drivers/net/ethernet/intel/e1000e/netdev.c

   1 /*******************************************************************************
   2
   3   Intel PRO/1000 Linux driver
   4   Copyright(c) 1999 - 2011 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/interrupt.h>
  40 #include <linux/tcp.h>
  41 #include <linux/ipv6.h>
  42 #include <linux/slab.h>
  43 #include <net/checksum.h>
  44 #include <net/ip6_checksum.h>
  45 #include <linux/mii.h>
  46 #include <linux/ethtool.h>
  47 #include <linux/if_vlan.h>
  48 #include <linux/cpu.h>
  49 #include <linux/smp.h>
  50 #include <linux/pm_qos.h>
  51 #include <linux/pm_runtime.h>
  52 #include <linux/aer.h>
  53 #include <linux/prefetch.h>
  54
  55 #include "e1000.h"
  56
  57 #define DRV_EXTRAVERSION "-k"
  58
  59 #define DRV_VERSION "1.5.1" DRV_EXTRAVERSION
  60 char e1000e_driver_name[] = "e1000e";
  61 const char e1000e_driver_version[] = DRV_VERSION;
  62
  63 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state);
  64
  65 static const struct e1000_info *e1000_info_tbl[] = {
  66         [board_82571]           = &e1000_82571_info,
  67         [board_82572]           = &e1000_82572_info,
  68         [board_82573]           = &e1000_82573_info,
  69         [board_82574]           = &e1000_82574_info,
  70         [board_82583]           = &e1000_82583_info,
  71         [board_80003es2lan]     = &e1000_es2_info,
  72         [board_ich8lan]         = &e1000_ich8_info,
  73         [board_ich9lan]         = &e1000_ich9_info,
  74         [board_ich10lan]        = &e1000_ich10_info,
  75         [board_pchlan]          = &e1000_pch_info,
  76         [board_pch2lan]         = &e1000_pch2_info,
  77 };
  78
  79 struct e1000_reg_info {
  80         u32 ofs;
  81         char *name;
  82 };
  83
  84 #define E1000_RDFH      0x02410 /* Rx Data FIFO Head - RW */
  85 #define E1000_RDFT      0x02418 /* Rx Data FIFO Tail - RW */
  86 #define E1000_RDFHS     0x02420 /* Rx Data FIFO Head Saved - RW */
  87 #define E1000_RDFTS     0x02428 /* Rx Data FIFO Tail Saved - RW */
  88 #define E1000_RDFPC     0x02430 /* Rx Data FIFO Packet Count - RW */
  89
  90 #define E1000_TDFH      0x03410 /* Tx Data FIFO Head - RW */
  91 #define E1000_TDFT      0x03418 /* Tx Data FIFO Tail - RW */
  92 #define E1000_TDFHS     0x03420 /* Tx Data FIFO Head Saved - RW */
  93 #define E1000_TDFTS     0x03428 /* Tx Data FIFO Tail Saved - RW */
  94 #define E1000_TDFPC     0x03430 /* Tx Data FIFO Packet Count - RW */
  95
  96 static const struct e1000_reg_info e1000_reg_info_tbl[] = {
  97
  98         /* General Registers */
  99         {E1000_CTRL, "CTRL"},
 100         {E1000_STATUS, "STATUS"},
 101         {E1000_CTRL_EXT, "CTRL_EXT"},
 102
 103         /* Interrupt Registers */
 104         {E1000_ICR, "ICR"},
 105
 106         /* Rx Registers */
 107         {E1000_RCTL, "RCTL"},
 108         {E1000_RDLEN, "RDLEN"},
 109         {E1000_RDH, "RDH"},
 110         {E1000_RDT, "RDT"},
 111         {E1000_RDTR, "RDTR"},
 112         {E1000_RXDCTL(0), "RXDCTL"},
 113         {E1000_ERT, "ERT"},
 114         {E1000_RDBAL, "RDBAL"},
 115         {E1000_RDBAH, "RDBAH"},
 116         {E1000_RDFH, "RDFH"},
 117         {E1000_RDFT, "RDFT"},
 118         {E1000_RDFHS, "RDFHS"},
 119         {E1000_RDFTS, "RDFTS"},
 120         {E1000_RDFPC, "RDFPC"},
 121
 122         /* Tx Registers */
 123         {E1000_TCTL, "TCTL"},
 124         {E1000_TDBAL, "TDBAL"},
 125         {E1000_TDBAH, "TDBAH"},
 126         {E1000_TDLEN, "TDLEN"},
 127         {E1000_TDH, "TDH"},
 128         {E1000_TDT, "TDT"},
 129         {E1000_TIDV, "TIDV"},
 130         {E1000_TXDCTL(0), "TXDCTL"},
 131         {E1000_TADV, "TADV"},
 132         {E1000_TARC(0), "TARC"},
 133         {E1000_TDFH, "TDFH"},
 134         {E1000_TDFT, "TDFT"},
 135         {E1000_TDFHS, "TDFHS"},
 136         {E1000_TDFTS, "TDFTS"},
 137         {E1000_TDFPC, "TDFPC"},
 138
 139         /* List Terminator */
 140         {}
 141 };
 142
 143 /*
 144  * e1000_regdump - register printout routine
 145  */
 146 static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
 147 {
 148         int n = 0;
 149         char rname[16];
 150         u32 regs[8];
 151
 152         switch (reginfo->ofs) {
 153         case E1000_RXDCTL(0):
 154                 for (n = 0; n < 2; n++)
 155                         regs[n] = __er32(hw, E1000_RXDCTL(n));
 156                 break;
 157         case E1000_TXDCTL(0):
 158                 for (n = 0; n < 2; n++)
 159                         regs[n] = __er32(hw, E1000_TXDCTL(n));
 160                 break;
 161         case E1000_TARC(0):
 162                 for (n = 0; n < 2; n++)
 163                         regs[n] = __er32(hw, E1000_TARC(n));
 164                 break;
 165         default:
 166                 printk(KERN_INFO "%-15s %08x\n",
 167                        reginfo->name, __er32(hw, reginfo->ofs));
 168                 return;
 169         }
 170
 171         snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
 172         printk(KERN_INFO "%-15s ", rname);
 173         for (n = 0; n < 2; n++)
 174                 printk(KERN_CONT "%08x ", regs[n]);
 175         printk(KERN_CONT "\n");
 176 }
 177
 178 /*
 179  * e1000e_dump - Print registers, Tx-ring and Rx-ring
 180  */
 181 static void e1000e_dump(struct e1000_adapter *adapter)
 182 {
 183         struct net_device *netdev = adapter->netdev;
 184         struct e1000_hw *hw = &adapter->hw;
 185         struct e1000_reg_info *reginfo;
 186         struct e1000_ring *tx_ring = adapter->tx_ring;
 187         struct e1000_tx_desc *tx_desc;
 188         struct my_u0 {
 189                 u64 a;
 190                 u64 b;
 191         } *u0;
 192         struct e1000_buffer *buffer_info;
 193         struct e1000_ring *rx_ring = adapter->rx_ring;
 194         union e1000_rx_desc_packet_split *rx_desc_ps;
 195         union e1000_rx_desc_extended *rx_desc;
 196         struct my_u1 {
 197                 u64 a;
 198                 u64 b;
 199                 u64 c;
 200                 u64 d;
 201         } *u1;
 202         u32 staterr;
 203         int i = 0;
 204
 205         if (!netif_msg_hw(adapter))
 206                 return;
 207
 208         /* Print netdevice Info */
 209         if (netdev) {
 210                 dev_info(&adapter->pdev->dev, "Net device Info\n");
 211                 printk(KERN_INFO "Device Name     state            "
 212                        "trans_start      last_rx\n");
 213                 printk(KERN_INFO "%-15s %016lX %016lX %016lX\n",
 214                        netdev->name, netdev->state, netdev->trans_start,
 215                        netdev->last_rx);
 216         }
 217
 218         /* Print Registers */
 219         dev_info(&adapter->pdev->dev, "Register Dump\n");
 220         printk(KERN_INFO " Register Name   Value\n");
 221         for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
 222              reginfo->name; reginfo++) {
 223                 e1000_regdump(hw, reginfo);
 224         }
 225
 226         /* Print Tx Ring Summary */
 227         if (!netdev || !netif_running(netdev))
 228                 goto exit;
 229
 230         dev_info(&adapter->pdev->dev, "Tx Ring Summary\n");
 231         printk(KERN_INFO "Queue [NTU] [NTC] [bi(ntc)->dma  ]"
 232                " leng ntw timestamp\n");
 233         buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
 234         printk(KERN_INFO " %5d %5X %5X %016llX %04X %3X %016llX\n",
 235                0, tx_ring->next_to_use, tx_ring->next_to_clean,
 236                (unsigned long long)buffer_info->dma,
 237                buffer_info->length,
 238                buffer_info->next_to_watch,
 239                (unsigned long long)buffer_info->time_stamp);
 240
 241         /* Print Tx Ring */
 242         if (!netif_msg_tx_done(adapter))
 243                 goto rx_ring_summary;
 244
 245         dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
 246
 247         /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
 248          *
 249          * Legacy Transmit Descriptor
 250          *   +--------------------------------------------------------------+
 251          * 0 |         Buffer Address [63:0] (Reserved on Write Back)       |
 252          *   +--------------------------------------------------------------+
 253          * 8 | Special  |    CSS     | Status |  CMD    |  CSO   |  Length  |
 254          *   +--------------------------------------------------------------+
 255          *   63       48 47        36 35    32 31     24 23    16 15        0
 256          *
 257          * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
 258          *   63      48 47    40 39       32 31             16 15    8 7      0
 259          *   +----------------------------------------------------------------+
 260          * 0 |  TUCSE  | TUCS0  |   TUCSS   |     IPCSE       | IPCS0 | IPCSS |
 261          *   +----------------------------------------------------------------+
 262          * 8 |   MSS   | HDRLEN | RSV | STA | TUCMD | DTYP |      PAYLEN      |
 263          *   +----------------------------------------------------------------+
 264          *   63      48 47    40 39 36 35 32 31   24 23  20 19                0
 265          *
 266          * Extended Data Descriptor (DTYP=0x1)
 267          *   +----------------------------------------------------------------+
 268          * 0 |                     Buffer Address [63:0]                      |
 269          *   +----------------------------------------------------------------+
 270          * 8 | VLAN tag |  POPTS  | Rsvd | Status | Command | DTYP |  DTALEN  |
 271          *   +----------------------------------------------------------------+
 272          *   63       48 47     40 39  36 35    32 31     24 23  20 19        0
 273          */
 274         printk(KERN_INFO "Tl[desc]     [address 63:0  ] [SpeCssSCmCsLen]"
 275                " [bi->dma       ] leng  ntw timestamp        bi->skb "
 276                "<-- Legacy format\n");
 277         printk(KERN_INFO "Tc[desc]     [Ce CoCsIpceCoS] [MssHlRSCm0Plen]"
 278                " [bi->dma       ] leng  ntw timestamp        bi->skb "
 279                "<-- Ext Context format\n");
 280         printk(KERN_INFO "Td[desc]     [address 63:0  ] [VlaPoRSCm1Dlen]"
 281                " [bi->dma       ] leng  ntw timestamp        bi->skb "
 282                "<-- Ext Data format\n");
 283         for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
 284                 tx_desc = E1000_TX_DESC(*tx_ring, i);
 285                 buffer_info = &tx_ring->buffer_info[i];
 286                 u0 = (struct my_u0 *)tx_desc;
 287                 printk(KERN_INFO "T%c[0x%03X]    %016llX %016llX %016llX "
 288                        "%04X  %3X %016llX %p",
 289                        (!(le64_to_cpu(u0->b) & (1 << 29)) ? 'l' :
 290                         ((le64_to_cpu(u0->b) & (1 << 20)) ? 'd' : 'c')), i,
 291                        (unsigned long long)le64_to_cpu(u0->a),
 292                        (unsigned long long)le64_to_cpu(u0->b),
 293                        (unsigned long long)buffer_info->dma,
 294                        buffer_info->length, buffer_info->next_to_watch,
 295                        (unsigned long long)buffer_info->time_stamp,
 296                        buffer_info->skb);
 297                 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
 298                         printk(KERN_CONT " NTC/U\n");
 299                 else if (i == tx_ring->next_to_use)
 300                         printk(KERN_CONT " NTU\n");
 301                 else if (i == tx_ring->next_to_clean)
 302                         printk(KERN_CONT " NTC\n");
 303                 else
 304                         printk(KERN_CONT "\n");
 305
 306                 if (netif_msg_pktdata(adapter) && buffer_info->dma != 0)
 307                         print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
 308                                        16, 1, phys_to_virt(buffer_info->dma),
 309                                        buffer_info->length, true);
 310         }
 311
 312         /* Print Rx Ring Summary */
 313 rx_ring_summary:
 314         dev_info(&adapter->pdev->dev, "Rx Ring Summary\n");
 315         printk(KERN_INFO "Queue [NTU] [NTC]\n");
 316         printk(KERN_INFO " %5d %5X %5X\n", 0,
 317                rx_ring->next_to_use, rx_ring->next_to_clean);
 318
 319         /* Print Rx Ring */
 320         if (!netif_msg_rx_status(adapter))
 321                 goto exit;
 322
 323         dev_info(&adapter->pdev->dev, "Rx Ring Dump\n");
 324         switch (adapter->rx_ps_pages) {
 325         case 1:
 326         case 2:
 327         case 3:
 328                 /* [Extended] Packet Split Receive Descriptor Format
 329                  *
