * for transmit and receive operations are allocated, the interrupt
3477  * handler is registered with the OS, the watchdog timer is started,
3478  * and the stack is notified that the interface is ready.
3479  **/
3480 static int e1000_open(struct net_device *netdev)
3481 {
3482         struct e1000_adapter *adapter = netdev_priv(netdev);
3483         struct e1000_hw *hw = &adapter->hw;
3484         struct pci_dev *pdev = adapter->pdev;
3485         int err;
3486
3487         /* disallow open during test */
3488         if (test_bit(__E1000_TESTING, &adapter->state))
3489                 return -EBUSY;
3490
3491         pm_runtime_get_sync(&pdev->dev);
3492
3493         netif_carrier_off(netdev);
3494
3495         /* allocate transmit descriptors */
3496         err = e1000e_setup_tx_resources(adapter);
3497         if (err)
3498                 goto err_setup_tx;
3499
3500         /* allocate receive descriptors */
3501         err = e1000e_setup_rx_resources(adapter);
3502         if (err)
3503                 goto err_setup_rx;
3504
3505         /*
3506          * If AMT is enabled, let the firmware know that the network
3507          * interface is now open and reset the part to a known state.
3508          */
3509         if (adapter->flags & FLAG_HAS_AMT) {
3510                 e1000_get_hw_control(adapter);
3511                 e1000e_reset(adapter);
3512         }
3513
3514         e1000e_power_up_phy(adapter);
3515
3516         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3517         if ((adapter->hw.mng_cookie.status &
3518              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3519                 e1000_update_mng_vlan(adapter);
3520
3521         /*
3522          * before we allocate an interrupt, we must be ready to handle it.
3523          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3524          * as soon as we call pci_request_irq, so we have to setup our
3525          * clean_rx handler before we do so.
3526          */
3527         e1000_configure(adapter);
3528
3529         err = e1000_request_irq(adapter);
3530         if (err)
3531                 goto err_req_irq;
3532
3533         /*
3534          * Work around PCIe errata with MSI interrupts causing some chipsets to
3535          * ignore e1000e MSI messages, which means we need to test our MSI
3536          * interrupt now
3537          */
3538         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3539                 err = e1000_test_msi(adapter);
3540                 if (err) {
3541                         e_err("Interrupt allocation failed\n");
3542                         goto err_req_irq;
3543                 }
3544         }
3545
3546         /* From here on the code is the same as e1000e_up() */
3547         clear_bit(__E1000_DOWN, &adapter->state);
3548
3549         napi_enable(&adapter->napi);
3550
3551         e1000_irq_enable(adapter);
3552
3553         netif_start_queue(netdev);
3554
3555         adapter->idle_check = true;
3556         pm_runtime_put(&pdev->dev);
3557
3558         /* fire a link status change interrupt to start the watchdog */
3559         if (adapter->msix_entries)
3560                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3561         else
3562                 ew32(ICS, E1000_ICS_LSC);
3563
3564         return 0;
3565
3566 err_req_irq:
3567         e1000_release_hw_control(adapter);
3568         e1000_power_down_phy(adapter);
3569         e1000e_free_rx_resources(adapter);
3570 err_setup_rx:
3571         e1000e_free_tx_resources(adapter);
3572 err_setup_tx:
3573         e1000e_reset(adapter);
3574         pm_runtime_put_sync(&pdev->dev);
3575
3576         return err;
3577 }
3578
3579 /**
3580  * e1000_close - Disables a network interface
3581  * @netdev: network interface device structure
3582  *
3583  * Returns 0, this is not allowed to fail
3584  *
3585  * The close entry point is called when an interface is de-activated
3586  * by the OS.  The hardware is still under the drivers control, but
3587  * needs to be disabled.  A global MAC reset is issued to stop the
3588  * hardware, and all transmit and receive resources are freed.
3589  **/
3590 static int e1000_close(struct net_device *netdev)
3591 {
3592         struct e1000_adapter *adapter = netdev_priv(netdev);
3593         struct pci_dev *pdev = adapter->pdev;
3594
3595         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3596
3597         pm_runtime_get_sync(&pdev->dev);
3598
3599         if (!test_bit(__E1000_DOWN, &adapter->state)) {
3600                 e1000e_down(adapter);
3601                 e1000_free_irq(adapter);
3602         }
3603         e1000_power_down_phy(adapter);
3604
3605         e1000e_free_tx_resources(adapter);
3606         e1000e_free_rx_resources(adapter);
3607
3608         /*
3609          * kill manageability vlan ID if supported, but not if a vlan with
3610          * the same ID is registered on the host OS (let 8021q kill it)
3611          */
3612         if ((adapter->hw.mng_cookie.status &
3613                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3614              !(adapter->vlgrp &&
3615                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3616                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3617
3618         /*
3619          * If AMT is enabled, let the firmware know that the network
3620          * interface is now closed
3621          */
3622         if (adapter->flags & FLAG_HAS_AMT)
3623                 e1000_release_hw_control(adapter);
3624
3625         pm_runtime_put_sync(&pdev->dev);
3626
3627         return 0;
3628 }
3629 /**
3630  * e1000_set_mac - Change the Ethernet Address of the NIC
3631  * @netdev: network interface device structure
3632  * @p: pointer to an address structure
3633  *
3634  * Returns 0 on success, negative on failure
3635  **/
3636 static int e1000_set_mac(struct net_device *netdev, void *p)
3637 {
3638         struct e1000_adapter *adapter = netdev_priv(netdev);
3639         struct sockaddr *addr = p;
3640
3641         if (!is_valid_ether_addr(addr->sa_data))
3642                 return -EADDRNOTAVAIL;
3643
3644         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3645         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3646
3647         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3648
3649         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3650                 /* activate the work around */
3651                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3652
3653                 /*
3654                  * Hold a copy of the LAA in RAR[14] This is done so that
3655                  * between the time RAR[0] gets clobbered  and the time it
3656                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3657                  * of the RARs and no incoming packets directed to this port
3658                  * are dropped. Eventually the LAA will be in RAR[0] and
3659                  * RAR[14]
3660                  */
3661                 e1000e_rar_set(&adapter->hw,
3662                               adapter->hw.mac.addr,
3663                               adapter->hw.mac.rar_entry_count - 1);
3664         }
3665
3666         return 0;
3667 }
3668
3669 /**
3670  * e1000e_update_phy_task - work thread to update phy
3671  * @work: pointer to our work struct
3672  *
3673  * this worker thread exists because we must acquire a
3674  * semaphore to read the phy, which we could msleep while
3675  * waiting for it, and we can't msleep in a timer.
3676  **/
3677 static void e1000e_update_phy_task(struct work_struct *work)
3678 {
3679         struct e1000_adapter *adapter = container_of(work,
3680                                         struct e1000_adapter, update_phy_task);
3681         e1000_get_phy_info(&adapter->hw);
3682 }
3683
3684 /*
3685  * Need to wait a few seconds after link up to get diagnostic information from
3686  * the phy
3687  */
3688 static void e1000_update_phy_info(unsigned long data)
3689 {
3690         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3691         schedule_work(&adapter->update_phy_task);
3692 }
3693
3694 /**
3695  * e1000e_update_phy_stats - Update the PHY statistics counters
3696  * @adapter: board private structure
3697  **/
3698 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
3699 {
3700         struct e1000_hw *hw = &adapter->hw;
3701         s32 ret_val;
3702         u16 phy_data;
3703
3704         ret_val = hw->phy.ops.acquire(hw);
3705         if (ret_val)
3706                 return;
3707
3708         hw->phy.addr = 1;
3709
3710 #define HV_PHY_STATS_PAGE       778
3711         /*
3712          * A page set is expensive so check if already on desired page.
3713          * If not, set to the page with the PHY status registers.
3714          */
3715         ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
3716                                            &phy_data);
3717         if (ret_val)
3718                 goto release;
3719         if (phy_data != (HV_PHY_STATS_PAGE << IGP_PAGE_SHIFT)) {
3720                 ret_val = e1000e_write_phy_reg_mdic(hw,
3721                                                     IGP01E1000_PHY_PAGE_SELECT,
3722                                                     (HV_PHY_STATS_PAGE <<
3723                                                      IGP_PAGE_SHIFT));
3724                 if (ret_val)
3725                         goto release;
3726         }
3727
3728         /* Read/clear the upper 16-bit registers and read/accumulate lower */
3729
3730         /* Single Collision Count */
3731         e1000e_read_phy_reg_mdic(hw, HV_SCC_UPPER & MAX_PHY_REG_ADDRESS,
3732                                  &phy_data);
3733         ret_val = e1000e_read_phy_reg_mdic(hw,
3734                                            HV_SCC_LOWER & MAX_PHY_REG_ADDRESS,
3735                                            &phy_data);
3736         if (!ret_val)
3737                 adapter->stats.scc += phy_data;
3738
3739         /* Excessive Collision Count */
3740         e1000e_read_phy_reg_mdic(hw, HV_ECOL_UPPER & MAX_PHY_REG_ADDRESS,
3741                                  &phy_data);
3742         ret_val = e1000e_read_phy_reg_mdic(hw,
3743                                            HV_ECOL_LOWER & MAX_PHY_REG_ADDRESS,
3744                                            &phy_data);
3745         if (!ret_val)
3746                 adapter->stats.ecol += phy_data;
3747
3748         /* Multiple Collision Count */
3749         e1000e_read_phy_reg_mdic(hw, HV_MCC_UPPER & MAX_PHY_REG_ADDRESS,
3750                                  &phy_data);
3751         ret_val = e1000e_read_phy_reg_mdic(hw,
3752                                            HV_MCC_LOWER & MAX_PHY_REG_ADDRESS,
3753                                            &phy_data);
3754         if (!ret_val)
3755                 adapter->stats.mcc += phy_data;
3756
3757         /* Late Collision Count */
3758         e1000e_read_phy_reg_mdic(hw, HV_LATECOL_UPPER & MAX_PHY_REG_ADDRESS,
3759                                  &phy_data);
3760         ret_val = e1000e_read_phy_reg_mdic(hw,
3761                                            HV_LATECOL_LOWER &
3762                                            MAX_PHY_REG_ADDRESS,
3763                                            &phy_data);
3764         if (!ret_val)
3765                 adapter->stats.latecol += phy_data;
3766
3767         /* Collision Count - also used for adaptive IFS */
3768         e1000e_read_phy_reg_mdic(hw, HV_COLC_UPPER & MAX_PHY_REG_ADDRESS,
3769                                  &phy_data);
3770         ret_val = e1000e_read_phy_reg_mdic(hw,
3771                                            HV_COLC_LOWER & MAX_PHY_REG_ADDRESS,
3772                                            &phy_data);
3773         if (!ret_val)
3774                 hw->mac.collision_delta = phy_data;
3775
3776         /* Defer Count */
3777         e1000e_read_phy_reg_mdic(hw, HV_DC_UPPER & MAX_PHY_REG_ADDRESS,
3778                                  &phy_data);
3779         ret_val = e1000e_read_phy_reg_mdic(hw,
3780                                            HV_DC_LOWER & MAX_PHY_REG_ADDRESS,
3781                                            &phy_data);
3782         if (!ret_val)
3783                 adapter->stats.dc += phy_data;
3784
3785         /* Transmit with no CRS */
3786         e1000e_read_phy_reg_mdic(hw, HV_TNCRS_UPPER & MAX_PHY_REG_ADDRESS,
3787                                  &phy_data);
3788         ret_val = e1000e_read_phy_reg_mdic(hw,
3789                                            HV_TNCRS_LOWER & MAX_PHY_REG_ADDRESS,
3790                                            &phy_data);
3791         if (!ret_val)
3792                 adapter->stats.tncrs += phy_data;
3793
3794 release:
3795         hw->phy.ops.release(hw);
3796 }
3797
3798 /**
3799  * e1000e_update_stats - Update the board statistics counters
3800  * @adapter: board private structure
3801  **/
3802 void e1000e_update_stats(struct e1000_adapter *adapter)
3803 {
3804         struct net_device *netdev = adapter->netdev;
3805         struct e1000_hw *hw = &adapter->hw;
3806         struct pci_dev *pdev = adapter->pdev;
3807
3808         /*
3809          * Prevent stats update while adapter is being reset, or if the pci
3810          * connection is down.
3811          */
3812         if (adapter->link_speed == 0)
3813                 return;
3814         if (pci_channel_offline(pdev))
3815                 return;
3816
3817         adapter->stats.crcerrs += er32(CRCERRS);
3818         adapter->stats.gprc += er32(GPRC);
3819         adapter->stats.gorc += er32(GORCL);
3820         er32(GORCH); /* Clear gorc */
3821         adapter->stats.bprc += er32(BPRC);
3822         adapter->stats.mprc += er32(MPRC);
3823         adapter->stats.roc += er32(ROC);
3824
3825         adapter->stats.mpc += er32(MPC);
3826
3827         /* Half-duplex statistics */
3828         if (adapter->link_duplex == HALF_DUPLEX) {
3829                 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
3830                         e1000e_update_phy_stats(adapter);
3831                 } else {
3832                         adapter->stats.scc += er32(SCC);
3833                         adapter->stats.ecol += er32(ECOL);
3834                         adapter->stats.mcc += er32(MCC);
3835                         adapter->stats.latecol += er32(LATECOL);
3836                         adapter->stats.dc += er32(DC);
3837
3838                         hw->mac.collision_delta = er32(COLC);
3839
3840                         if ((hw->mac.type != e1000_82574) &&
3841                             (hw->mac.type != e1000_82583))
3842                                 adapter->stats.tncrs += er32(TNCRS);
3843                 }
3844                 adapter->stats.colc += hw->mac.collision_delta;
3845         }
3846
3847         adapter->stats.xonrxc += er32(XONRXC);
3848         adapter->stats.xontxc += er32(XONTXC);
3849         adapter->stats.xoffrxc += er32(XOFFRXC);
3850         adapter->stats.xofftxc += er32(XOFFTXC);
3851         adapter->stats.gptc += er32(GPTC);
3852         adapter->stats.gotc += er32(GOTCL);
3853         er32(GOTCH); /* Clear gotc */
3854         adapter->stats.rnbc += er32(RNBC);
3855         adapter->stats.ruc += er32(RUC);
3856
3857         adapter->stats.mptc += er32(MPTC);
3858         adapter->stats.bptc += er32(BPTC);
3859
3860         /* used for adaptive IFS */
3861
3862         hw->mac.tx_packet_delta = er32(TPT);
3863         adapter->stats.tpt += hw->mac.tx_packet_delta;
3864
3865         adapter->stats.algnerrc += er32(ALGNERRC);
3866         adapter->stats.rxerrc += er32(RXERRC);
3867         adapter->stats.cexterr += er32(CEXTERR);
3868         adapter->stats.tsctc += er32(TSCTC);
3869         adapter->stats.tsctfc += er32(TSCTFC);
3870
3871         /* Fill out the OS statistics structure */
3872         netdev->stats.multicast = adapter->stats.mprc;
3873         netdev->stats.collisions = adapter->stats.colc;
3874
3875         /* Rx Errors */
3876
3877         /*
3878          * RLEC on some newer hardware can be incorrect so build
3879          * our own version based on RUC and ROC
3880          */
3881         netdev->stats.rx_errors = adapter->stats.rxerrc +
3882                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3883                 adapter->stats.ruc + adapter->stats.roc +
3884                 adapter->stats.cexterr;
3885         netdev->stats.rx_length_errors = adapter->stats.ruc +
3886                                               adapter->stats.roc;
3887         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3888         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3889         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3890
3891         /* Tx Errors */
3892         netdev->stats.tx_errors = adapter->stats.ecol +
3893                                        adapter->stats.latecol;
3894         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3895         netdev->stats.tx_window_errors = adapter->stats.latecol;
3896         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3897
3898         /* Tx Dropped needs to be maintained elsewhere */
3899
3900         /* Management Stats */
3901         adapter->stats.mgptc += er32(MGTPTC);
3902         adapter->stats.mgprc += er32(MGTPRC);
3903         adapter->stats.mgpdc += er32(MGTPDC);
3904 }
3905
3906 /**
3907  * e1000_phy_read_status - Update the PHY register status snapshot
3908  * @adapter: board private structure
3909  **/
3910 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3911 {
3912         struct e1000_hw *hw = &adapter->hw;
3913         struct e1000_phy_regs *phy = &adapter->phy_regs;
3914         int ret_val;
3915
3916         if ((er32(STATUS) & E1000_STATUS_LU) &&
3917             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3918                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3919                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3920                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3921                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3922                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3923                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3924                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3925                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3926                 if (ret_val)
3927                         e_warn("Error reading PHY register\n");
3928         } else {
3929                 /*
3930                  * Do not read PHY registers if link is not up
3931                  * Set values to typical power-on defaults
3932                  */
3933                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3934                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3935                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3936                              BMSR_ERCAP);
3937                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3938                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3939                 phy->lpa = 0;
3940                 phy->expansion = EXPANSION_ENABLENPAGE;
3941                 phy->ctrl1000 = ADVERTISE_1000FULL;
3942                 phy->stat1000 = 0;
3943                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3944         }
3945 }
3946
3947 static void e1000_print_link_info(struct e1000_adapter *adapter)
3948 {
3949         struct e1000_hw *hw = &adapter->hw;
3950         u32 ctrl = er32(CTRL);
3951
3952         /* Link status message must follow this format for user tools */
3953         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
3954                "Flow Control: %s\n",
3955                adapter->netdev->name,
3956                adapter->link_speed,
3957                (adapter->link_duplex == FULL_DUPLEX) ?
3958                                 "Full Duplex" : "Half Duplex",
3959                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3960                                 "RX/TX" :
3961                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3962                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3963 }
3964
3965 bool e1000e_has_link(struct e1000_adapter *adapter)
3966 {
3967         struct e1000_hw *hw = &adapter->hw;
3968         bool link_active = 0;
3969         s32 ret_val = 0;
3970
3971         /*
3972          * get_link_status is set on LSC (link status) interrupt or
3973          * Rx sequence error interrupt.  get_link_status will stay
3974          * false until the check_for_link establishes link
3975          * for copper adapters ONLY
3976          */
3977         switch (hw->phy.media_type) {
3978         case e1000_media_type_copper:
3979                 if (hw->mac.get_link_status) {
3980                         ret_val = hw->mac.ops.check_for_link(hw);
3981                         link_active = !hw->mac.get_link_status;
3982                 } else {
3983                         link_active = 1;
3984                 }
3985                 break;
3986         case e1000_media_type_fiber:
3987                 ret_val = hw->mac.ops.check_for_link(hw);
3988                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3989                 break;
3990         case e1000_media_type_internal_serdes:
3991                 ret_val = hw->mac.ops.check_for_link(hw);
3992                 link_active = adapter->hw.mac.serdes_has_link;
3993                 break;
3994         default:
3995         case e1000_media_type_unknown:
3996                 break;
3997         }
3998
3999         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4000             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4001                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4002                 e_info("Gigabit has been disabled, downgrading speed\n");
4003         }
4004
4005         return link_active;
4006 }
4007
4008 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4009 {
4010         /* make sure the receive unit is started */
4011         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4012             (adapter->flags & FLAG_RX_RESTART_NOW)) {
4013                 struct e1000_hw *hw = &adapter->hw;
4014                 u32 rctl = er32(RCTL);
4015                 ew32(RCTL, rctl | E1000_RCTL_EN);
4016                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
4017         }
4018 }
4019
4020 /**
4021  * e1000_watchdog - Timer Call-back
4022  * @data: pointer to adapter cast into an unsigned long
4023  **/
4024 static void e1000_watchdog(unsigned long data)
4025 {
4026         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4027
4028         /* Do the rest outside of interrupt context */
4029         schedule_work(&adapter->watchdog_task);
4030
4031         /* TODO: make this use queue_delayed_work() */
4032 }
4033
4034 static void e1000_watchdog_task(struct work_struct *work)
4035 {
4036         struct e1000_adapter *adapter = container_of(work,
4037                                         struct e1000_adapter, watchdog_task);
4038         struct net_device *netdev = adapter->netdev;
4039         struct e1000_mac_info *mac = &adapter->hw.mac;
4040         struct e1000_phy_info *phy = &adapter->hw.phy;
4041         struct e1000_ring *tx_ring = adapter->tx_ring;
4042         struct e1000_hw *hw = &adapter->hw;
4043         u32 link, tctl;
4044         int tx_pending = 0;
4045
4046         link = e1000e_has_link(adapter);
4047         if ((netif_carrier_ok(netdev)) && link) {
4048                 /* Cancel scheduled suspend requests. */
4049                 pm_runtime_resume(netdev->dev.parent);
4050
4051                 e1000e_enable_receives(adapter);
4052                 goto link_up;
4053         }
4054
4055         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4056             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4057                 e1000_update_mng_vlan(adapter);
4058
4059         if (link) {
4060                 if (!netif_carrier_ok(netdev)) {
4061                         bool txb2b = 1;
4062
4063                         /* Cancel scheduled suspend requests. */
4064                         pm_runtime_resume(netdev->dev.parent);
4065
4066                         /* update snapshot of PHY registers on LSC */
4067                         e1000_phy_read_status(adapter);
4068                         mac->ops.get_link_up_info(&adapter->hw,
4069                                                    &adapter->link_speed,
4070                                                    &adapter->link_duplex);
4071                         e1000_print_link_info(adapter);
4072                         /*
4073                          * On supported PHYs, check for duplex mismatch only
4074                          * if link has autonegotiated at 10/100 half
4075                          */
4076                         if ((hw->phy.type == e1000_phy_igp_3 ||
4077                              hw->phy.type == e1000_phy_bm) &&
4078                             (hw->mac.autoneg == true) &&
4079                             (adapter->link_speed == SPEED_10 ||
4080                              adapter->link_speed == SPEED_100) &&
4081                             (adapter->link_duplex == HALF_DUPLEX)) {
4082                                 u16 autoneg_exp;
4083
4084                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4085
4086                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4087                                         e_info("Autonegotiated half duplex but"
4088                                                " link partner cannot autoneg. "
4089                                                " Try forcing full duplex if "
4090                                                "link gets many collisions.\n");
4091                         }
4092
4093                         /* adjust timeout factor according to speed/duplex */
4094                         adapter->tx_timeout_factor = 1;
4095                         switch (adapter->link_speed) {
4096                         case SPEED_10:
4097                                 txb2b = 0;
4098                                 adapter->tx_timeout_factor = 16;
4099                                 break;
4100                         case SPEED_100:
4101                                 txb2b = 0;
4102                                 adapter->tx_timeout_factor = 10;
4103                                 break;
4104                         }
4105
4106                         /*
4107                          * workaround: re-program speed mode bit after
4108                          * link-up event
4109                          */
4110                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4111                             !txb2b) {
4112                                 u32 tarc0;
4113                                 tarc0 = er32(TARC(0));
4114                                 tarc0 &= ~SPEED_MODE_BIT;
4115                                 ew32(TARC(0), tarc0);
4116                         }
4117
4118                         /*
4119                          * disable TSO for pcie and 10/100 speeds, to avoid
4120                          * some hardware issues
4121                          */
4122                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
4123                                 switch (adapter->link_speed) {
4124                                 case SPEED_10:
4125                                 case SPEED_100:
4126                                         e_info("10/100 speed: disabling TSO\n");
4127                                         netdev->features &= ~NETIF_F_TSO;
4128                                         netdev->features &= ~NETIF_F_TSO6;
4129                                         break;
4130                                 case SPEED_1000:
4131                                         netdev->features |= NETIF_F_TSO;
4132                                         netdev->features |= NETIF_F_TSO6;
4133                                         break;
4134                                 default:
4135                                         /* oops */
4136                                         break;
4137                                 }
4138                         }
4139
4140                         /*
4141                          * enable transmits in the hardware, need to do this
4142                          * after setting TARC(0)
4143                          */
4144                         tctl = er32(TCTL);
4145                         tctl |= E1000_TCTL_EN;
4146                         ew32(TCTL, tctl);
4147
4148                         /*
4149                          * Perform any post-link-up configuration before
4150                          * reporting link up.