 330                  *    +-----------------------------------------------------+
 331                  *  0 |                Buffer Address 0 [63:0]              |
 332                  *    +-----------------------------------------------------+
 333                  *  8 |                Buffer Address 1 [63:0]              |
 334                  *    +-----------------------------------------------------+
 335                  * 16 |                Buffer Address 2 [63:0]              |
 336                  *    +-----------------------------------------------------+
 337                  * 24 |                Buffer Address 3 [63:0]              |
 338                  *    +-----------------------------------------------------+
 339                  */
 340                 printk(KERN_INFO "R  [desc]      [buffer 0 63:0 ] "
 341                        "[buffer 1 63:0 ] "
 342                        "[buffer 2 63:0 ] [buffer 3 63:0 ] [bi->dma       ] "
 343                        "[bi->skb] <-- Ext Pkt Split format\n");
 344                 /* [Extended] Receive Descriptor (Write-Back) Format
 345                  *
 346                  *   63       48 47    32 31     13 12    8 7    4 3        0
 347                  *   +------------------------------------------------------+
 348                  * 0 | Packet   | IP     |  Rsvd   | MRQ   | Rsvd | MRQ RSS |
 349                  *   | Checksum | Ident  |         | Queue |      |  Type   |
 350                  *   +------------------------------------------------------+
 351                  * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 352                  *   +------------------------------------------------------+
 353                  *   63       48 47    32 31            20 19               0
 354                  */
 355                 printk(KERN_INFO "RWB[desc]      [ck ipid mrqhsh] "
 356                        "[vl   l0 ee  es] "
 357                        "[ l3  l2  l1 hs] [reserved      ] ---------------- "
 358                        "[bi->skb] <-- Ext Rx Write-Back format\n");
 359                 for (i = 0; i < rx_ring->count; i++) {
 360                         buffer_info = &rx_ring->buffer_info[i];
 361                         rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
 362                         u1 = (struct my_u1 *)rx_desc_ps;
 363                         staterr =
 364                             le32_to_cpu(rx_desc_ps->wb.middle.status_error);
 365                         if (staterr & E1000_RXD_STAT_DD) {
 366                                 /* Descriptor Done */
 367                                 printk(KERN_INFO "RWB[0x%03X]     %016llX "
 368                                        "%016llX %016llX %016llX "
 369                                        "---------------- %p", i,
 370                                        (unsigned long long)le64_to_cpu(u1->a),
 371                                        (unsigned long long)le64_to_cpu(u1->b),
 372                                        (unsigned long long)le64_to_cpu(u1->c),
 373                                        (unsigned long long)le64_to_cpu(u1->d),
 374                                        buffer_info->skb);
 375                         } else {
 376                                 printk(KERN_INFO "R  [0x%03X]     %016llX "
 377                                        "%016llX %016llX %016llX %016llX %p", i,
 378                                        (unsigned long long)le64_to_cpu(u1->a),
 379                                        (unsigned long long)le64_to_cpu(u1->b),
 380                                        (unsigned long long)le64_to_cpu(u1->c),
 381                                        (unsigned long long)le64_to_cpu(u1->d),
 382                                        (unsigned long long)buffer_info->dma,
 383                                        buffer_info->skb);
 384
 385                                 if (netif_msg_pktdata(adapter))
 386                                         print_hex_dump(KERN_INFO, "",
 387                                                 DUMP_PREFIX_ADDRESS, 16, 1,
 388                                                 phys_to_virt(buffer_info->dma),
 389                                                 adapter->rx_ps_bsize0, true);
 390                         }
 391
 392                         if (i == rx_ring->next_to_use)
 393                                 printk(KERN_CONT " NTU\n");
 394                         else if (i == rx_ring->next_to_clean)
 395                                 printk(KERN_CONT " NTC\n");
 396                         else
 397                                 printk(KERN_CONT "\n");
 398                 }
 399                 break;
 400         default:
 401         case 0:
 402                 /* Extended Receive Descriptor (Read) Format
 403                  *
 404                  *   +-----------------------------------------------------+
 405                  * 0 |                Buffer Address [63:0]                |
 406                  *   +-----------------------------------------------------+
 407                  * 8 |                      Reserved                       |
 408                  *   +-----------------------------------------------------+
 409                  */
 410                 printk(KERN_INFO "R  [desc]      [buf addr 63:0 ] "
 411                        "[reserved 63:0 ] [bi->dma       ] "
 412                        "[bi->skb] <-- Ext (Read) format\n");
 413                 /* Extended Receive Descriptor (Write-Back) Format
 414                  *
 415                  *   63       48 47    32 31    24 23            4 3        0
 416                  *   +------------------------------------------------------+
 417                  *   |     RSS Hash      |        |               |         |
 418                  * 0 +-------------------+  Rsvd  |   Reserved    | MRQ RSS |
 419                  *   | Packet   | IP     |        |               |  Type   |
 420                  *   | Checksum | Ident  |        |               |         |
 421                  *   +------------------------------------------------------+
 422                  * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 423                  *   +------------------------------------------------------+
 424                  *   63       48 47    32 31            20 19               0
 425                  */
 426                 printk(KERN_INFO "RWB[desc]      [cs ipid    mrq] "
 427                        "[vt   ln xe  xs] "
 428                        "[bi->skb] <-- Ext (Write-Back) format\n");
 429
 430                 for (i = 0; i < rx_ring->count; i++) {
 431                         buffer_info = &rx_ring->buffer_info[i];
 432                         rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 433                         u1 = (struct my_u1 *)rx_desc;
 434                         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 435                         if (staterr & E1000_RXD_STAT_DD) {
 436                                 /* Descriptor Done */
 437                                 printk(KERN_INFO "RWB[0x%03X]     %016llX "
 438                                        "%016llX ---------------- %p", i,
 439                                        (unsigned long long)le64_to_cpu(u1->a),
 440                                        (unsigned long long)le64_to_cpu(u1->b),
 441                                        buffer_info->skb);
 442                         } else {
 443                                 printk(KERN_INFO "R  [0x%03X]     %016llX "
 444                                        "%016llX %016llX %p", i,
 445                                        (unsigned long long)le64_to_cpu(u1->a),
 446                                        (unsigned long long)le64_to_cpu(u1->b),
 447                                        (unsigned long long)buffer_info->dma,
 448                                        buffer_info->skb);
 449
 450                                 if (netif_msg_pktdata(adapter))
 451                                         print_hex_dump(KERN_INFO, "",
 452                                                        DUMP_PREFIX_ADDRESS, 16,
 453                                                        1,
 454                                                        phys_to_virt
 455                                                        (buffer_info->dma),
 456                                                        adapter->rx_buffer_len,
 457                                                        true);
 458                         }
 459
 460                         if (i == rx_ring->next_to_use)
 461                                 printk(KERN_CONT " NTU\n");
 462                         else if (i == rx_ring->next_to_clean)
 463                                 printk(KERN_CONT " NTC\n");
 464                         else
 465                                 printk(KERN_CONT "\n");
 466                 }
 467         }
 468
 469 exit:
 470         return;
 471 }
 472
 473 /**
 474  * e1000_desc_unused - calculate if we have unused descriptors
 475  **/
 476 static int e1000_desc_unused(struct e1000_ring *ring)
 477 {
 478         if (ring->next_to_clean > ring->next_to_use)
 479                 return ring->next_to_clean - ring->next_to_use - 1;
 480
 481         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
 482 }
 483
 484 /**
 485  * e1000_receive_skb - helper function to handle Rx indications
 486  * @adapter: board private structure
 487  * @status: descriptor status field as written by hardware
 488  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
 489  * @skb: pointer to sk_buff to be indicated to stack
 490  **/
 491 static void e1000_receive_skb(struct e1000_adapter *adapter,
 492                               struct net_device *netdev, struct sk_buff *skb,
 493                               u8 status, __le16 vlan)
 494 {
 495         u16 tag = le16_to_cpu(vlan);
 496         skb->protocol = eth_type_trans(skb, netdev);
 497
 498         if (status & E1000_RXD_STAT_VP)
 499                 __vlan_hwaccel_put_tag(skb, tag);
 500
 501         napi_gro_receive(&adapter->napi, skb);
 502 }
 503
 504 /**
 505  * e1000_rx_checksum - Receive Checksum Offload
 506  * @adapter:     board private structure
 507  * @status_err:  receive descriptor status and error fields
 508  * @csum:       receive descriptor csum field
 509  * @sk_buff:     socket buffer with received data
 510  **/
 511 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
 512                               u32 csum, struct sk_buff *skb)
 513 {
 514         u16 status = (u16)status_err;
 515         u8 errors = (u8)(status_err >> 24);
 516
 517         skb_checksum_none_assert(skb);
 518
 519         /* Ignore Checksum bit is set */
 520         if (status & E1000_RXD_STAT_IXSM)
 521                 return;
 522         /* TCP/UDP checksum error bit is set */
 523         if (errors & E1000_RXD_ERR_TCPE) {
 524                 /* let the stack verify checksum errors */
 525                 adapter->hw_csum_err++;
 526                 return;
 527         }
 528
 529         /* TCP/UDP Checksum has not been calculated */
 530         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
 531                 return;
 532
 533         /* It must be a TCP or UDP packet with a valid checksum */
 534         if (status & E1000_RXD_STAT_TCPCS) {
 535                 /* TCP checksum is good */
 536                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 537         } else {
 538                 /*
 539                  * IP fragment with UDP payload
 540                  * Hardware complements the payload checksum, so we undo it
 541                  * and then put the value in host order for further stack use.
 542                  */
 543                 __sum16 sum = (__force __sum16)htons(csum);
 544                 skb->csum = csum_unfold(~sum);
 545                 skb->ip_summed = CHECKSUM_COMPLETE;
 546         }
 547         adapter->hw_csum_good++;
 548 }
 549
 550 /**
 551  * e1000e_update_tail_wa - helper function for e1000e_update_[rt]dt_wa()
 552  * @hw: pointer to the HW structure
 553  * @tail: address of tail descriptor register
 554  * @i: value to write to tail descriptor register
 555  *
 556  * When updating the tail register, the ME could be accessing Host CSR
 557  * registers at the same time.  Normally, this is handled in h/w by an
 558  * arbiter but on some parts there is a bug that acknowledges Host accesses
 559  * later than it should which could result in the descriptor register to
 560  * have an incorrect value.  Workaround this by checking the FWSM register
 561  * which has bit 24 set while ME is accessing Host CSR registers, wait
 562  * if it is set and try again a number of times.