4151                          */
4152                         if (phy->ops.cfg_on_link_up)
4153                                 phy->ops.cfg_on_link_up(hw);
4154
4155                         netif_carrier_on(netdev);
4156
4157                         if (!test_bit(__E1000_DOWN, &adapter->state))
4158                                 mod_timer(&adapter->phy_info_timer,
4159                                           round_jiffies(jiffies + 2 * HZ));
4160                 }
4161         } else {
4162                 if (netif_carrier_ok(netdev)) {
4163                         adapter->link_speed = 0;
4164                         adapter->link_duplex = 0;
4165                         /* Link status message must follow this format */
4166                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4167                                adapter->netdev->name);
4168                         netif_carrier_off(netdev);
4169                         if (!test_bit(__E1000_DOWN, &adapter->state))
4170                                 mod_timer(&adapter->phy_info_timer,
4171                                           round_jiffies(jiffies + 2 * HZ));
4172
4173                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4174                                 schedule_work(&adapter->reset_task);
4175                         else
4176                                 pm_schedule_suspend(netdev->dev.parent,
4177                                                         LINK_TIMEOUT);
4178                 }
4179         }
4180
4181 link_up:
4182         e1000e_update_stats(adapter);
4183
4184         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4185         adapter->tpt_old = adapter->stats.tpt;
4186         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4187         adapter->colc_old = adapter->stats.colc;
4188
4189         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4190         adapter->gorc_old = adapter->stats.gorc;
4191         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4192         adapter->gotc_old = adapter->stats.gotc;
4193
4194         e1000e_update_adaptive(&adapter->hw);
4195
4196         if (!netif_carrier_ok(netdev)) {
4197                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
4198                                tx_ring->count);
4199                 if (tx_pending) {
4200                         /*
4201                          * We've lost link, so the controller stops DMA,
4202                          * but we've got queued Tx work that's never going
4203                          * to get done, so reset controller to flush Tx.
4204                          * (Do the reset outside of interrupt context).
4205                          */
4206                         adapter->tx_timeout_count++;
4207                         schedule_work(&adapter->reset_task);
4208                         /* return immediately since reset is imminent */
4209                         return;
4210                 }
4211         }
4212
4213         /* Simple mode for Interrupt Throttle Rate (ITR) */
4214         if (adapter->itr_setting == 4) {
4215                 /*
4216                  * Symmetric Tx/Rx gets a reduced ITR=2000;
4217                  * Total asymmetrical Tx or Rx gets ITR=8000;
4218                  * everyone else is between 2000-8000.
4219                  */
4220                 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4221                 u32 dif = (adapter->gotc > adapter->gorc ?
4222                             adapter->gotc - adapter->gorc :
4223                             adapter->gorc - adapter->gotc) / 10000;
4224                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4225
4226                 ew32(ITR, 1000000000 / (itr * 256));
4227         }
4228
4229         /* Cause software interrupt to ensure Rx ring is cleaned */
4230         if (adapter->msix_entries)
4231                 ew32(ICS, adapter->rx_ring->ims_val);
4232         else
4233                 ew32(ICS, E1000_ICS_RXDMT0);
4234
4235         /* Force detection of hung controller every watchdog period */
4236         adapter->detect_tx_hung = 1;
4237
4238         /*
4239          * With 82571 controllers, LAA may be overwritten due to controller
4240          * reset from the other port. Set the appropriate LAA in RAR[0]
4241          */
4242         if (e1000e_get_laa_state_82571(hw))
4243                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4244
4245         /* Reset the timer */
4246         if (!test_bit(__E1000_DOWN, &adapter->state))
4247                 mod_timer(&adapter->watchdog_timer,
4248                           round_jiffies(jiffies + 2 * HZ));
4249 }
4250
4251 #define E1000_TX_FLAGS_CSUM             0x00000001
4252 #define E1000_TX_FLAGS_VLAN             0x00000002
4253 #define E1000_TX_FLAGS_TSO              0x00000004
4254 #define E1000_TX_FLAGS_IPV4             0x00000008
4255 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
4256 #define E1000_TX_FLAGS_VLAN_SHIFT       16
4257
4258 static int e1000_tso(struct e1000_adapter *adapter,
4259                      struct sk_buff *skb)
4260 {
4261         struct e1000_ring *tx_ring = adapter->tx_ring;
4262         struct e1000_context_desc *context_desc;
4263         struct e1000_buffer *buffer_info;
4264         unsigned int i;
4265         u32 cmd_length = 0;
4266         u16 ipcse = 0, tucse, mss;
4267         u8 ipcss, ipcso, tucss, tucso, hdr_len;
4268         int err;
4269
4270         if (!skb_is_gso(skb))
4271                 return 0;
4272
4273         if (skb_header_cloned(skb)) {
4274                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4275                 if (err)
4276                         return err;
4277         }
4278
4279         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4280         mss = skb_shinfo(skb)->gso_size;
4281         if (skb->protocol == htons(ETH_P_IP)) {
4282                 struct iphdr *iph = ip_hdr(skb);
4283                 iph->tot_len = 0;
4284                 iph->check = 0;
4285                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4286                                                          0, IPPROTO_TCP, 0);
4287                 cmd_length = E1000_TXD_CMD_IP;
4288                 ipcse = skb_transport_offset(skb) - 1;
4289         } else if (skb_is_gso_v6(skb)) {
4290                 ipv6_hdr(skb)->payload_len = 0;
4291                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4292                                                        &ipv6_hdr(skb)->daddr,
4293                                                        0, IPPROTO_TCP, 0);
4294                 ipcse = 0;
4295         }
4296         ipcss = skb_network_offset(skb);
4297         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4298         tucss = skb_transport_offset(skb);
4299         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4300         tucse = 0;
4301
4302         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4303                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4304
4305         i = tx_ring->next_to_use;
4306         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4307         buffer_info = &tx_ring->buffer_info[i];
4308
4309         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4310         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4311         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4312         context_desc->upper_setup.tcp_fields.tucss = tucss;
4313         context_desc->upper_setup.tcp_fields.tucso = tucso;
4314         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4315         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4316         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4317         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4318
4319         buffer_info->time_stamp = jiffies;
4320         buffer_info->next_to_watch = i;
4321
4322         i++;
4323         if (i == tx_ring->count)
4324                 i = 0;
4325         tx_ring->next_to_use = i;
4326
4327         return 1;
4328 }
4329
4330 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
4331 {
4332         struct e1000_ring *tx_ring = adapter->tx_ring;
4333         struct e1000_context_desc *context_desc;
4334         struct e1000_buffer *buffer_info;
4335         unsigned int i;
4336         u8 css;
4337         u32 cmd_len = E1000_TXD_CMD_DEXT;
4338         __be16 protocol;
4339
4340         if (skb->ip_summed != CHECKSUM_PARTIAL)
4341                 return 0;
4342
4343         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4344                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4345         else
4346                 protocol = skb->protocol;
4347
4348         switch (protocol) {
4349         case cpu_to_be16(ETH_P_IP):
4350                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4351                         cmd_len |= E1000_TXD_CMD_TCP;
4352                 break;
4353         case cpu_to_be16(ETH_P_IPV6):
4354                 /* XXX not handling all IPV6 headers */
4355                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4356                         cmd_len |= E1000_TXD_CMD_TCP;
4357                 break;
4358         default:
4359                 if (unlikely(net_ratelimit()))
4360                         e_warn("checksum_partial proto=%x!\n",
4361                                be16_to_cpu(protocol));
4362                 break;
4363         }
4364
4365         css = skb_transport_offset(skb);
4366
4367         i = tx_ring->next_to_use;
4368         buffer_info = &tx_ring->buffer_info[i];
4369         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4370
4371         context_desc->lower_setup.ip_config = 0;
4372         context_desc->upper_setup.tcp_fields.tucss = css;
4373         context_desc->upper_setup.tcp_fields.tucso =
4374                                 css + skb->csum_offset;
4375         context_desc->upper_setup.tcp_fields.tucse = 0;
4376         context_desc->tcp_seg_setup.data = 0;
4377         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4378
4379         buffer_info->time_stamp = jiffies;
4380         buffer_info->next_to_watch = i;
4381
4382         i++;
4383         if (i == tx_ring->count)
4384                 i = 0;
4385         tx_ring->next_to_use = i;
4386
4387         return 1;
4388 }
4389
4390 #define E1000_MAX_PER_TXD       8192
4391 #define E1000_MAX_TXD_PWR       12
4392
4393 static int e1000_tx_map(struct e1000_adapter *adapter,
4394                         struct sk_buff *skb, unsigned int first,
4395                         unsigned int max_per_txd, unsigned int nr_frags,
4396                         unsigned int mss)
4397 {
4398         struct e1000_ring *tx_ring = adapter->tx_ring;
4399         struct pci_dev *pdev = adapter->pdev;
4400         struct e1000_buffer *buffer_info;
4401         unsigned int len = skb_headlen(skb);
4402         unsigned int offset = 0, size, count = 0, i;
4403         unsigned int f, bytecount, segs;
4404
4405         i = tx_ring->next_to_use;
4406
4407         while (len) {
4408                 buffer_info = &tx_ring->buffer_info[i];
4409                 size = min(len, max_per_txd);
4410
4411                 buffer_info->length = size;
4412                 buffer_info->time_stamp = jiffies;
4413                 buffer_info->next_to_watch = i;
4414                 buffer_info->dma = dma_map_single(&pdev->dev,
4415                                                   skb->data + offset,
4416                                                   size, DMA_TO_DEVICE);
4417                 buffer_info->mapped_as_page = false;
4418                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4419                         goto dma_error;
4420