 563  **/
 564 static inline s32 e1000e_update_tail_wa(struct e1000_hw *hw, u8 __iomem * tail,
 565                                         unsigned int i)
 566 {
 567         unsigned int j = 0;
 568
 569         while ((j++ < E1000_ICH_FWSM_PCIM2PCI_COUNT) &&
 570                (er32(FWSM) & E1000_ICH_FWSM_PCIM2PCI))
 571                 udelay(50);
 572
 573         writel(i, tail);
 574
 575         if ((j == E1000_ICH_FWSM_PCIM2PCI_COUNT) && (i != readl(tail)))
 576                 return E1000_ERR_SWFW_SYNC;
 577
 578         return 0;
 579 }
 580
 581 static void e1000e_update_rdt_wa(struct e1000_adapter *adapter, unsigned int i)
 582 {
 583         u8 __iomem *tail = (adapter->hw.hw_addr + adapter->rx_ring->tail);
 584         struct e1000_hw *hw = &adapter->hw;
 585
 586         if (e1000e_update_tail_wa(hw, tail, i)) {
 587                 u32 rctl = er32(RCTL);
 588                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
 589                 e_err("ME firmware caused invalid RDT - resetting\n");
 590                 schedule_work(&adapter->reset_task);
 591         }
 592 }
 593
 594 static void e1000e_update_tdt_wa(struct e1000_adapter *adapter, unsigned int i)
 595 {
 596         u8 __iomem *tail = (adapter->hw.hw_addr + adapter->tx_ring->tail);
 597         struct e1000_hw *hw = &adapter->hw;
 598
 599         if (e1000e_update_tail_wa(hw, tail, i)) {
 600                 u32 tctl = er32(TCTL);
 601                 ew32(TCTL, tctl & ~E1000_TCTL_EN);
 602                 e_err("ME firmware caused invalid TDT - resetting\n");
 603                 schedule_work(&adapter->reset_task);
 604         }
 605 }
 606
 607 /**
 608  * e1000_alloc_rx_buffers - Replace used receive buffers
 609  * @adapter: address of board private structure
 610  **/
 611 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
 612                                    int cleaned_count, gfp_t gfp)
 613 {
 614         struct net_device *netdev = adapter->netdev;
 615         struct pci_dev *pdev = adapter->pdev;
 616         struct e1000_ring *rx_ring = adapter->rx_ring;
 617         union e1000_rx_desc_extended *rx_desc;
 618         struct e1000_buffer *buffer_info;
 619         struct sk_buff *skb;
 620         unsigned int i;
 621         unsigned int bufsz = adapter->rx_buffer_len;
 622
 623         i = rx_ring->next_to_use;
 624         buffer_info = &rx_ring->buffer_info[i];
 625
 626         while (cleaned_count--) {
 627                 skb = buffer_info->skb;
 628                 if (skb) {
 629                         skb_trim(skb, 0);
 630                         goto map_skb;
 631                 }
 632
 633                 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
 634                 if (!skb) {
 635                         /* Better luck next round */
 636                         adapter->alloc_rx_buff_failed++;
 637                         break;
 638                 }
 639
 640                 buffer_info->skb = skb;
 641 map_skb:
 642                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 643                                                   adapter->rx_buffer_len,
 644                                                   DMA_FROM_DEVICE);
 645                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 646                         dev_err(&pdev->dev, "Rx DMA map failed\n");
 647                         adapter->rx_dma_failed++;
 648                         break;
 649                 }
 650
 651                 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 652                 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
 653
 654                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 655                         /*
 656                          * Force memory writes to complete before letting h/w
 657                          * know there are new descriptors to fetch.  (Only
 658                          * applicable for weak-ordered memory model archs,
 659                          * such as IA-64).
 660                          */
 661                         wmb();
 662                         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
 663                                 e1000e_update_rdt_wa(adapter, i);
 664                         else
 665                                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 666                 }
 667                 i++;
 668                 if (i == rx_ring->count)
 669                         i = 0;
 670                 buffer_info = &rx_ring->buffer_info[i];
 671         }
 672
 673         rx_ring->next_to_use = i;
 674 }
 675
 676 /**
 677  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
 678  * @adapter: address of board private structure
 679  **/
 680 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 681                                       int cleaned_count, gfp_t gfp)
 682 {
 683         struct net_device *netdev = adapter->netdev;
 684         struct pci_dev *pdev = adapter->pdev;
 685         union e1000_rx_desc_packet_split *rx_desc;
 686         struct e1000_ring *rx_ring = adapter->rx_ring;
 687         struct e1000_buffer *buffer_info;
 688         struct e1000_ps_page *ps_page;
 689         struct sk_buff *skb;
 690         unsigned int i, j;
 691
 692         i = rx_ring->next_to_use;
 693         buffer_info = &rx_ring->buffer_info[i];
 694
 695         while (cleaned_count--) {
 696                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
 697
 698                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
 699                         ps_page = &buffer_info->ps_pages[j];
 700                         if (j >= adapter->rx_ps_pages) {
 701                                 /* all unused desc entries get hw null ptr */
 702                                 rx_desc->read.buffer_addr[j + 1] =
 703                                     ~cpu_to_le64(0);
 704                                 continue;
 705                         }
 706                         if (!ps_page->page) {
 707                                 ps_page->page = alloc_page(gfp);
 708                                 if (!ps_page->page) {
 709                                         adapter->alloc_rx_buff_failed++;
 710                                         goto no_buffers;
 711                                 }
 712                                 ps_page->dma = dma_map_page(&pdev->dev,
 713                                                             ps_page->page,
 714                                                             0, PAGE_SIZE,
 715                                                             DMA_FROM_DEVICE);
 716                                 if (dma_mapping_error(&pdev->dev,
 717                                                       ps_page->dma)) {
 718                                         dev_err(&adapter->pdev->dev,
 719                                                 "Rx DMA page map failed\n");
 720                                         adapter->rx_dma_failed++;
 721                                         goto no_buffers;
 722                                 }
 723                         }
 724                         /*
 725                          * Refresh the desc even if buffer_addrs
 726                          * didn't change because each write-back
 727                          * erases this info.
 728                          */
 729                         rx_desc->read.buffer_addr[j + 1] =
 730                             cpu_to_le64(ps_page->dma);
 731                 }
 732
 733                 skb = __netdev_alloc_skb_ip_align(netdev,
 734                                                   adapter->rx_ps_bsize0,
 735                                                   gfp);
 736
 737                 if (!skb) {
 738                         adapter->alloc_rx_buff_failed++;
 739                         break;
 740                 }
 741
 742                 buffer_info->skb = skb;
 743                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
 744                                                   adapter->rx_ps_bsize0,
 745                                                   DMA_FROM_DEVICE);
 746                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
 747                         dev_err(&pdev->dev, "Rx DMA map failed\n");
 748                         adapter->rx_dma_failed++;
 749                         /* cleanup skb */
 750                         dev_kfree_skb_any(skb);
 751                         buffer_info->skb = NULL;
 752                         break;
 753                 }
 754
 755                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
 756
 757                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
 758                         /*
 759                          * Force memory writes to complete before letting h/w
 760                          * know there are new descriptors to fetch.  (Only
 761                          * applicable for weak-ordered memory model archs,
 762                          * such as IA-64).
 763                          */
 764                         wmb();
 765                         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
 766                                 e1000e_update_rdt_wa(adapter, i << 1);
 767                         else
 768                                 writel(i << 1,
 769                                        adapter->hw.hw_addr + rx_ring->tail);
 770                 }
 771
 772                 i++;
 773                 if (i == rx_ring->count)
 774                         i = 0;
 775                 buffer_info = &rx_ring->buffer_info[i];
 776         }
 777
 778 no_buffers:
 779         rx_ring->next_to_use = i;
 780 }
 781
 782 /**
 783  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
 784  * @adapter: address of board private structure
 785  * @cleaned_count: number of buffers to allocate this pass
 786  **/
 787
 788 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
 789                                          int cleaned_count, gfp_t gfp)
 790 {
 791         struct net_device *netdev = adapter->netdev;
 792         struct pci_dev *pdev = adapter->pdev;
 793         union e1000_rx_desc_extended *rx_desc;
 794         struct e1000_ring *rx_ring = adapter->rx_ring;
 795         struct e1000_buffer *buffer_info;
 796         struct sk_buff *skb;
 797         unsigned int i;
 798         unsigned int bufsz = 256 - 16 /* for skb_reserve */;
 799
 800         i = rx_ring->next_to_use;
 801         buffer_info = &rx_ring->buffer_info[i];
 802
 803         while (cleaned_count--) {
 804                 skb = buffer_info->skb;
 805                 if (skb) {
 806                         skb_trim(skb, 0);
 807                         goto check_page;
 808                 }
 809
 810                 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
 811                 if (unlikely(!skb)) {
 812                         /* Better luck next round */
 813                         adapter->alloc_rx_buff_failed++;
 814                         break;
 815                 }
 816
 817                 buffer_info->skb = skb;
 818 check_page:
 819                 /* allocate a new page if necessary */
 820                 if (!buffer_info->page) {
 821                         buffer_info->page = alloc_page(gfp);
 822                         if (unlikely(!buffer_info->page)) {
 823                                 adapter->alloc_rx_buff_failed++;
 824                                 break;
 825                         }
 826                 }
 827
 828                 if (!buffer_info->dma)
 829                         buffer_info->dma = dma_map_page(&pdev->dev,
 830                                                         buffer_info->page, 0,
 831                                                         PAGE_SIZE,
 832                                                         DMA_FROM_DEVICE);
 833
 834                 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 835                 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
 836
 837                 if (unlikely(++i == rx_ring->count))
 838                         i = 0;
 839                 buffer_info = &rx_ring->buffer_info[i];
 840         }
 841
 842         if (likely(rx_ring->next_to_use != i)) {
 843                 rx_ring->next_to_use = i;
 844                 if (unlikely(i-- == 0))
 845                         i = (rx_ring->count - 1);
 846
 847                 /* Force memory writes to complete before letting h/w
 848                  * know there are new descriptors to fetch.  (Only
 849                  * applicable for weak-ordered memory model archs,
 850                  * such as IA-64). */
 851                 wmb();
 852                 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
 853                         e1000e_update_rdt_wa(adapter, i);
 854                 else
 855                         writel(i, adapter->hw.hw_addr + rx_ring->tail);
 856         }
 857 }
 858
 859 /**
 860  * e1000_clean_rx_irq - Send received data up the network stack; legacy
 861  * @adapter: board private structure
 862  *
 863  * the return value indicates whether actual cleaning was done, there
 864  * is no guarantee that everything was cleaned
 865  **/
 866 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
 867                                int *work_done, int work_to_do)
 868 {
 869         struct net_device *netdev = adapter->netdev;
 870         struct pci_dev *pdev = adapter->pdev;
 871         struct e1000_hw *hw = &adapter->hw;
 872         struct e1000_ring *rx_ring = adapter->rx_ring;
 873         union e1000_rx_desc_extended *rx_desc, *next_rxd;
 874         struct e1000_buffer *buffer_info, *next_buffer;
 875         u32 length, staterr;
 876         unsigned int i;
 877         int cleaned_count = 0;
 878         bool cleaned = 0;
 879         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 880
 881         i = rx_ring->next_to_clean;
 882         rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
 883         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 884         buffer_info = &rx_ring->buffer_info[i];
 885
 886         while (staterr & E1000_RXD_STAT_DD) {
 887                 struct sk_buff *skb;
 888
 889                 if (*work_done >= work_to_do)
 890                         break;
 891                 (*work_done)++;
 892                 rmb();  /* read descriptor and rx_buffer_info after status DD */
 893
 894                 skb = buffer_info->skb;
 895                 buffer_info->skb = NULL;
 896
 897                 prefetch(skb->data - NET_IP_ALIGN);
 898
 899                 i++;
 900                 if (i == rx_ring->count)
 901                         i = 0;
 902                 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
 903                 prefetch(next_rxd);
 904
 905                 next_buffer = &rx_ring->buffer_info[i];
 906
 907                 cleaned = 1;
 908                 cleaned_count++;
 909                 dma_unmap_single(&pdev->dev,
 910                                  buffer_info->dma,
 911                                  adapter->rx_buffer_len,
 912                                  DMA_FROM_DEVICE);
 913                 buffer_info->dma = 0;
 914
 915                 length = le16_to_cpu(rx_desc->wb.upper.length);
 916
 917                 /*
 918                  * !EOP means multiple descriptors were used to store a single
 919                  * packet, if that's the case we need to toss it.  In fact, we
 920                  * need to toss every packet with the EOP bit clear and the
 921                  * next frame that _does_ have the EOP bit set, as it is by
 922                  * definition only a frame fragment
 923                  */
 924                 if (unlikely(!(staterr & E1000_RXD_STAT_EOP)))
 925                         adapter->flags2 |= FLAG2_IS_DISCARDING;
 926
 927                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
 928                         /* All receives must fit into a single buffer */
 929                         e_dbg("Receive packet consumed multiple buffers\n");
 930                         /* recycle */
 931                         buffer_info->skb = skb;
 932                         if (staterr & E1000_RXD_STAT_EOP)
 933                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
 934                         goto next_desc;
 935                 }
 936
 937                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
 938                         /* recycle */
 939                         buffer_info->skb = skb;
 940                         goto next_desc;
 941                 }
 942
 943                 /* adjust length to remove Ethernet CRC */
 944                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
 945                         length -= 4;
 946
 947                 total_rx_bytes += length;
 948                 total_rx_packets++;
 949
 950                 /*
 951                  * code added for copybreak, this should improve
 952                  * performance for small packets with large amounts
 953                  * of reassembly being done in the stack
 954                  */
 955                 if (length < copybreak) {
 956                         struct sk_buff *new_skb =
 957                             netdev_alloc_skb_ip_align(netdev, length);
 958                         if (new_skb) {
 959                                 skb_copy_to_linear_data_offset(new_skb,
 960                                                                -NET_IP_ALIGN,
 961                                                                (skb->data -
 962                                                                 NET_IP_ALIGN),
 963                                                                (length +
 964                                                                 NET_IP_ALIGN));
 965                                 /* save the skb in buffer_info as good */
 966                                 buffer_info->skb = skb;
 967                                 skb = new_skb;
 968                         }
 969                         /* else just continue with the old one */
 970                 }
 971                 /* end copybreak code */
 972                 skb_put(skb, length);
 973
 974                 /* Receive Checksum Offload */
 975                 e1000_rx_checksum(adapter, staterr,
 976                                   le16_to_cpu(rx_desc->wb.lower.hi_dword.