4421                 len -= size;
4422                 offset += size;
4423                 count++;
4424
4425                 if (len) {
4426                         i++;
4427                         if (i == tx_ring->count)
4428                                 i = 0;
4429                 }
4430         }
4431
4432         for (f = 0; f < nr_frags; f++) {
4433                 struct skb_frag_struct *frag;
4434
4435                 frag = &skb_shinfo(skb)->frags[f];
4436                 len = frag->size;
4437                 offset = frag->page_offset;
4438
4439                 while (len) {
4440                         i++;
4441                         if (i == tx_ring->count)
4442                                 i = 0;
4443
4444                         buffer_info = &tx_ring->buffer_info[i];
4445                         size = min(len, max_per_txd);
4446
4447                         buffer_info->length = size;
4448                         buffer_info->time_stamp = jiffies;
4449                         buffer_info->next_to_watch = i;
4450                         buffer_info->dma = dma_map_page(&pdev->dev, frag->page,
4451                                                         offset, size,
4452                                                         DMA_TO_DEVICE);
4453                         buffer_info->mapped_as_page = true;
4454                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4455                                 goto dma_error;
4456
4457                         len -= size;
4458                         offset += size;
4459                         count++;
4460                 }
4461         }
4462
4463         segs = skb_shinfo(skb)->gso_segs ?: 1;
4464         /* multiply data chunks by size of headers */
4465         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4466
4467         tx_ring->buffer_info[i].skb = skb;
4468         tx_ring->buffer_info[i].segs = segs;
4469         tx_ring->buffer_info[i].bytecount = bytecount;
4470         tx_ring->buffer_info[first].next_to_watch = i;
4471
4472         return count;
4473
4474 dma_error:
4475         dev_err(&pdev->dev, "TX DMA map failed\n");
4476         buffer_info->dma = 0;
4477         if (count)
4478                 count--;
4479
4480         while (count--) {
4481                 if (i==0)
4482                         i += tx_ring->count;
4483                 i--;
4484                 buffer_info = &tx_ring->buffer_info[i];
4485                 e1000_put_txbuf(adapter, buffer_info);;
4486         }
4487
4488         return 0;
4489 }
4490
4491 static void e1000_tx_queue(struct e1000_adapter *adapter,
4492                            int tx_flags, int count)
4493 {
4494         struct e1000_ring *tx_ring = adapter->tx_ring;
4495         struct e1000_tx_desc *tx_desc = NULL;
4496         struct e1000_buffer *buffer_info;
4497         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4498         unsigned int i;
4499
4500         if (tx_flags & E1000_TX_FLAGS_TSO) {
4501                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4502                              E1000_TXD_CMD_TSE;
4503                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4504
4505                 if (tx_flags & E1000_TX_FLAGS_IPV4)
4506                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4507         }
4508
4509         if (tx_flags & E1000_TX_FLAGS_CSUM) {
4510                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4511                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4512         }
4513
4514         if (tx_flags & E1000_TX_FLAGS_VLAN) {
4515                 txd_lower |= E1000_TXD_CMD_VLE;
4516                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4517         }
4518
4519         i = tx_ring->next_to_use;
4520
4521         while (count--) {
4522                 buffer_info = &tx_ring->buffer_info[i];
4523                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4524                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4525                 tx_desc->lower.data =
4526                         cpu_to_le32(txd_lower | buffer_info->length);
4527                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4528
4529                 i++;
4530                 if (i == tx_ring->count)
4531                         i = 0;
4532         }
4533
4534         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4535
4536         /*
4537          * Force memory writes to complete before letting h/w
4538          * know there are new descriptors to fetch.  (Only
4539          * applicable for weak-ordered memory model archs,
4540          * such as IA-64).
4541          */
4542         wmb();
4543
4544         tx_ring->next_to_use = i;
4545         writel(i, adapter->hw.hw_addr + tx_ring->tail);
4546         /*
4547          * we need this if more than one processor can write to our tail
4548          * at a time, it synchronizes IO on IA64/Altix systems
4549          */
4550         mmiowb();
4551 }
4552
4553 #define MINIMUM_DHCP_PACKET_SIZE 282
4554 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4555                                     struct sk_buff *skb)
4556 {
4557         struct e1000_hw *hw =  &adapter->hw;
4558         u16 length, offset;
4559
4560         if (vlan_tx_tag_present(skb)) {
4561                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4562                     (adapter->hw.mng_cookie.status &
4563                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4564                         return 0;
4565         }
4566
4567         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4568                 return 0;
4569
4570         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4571                 return 0;
4572
4573         {
4574                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4575                 struct udphdr *udp;
4576
4577                 if (ip->protocol != IPPROTO_UDP)
4578                         return 0;
4579
4580                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4581                 if (ntohs(udp->dest) != 67)
4582                         return 0;
4583
4584                 offset = (u8 *)udp + 8 - skb->data;
4585                 length = skb->len - offset;
4586                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4587         }
4588
4589         return 0;
4590 }
4591
4592 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4593 {
4594         struct e1000_adapter *adapter = netdev_priv(netdev);
4595
4596         netif_stop_queue(netdev);
4597         /*
4598          * Herbert's original patch had:
4599          *  smp_mb__after_netif_stop_queue();
4600          * but since that doesn't exist yet, just open code it.
4601          */
4602         smp_mb();
4603
4604         /*
4605          * We need to check again in a case another CPU has just
4606          * made room available.
4607          */
4608         if (e1000_desc_unused(adapter->tx_ring) < size)
4609                 return -EBUSY;
4610
4611         /* A reprieve! */
4612         netif_start_queue(netdev);
4613         ++adapter->restart_queue;
4614         return 0;
4615 }
4616
4617 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4618 {
4619         struct e1000_adapter *adapter = netdev_priv(netdev);
4620
4621         if (e1000_desc_unused(adapter->tx_ring) >= size)
4622                 return 0;
4623         return __e1000_maybe_stop_tx(netdev, size);
4624 }
4625
4626 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4627 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4628                                     struct net_device *netdev)
4629 {
4630         struct e1000_adapter *adapter = netdev_priv(netdev);
4631         struct e1000_ring *tx_ring = adapter->tx_ring;
4632         unsigned int first;
4633         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4634         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4635         unsigned int tx_flags = 0;
4636         unsigned int len = skb_headlen(skb);
4637         unsigned int nr_frags;
4638         unsigned int mss;
4639         int count = 0;
4640         int tso;
4641         unsigned int f;
4642
4643         if (test_bit(__E1000_DOWN, &adapter->state)) {
4644                 dev_kfree_skb_any(skb);
4645                 return NETDEV_TX_OK;
4646         }
4647
4648         if (skb->len <= 0) {
4649                 dev_kfree_skb_any(skb);
4650                 return NETDEV_TX_OK;
4651         }
4652
4653         mss = skb_shinfo(skb)->gso_size;
4654         /*
4655          * The controller does a simple calculation to
4656          * make sure there is enough room in the FIFO before
4657          * initiating the DMA for each buffer.  The calc is:
4658          * 4 = ceil(buffer len/mss).  To make sure we don't
4659          * overrun the FIFO, adjust the max buffer len if mss
4660          * drops.
4661          */
4662         if (mss) {
4663                 u8 hdr_len;
4664                 max_per_txd = min(mss << 2, max_per_txd);
4665                 max_txd_pwr = fls(max_per_txd) - 1;
4666
4667                 /*
4668                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4669                  * points to just header, pull a few bytes of payload from
4670                  * frags into skb->data
4671                  */
4672                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4673                 /*
4674                  * we do this workaround for ES2LAN, but it is un-necessary,
4675                  * avoiding it could save a lot of cycles
4676                  */
4677                 if (skb->data_len && (hdr_len == len)) {
4678                         unsigned int pull_size;
4679
4680                         pull_size = min((unsigned int)4, skb->data_len);
4681                         if (!__pskb_pull_tail(skb, pull_size)) {
4682                                 e_err("__pskb_pull_tail failed.\n");
4683                                 dev_kfree_skb_any(skb);
4684                                 return NETDEV_TX_OK;
4685                         }
4686                         len = skb_headlen(skb);
4687                 }
4688         }
4689
4690         /* reserve a descriptor for the offload context */
4691         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4692                 count++;
4693         count++;
4694
4695         count += TXD_USE_COUNT(len, max_txd_pwr);
4696
4697         nr_frags = skb_shinfo(skb)->nr_frags;
4698         for (f = 0; f < nr_frags; f++)
4699                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4700                                        max_txd_pwr);
4701
4702         if (adapter->hw.mac.tx_pkt_filtering)
4703                 e1000_transfer_dhcp_info(adapter, skb);
4704
4705         /*
4706          * need: count + 2 desc gap to keep tail from touching
4707          * head, otherwise try next time
4708          */
4709         if (e1000_maybe_stop_tx(netdev, count + 2))
4710                 return NETDEV_TX_BUSY;
4711
4712         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4713                 tx_flags |= E1000_TX_FLAGS_VLAN;
4714                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4715         }
4716
4717         first = tx_ring->next_to_use;
4718
4719         tso = e1000_tso(adapter, skb);
4720         if (tso < 0) {
4721                 dev_kfree_skb_any(skb);
4722                 return NETDEV_TX_OK;
4723         }
4724
4725         if (tso)
4726                 tx_flags |= E1000_TX_FLAGS_TSO;
4727         else if (e1000_tx_csum(adapter, skb))
4728                 tx_flags |= E1000_TX_FLAGS_CSUM;
4729
4730         /*
4731          * Old method was to assume IPv4 packet by default if TSO was enabled.