 977                                               csum_ip.csum), skb);
 978
 979                 e1000_receive_skb(adapter, netdev, skb, staterr,
 980                                   rx_desc->wb.upper.vlan);
 981
 982 next_desc:
 983                 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
 984
 985                 /* return some buffers to hardware, one at a time is too slow */
 986                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
 987                         adapter->alloc_rx_buf(adapter, cleaned_count,
 988                                               GFP_ATOMIC);
 989                         cleaned_count = 0;
 990                 }
 991
 992                 /* use prefetched values */
 993                 rx_desc = next_rxd;
 994                 buffer_info = next_buffer;
 995
 996                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 997         }
 998         rx_ring->next_to_clean = i;
 999
1000         cleaned_count = e1000_desc_unused(rx_ring);
1001         if (cleaned_count)
1002                 adapter->alloc_rx_buf(adapter, cleaned_count, GFP_ATOMIC);
1003
1004         adapter->total_rx_bytes += total_rx_bytes;
1005         adapter->total_rx_packets += total_rx_packets;
1006         return cleaned;
1007 }
1008
1009 static void e1000_put_txbuf(struct e1000_adapter *adapter,
1010                              struct e1000_buffer *buffer_info)
1011 {
1012         if (buffer_info->dma) {
1013                 if (buffer_info->mapped_as_page)
1014                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1015                                        buffer_info->length, DMA_TO_DEVICE);
1016                 else
1017                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1018                                          buffer_info->length, DMA_TO_DEVICE);
1019                 buffer_info->dma = 0;
1020         }
1021         if (buffer_info->skb) {
1022                 dev_kfree_skb_any(buffer_info->skb);
1023                 buffer_info->skb = NULL;
1024         }
1025         buffer_info->time_stamp = 0;
1026 }
1027
1028 static void e1000_print_hw_hang(struct work_struct *work)
1029 {
1030         struct e1000_adapter *adapter = container_of(work,
1031                                                      struct e1000_adapter,
1032                                                      print_hang_task);
1033         struct net_device *netdev = adapter->netdev;
1034         struct e1000_ring *tx_ring = adapter->tx_ring;
1035         unsigned int i = tx_ring->next_to_clean;
1036         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
1037         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
1038         struct e1000_hw *hw = &adapter->hw;
1039         u16 phy_status, phy_1000t_status, phy_ext_status;
1040         u16 pci_status;
1041
1042         if (test_bit(__E1000_DOWN, &adapter->state))
1043                 return;
1044
1045         if (!adapter->tx_hang_recheck &&
1046             (adapter->flags2 & FLAG2_DMA_BURST)) {
1047                 /* May be block on write-back, flush and detect again
1048                  * flush pending descriptor writebacks to memory
1049                  */
1050                 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
1051                 /* execute the writes immediately */
1052                 e1e_flush();
1053                 adapter->tx_hang_recheck = true;
1054                 return;
1055         }
1056         /* Real hang detected */
1057         adapter->tx_hang_recheck = false;
1058         netif_stop_queue(netdev);
1059
1060         e1e_rphy(hw, PHY_STATUS, &phy_status);
1061         e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
1062         e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
1063
1064         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
1065
1066         /* detected Hardware unit hang */
1067         e_err("Detected Hardware Unit Hang:\n"
1068               "  TDH                  <%x>\n"
1069               "  TDT                  <%x>\n"
1070               "  next_to_use          <%x>\n"
1071               "  next_to_clean        <%x>\n"
1072               "buffer_info[next_to_clean]:\n"
1073               "  time_stamp           <%lx>\n"
1074               "  next_to_watch        <%x>\n"
1075               "  jiffies              <%lx>\n"
1076               "  next_to_watch.status <%x>\n"
1077               "MAC Status             <%x>\n"
1078               "PHY Status             <%x>\n"
1079               "PHY 1000BASE-T Status  <%x>\n"
1080               "PHY Extended Status    <%x>\n"
1081               "PCI Status             <%x>\n",
1082               readl(adapter->hw.hw_addr + tx_ring->head),
1083               readl(adapter->hw.hw_addr + tx_ring->tail),
1084               tx_ring->next_to_use,
1085               tx_ring->next_to_clean,
1086               tx_ring->buffer_info[eop].time_stamp,
1087               eop,
1088               jiffies,
1089               eop_desc->upper.fields.status,
1090               er32(STATUS),
1091               phy_status,
1092               phy_1000t_status,
1093               phy_ext_status,
1094               pci_status);
1095 }
1096
1097 /**
1098  * e1000_clean_tx_irq - Reclaim resources after transmit completes
1099  * @adapter: board private structure
1100  *
1101  * the return value indicates whether actual cleaning was done, there
1102  * is no guarantee that everything was cleaned
1103  **/
1104 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
1105 {
1106         struct net_device *netdev = adapter->netdev;
1107         struct e1000_hw *hw = &adapter->hw;
1108         struct e1000_ring *tx_ring = adapter->tx_ring;
1109         struct e1000_tx_desc *tx_desc, *eop_desc;
1110         struct e1000_buffer *buffer_info;
1111         unsigned int i, eop;
1112         unsigned int count = 0;
1113         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
1114
1115         i = tx_ring->next_to_clean;
1116         eop = tx_ring->buffer_info[i].next_to_watch;
1117         eop_desc = E1000_TX_DESC(*tx_ring, eop);
1118
1119         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1120                (count < tx_ring->count)) {
1121                 bool cleaned = false;
1122                 rmb(); /* read buffer_info after eop_desc */
1123                 for (; !cleaned; count++) {
1124                         tx_desc = E1000_TX_DESC(*tx_ring, i);
1125                         buffer_info = &tx_ring->buffer_info[i];
1126                         cleaned = (i == eop);
1127
1128                         if (cleaned) {
1129                                 total_tx_packets += buffer_info->segs;
1130                                 total_tx_bytes += buffer_info->bytecount;
1131                         }
1132
1133                         e1000_put_txbuf(adapter, buffer_info);
1134                         tx_desc->upper.data = 0;
1135
1136                         i++;
1137                         if (i == tx_ring->count)
1138                                 i = 0;
1139                 }
1140
1141                 if (i == tx_ring->next_to_use)
1142                         break;
1143                 eop = tx_ring->buffer_info[i].next_to_watch;
1144                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1145         }
1146
1147         tx_ring->next_to_clean = i;
1148
1149 #define TX_WAKE_THRESHOLD 32
1150         if (count && netif_carrier_ok(netdev) &&
1151             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1152                 /* Make sure that anybody stopping the queue after this
1153                  * sees the new next_to_clean.
1154                  */
1155                 smp_mb();
1156
1157                 if (netif_queue_stopped(netdev) &&
1158                     !(test_bit(__E1000_DOWN, &adapter->state))) {
1159                         netif_wake_queue(netdev);
1160                         ++adapter->restart_queue;
1161                 }
1162         }
1163
1164         if (adapter->detect_tx_hung) {
1165                 /*
1166                  * Detect a transmit hang in hardware, this serializes the
1167                  * check with the clearing of time_stamp and movement of i
1168                  */
1169                 adapter->detect_tx_hung = 0;
1170                 if (tx_ring->buffer_info[i].time_stamp &&
1171                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1172                                + (adapter->tx_timeout_factor * HZ)) &&
1173                     !(er32(STATUS) & E1000_STATUS_TXOFF))
1174                         schedule_work(&adapter->print_hang_task);
1175                 else
1176                         adapter->tx_hang_recheck = false;
1177         }
1178         adapter->total_tx_bytes += total_tx_bytes;
1179         adapter->total_tx_packets += total_tx_packets;
1180         return count < tx_ring->count;
1181 }
1182
1183 /**
1184  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1185  * @adapter: board private structure
1186  *
1187  * the return value indicates whether actual cleaning was done, there
1188  * is no guarantee that everything was cleaned
1189  **/
1190 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
1191                                   int *work_done, int work_to_do)
1192 {
1193         struct e1000_hw *hw = &adapter->hw;
1194         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1195         struct net_device *netdev = adapter->netdev;
1196         struct pci_dev *pdev = adapter->pdev;
1197         struct e1000_ring *rx_ring = adapter->rx_ring;
1198         struct e1000_buffer *buffer_info, *next_buffer;
1199         struct e1000_ps_page *ps_page;
1200         struct sk_buff *skb;
1201         unsigned int i, j;
1202         u32 length, staterr;
1203         int cleaned_count = 0;
1204         bool cleaned = 0;
1205         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1206
1207         i = rx_ring->next_to_clean;
1208         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1209         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1210         buffer_info = &rx_ring->buffer_info[i];
1211
1212         while (staterr & E1000_RXD_STAT_DD) {
1213                 if (*work_done >= work_to_do)
1214                         break;
1215                 (*work_done)++;
1216                 skb = buffer_info->skb;
1217                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1218
1219                 /* in the packet split case this is header only */
1220                 prefetch(skb->data - NET_IP_ALIGN);
1221
1222                 i++;
1223                 if (i == rx_ring->count)
1224                         i = 0;
1225                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1226                 prefetch(next_rxd);
1227
1228                 next_buffer = &rx_ring->buffer_info[i];
1229
1230                 cleaned = 1;
1231                 cleaned_count++;
1232                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1233                                  adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1234                 buffer_info->dma = 0;
1235
1236                 /* see !EOP comment in other Rx routine */
1237                 if (!(staterr & E1000_RXD_STAT_EOP))
1238                         adapter->flags2 |= FLAG2_IS_DISCARDING;
1239
1240                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1241                         e_dbg("Packet Split buffers didn't pick up the full "
1242                               "packet\n");
1243                         dev_kfree_skb_irq(skb);
1244                         if (staterr & E1000_RXD_STAT_EOP)
1245                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1246                         goto next_desc;
1247                 }
1248
1249                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
1250                         dev_kfree_skb_irq(skb);
1251                         goto next_desc;
1252                 }
1253
1254                 length = le16_to_cpu(rx_desc->wb.middle.length0);
1255
1256                 if (!length) {
1257                         e_dbg("Last part of the packet spanning multiple "
1258                               "descriptors\n");
1259                         dev_kfree_skb_irq(skb);
1260                         goto next_desc;
1261                 }
1262
1263                 /* Good Receive */
1264                 skb_put(skb, length);
1265
1266                 {
1267                 /*
1268                  * this looks ugly, but it seems compiler issues make it
1269                  * more efficient than reusing j
1270                  */
1271                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1272
1273                 /*
1274                  * page alloc/put takes too long and effects small packet
1275                  * throughput, so unsplit small packets and save the alloc/put
1276                  * only valid in softirq (napi) context to call kmap_*
1277                  */
1278                 if (l1 && (l1 <= copybreak) &&
1279                     ((length + l1) <= adapter->rx_ps_bsize0)) {
1280                         u8 *vaddr;
1281
1282                         ps_page = &buffer_info->ps_pages[0];
1283
1284                         /*
1285                          * there is no documentation about how to call
1286                          * kmap_atomic, so we can't hold the mapping
1287                          * very long
1288                          */
1289                         dma_sync_single_for_cpu(&pdev->dev, ps_page->dma,
1290                                                 PAGE_SIZE, DMA_FROM_DEVICE);
1291                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
1292                         memcpy(skb_tail_pointer(skb), vaddr, l1);
1293                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
1294                         dma_sync_single_for_device(&pdev->dev, ps_page->dma,
1295                                                    PAGE_SIZE, DMA_FROM_DEVICE);
1296
1297                         /* remove the CRC */
1298                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1299                                 l1 -= 4;
1300
1301                         skb_put(skb, l1);
1302                         goto copydone;
1303                 } /* if */
1304                 }
1305
1306                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1307                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1308                         if (!length)
1309                                 break;
1310
1311                         ps_page = &buffer_info->ps_pages[j];
1312                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1313                                        DMA_FROM_DEVICE);
1314                         ps_page->dma = 0;
1315                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1316                         ps_page->page = NULL;
1317                         skb->len += length;
1318                         skb->data_len += length;
1319                         skb->truesize += PAGE_SIZE;
1320                 }
1321
1322                 /* strip the ethernet crc, problem is we're using pages now so
1323                  * this whole operation can get a little cpu intensive
1324                  */
1325                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1326                         pskb_trim(skb, skb->len - 4);
1327
1328 copydone:
1329                 total_rx_bytes += skb->len;
1330                 total_rx_packets++;
1331
1332                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
1333                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
1334
1335                 if (rx_desc->wb.upper.header_status &
1336                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1337                         adapter->rx_hdr_split++;
1338
1339                 e1000_receive_skb(adapter, netdev, skb,
1340                                   staterr, rx_desc->wb.middle.vlan);
1341
1342 next_desc:
1343                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1344                 buffer_info->skb = NULL;
1345
1346                 /* return some buffers to hardware, one at a time is too slow */
1347                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1348                         adapter->alloc_rx_buf(adapter, cleaned_count,
1349                                               GFP_ATOMIC);
1350                         cleaned_count = 0;
1351                 }
1352
1353                 /* use prefetched values */
1354                 rx_desc = next_rxd;
1355                 buffer_info = next_buffer;
1356
1357                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1358         }
1359         rx_ring->next_to_clean = i;
1360
1361         cleaned_count = e1000_desc_unused(rx_ring);
1362         if (cleaned_count)
1363                 adapter->alloc_rx_buf(adapter, cleaned_count, GFP_ATOMIC);
1364
1365         adapter->total_rx_bytes += total_rx_bytes;
1366         adapter->total_rx_packets += total_rx_packets;
1367         return cleaned;
1368 }
1369
1370 /**
1371  * e1000_consume_page - helper function
1372  **/
1373 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1374                                u16 length)
1375 {
1376         bi->page = NULL;
1377         skb->len += length;
1378         skb->data_len += length;
1379         skb->truesize += PAGE_SIZE;
1380 }
1381
1382 /**
1383  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1384  * @adapter: board private structure
1385  *
1386  * the return value indicates whether actual cleaning was done, there
1387  * is no guarantee that everything was cleaned
1388  **/
1389
1390 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
1391                                      int *work_done, int work_to_do)
1392 {
1393         struct net_device *netdev = adapter->netdev;
1394         struct pci_dev *pdev = adapter->pdev;
1395         struct e1000_ring *rx_ring = adapter->rx_ring;
1396         union e1000_rx_desc_extended *rx_desc, *next_rxd;
1397         struct e1000_buffer *buffer_info, *next_buffer;
1398         u32 length, staterr;
1399         unsigned int i;
1400         int cleaned_count = 0;
1401         bool cleaned = false;
1402         unsigned int total_rx_bytes=0, total_rx_packets=0;
1403
1404         i = rx_ring->next_to_clean;
1405         rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
1406         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1407         buffer_info = &rx_ring->buffer_info[i];
1408
1409         while (staterr & E1000_RXD_STAT_DD) {
1410                 struct sk_buff *skb;
1411
1412                 if (*work_done >= work_to_do)
1413                         break;
1414                 (*work_done)++;
1415                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1416
1417                 skb = buffer_info->skb;
1418                 buffer_info->skb = NULL;
1419
1420                 ++i;
1421                 if (i == rx_ring->count)
1422                         i = 0;
1423                 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
1424                 prefetch(next_rxd);
1425
1426                 next_buffer = &rx_ring->buffer_info[i];
1427
1428                 cleaned = true;
1429                 cleaned_count++;
1430                 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1431                                DMA_FROM_DEVICE);
1432                 buffer_info->dma = 0;
1433
1434                 length = le16_to_cpu(rx_desc->wb.upper.length);
1435
1436                 /* errors is only valid for DD + EOP descriptors */
1437                 if (unlikely((staterr & E1000_RXD_STAT_EOP) &&
1438                              (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK))) {
1439                         /* recycle both page and skb */
1440                         buffer_info->skb = skb;
1441                         /* an error means any chain goes out the window too */
1442                         if (rx_ring->rx_skb_top)
1443                                 dev_kfree_skb_irq(rx_ring->rx_skb_top);
1444                         rx_ring->rx_skb_top = NULL;
1445                         goto next_desc;
1446                 }
1447
1448 #define rxtop (rx_ring->rx_skb_top)
1449                 if (!(staterr & E1000_RXD_STAT_EOP)) {
1450                         /* this descriptor is only the beginning (or middle) */
1451                         if (!rxtop) {
1452                                 /* this is the beginning of a chain */
1453                                 rxtop = skb;
1454                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1455                                                    0, length);
1456                         } else {
1457                                 /* this is the middle of a chain */
1458                                 skb_fill_page_desc(rxtop,
1459                                     skb_shinfo(rxtop)->nr_frags,
1460                                     buffer_info->page, 0, length);
1461                                 /* re-use the skb, only consumed the page */
1462                                 buffer_info->skb = skb;
1463                         }
1464                         e1000_consume_page(buffer_info, rxtop, length);
1465                         goto next_desc;
1466                 } else {
1467                         if (rxtop) {
1468                                 /* end of the chain */
1469                                 skb_fill_page_desc(rxtop,
1470                                     skb_shinfo(rxtop)->nr_frags,
1471                                     buffer_info->page, 0, length);
1472                                 /* re-use the current skb, we only consumed the
1473                                  * page */
1474                                 buffer_info->skb = skb;
1475                                 skb = rxtop;
1476                                 rxtop = NULL;
1477                                 e1000_consume_page(buffer_info, skb, length);
1478                         } else {
1479                                 /* no chain, got EOP, this buf is the packet
1480                                  * copybreak to save the put_page/alloc_page */
1481                                 if (length <= copybreak &&
1482                                     skb_tailroom(skb) >= length) {
1483                                         u8 *vaddr;
1484                                         vaddr = kmap_atomic(buffer_info->page,
1485                                                            KM_SKB_DATA_SOFTIRQ);
1486                                         memcpy(skb_tail_pointer(skb), vaddr,
1487                                                length);
1488                                         kunmap_atomic(vaddr,
1489                                                       KM_SKB_DATA_SOFTIRQ);
1490                                         /* re-use the page, so don't erase
1491                                          * buffer_info->page */
1492                                         skb_put(skb, length);
1493                                 } else {
1494                                         skb_fill_page_desc(skb, 0,
1495                                                            buffer_info->page, 0,
1496                                                            length);
1497                                         e1000_consume_page(buffer_info, skb,
1498                                                            length);
1499                                 }
1500                         }
1501                 }
1502
1503                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1504                 e1000_rx_checksum(adapter, staterr,
1505                                   le16_to_cpu(rx_desc->wb.lower.hi_dword.