4732          * 82571 hardware supports TSO capabilities for IPv6 as well...
4733          * no longer assume, we must.
4734          */
4735         if (skb->protocol == htons(ETH_P_IP))
4736                 tx_flags |= E1000_TX_FLAGS_IPV4;
4737
4738         /* if count is 0 then mapping error has occured */
4739         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4740         if (count) {
4741                 e1000_tx_queue(adapter, tx_flags, count);
4742                 /* Make sure there is space in the ring for the next send. */
4743                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4744
4745         } else {
4746                 dev_kfree_skb_any(skb);
4747                 tx_ring->buffer_info[first].time_stamp = 0;
4748                 tx_ring->next_to_use = first;
4749         }
4750
4751         return NETDEV_TX_OK;
4752 }
4753
4754 /**
4755  * e1000_tx_timeout - Respond to a Tx Hang
4756  * @netdev: network interface device structure
4757  **/
4758 static void e1000_tx_timeout(struct net_device *netdev)
4759 {
4760         struct e1000_adapter *adapter = netdev_priv(netdev);
4761
4762         /* Do the reset outside of interrupt context */
4763         adapter->tx_timeout_count++;
4764         schedule_work(&adapter->reset_task);
4765 }
4766
4767 static void e1000_reset_task(struct work_struct *work)
4768 {
4769         struct e1000_adapter *adapter;
4770         adapter = container_of(work, struct e1000_adapter, reset_task);
4771
4772         e1000e_dump(adapter);
4773         e_err("Reset adapter\n");
4774         e1000e_reinit_locked(adapter);
4775 }
4776
4777 /**
4778  * e1000_get_stats - Get System Network Statistics
4779  * @netdev: network interface device structure
4780  *
4781  * Returns the address of the device statistics structure.
4782  * The statistics are actually updated from the timer callback.
4783  **/
4784 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4785 {
4786         /* only return the current stats */
4787         return &netdev->stats;
4788 }
4789
4790 /**
4791  * e1000_change_mtu - Change the Maximum Transfer Unit
4792  * @netdev: network interface device structure
4793  * @new_mtu: new value for maximum frame size
4794  *
4795  * Returns 0 on success, negative on failure
4796  **/
4797 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4798 {
4799         struct e1000_adapter *adapter = netdev_priv(netdev);
4800         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4801
4802         /* Jumbo frame support */
4803         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
4804             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4805                 e_err("Jumbo Frames not supported.\n");
4806                 return -EINVAL;
4807         }
4808
4809         /* Supported frame sizes */
4810         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4811             (max_frame > adapter->max_hw_frame_size)) {
4812                 e_err("Unsupported MTU setting\n");
4813                 return -EINVAL;
4814         }
4815
4816         /* 82573 Errata 17 */
4817         if (((adapter->hw.mac.type == e1000_82573) ||
4818              (adapter->hw.mac.type == e1000_82574)) &&
4819             (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
4820                 adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
4821                 e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
4822         }
4823
4824         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4825                 msleep(1);
4826         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
4827         adapter->max_frame_size = max_frame;
4828         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4829         netdev->mtu = new_mtu;
4830         if (netif_running(netdev))
4831                 e1000e_down(adapter);
4832
4833         /*
4834          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4835          * means we reserve 2 more, this pushes us to allocate from the next
4836          * larger slab size.
4837          * i.e. RXBUFFER_2048 --> size-4096 slab
4838          * However with the new *_jumbo_rx* routines, jumbo receives will use
4839          * fragmented skbs
4840          */
4841
4842         if (max_frame <= 2048)
4843                 adapter->rx_buffer_len = 2048;
4844         else
4845                 adapter->rx_buffer_len = 4096;
4846
4847         /* adjust allocation if LPE protects us, and we aren't using SBP */
4848         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4849              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4850                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4851                                          + ETH_FCS_LEN;
4852
4853         if (netif_running(netdev))
4854                 e1000e_up(adapter);
4855         else
4856                 e1000e_reset(adapter);
4857
4858         clear_bit(__E1000_RESETTING, &adapter->state);
4859
4860         return 0;
4861 }
4862
4863 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4864                            int cmd)
4865 {
4866         struct e1000_adapter *adapter = netdev_priv(netdev);
4867         struct mii_ioctl_data *data = if_mii(ifr);
4868
4869         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4870                 return -EOPNOTSUPP;
4871
4872         switch (cmd) {
4873         case SIOCGMIIPHY:
4874                 data->phy_id = adapter->hw.phy.addr;
4875                 break;
4876         case SIOCGMIIREG:
4877                 e1000_phy_read_status(adapter);
4878
4879                 switch (data->reg_num & 0x1F) {
4880                 case MII_BMCR:
4881                         data->val_out = adapter->phy_regs.bmcr;
4882                         break;
4883                 case MII_BMSR:
4884                         data->val_out = adapter->phy_regs.bmsr;
4885                         break;
4886                 case MII_PHYSID1:
4887                         data->val_out = (adapter->hw.phy.id >> 16);
4888                         break;
4889                 case MII_PHYSID2:
4890                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
4891                         break;
4892                 case MII_ADVERTISE:
4893                         data->val_out = adapter->phy_regs.advertise;
4894                         break;
4895                 case MII_LPA:
4896                         data->val_out = adapter->phy_regs.lpa;
4897                         break;
4898                 case MII_EXPANSION:
4899                         data->val_out = adapter->phy_regs.expansion;
4900                         break;
4901                 case MII_CTRL1000:
4902                         data->val_out = adapter->phy_regs.ctrl1000;
4903                         break;
4904                 case MII_STAT1000:
4905                         data->val_out = adapter->phy_regs.stat1000;
4906                         break;
4907                 case MII_ESTATUS:
4908                         data->val_out = adapter->phy_regs.estatus;
4909                         break;
4910                 default:
4911                         return -EIO;
4912                 }
4913                 break;
4914         case SIOCSMIIREG:
4915         default:
4916                 return -EOPNOTSUPP;
4917         }
4918         return 0;
4919 }
4920
4921 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4922 {
4923         switch (cmd) {
4924         case SIOCGMIIPHY:
4925         case SIOCGMIIREG:
4926         case SIOCSMIIREG:
4927                 return e1000_mii_ioctl(netdev, ifr, cmd);
4928         default:
4929                 return -EOPNOTSUPP;
4930         }
4931 }
4932
4933 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
4934 {
4935         struct e1000_hw *hw = &adapter->hw;
4936         u32 i, mac_reg;
4937         u16 phy_reg;
4938         int retval = 0;
4939
4940         /* copy MAC RARs to PHY RARs */
4941         e1000_copy_rx_addrs_to_phy_ich8lan(hw);
4942
4943         /* copy MAC MTA to PHY MTA */
4944         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
4945                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
4946                 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
4947                 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
4948         }
4949
4950         /* configure PHY Rx Control register */
4951         e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
4952         mac_reg = er32(RCTL);
4953         if (mac_reg & E1000_RCTL_UPE)
4954                 phy_reg |= BM_RCTL_UPE;
4955         if (mac_reg & E1000_RCTL_MPE)
4956                 phy_reg |= BM_RCTL_MPE;
4957         phy_reg &= ~(BM_RCTL_MO_MASK);
4958         if (mac_reg & E1000_RCTL_MO_3)
4959                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
4960                                 << BM_RCTL_MO_SHIFT);
4961         if (mac_reg & E1000_RCTL_BAM)
4962                 phy_reg |= BM_RCTL_BAM;
4963         if (mac_reg & E1000_RCTL_PMCF)
4964                 phy_reg |= BM_RCTL_PMCF;
4965         mac_reg = er32(CTRL);
4966         if (mac_reg & E1000_CTRL_RFCE)
4967                 phy_reg |= BM_RCTL_RFCE;
4968         e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
4969
4970         /* enable PHY wakeup in MAC register */
4971         ew32(WUFC, wufc);
4972         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
4973
4974         /* configure and enable PHY wakeup in PHY registers */
4975         e1e_wphy(&adapter->hw, BM_WUFC, wufc);
4976         e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
4977
4978         /* activate PHY wakeup */
4979         retval = hw->phy.ops.acquire(hw);
4980         if (retval) {
4981                 e_err("Could not acquire PHY\n");
4982                 return retval;
4983         }
4984         e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4985                                  (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
4986         retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
4987         if (retval) {
4988                 e_err("Could not read PHY page 769\n");
4989                 goto out;
4990         }
4991         phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
4992         retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
4993         if (retval)
4994                 e_err("Could not set PHY Host Wakeup bit\n");
4995 out:
4996         hw->phy.ops.release(hw);
4997
4998         return retval;
4999 }
5000
5001 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5002                             bool runtime)
5003 {
5004         struct net_device *netdev = pci_get_drvdata(pdev);
5005         struct e1000_adapter *adapter = netdev_priv(netdev);
5006         struct e1000_hw *hw = &adapter->hw;
5007         u32 ctrl, ctrl_ext, rctl, status;
5008         /* Runtime suspend should only enable wakeup for link changes */
5009         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5010         int retval = 0;
5011
5012         netif_device_detach(netdev);
5013
5014         if (netif_running(netdev)) {
5015                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5016                 e1000e_down(adapter);
5017                 e1000_free_irq(adapter);
5018         }
5019         e1000e_reset_interrupt_capability(adapter);
5020
5021         retval = pci_save_state(pdev);
5022         if (retval)
5023                 return retval;
5024
5025         status = er32(STATUS);
5026         if (status & E1000_STATUS_LU)
5027                 wufc &= ~E1000_WUFC_LNKC;
5028
5029         if (wufc) {
5030                 e1000_setup_rctl(adapter);
5031                 e1000_set_multi(netdev);
5032
5033                 /* turn on all-multi mode if wake on multicast is enabled */
5034                 if (wufc & E1000_WUFC_MC) {
5035                         rctl = er32(RCTL);
5036                         rctl |= E1000_RCTL_MPE;
5037                         ew32(RCTL, rctl);
5038                 }
5039
5040                 ctrl = er32(CTRL);
5041                 /* advertise wake from D3Cold */
5042                 #define E1000_CTRL_ADVD3WUC 0x00100000
5043                 /* phy power management enable */
5044                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5045                 ctrl |= E1000_CTRL_ADVD3WUC;
5046                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5047                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5048                 ew32(CTRL, ctrl);
5049
5050                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5051                     adapter->hw.phy.media_type ==
5052                     e1000_media_type_internal_serdes) {
5053                         /* keep the laser running in D3 */
5054                         ctrl_ext = er32(CTRL_EXT);
5055                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5056                         ew32(CTRL_EXT, ctrl_ext);
5057                 }
5058
5059                 if (adapter->flags & FLAG_IS_ICH)
5060                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
5061
5062                 /* Allow time for pending master requests to run */
5063                 e1000e_disable_pcie_master(&adapter->hw);
5064
5065                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5066                         /* enable wakeup by the PHY */
5067                         retval = e1000_init_phy_wakeup(adapter, wufc);
5068                         if (retval)
5069                                 return retval;
5070                 } else {
5071                         /* enable wakeup by the MAC */
5072                         ew32(WUFC, wufc);
5073                         ew32(WUC, E1000_WUC_PME_EN);
5074                 }
5075         } else {
5076                 ew32(WUC, 0);
5077                 ew32(WUFC, 0);
5078         }
5079
5080         *enable_wake = !!wufc;
5081
5082         /* make sure adapter isn't asleep if manageability is enabled */
5083         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5084             (hw->mac.ops.check_mng_mode(hw)))
5085                 *enable_wake = true;
5086
5087         if (adapter->hw.phy.type == e1000_phy_igp_3)
5088                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5089
5090         /*
5091          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5092          * would have already happened in close and is redundant.