1506                                               csum_ip.csum), skb);
1507
1508                 /* probably a little skewed due to removing CRC */
1509                 total_rx_bytes += skb->len;
1510                 total_rx_packets++;
1511
1512                 /* eth type trans needs skb->data to point to something */
1513                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1514                         e_err("pskb_may_pull failed.\n");
1515                         dev_kfree_skb_irq(skb);
1516                         goto next_desc;
1517                 }
1518
1519                 e1000_receive_skb(adapter, netdev, skb, staterr,
1520                                   rx_desc->wb.upper.vlan);
1521
1522 next_desc:
1523                 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1524
1525                 /* return some buffers to hardware, one at a time is too slow */
1526                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1527                         adapter->alloc_rx_buf(adapter, cleaned_count,
1528                                               GFP_ATOMIC);
1529                         cleaned_count = 0;
1530                 }
1531
1532                 /* use prefetched values */
1533                 rx_desc = next_rxd;
1534                 buffer_info = next_buffer;
1535
1536                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1537         }
1538         rx_ring->next_to_clean = i;
1539
1540         cleaned_count = e1000_desc_unused(rx_ring);
1541         if (cleaned_count)
1542                 adapter->alloc_rx_buf(adapter, cleaned_count, GFP_ATOMIC);
1543
1544         adapter->total_rx_bytes += total_rx_bytes;
1545         adapter->total_rx_packets += total_rx_packets;
1546         return cleaned;
1547 }
1548
1549 /**
1550  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1551  * @adapter: board private structure
1552  **/
1553 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1554 {
1555         struct e1000_ring *rx_ring = adapter->rx_ring;
1556         struct e1000_buffer *buffer_info;
1557         struct e1000_ps_page *ps_page;
1558         struct pci_dev *pdev = adapter->pdev;
1559         unsigned int i, j;
1560
1561         /* Free all the Rx ring sk_buffs */
1562         for (i = 0; i < rx_ring->count; i++) {
1563                 buffer_info = &rx_ring->buffer_info[i];
1564                 if (buffer_info->dma) {
1565                         if (adapter->clean_rx == e1000_clean_rx_irq)
1566                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1567                                                  adapter->rx_buffer_len,
1568                                                  DMA_FROM_DEVICE);
1569                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1570                                 dma_unmap_page(&pdev->dev, buffer_info->dma,
1571                                                PAGE_SIZE,
1572                                                DMA_FROM_DEVICE);
1573                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1574                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1575                                                  adapter->rx_ps_bsize0,
1576                                                  DMA_FROM_DEVICE);
1577                         buffer_info->dma = 0;
1578                 }
1579
1580                 if (buffer_info->page) {
1581                         put_page(buffer_info->page);
1582                         buffer_info->page = NULL;
1583                 }
1584
1585                 if (buffer_info->skb) {
1586                         dev_kfree_skb(buffer_info->skb);
1587                         buffer_info->skb = NULL;
1588                 }
1589
1590                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1591                         ps_page = &buffer_info->ps_pages[j];
1592                         if (!ps_page->page)
1593                                 break;
1594                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1595                                        DMA_FROM_DEVICE);
1596                         ps_page->dma = 0;
1597                         put_page(ps_page->page);
1598                         ps_page->page = NULL;
1599                 }
1600         }
1601
1602         /* there also may be some cached data from a chained receive */
1603         if (rx_ring->rx_skb_top) {
1604                 dev_kfree_skb(rx_ring->rx_skb_top);
1605                 rx_ring->rx_skb_top = NULL;
1606         }
1607
1608         /* Zero out the descriptor ring */
1609         memset(rx_ring->desc, 0, rx_ring->size);
1610
1611         rx_ring->next_to_clean = 0;
1612         rx_ring->next_to_use = 0;
1613         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1614
1615         writel(0, adapter->hw.hw_addr + rx_ring->head);
1616         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1617 }
1618
1619 static void e1000e_downshift_workaround(struct work_struct *work)
1620 {
1621         struct e1000_adapter *adapter = container_of(work,
1622                                         struct e1000_adapter, downshift_task);
1623
1624         if (test_bit(__E1000_DOWN, &adapter->state))
1625                 return;
1626
1627         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1628 }
1629
1630 /**
1631  * e1000_intr_msi - Interrupt Handler
1632  * @irq: interrupt number
1633  * @data: pointer to a network interface device structure
1634  **/
1635 static irqreturn_t e1000_intr_msi(int irq, void *data)
1636 {
1637         struct net_device *netdev = data;
1638         struct e1000_adapter *adapter = netdev_priv(netdev);
1639         struct e1000_hw *hw = &adapter->hw;
1640         u32 icr = er32(ICR);
1641
1642         /*
1643          * read ICR disables interrupts using IAM
1644          */
1645
1646         if (icr & E1000_ICR_LSC) {
1647                 hw->mac.get_link_status = 1;
1648                 /*
1649                  * ICH8 workaround-- Call gig speed drop workaround on cable
1650                  * disconnect (LSC) before accessing any PHY registers
1651                  */
1652                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1653                     (!(er32(STATUS) & E1000_STATUS_LU)))
1654                         schedule_work(&adapter->downshift_task);
1655
1656                 /*
1657                  * 80003ES2LAN workaround-- For packet buffer work-around on
1658                  * link down event; disable receives here in the ISR and reset
1659                  * adapter in watchdog
1660                  */
1661                 if (netif_carrier_ok(netdev) &&
1662                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1663                         /* disable receives */
1664                         u32 rctl = er32(RCTL);
1665                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1666                         adapter->flags |= FLAG_RX_RESTART_NOW;
1667                 }
1668                 /* guard against interrupt when we're going down */
1669                 if (!test_bit(__E1000_DOWN, &adapter->state))
1670                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1671         }
1672
1673         if (napi_schedule_prep(&adapter->napi)) {
1674                 adapter->total_tx_bytes = 0;
1675                 adapter->total_tx_packets = 0;
1676                 adapter->total_rx_bytes = 0;
1677                 adapter->total_rx_packets = 0;
1678                 __napi_schedule(&adapter->napi);
1679         }
1680
1681         return IRQ_HANDLED;
1682 }
1683
1684 /**
1685  * e1000_intr - Interrupt Handler
1686  * @irq: interrupt number
1687  * @data: pointer to a network interface device structure
1688  **/
1689 static irqreturn_t e1000_intr(int irq, void *data)
1690 {
1691         struct net_device *netdev = data;
1692         struct e1000_adapter *adapter = netdev_priv(netdev);
1693         struct e1000_hw *hw = &adapter->hw;
1694         u32 rctl, icr = er32(ICR);
1695
1696         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1697                 return IRQ_NONE;  /* Not our interrupt */
1698
1699         /*
1700          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1701          * not set, then the adapter didn't send an interrupt
1702          */
1703         if (!(icr & E1000_ICR_INT_ASSERTED))
1704                 return IRQ_NONE;
1705
1706         /*
1707          * Interrupt Auto-Mask...upon reading ICR,
1708          * interrupts are masked.  No need for the
1709          * IMC write
1710          */
1711
1712         if (icr & E1000_ICR_LSC) {
1713                 hw->mac.get_link_status = 1;
1714                 /*
1715                  * ICH8 workaround-- Call gig speed drop workaround on cable
1716                  * disconnect (LSC) before accessing any PHY registers
1717                  */
1718                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1719                     (!(er32(STATUS) & E1000_STATUS_LU)))
1720                         schedule_work(&adapter->downshift_task);
1721
1722                 /*
1723                  * 80003ES2LAN workaround--
1724                  * For packet buffer work-around on link down event;
1725                  * disable receives here in the ISR and
1726                  * reset adapter in watchdog
1727                  */
1728                 if (netif_carrier_ok(netdev) &&
1729                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1730                         /* disable receives */
1731                         rctl = er32(RCTL);
1732                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1733                         adapter->flags |= FLAG_RX_RESTART_NOW;
1734                 }
1735                 /* guard against interrupt when we're going down */
1736                 if (!test_bit(__E1000_DOWN, &adapter->state))
1737                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1738         }
1739
1740         if (napi_schedule_prep(&adapter->napi)) {
1741                 adapter->total_tx_bytes = 0;
1742                 adapter->total_tx_packets = 0;
1743                 adapter->total_rx_bytes = 0;
1744                 adapter->total_rx_packets = 0;
1745                 __napi_schedule(&adapter->napi);
1746         }
1747
1748         return IRQ_HANDLED;
1749 }
1750
1751 static irqreturn_t e1000_msix_other(int irq, void *data)
1752 {
1753         struct net_device *netdev = data;
1754         struct e1000_adapter *adapter = netdev_priv(netdev);
1755         struct e1000_hw *hw = &adapter->hw;
1756         u32 icr = er32(ICR);
1757
1758         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1759                 if (!test_bit(__E1000_DOWN, &adapter->state))
1760                         ew32(IMS, E1000_IMS_OTHER);
1761                 return IRQ_NONE;
1762         }
1763
1764         if (icr & adapter->eiac_mask)
1765                 ew32(ICS, (icr & adapter->eiac_mask));
1766
1767         if (icr & E1000_ICR_OTHER) {
1768                 if (!(icr & E1000_ICR_LSC))
1769                         goto no_link_interrupt;
1770                 hw->mac.get_link_status = 1;
1771                 /* guard against interrupt when we're going down */
1772                 if (!test_bit(__E1000_DOWN, &adapter->state))
1773                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1774         }
1775
1776 no_link_interrupt:
1777         if (!test_bit(__E1000_DOWN, &adapter->state))
1778                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1779
1780         return IRQ_HANDLED;
1781 }
1782
1783
1784 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1785 {
1786         struct net_device *netdev = data;
1787         struct e1000_adapter *adapter = netdev_priv(netdev);
1788         struct e1000_hw *hw = &adapter->hw;
1789         struct e1000_ring *tx_ring = adapter->tx_ring;
1790
1791
1792         adapter->total_tx_bytes = 0;
1793         adapter->total_tx_packets = 0;
1794
1795         if (!e1000_clean_tx_irq(adapter))
1796                 /* Ring was not completely cleaned, so fire another interrupt */
1797                 ew32(ICS, tx_ring->ims_val);
1798
1799         return IRQ_HANDLED;
1800 }
1801
1802 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1803 {
1804         struct net_device *netdev = data;
1805         struct e1000_adapter *adapter = netdev_priv(netdev);
1806
1807         /* Write the ITR value calculated at the end of the
1808          * previous interrupt.
1809          */
1810         if (adapter->rx_ring->set_itr) {
1811                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1812                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1813                 adapter->rx_ring->set_itr = 0;
1814         }
1815
1816         if (napi_schedule_prep(&adapter->napi)) {
1817                 adapter->total_rx_bytes = 0;
1818                 adapter->total_rx_packets = 0;
1819                 __napi_schedule(&adapter->napi);
1820         }
1821         return IRQ_HANDLED;
1822 }
1823
1824 /**
1825  * e1000_configure_msix - Configure MSI-X hardware
1826  *
1827  * e1000_configure_msix sets up the hardware to properly
1828  * generate MSI-X interrupts.
1829  **/
1830 static void e1000_configure_msix(struct e1000_adapter *adapter)
1831 {
1832         struct e1000_hw *hw = &adapter->hw;
1833         struct e1000_ring *rx_ring = adapter->rx_ring;
1834         struct e1000_ring *tx_ring = adapter->tx_ring;
1835         int vector = 0;
1836         u32 ctrl_ext, ivar = 0;
1837
1838         adapter->eiac_mask = 0;
1839
1840         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1841         if (hw->mac.type == e1000_82574) {
1842                 u32 rfctl = er32(RFCTL);
1843                 rfctl |= E1000_RFCTL_ACK_DIS;
1844                 ew32(RFCTL, rfctl);
1845         }
1846
1847 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1848         /* Configure Rx vector */
1849         rx_ring->ims_val = E1000_IMS_RXQ0;
1850         adapter->eiac_mask |= rx_ring->ims_val;
1851         if (rx_ring->itr_val)
1852                 writel(1000000000 / (rx_ring->itr_val * 256),
1853                        hw->hw_addr + rx_ring->itr_register);
1854         else
1855                 writel(1, hw->hw_addr + rx_ring->itr_register);
1856         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1857
1858         /* Configure Tx vector */
1859         tx_ring->ims_val = E1000_IMS_TXQ0;
1860         vector++;
1861         if (tx_ring->itr_val)
1862                 writel(1000000000 / (tx_ring->itr_val * 256),
1863                        hw->hw_addr + tx_ring->itr_register);
1864         else
1865                 writel(1, hw->hw_addr + tx_ring->itr_register);
1866         adapter->eiac_mask |= tx_ring->ims_val;
1867         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1868
1869         /* set vector for Other Causes, e.g. link changes */
1870         vector++;
1871         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1872         if (rx_ring->itr_val)
1873                 writel(1000000000 / (rx_ring->itr_val * 256),
1874                        hw->hw_addr + E1000_EITR_82574(vector));
1875         else
1876                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1877
1878         /* Cause Tx interrupts on every write back */
1879         ivar |= (1 << 31);
1880
1881         ew32(IVAR, ivar);
1882
1883         /* enable MSI-X PBA support */
1884         ctrl_ext = er32(CTRL_EXT);
1885         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1886
1887         /* Auto-Mask Other interrupts upon ICR read */
1888 #define E1000_EIAC_MASK_82574   0x01F00000
1889         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1890         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1891         ew32(CTRL_EXT, ctrl_ext);
1892         e1e_flush();
1893 }
1894
1895 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1896 {
1897         if (adapter->msix_entries) {
1898                 pci_disable_msix(adapter->pdev);
1899                 kfree(adapter->msix_entries);
1900                 adapter->msix_entries = NULL;
1901         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1902                 pci_disable_msi(adapter->pdev);
1903                 adapter->flags &= ~FLAG_MSI_ENABLED;
1904         }
1905 }
1906
1907 /**
1908  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1909  *
1910  * Attempt to configure interrupts using the best available
1911  * capabilities of the hardware and kernel.
1912  **/
1913 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1914 {
1915         int err;
1916         int i;
1917
1918         switch (adapter->int_mode) {
1919         case E1000E_INT_MODE_MSIX:
1920                 if (adapter->flags & FLAG_HAS_MSIX) {
1921                         adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1922                         adapter->msix_entries = kcalloc(adapter->num_vectors,
1923                                                       sizeof(struct msix_entry),
1924                                                       GFP_KERNEL);
1925                         if (adapter->msix_entries) {
1926                                 for (i = 0; i < adapter->num_vectors; i++)
1927                                         adapter->msix_entries[i].entry = i;
1928
1929                                 err = pci_enable_msix(adapter->pdev,
1930                                                       adapter->msix_entries,
1931                                                       adapter->num_vectors);
1932                                 if (err == 0)
1933                                         return;
1934                         }
1935                         /* MSI-X failed, so fall through and try MSI */
1936                         e_err("Failed to initialize MSI-X interrupts.  "
1937                               "Falling back to MSI interrupts.\n");
1938                         e1000e_reset_interrupt_capability(adapter);
1939                 }
1940                 adapter->int_mode = E1000E_INT_MODE_MSI;
1941                 /* Fall through */
1942         case E1000E_INT_MODE_MSI:
1943                 if (!pci_enable_msi(adapter->pdev)) {
1944                         adapter->flags |= FLAG_MSI_ENABLED;
1945                 } else {
1946                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1947                         e_err("Failed to initialize MSI interrupts.  Falling "
1948                               "back to legacy interrupts.\n");
1949                 }
1950                 /* Fall through */
1951         case E1000E_INT_MODE_LEGACY:
1952                 /* Don't do anything; this is the system default */
1953                 break;
1954         }
1955
1956         /* store the number of vectors being used */
1957         adapter->num_vectors = 1;
1958 }
1959
1960 /**
1961  * e1000_request_msix - Initialize MSI-X interrupts
1962  *
1963  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1964  * kernel.
1965  **/
1966 static int e1000_request_msix(struct e1000_adapter *adapter)
1967 {
1968         struct net_device *netdev = adapter->netdev;
1969         int err = 0, vector = 0;
1970
1971         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1972                 snprintf(adapter->rx_ring->name,
1973                          sizeof(adapter->rx_ring->name) - 1,
1974                          "%s-rx-0", netdev->name);
1975         else
1976                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1977         err = request_irq(adapter->msix_entries[vector].vector,
1978                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1979                           netdev);
1980         if (err)
1981                 goto out;
1982         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1983         adapter->rx_ring->itr_val = adapter->itr;
1984         vector++;
1985
1986         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1987                 snprintf(adapter->tx_ring->name,
1988                          sizeof(adapter->tx_ring->name) - 1,
1989                          "%s-tx-0", netdev->name);
1990         else
1991                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1992         err = request_irq(adapter->msix_entries[vector].vector,
1993                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1994                           netdev);
1995         if (err)
1996                 goto out;
1997         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1998         adapter->tx_ring->itr_val = adapter->itr;
1999         vector++;
2000
2001         err = request_irq(adapter->msix_entries[vector].vector,
2002                           e1000_msix_other, 0, netdev->name, netdev);
2003         if (err)
2004                 goto out;
2005
2006         e1000_configure_msix(adapter);
2007         return 0;
2008 out:
2009         return err;
2010 }
2011
2012 /**
2013  * e1000_request_irq - initialize interrupts
2014  *
2015  * Attempts to configure interrupts using the best available
2016  * capabilities of the hardware and kernel.
2017  **/
2018 static int e1000_request_irq(struct e1000_adapter *adapter)
2019 {
2020         struct net_device *netdev = adapter->netdev;
2021         int err;
2022
2023         if (adapter->msix_entries) {
2024                 err = e1000_request_msix(adapter);
2025                 if (!err)
2026                         return err;
2027                 /* fall back to MSI */
2028                 e1000e_reset_interrupt_capability(adapter);
2029                 adapter->int_mode = E1000E_INT_MODE_MSI;
2030                 e1000e_set_interrupt_capability(adapter);
2031         }
2032         if (adapter->flags & FLAG_MSI_ENABLED) {
2033                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
2034                                   netdev->name, netdev);
2035                 if (!err)
2036                         return err;
2037
2038                 /* fall back to legacy interrupt */
2039                 e1000e_reset_interrupt_capability(adapter);
2040                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2041         }
2042
2043         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
2044                           netdev->name, netdev);
2045         if (err)
2046                 e_err("Unable to allocate interrupt, Error: %d\n", err);
2047
2048         return err;
2049 }
2050
2051 static void e1000_free_irq(struct e1000_adapter *adapter)
2052 {
2053         struct net_device *netdev = adapter->netdev;
2054
2055         if (adapter->msix_entries) {
2056                 int vector = 0;
2057
2058                 free_irq(adapter->msix_entries[vector].vector, netdev);
2059                 vector++;
2060
2061                 free_irq(adapter->msix_entries[vector].vector, netdev);
2062                 vector++;
2063
2064                 /* Other Causes interrupt vector */
2065                 free_irq(adapter->msix_entries[vector].vector, netdev);
2066                 return;
2067         }
2068
2069         free_irq(adapter->pdev->irq, netdev);
2070 }
2071
2072 /**
2073  * e1000_irq_disable - Mask off interrupt generation on the NIC
2074  **/
2075 static void e1000_irq_disable(struct e1000_adapter *adapter)
2076 {
2077         struct e1000_hw *hw = &adapter->hw;
2078
2079         ew32(IMC, ~0);
2080         if (adapter->msix_entries)
2081                 ew32(EIAC_82574, 0);
2082         e1e_flush();
2083
2084         if (adapter->msix_entries) {
2085                 int i;
2086                 for (i = 0; i < adapter->num_vectors; i++)
2087                         synchronize_irq(adapter->msix_entries[i].vector);
2088         } else {
2089                 synchronize_irq(adapter->pdev->irq);
2090         }
2091 }
2092
2093 /**
2094  * e1000_irq_enable - Enable default interrupt generation settings
2095  **/
2096 static void e1000_irq_enable(struct e1000_adapter *adapter)
2097 {
2098         struct e1000_hw *hw = &adapter->hw;
2099
2100         if (adapter->msix_entries) {
2101                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
2102                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
2103         } else {
2104                 ew32(IMS, IMS_ENABLE_MASK);
2105         }
2106         e1e_flush();
2107 }
2108
2109 /**
2110  * e1000e_get_hw_control - get control of the h/w from f/w
2111  * @adapter: address of board private structure
2112  *
2113  * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2114  * For ASF and Pass Through versions of f/w this means that
2115  * the driver is loaded. For AMT version (only with 82573)
2116  * of the f/w this means that the network i/f is open.
2117  **/
2118 void e1000e_get_hw_control(struct e1000_adapter *adapter)
2119 {
2120         struct e1000_hw *hw = &adapter->hw;
2121         u32 ctrl_ext;
2122         u32 swsm;
2123
2124         /* Let firmware know the driver has taken over */
2125         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2126                 swsm = er32(SWSM);
2127                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2128         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2129                 ctrl_ext = er32(CTRL_EXT);
2130                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2131         }
2132 }
2133
2134 /**
2135  * e1000e_release_hw_control - release control of the h/w to f/w
2136  * @adapter: address of board private structure
2137  *
2138  * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2139  * For ASF and Pass Through versions of f/w this means that the
2140  * driver is no longer loaded. For AMT version (only with 82573) i
2141  * of the f/w this means that the network i/f is closed.
2142  *
2143  **/
2144 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2145 {
2146         struct e1000_hw *hw = &adapter->hw;
2147         u32 ctrl_ext;
2148         u32 swsm;
2149
2150         /* Let firmware taken over control of h/w */
2151         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2152                 swsm = er32(SWSM);
2153                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2154         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2155                 ctrl_ext = er32(CTRL_EXT);
2156                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2157         }
2158 }
2159
2160 /**
2161  * @e1000_alloc_ring - allocate memory for a ring structure
2162  **/
2163 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2164                                 struct e1000_ring *ring)
2165 {
2166         struct pci_dev *pdev = adapter->pdev;
2167
2168         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2169                                         GFP_KERNEL);
2170         if (!ring->desc)
2171                 return -ENOMEM;
2172
2173         return 0;
2174 }
2175
2176 /**
2177  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2178  * @adapter: board private structure
2179  *
2180  * Return 0 on success, negative on failure
2181  **/
2182 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
2183 {
2184         struct e1000_ring *tx_ring = adapter->tx_ring;
2185         int err = -ENOMEM, size;
2186
2187         size = sizeof(struct e1000_buffer) * tx_ring->count;
2188         tx_ring->buffer_info = vzalloc(size);
2189         if (!tx_ring->buffer_info)
2190                 goto err;
2191
2192         /* round up to nearest 4K */
2193         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2194         tx_ring->size = ALIGN(tx_ring->size, 4096);
2195
2196         err = e1000_alloc_ring_dma(adapter, tx_ring);
2197         if (err)
2198                 goto err;
2199
2200         tx_ring->next_to_use = 0;
2201         tx_ring->next_to_clean = 0;
2202
2203         return 0;
2204 err:
2205         vfree(tx_ring->buffer_info);
2206         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2207         return err;
2208 }
2209
2210 /**
2211  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2212  * @adapter: board private structure
2213  *
2214  * Returns 0 on success, negative on failure
2215  **/
2216 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
2217 {
2218         struct e1000_ring *rx_ring = adapter->rx_ring;
2219         struct e1000_buffer *buffer_info;
2220         int i, size, desc_len, err = -ENOMEM;
2221
2222         size = sizeof(struct e1000_buffer) * rx_ring->count;
2223         rx_ring->buffer_info = vzalloc(size);
2224         if (!rx_ring->buffer_info)
2225                 goto err;
2226
2227         for (i = 0; i < rx_ring->count; i++) {
2228                 buffer_info = &rx_ring->buffer_info[i];
2229                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2230                                                 sizeof(struct e1000_ps_page),
2231                                                 GFP_KERNEL);
2232                 if (!buffer_info->ps_pages)
2233                         goto err_pages;
2234         }
2235
2236         desc_len = sizeof(union e1000_rx_desc_packet_split);
2237
2238         /* Round up to nearest 4K */
2239         rx_ring->size = rx_ring->count * desc_len;
2240         rx_ring->size = ALIGN(rx_ring->size, 4096);
2241
2242         err = e1000_alloc_ring_dma(adapter, rx_ring);
2243         if (err)
2244                 goto err_pages;
2245
2246         rx_ring->next_to_clean = 0;
2247         rx_ring->next_to_use = 0;
2248         rx_ring->rx_skb_top = NULL;
2249
2250         return 0;
2251
2252 err_pages:
2253         for (i = 0; i < rx_ring->count; i++) {
2254                 buffer_info = &rx_ring->buffer_info[i];
2255                 kfree(buffer_info->ps_pages);
2256         }
2257 err:
2258         vfree(rx_ring->buffer_info);
2259         e_err("Unable to allocate memory for the receive descriptor ring\n");
2260         return err;
2261 }
2262
2263 /**
2264  * e1000_clean_tx_ring - Free Tx Buffers
2265  * @adapter: board private structure
2266  **/
2267 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
2268 {
2269         struct e1000_ring *tx_ring = adapter->tx_ring;
2270         struct e1000_buffer *buffer_info;
2271         unsigned long size;
2272         unsigned int i;
2273
2274         for (i = 0; i < tx_ring->count; i++) {
2275                 buffer_info = &tx_ring->buffer_info[i];
2276                 e1000_put_txbuf(adapter, buffer_info);
2277         }
2278
2279         size = sizeof(struct e1000_buffer) * tx_ring->count;
2280         memset(tx_ring->buffer_info, 0, size);
2281
2282         memset(tx_ring->desc, 0, tx_ring->size);
2283
2284         tx_ring->next_to_use = 0;
2285         tx_ring->next_to_clean = 0;
2286
2287         writel(0, adapter->hw.hw_addr + tx_ring->head);
2288         writel(0, adapter->hw.hw_addr + tx_ring->tail);
2289 }
2290
2291 /**
2292  * e1000e_free_tx_resources - Free Tx Resources per Queue
2293  * @adapter: board private structure
2294  *
2295  * Free all transmit software resources
2296  **/
2297 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
2298 {
2299         struct pci_dev *pdev = adapter->pdev;
2300         struct e1000_ring *tx_ring = adapter->tx_ring;
2301
2302         e1000_clean_tx_ring(adapter);
2303
2304         vfree(tx_ring->buffer_info);
2305         tx_ring->buffer_info = NULL;
2306
2307         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2308                           tx_ring->dma);
2309         tx_ring->desc = NULL;
2310 }
2311
2312 /**
2313  * e1000e_free_rx_resources - Free Rx Resources
2314  * @adapter: board private structure
2315  *
2316  * Free all receive software resources
2317  **/
2318
2319 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
2320 {
2321         struct pci_dev *pdev = adapter->pdev;
2322         struct e1000_ring *rx_ring = adapter->rx_ring;
2323         int i;
2324
2325         e1000_clean_rx_ring(adapter);
2326
2327         for (i = 0; i < rx_ring->count; i++)
2328                 kfree(rx_ring->buffer_info[i].ps_pages);
2329
2330         vfree(rx_ring->buffer_info);
2331         rx_ring->buffer_info = NULL;
2332
2333         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2334                           rx_ring->dma);
2335         rx_ring->desc = NULL;
2336 }
2337
2338 /**
2339  * e1000_update_itr - update the dynamic ITR value based on statistics
2340  * @adapter: pointer to adapter
2341  * @itr_setting: current adapter->itr
2342  * @packets: the number of packets during this measurement interval
2343  * @bytes: the number of bytes during this measurement interval
2344  *
2345  *      Stores a new ITR value based on packets and byte
2346  *      counts during the last interrupt.  The advantage of per interrupt
2347  *      computation is faster updates and more accurate ITR for the current
2348  *      traffic pattern.  Constants in this function were computed
2349  *      based on theoretical maximum wire speed and thresholds were set based
2350  *      on testing data as well as attempting to minimize response time
2351  *      while increasing bulk throughput.  This functionality is controlled
2352  *      by the InterruptThrottleRate module parameter.