5093          */
5094         e1000_release_hw_control(adapter);
5095
5096         pci_disable_device(pdev);
5097
5098         return 0;
5099 }
5100
5101 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5102 {
5103         if (sleep && wake) {
5104                 pci_prepare_to_sleep(pdev);
5105                 return;
5106         }
5107
5108         pci_wake_from_d3(pdev, wake);
5109         pci_set_power_state(pdev, PCI_D3hot);
5110 }
5111
5112 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5113                                     bool wake)
5114 {
5115         struct net_device *netdev = pci_get_drvdata(pdev);
5116         struct e1000_adapter *adapter = netdev_priv(netdev);
5117
5118         /*
5119          * The pci-e switch on some quad port adapters will report a
5120          * correctable error when the MAC transitions from D0 to D3.  To
5121          * prevent this we need to mask off the correctable errors on the
5122          * downstream port of the pci-e switch.
5123          */
5124         if (adapter->flags & FLAG_IS_QUAD_PORT) {
5125                 struct pci_dev *us_dev = pdev->bus->self;
5126                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
5127                 u16 devctl;
5128
5129                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5130                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5131                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
5132
5133                 e1000_power_off(pdev, sleep, wake);
5134
5135                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5136         } else {
5137                 e1000_power_off(pdev, sleep, wake);
5138         }
5139 }
5140
5141 #ifdef CONFIG_PCIEASPM
5142 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5143 {
5144         pci_disable_link_state(pdev, state);
5145 }
5146 #else
5147 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5148 {
5149         int pos;
5150         u16 reg16;
5151
5152         /*
5153          * Both device and parent should have the same ASPM setting.
5154          * Disable ASPM in downstream component first and then upstream.
5155          */
5156         pos = pci_pcie_cap(pdev);
5157         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5158         reg16 &= ~state;
5159         pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5160
5161         if (!pdev->bus->self)
5162                 return;
5163
5164         pos = pci_pcie_cap(pdev->bus->self);
5165         pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5166         reg16 &= ~state;
5167         pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5168 }
5169 #endif
5170 void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5171 {
5172         dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5173                  (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5174                  (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5175
5176         __e1000e_disable_aspm(pdev, state);
5177 }
5178
5179 #ifdef CONFIG_PM_OPS
5180 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5181 {
5182         return !!adapter->tx_ring->buffer_info;
5183 }
5184
5185 static int __e1000_resume(struct pci_dev *pdev)
5186 {
5187         struct net_device *netdev = pci_get_drvdata(pdev);
5188         struct e1000_adapter *adapter = netdev_priv(netdev);
5189         struct e1000_hw *hw = &adapter->hw;
5190         u32 err;
5191
5192         pci_set_power_state(pdev, PCI_D0);
5193         pci_restore_state(pdev);
5194         pci_save_state(pdev);
5195         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5196                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5197
5198         e1000e_set_interrupt_capability(adapter);
5199         if (netif_running(netdev)) {
5200                 err = e1000_request_irq(adapter);
5201                 if (err)
5202                         return err;
5203         }
5204
5205         e1000e_power_up_phy(adapter);
5206
5207         /* report the system wakeup cause from S3/S4 */
5208         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5209                 u16 phy_data;
5210
5211                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5212                 if (phy_data) {
5213                         e_info("PHY Wakeup cause - %s\n",
5214                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5215                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5216                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5217                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5218                                 phy_data & E1000_WUS_LNKC ? "Link Status "
5219                                 " Change" : "other");
5220                 }
5221                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5222         } else {
5223                 u32 wus = er32(WUS);
5224                 if (wus) {
5225                         e_info("MAC Wakeup cause - %s\n",
5226                                 wus & E1000_WUS_EX ? "Unicast Packet" :
5227                                 wus & E1000_WUS_MC ? "Multicast Packet" :
5228                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
5229                                 wus & E1000_WUS_MAG ? "Magic Packet" :
5230                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
5231                                 "other");
5232                 }
5233                 ew32(WUS, ~0);
5234         }
5235
5236         e1000e_reset(adapter);
5237
5238         e1000_init_manageability_pt(adapter);
5239
5240         if (netif_running(netdev))
5241                 e1000e_up(adapter);
5242
5243         netif_device_attach(netdev);
5244
5245         /*
5246          * If the controller has AMT, do not set DRV_LOAD until the interface
5247          * is up.  For all other cases, let the f/w know that the h/w is now
5248          * under the control of the driver.
5249          */
5250         if (!(adapter->flags & FLAG_HAS_AMT))
5251                 e1000_get_hw_control(adapter);
5252
5253         return 0;
5254 }
5255
5256 #ifdef CONFIG_PM_SLEEP
5257 static int e1000_suspend(struct device *dev)
5258 {
5259         struct pci_dev *pdev = to_pci_dev(dev);
5260         int retval;
5261         bool wake;
5262
5263         retval = __e1000_shutdown(pdev, &wake, false);
5264         if (!retval)
5265                 e1000_complete_shutdown(pdev, true, wake);
5266
5267         return retval;
5268 }
5269
5270 static int e1000_resume(struct device *dev)
5271 {
5272         struct pci_dev *pdev = to_pci_dev(dev);
5273         struct net_device *netdev = pci_get_drvdata(pdev);
5274         struct e1000_adapter *adapter = netdev_priv(netdev);
5275
5276         if (e1000e_pm_ready(adapter))
5277                 adapter->idle_check = true;
5278
5279         return __e1000_resume(pdev);
5280 }
5281 #endif /* CONFIG_PM_SLEEP */
5282
5283 #ifdef CONFIG_PM_RUNTIME
5284 static int e1000_runtime_suspend(struct device *dev)
5285 {
5286         struct pci_dev *pdev = to_pci_dev(dev);
5287         struct net_device *netdev = pci_get_drvdata(pdev);
5288         struct e1000_adapter *adapter = netdev_priv(netdev);
5289
5290         if (e1000e_pm_ready(adapter)) {
5291                 bool wake;
5292
5293                 __e1000_shutdown(pdev, &wake, true);
5294         }
5295
5296         return 0;
5297 }
5298
5299 static int e1000_idle(struct device *dev)
5300 {
5301         struct pci_dev *pdev = to_pci_dev(dev);
5302         struct net_device *netdev = pci_get_drvdata(pdev);
5303         struct e1000_adapter *adapter = netdev_priv(netdev);
5304
5305         if (!e1000e_pm_ready(adapter))
5306                 return 0;
5307
5308         if (adapter->idle_check) {
5309                 adapter->idle_check = false;
5310                 if (!e1000e_has_link(adapter))
5311                         pm_schedule_suspend(dev, MSEC_PER_SEC);
5312         }
5313
5314         return -EBUSY;
5315 }
5316
5317 static int e1000_runtime_resume(struct device *dev)
5318 {
5319         struct pci_dev *pdev = to_pci_dev(dev);
5320         struct net_device *netdev = pci_get_drvdata(pdev);
5321         struct e1000_adapter *adapter = netdev_priv(netdev);
5322
5323         if (!e1000e_pm_ready(adapter))
5324                 return 0;
5325
5326         adapter->idle_check = !dev->power.runtime_auto;
5327         return __e1000_resume(pdev);
5328 }
5329 #endif /* CONFIG_PM_RUNTIME */
5330 #endif /* CONFIG_PM_OPS */
5331
5332 static void e1000_shutdown(struct pci_dev *pdev)
5333 {
5334         bool wake = false;
5335
5336         __e1000_shutdown(pdev, &wake, false);
5337
5338         if (system_state == SYSTEM_POWER_OFF)
5339                 e1000_complete_shutdown(pdev, false, wake);
5340 }
5341
5342 #ifdef CONFIG_NET_POLL_CONTROLLER
5343 /*
5344  * Polling 'interrupt' - used by things like netconsole to send skbs
5345  * without having to re-enable interrupts. It's not called while
5346  * the interrupt routine is executing.
5347  */
5348 static void e1000_netpoll(struct net_device *netdev)
5349 {
5350         struct e1000_adapter *adapter = netdev_priv(netdev);
5351
5352         disable_irq(adapter->pdev->irq);
5353         e1000_intr(adapter->pdev->irq, netdev);
5354
5355         enable_irq(adapter->pdev->irq);
5356 }
5357 #endif
5358
5359 /**
5360  * e1000_io_error_detected - called when PCI error is detected
5361  * @pdev: Pointer to PCI device
5362  * @state: The current pci connection state
5363  *
5364  * This function is called after a PCI bus error affecting
5365  * this device has been detected.
5366  */
5367 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5368                                                 pci_channel_state_t state)
5369 {
5370         struct net_device *netdev = pci_get_drvdata(pdev);
5371         struct e1000_adapter *adapter = netdev_priv(netdev);
5372
5373         netif_device_detach(netdev);
5374
5375         if (state == pci_channel_io_perm_failure)
5376                 return PCI_ERS_RESULT_DISCONNECT;
5377
5378         if (netif_running(netdev))
5379                 e1000e_down(adapter);
5380         pci_disable_device(pdev);
5381
5382         /* Request a slot slot reset. */
5383         return PCI_ERS_RESULT_NEED_RESET;
5384 }
5385
5386 /**
5387  * e1000_io_slot_reset - called after the pci bus has been reset.
5388  * @pdev: Pointer to PCI device
5389  *
5390  * Restart the card from scratch, as if from a cold-boot. Implementation
5391  * resembles the first-half of the e1000_resume routine.
5392  */
5393 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5394 {
5395         struct net_device *netdev = pci_get_drvdata(pdev);
5396         struct e1000_adapter *adapter = netdev_priv(netdev);
5397         struct e1000_hw *hw = &adapter->hw;
5398         int err;
5399         pci_ers_result_t result;
5400
5401         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5402                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5403         err = pci_enable_device_mem(pdev);
5404         if (err) {
5405                 dev_err(&pdev->dev,
5406                         "Cannot re-enable PCI device after reset.\n");
5407                 result = PCI_ERS_RESULT_DISCONNECT;
5408         } else {
5409                 pci_set_master(pdev);
5410                 pdev->state_saved = true;
5411                 pci_restore_state(pdev);
5412
5413                 pci_enable_wake(pdev, PCI_D3hot, 0);
5414                 pci_enable_wake(pdev, PCI_D3cold, 0);
5415
5416                 e1000e_reset(adapter);
5417                 ew32(WUS, ~0);
5418                 result = PCI_ERS_RESULT_RECOVERED;
5419         }
5420
5421         pci_cleanup_aer_uncorrect_error_status(pdev);
5422
5423         return result;
5424 }
5425
5426 /**
5427  * e1000_io_resume - called when traffic can start flowing again.
5428  * @pdev: Pointer to PCI device
5429  *
5430  * This callback is called when the error recovery driver tells us that
5431  * its OK to resume normal operation. Implementation resembles the
5432  * second-half of the e1000_resume routine.
5433  */
5434 static void e1000_io_resume(struct pci_dev *pdev)
5435 {
5436         struct net_device *netdev = pci_get_drvdata(pdev);
5437         struct e1000_adapter *adapter = netdev_priv(netdev);
5438
5439         e1000_init_manageability_pt(adapter);
5440
5441         if (netif_running(netdev)) {
5442                 if (e1000e_up(adapter)) {
5443                         dev_err(&pdev->dev,
5444                                 "can't bring device back up after reset\n");
5445                         return;
5446                 }
5447         }
5448
5449         netif_device_attach(netdev);
5450
5451         /*
5452          * If the controller has AMT, do not set DRV_LOAD until the interface
5453          * is up.  For all other cases, let the f/w know that the h/w is now
5454          * under the control of the driver.
5455          */
5456         if (!(adapter->flags & FLAG_HAS_AMT))
5457                 e1000_get_hw_control(adapter);
5458
5459 }
5460
5461 static void e1000_print_device_info(struct e1000_adapter *adapter)
5462 {
5463         struct e1000_hw *hw = &adapter->hw;
5464         struct net_device *netdev = adapter->netdev;
5465         u32 pba_num;
5466
5467         /* print bus type/speed/width info */
5468         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
5469                /* bus width */
5470                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5471                 "Width x1"),
5472                /* MAC address */
5473                netdev->dev_addr);
5474         e_info("Intel(R) PRO/%s Network Connection\n",
5475                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5476         e1000e_read_pba_num(hw, &pba_num);
5477         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
5478                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
5479 }
5480
5481 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5482 {
5483         struct e1000_hw *hw = &adapter->hw;
5484         int ret_val;
5485         u16 buf = 0;
5486
5487         if (hw->mac.type != e1000_82573)
5488                 return;
5489
5490         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
5491         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
5492                 /* Deep Smart Power Down (DSPD) */
5493                 dev_warn(&adapter->pdev->dev,
5494                          "Warning: detected DSPD enabled in EEPROM\n");
5495         }
5496 }
5497
5498 static const struct net_device_ops e1000e_netdev_ops = {
5499         .ndo_open               = e1000_open,
5500         .ndo_stop               = e1000_close,
5501         .ndo_start_xmit         = e1000_xmit_frame,
5502         .ndo_get_stats          = e1000_get_stats,
5503         .ndo_set_multicast_list = e1000_set_multi,
5504         .ndo_set_mac_address    = e1000_set_mac,
5505         .ndo_change_mtu         = e1000_change_mtu,
5506         .ndo_do_ioctl           = e1000_ioctl,
5507         .ndo_tx_timeout         = e1000_tx_timeout,
5508         .ndo_validate_addr      = eth_validate_addr,
5509
5510         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
5511         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
5512         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
5513 #ifdef CONFIG_NET_POLL_CONTROLLER
5514         .ndo_poll_controller    = e1000_netpoll,
5515 #endif
5516 };
5517
5518 /**
5519  * e1000_probe - Device Initialization Routine
5520  * @pdev: PCI device information struct
5521  * @ent: entry in e1000_pci_tbl
5522  *
5523  * Returns 0 on success, negative on failure
5524  *
5525  * e1000_probe initializes an adapter identified by a pci_dev structure.
5526  * The OS initialization, configuring of the adapter private structure,
5527  * and a hardware reset occur.
5528  **/
5529 static int __devinit e1000_probe(struct pci_dev *pdev,
5530                                  const struct pci_device_id *ent)
5531 {
5532         struct net_device *netdev;
5533         struct e1000_adapter *adapter;
5534         struct e1000_hw *hw;
5535         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
5536         resource_size_t mmio_start, mmio_len;
5537         resource_size_t flash_start, flash_len;
5538
5539         static int cards_found;
5540         int i, err, pci_using_dac;
5541         u16 eeprom_data = 0;
5542         u16 eeprom_apme_mask = E1000_EEPROM_APME;
5543
5544         if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
5545                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5546
5547         err = pci_enable_device_mem(pdev);
5548         if (err)
5549                 return err;
5550
5551         pci_using_dac = 0;
5552         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
5553         if (!err) {
5554                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
5555                 if (!err)
5556                         pci_using_dac = 1;
5557         } else {
5558                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
5559                 if (err) {
5560                         err = dma_set_coherent_mask(&pdev->dev,
5561                                                     DMA_BIT_MASK(32));
5562                         if (err) {
5563                                 dev_err(&pdev->dev, "No usable DMA "
5564                                         "configuration, aborting\n");
5565                                 goto err_dma;
5566                         }
5567                 }
5568         }
5569
5570         err = pci_request_selected_regions_exclusive(pdev,
5571                                           pci_select_bars(pdev, IORESOURCE_MEM),
5572                                           e1000e_driver_name);
5573         if (err)
5574                 goto err_pci_reg;
5575
5576         /* AER (Advanced Error Reporting) hooks */
5577         pci_enable_pcie_error_reporting(pdev);
5578
5579         pci_set_master(pdev);
5580         /* PCI config space info */
5581         err = pci_save_state(pdev);
5582         if (err)
5583                 goto err_alloc_etherdev;
5584
5585         err = -ENOMEM;
5586         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
5587         if (!netdev)
5588                 goto err_alloc_etherdev;
5589
5590         SET_NETDEV_DEV(netdev, &pdev->dev);
5591
5592         netdev->irq = pdev->irq;
5593
5594         pci_set_drvdata(pdev, netdev);
5595         adapter = netdev_priv(netdev);
5596         hw = &adapter->hw;
5597         adapter->netdev = netdev;
5598         adapter->pdev = pdev;
5599         adapter->ei = ei;
5600         adapter->pba = ei->pba;
5601         adapter->flags = ei->flags;
5602         adapter->flags2 = ei->flags2;
5603         adapter->hw.adapter = adapter;
5604         adapter->hw.mac.type = ei->mac;
5605         adapter->max_hw_frame_size = ei->max_hw_frame_size;
5606         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5607
5608         mmio_start = pci_resource_start(pdev, 0);
5609         mmio_len = pci_resource_len(pdev, 0);
5610
5611         err = -EIO;
5612         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5613         if (!adapter->hw.hw_addr)
5614                 goto err_ioremap;
5615
5616         if ((adapter->flags & FLAG_HAS_FLASH) &&
5617             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5618                 flash_start = pci_resource_start(pdev, 1);
5619                 flash_len = pci_resource_len(pdev, 1);
5620                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5621                 if (!adapter->hw.flash_address)
5622                         goto err_flashmap;
5623         }
5624
5625         /* construct the net_device struct */
5626         netdev->netdev_ops              = &e1000e_netdev_ops;
5627         e1000e_set_ethtool_ops(netdev);
5628         netdev->watchdog_timeo          = 5 * HZ;
5629         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5630         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5631
5632         netdev->mem_start = mmio_start;
5633         netdev->mem_end = mmio_start + mmio_len;
5634
5635         adapter->bd_number = cards_found++;
5636
5637         e1000e_check_options(adapter);
5638
5639         /* setup adapter struct */
5640         err = e1000_sw_init(adapter);
5641         if (err)
5642                 goto err_sw_init;
5643
5644         err = -EIO;
5645
5646         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5647         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5648         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5649
5650         err = ei->get_variants(adapter);
5651         if (err)
5652                 goto err_hw_init;
5653
5654         if ((adapter->flags & FLAG_IS_ICH) &&
5655             (adapter->flags & FLAG_READ_ONLY_NVM))
5656                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5657
5658         hw->mac.ops.get_bus_info(&adapter->hw);
5659
5660         adapter->hw.phy.autoneg_wait_to_complete = 0;
5661
5662         /* Copper options */
5663         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5664                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5665                 adapter->hw.phy.disable_polarity_correction = 0;
5666                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5667         }
5668
5669         if (e1000_check_reset_block(&adapter->hw))
5670                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5671
5672         netdev->features = NETIF_F_SG |
5673                            NETIF_F_HW_CSUM |
5674                            NETIF_F_HW_VLAN_TX |
5675                            NETIF_F_HW_VLAN_RX;
5676
5677         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5678                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5679
5680         netdev->features |= NETIF_F_TSO;
5681         netdev->features |= NETIF_F_TSO6;
5682
5683         netdev->vlan_features |= NETIF_F_TSO;
5684         netdev->vlan_features |= NETIF_F_TSO6;
5685         netdev->vlan_features |= NETIF_F_HW_CSUM;
5686         netdev->vlan_features |= NETIF_F_SG;
5687
5688         if (pci_using_dac)
5689                 netdev->features |= NETIF_F_HIGHDMA;
5690
5691         if (e1000e_enable_mng_pass_thru(&adapter->hw))
5692                 adapter->flags |= FLAG_MNG_PT_ENABLED;
5693
5694         /*
5695          * before reading the NVM, reset the controller to
5696          * put the device in a known good starting state
5697          */
5698         adapter->hw.mac.ops.reset_hw(&adapter->hw);
5699
5700         /*
5701          * systems with ASPM and others may see the checksum fail on the first
5702          * attempt. Let's give it a few tries
5703          */
5704         for (i = 0;; i++) {
5705                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5706                         break;
5707                 if (i == 2) {
5708                         e_err("The NVM Checksum Is Not Valid\n");
5709                         err = -EIO;
5710                         goto err_eeprom;
5711                 }
5712         }
5713
5714         e1000_eeprom_checks(adapter);
5715
5716         /* copy the MAC address */
5717         if (e1000e_read_mac_addr(&adapter->hw))
5718                 e_err("NVM Read Error while reading MAC address\n");
5719
5720         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5721         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5722
5723         if (!is_valid_ether_addr(netdev->perm_addr)) {
5724                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5725                 err = -EIO;
5726                 goto err_eeprom;
5727         }
5728
5729         init_timer(&adapter->watchdog_timer);
5730         adapter->watchdog_timer.function = &e1000_watchdog;
5731         adapter->watchdog_timer.data = (unsigned long) adapter;
5732
5733         init_timer(&adapter->phy_info_timer);
5734         adapter->phy_info_timer.function = &e1000_update_phy_info;
5735         adapter->phy_info_timer.data = (unsigned long) adapter;
5736
5737         INIT_WORK(&adapter->reset_task, e1000_reset_task);
5738         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
5739         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
5740         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
5741         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
5742
5743         /* Initialize link parameters. User can change them with ethtool */
5744         adapter->hw.mac.autoneg = 1;
5745         adapter->fc_autoneg = 1;
5746         adapter->hw.fc.requested_mode = e1000_fc_default;
5747         adapter->hw.fc.current_mode = e1000_fc_default;
5748         adapter->hw.phy.autoneg_advertised = 0x2f;
5749
5750         /* ring size defaults */
5751         adapter->rx_ring->count = 256;
5752         adapter->tx_ring->count = 256;
5753
5754         /*
5755          * Initial Wake on LAN setting - If APM wake is enabled in
5756          * the EEPROM, enable the ACPI Magic Packet filter
5757          */
5758         if (adapter->flags & FLAG_APME_IN_WUC) {
5759                 /* APME bit in EEPROM is mapped to WUC.APME */
5760                 eeprom_data = er32(WUC);
5761                 eeprom_apme_mask = E1000_WUC_APME;
5762                 if (eeprom_data & E1000_WUC_PHY_WAKE)
5763                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
5764         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5765                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5766                     (adapter->hw.bus.func == 1))
5767                         e1000_read_nvm(&adapter->hw,
5768                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5769                 else
5770                         e1000_read_nvm(&adapter->hw,
5771                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5772         }
5773
5774         /* fetch WoL from EEPROM */
5775         if (eeprom_data & eeprom_apme_mask)
5776                 adapter->eeprom_wol |= E1000_WUFC_MAG;
5777
5778         /*
5779          * now that we have the eeprom settings, apply the special cases
5780          * where the eeprom may be wrong or the board simply won't support
5781          * wake on lan on a particular port
5782          */
5783         if (!(adapter->flags & FLAG_HAS_WOL))
5784                 adapter->eeprom_wol = 0;
5785
5786         /* initialize the wol settings based on the eeprom settings */
5787         adapter->wol = adapter->eeprom_wol;
5788         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5789
5790         /* save off EEPROM version number */
5791         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
5792
5793         /* reset the hardware with the new settings */
5794         e1000e_reset(adapter);
5795
5796         /*
5797          * If the controller has AMT, do not set DRV_LOAD until the interface
5798          * is up.  For all other cases, let the f/w know that the h/w is now
5799          * under the control of the driver.
5800          */
5801         if (!(adapter->flags & FLAG_HAS_AMT))
5802                 e1000_get_hw_control(adapter);
5803
5804         strcpy(netdev->name, "eth%d");
5805         err = register_netdev(netdev);
5806         if (err)
5807                 goto err_register;
5808
5809         /* carrier off reporting is important to ethtool even BEFORE open */
5810         netif_carrier_off(netdev);
5811
5812         e1000_print_device_info(adapter);
5813
5814         if (pci_dev_run_wake(pdev)) {
5815                 pm_runtime_set_active(&pdev->dev);
5816                 pm_runtime_enable(&pdev->dev);
5817         }
5818         pm_schedule_suspend(&pdev->dev, MSEC_PER_SEC);
5819
5820         return 0;
5821
5822 err_register:
5823         if (!(adapter->flags & FLAG_HAS_AMT))
5824                 e1000_release_hw_control(adapter);
5825 err_eeprom:
5826         if (!e1000_check_reset_block(&adapter->hw))
5827                 e1000_phy_hw_reset(&adapter->hw);
5828 err_hw_init:
5829
5830         kfree(adapter->tx_ring);
5831         kfree(adapter->rx_ring);
5832 err_sw_init:
5833         if (adapter->hw.flash_address)
5834                 iounmap(adapter->hw.flash_address);
5835         e1000e_reset_interrupt_capability(adapter);
5836 err_flashmap:
5837         iounmap(adapter->hw.hw_addr);
5838 err_ioremap:
5839         free_netdev(netdev);
5840 err_alloc_etherdev:
5841         pci_release_selected_regions(pdev,
5842                                      pci_select_bars(pdev, IORESOURCE_MEM));
5843 err_pci_reg:
5844 err_dma:
5845         pci_disable_device(pdev);
5846         return err;
5847 }
5848
5849 /**
5850  * e1000_remove - Device Removal Routine
5851  * @pdev: PCI device information struct
5852  *
5853  * e1000_remove is called by the PCI subsystem to alert the driver
5854  * that it should release a PCI device.  The could be caused by a
5855  * Hot-Plug event, or because the driver is going to be removed from
5856  * memory.
5857  **/
5858 static void __devexit e1000_remove(struct pci_dev *pdev)
5859 {
5860         struct net_device *netdev = pci_get_drvdata(pdev);
5861         struct e1000_adapter *adapter = netdev_priv(netdev);
5862         bool down = test_bit(__E1000_DOWN, &adapter->state);
5863
5864         pm_runtime_get_sync(&pdev->dev);
5865
5866         /*
5867          * flush_scheduled work may reschedule our watchdog task, so
5868          * explicitly disable watchdog tasks from being rescheduled
5869          */
5870         if (!down)
5871                 set_bit(__E1000_DOWN, &adapter->state);
5872         del_timer_sync(&adapter->watchdog_timer);
5873         del_timer_sync(&adapter->phy_info_timer);
5874
5875         cancel_work_sync(&adapter->reset_task);
5876         cancel_work_sync(&adapter->watchdog_task);
5877         cancel_work_sync(&adapter->downshift_task);
5878         cancel_work_sync(&adapter->update_phy_task);
5879         cancel_work_sync(&adapter->print_hang_task);
5880         flush_scheduled_work();
5881
5882         if (!(netdev->flags & IFF_UP))
5883                 e1000_power_down_phy(adapter);
5884
5885         /* Don't lie to e1000_close() down the road. */
5886         if (!down)
5887                 clear_bit(__E1000_DOWN, &adapter->state);
5888         unregister_netdev(netdev);
5889
5890         if (pci_dev_run_wake(pdev)) {
5891                 pm_runtime_disable(&pdev->dev);
5892                 pm_runtime_set_suspended(&pdev->dev);
5893         }
5894         pm_runtime_put_noidle(&pdev->dev);
5895
5896         /*
5897          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5898          * would have already happened in close and is redundant.
5899          */
5900         e1000_release_hw_control(adapter);
5901
5902         e1000e_reset_interrupt_capability(adapter);
5903         kfree(adapter->tx_ring);
5904         kfree(adapter->rx_ring);
5905
5906         iounmap(adapter->hw.hw_addr);
5907         if (adapter->hw.flash_address)
5908                 iounmap(adapter->hw.flash_address);
5909         pci_release_selected_regions(pdev,
5910                                      pci_select_bars(pdev, IORESOURCE_MEM));
5911
5912         free_netdev(netdev);
5913
5914         /* AER disable */
5915         pci_disable_pcie_error_reporting(pdev);
5916
5917         pci_disable_device(pdev);
5918 }
5919
5920 /* PCI Error Recovery (ERS) */
5921 static struct pci_error_handlers e1000_err_handler = {
5922         .error_detected = e1000_io_error_detected,
5923         .slot_reset = e1000_io_slot_reset,
5924         .resume = e1000_io_resume,
5925 };
5926
5927 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
5928         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
5929         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
5930         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
5931         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
5932         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
5933         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
5934         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
5935         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
5936         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
5937
5938         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
5939         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
5940         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
5941         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
5942
5943         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
5944         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
5945         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
5946
5947         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
5948         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
5949         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
5950
5951         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
5952           board_80003es2lan },
5953         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
5954           board_80003es2lan },
5955         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
5956           board_80003es2lan },
5957         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
5958           board_80003es2lan },
5959
5960         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
5961         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
5962         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
5963         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
5964         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
5965         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
5966         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
5967         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
5968
5969         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
5970         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
5971         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
5972         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
5973         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
5974         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
5975         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
5976         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
5977         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
5978
5979         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
5980         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
5981         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
5982
5983         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
5984         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
5985         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
5986
5987         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
5988         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
5989         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
5990         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
5991
5992         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
5993         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
5994
5995         { }     /* terminate list */
5996 };
5997 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
5998
5999 #ifdef CONFIG_PM_OPS
6000 static const struct dev_pm_ops e1000_pm_ops = {
6001         SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
6002         SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
6003                                 e1000_runtime_resume, e1000_idle)
6004 };
6005 #endif
6006
6007 /* PCI Device API Driver */
6008 static struct pci_driver e1000_driver = {
6009         .name     = e1000e_driver_name,
6010         .id_table = e1000_pci_tbl,
6011         .probe    = e1000_probe,
6012         .remove   = __devexit_p(e1000_remove),
6013 #ifdef CONFIG_PM_OPS
6014         .driver.pm = &e1000_pm_ops,
6015 #endif
6016         .shutdown = e1000_shutdown,
6017         .err_handler = &e1000_err_handler
6018 };
6019
6020 /**
6021  * e1000_init_module - Driver Registration Routine
6022  *
6023  * e1000_init_module is the first routine called when the driver is
6024  * loaded. All it does is register with the PCI subsystem.
6025  **/
6026 static int __init e1000_init_module(void)
6027 {
6028         int ret;
6029         pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
6030                 e1000e_driver_version);
6031         pr_info("Copyright (c) 1999 - 2010 Intel Corporation.\n");
6032         ret = pci_register_driver(&e1000_driver);
6033
6034         return ret;
6035 }
6036 module_init(e1000_init_module);
6037
6038 /**
6039  * e1000_exit_module - Driver Exit Cleanup Routine
6040  *
6041  * e1000_exit_module is called just before the driver is removed
6042  * from memory.
6043  **/
6044 static void __exit e1000_exit_module(void)
6045 {
6046         pci_unregister_driver(&e1000_driver);
6047 }
6048 module_exit(e1000_exit_module);
6049
6050
6051 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
6052 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
6053 MODULE_LICENSE("GPL");
6054 MODULE_VERSION(DRV_VERSION);
6055
6056 /* e1000_main.c */