2353  **/
2354 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2355                                      u16 itr_setting, int packets,
2356                                      int bytes)
2357 {
2358         unsigned int retval = itr_setting;
2359
2360         if (packets == 0)
2361                 goto update_itr_done;
2362
2363         switch (itr_setting) {
2364         case lowest_latency:
2365                 /* handle TSO and jumbo frames */
2366                 if (bytes/packets > 8000)
2367                         retval = bulk_latency;
2368                 else if ((packets < 5) && (bytes > 512))
2369                         retval = low_latency;
2370                 break;
2371         case low_latency:  /* 50 usec aka 20000 ints/s */
2372                 if (bytes > 10000) {
2373                         /* this if handles the TSO accounting */
2374                         if (bytes/packets > 8000)
2375                                 retval = bulk_latency;
2376                         else if ((packets < 10) || ((bytes/packets) > 1200))
2377                                 retval = bulk_latency;
2378                         else if ((packets > 35))
2379                                 retval = lowest_latency;
2380                 } else if (bytes/packets > 2000) {
2381                         retval = bulk_latency;
2382                 } else if (packets <= 2 && bytes < 512) {
2383                         retval = lowest_latency;
2384                 }
2385                 break;
2386         case bulk_latency: /* 250 usec aka 4000 ints/s */
2387                 if (bytes > 25000) {
2388                         if (packets > 35)
2389                                 retval = low_latency;
2390                 } else if (bytes < 6000) {
2391                         retval = low_latency;
2392                 }
2393                 break;
2394         }
2395
2396 update_itr_done:
2397         return retval;
2398 }
2399
2400 static void e1000_set_itr(struct e1000_adapter *adapter)
2401 {
2402         struct e1000_hw *hw = &adapter->hw;
2403         u16 current_itr;
2404         u32 new_itr = adapter->itr;
2405
2406         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2407         if (adapter->link_speed != SPEED_1000) {
2408                 current_itr = 0;
2409                 new_itr = 4000;
2410                 goto set_itr_now;
2411         }
2412
2413         if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2414                 new_itr = 0;
2415                 goto set_itr_now;
2416         }
2417
2418         adapter->tx_itr = e1000_update_itr(adapter,
2419                                     adapter->tx_itr,
2420                                     adapter->total_tx_packets,
2421                                     adapter->total_tx_bytes);
2422         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2423         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2424                 adapter->tx_itr = low_latency;
2425
2426         adapter->rx_itr = e1000_update_itr(adapter,
2427                                     adapter->rx_itr,
2428                                     adapter->total_rx_packets,
2429                                     adapter->total_rx_bytes);
2430         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2431         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2432                 adapter->rx_itr = low_latency;
2433
2434         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2435
2436         switch (current_itr) {
2437         /* counts and packets in update_itr are dependent on these numbers */
2438         case lowest_latency:
2439                 new_itr = 70000;
2440                 break;
2441         case low_latency:
2442                 new_itr = 20000; /* aka hwitr = ~200 */
2443                 break;
2444         case bulk_latency:
2445                 new_itr = 4000;
2446                 break;
2447         default:
2448                 break;
2449         }
2450
2451 set_itr_now:
2452         if (new_itr != adapter->itr) {
2453                 /*
2454                  * this attempts to bias the interrupt rate towards Bulk
2455                  * by adding intermediate steps when interrupt rate is
2456                  * increasing
2457                  */
2458                 new_itr = new_itr > adapter->itr ?
2459                              min(adapter->itr + (new_itr >> 2), new_itr) :
2460                              new_itr;
2461                 adapter->itr = new_itr;
2462                 adapter->rx_ring->itr_val = new_itr;
2463                 if (adapter->msix_entries)
2464                         adapter->rx_ring->set_itr = 1;
2465                 else
2466                         if (new_itr)
2467                                 ew32(ITR, 1000000000 / (new_itr * 256));
2468                         else
2469                                 ew32(ITR, 0);
2470         }
2471 }
2472
2473 /**
2474  * e1000_alloc_queues - Allocate memory for all rings
2475  * @adapter: board private structure to initialize
2476  **/
2477 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2478 {
2479         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2480         if (!adapter->tx_ring)
2481                 goto err;
2482
2483         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2484         if (!adapter->rx_ring)
2485                 goto err;
2486
2487         return 0;
2488 err:
2489         e_err("Unable to allocate memory for queues\n");
2490         kfree(adapter->rx_ring);
2491         kfree(adapter->tx_ring);
2492         return -ENOMEM;
2493 }
2494
2495 /**
2496  * e1000_clean - NAPI Rx polling callback
2497  * @napi: struct associated with this polling callback
2498  * @budget: amount of packets driver is allowed to process this poll
2499  **/
2500 static int e1000_clean(struct napi_struct *napi, int budget)
2501 {
2502         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2503         struct e1000_hw *hw = &adapter->hw;
2504         struct net_device *poll_dev = adapter->netdev;
2505         int tx_cleaned = 1, work_done = 0;
2506
2507         adapter = netdev_priv(poll_dev);
2508
2509         if (adapter->msix_entries &&
2510             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2511                 goto clean_rx;
2512
2513         tx_cleaned = e1000_clean_tx_irq(adapter);
2514
2515 clean_rx:
2516         adapter->clean_rx(adapter, &work_done, budget);
2517
2518         if (!tx_cleaned)
2519                 work_done = budget;
2520
2521         /* If budget not fully consumed, exit the polling mode */
2522         if (work_done < budget) {
2523                 if (adapter->itr_setting & 3)
2524                         e1000_set_itr(adapter);
2525                 napi_complete(napi);
2526                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2527                         if (adapter->msix_entries)
2528                                 ew32(IMS, adapter->rx_ring->ims_val);
2529                         else
2530                                 e1000_irq_enable(adapter);
2531                 }
2532         }
2533
2534         return work_done;
2535 }
2536
2537 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2538 {
2539         struct e1000_adapter *adapter = netdev_priv(netdev);
2540         struct e1000_hw *hw = &adapter->hw;
2541         u32 vfta, index;
2542
2543         /* don't update vlan cookie if already programmed */
2544         if ((adapter->hw.mng_cookie.status &
2545              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2546             (vid == adapter->mng_vlan_id))
2547                 return;
2548
2549         /* add VID to filter table */
2550         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2551                 index = (vid >> 5) & 0x7F;
2552                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2553                 vfta |= (1 << (vid & 0x1F));
2554                 hw->mac.ops.write_vfta(hw, index, vfta);
2555         }
2556
2557         set_bit(vid, adapter->active_vlans);
2558 }
2559
2560 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2561 {
2562         struct e1000_adapter *adapter = netdev_priv(netdev);
2563         struct e1000_hw *hw = &adapter->hw;
2564         u32 vfta, index;
2565
2566         if ((adapter->hw.mng_cookie.status &
2567              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2568             (vid == adapter->mng_vlan_id)) {
2569                 /* release control to f/w */
2570                 e1000e_release_hw_control(adapter);
2571                 return;
2572         }
2573
2574         /* remove VID from filter table */
2575         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2576                 index = (vid >> 5) & 0x7F;
2577                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2578                 vfta &= ~(1 << (vid & 0x1F));
2579                 hw->mac.ops.write_vfta(hw, index, vfta);
2580         }
2581
2582         clear_bit(vid, adapter->active_vlans);
2583 }
2584
2585 /**
2586  * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2587  * @adapter: board private structure to initialize
2588  **/
2589 static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2590 {
2591         struct net_device *netdev = adapter->netdev;
2592         struct e1000_hw *hw = &adapter->hw;
2593         u32 rctl;
2594
2595         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2596                 /* disable VLAN receive filtering */
2597                 rctl = er32(RCTL);
2598                 rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2599                 ew32(RCTL, rctl);
2600
2601                 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2602                         e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2603                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2604                 }
2605         }
2606 }
2607
2608 /**
2609  * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2610  * @adapter: board private structure to initialize
2611  **/
2612 static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2613 {
2614         struct e1000_hw *hw = &adapter->hw;
2615         u32 rctl;
2616
2617         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2618                 /* enable VLAN receive filtering */
2619                 rctl = er32(RCTL);
2620                 rctl |= E1000_RCTL_VFE;
2621                 rctl &= ~E1000_RCTL_CFIEN;
2622                 ew32(RCTL, rctl);
2623         }
2624 }
2625
2626 /**
2627  * e1000e_vlan_strip_enable - helper to disable HW VLAN stripping
2628  * @adapter: board private structure to initialize
2629  **/
2630 static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2631 {
2632         struct e1000_hw *hw = &adapter->hw;
2633         u32 ctrl;
2634
2635         /* disable VLAN tag insert/strip */
2636         ctrl = er32(CTRL);
2637         ctrl &= ~E1000_CTRL_VME;
2638         ew32(CTRL, ctrl);
2639 }
2640
2641 /**
2642  * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2643  * @adapter: board private structure to initialize
2644  **/
2645 static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2646 {
2647         struct e1000_hw *hw = &adapter->hw;
2648         u32 ctrl;
2649
2650         /* enable VLAN tag insert/strip */
2651         ctrl = er32(CTRL);
2652         ctrl |= E1000_CTRL_VME;
2653         ew32(CTRL, ctrl);
2654 }
2655
2656 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2657 {
2658         struct net_device *netdev = adapter->netdev;
2659         u16 vid = adapter->hw.mng_cookie.vlan_id;
2660         u16 old_vid = adapter->mng_vlan_id;
2661
2662         if (adapter->hw.mng_cookie.status &
2663             E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2664                 e1000_vlan_rx_add_vid(netdev, vid);
2665                 adapter->mng_vlan_id = vid;
2666         }
2667
2668         if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2669                 e1000_vlan_rx_kill_vid(netdev, old_vid);
2670 }
2671
2672 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2673 {
2674         u16 vid;
2675
2676         e1000_vlan_rx_add_vid(adapter->netdev, 0);
2677
2678         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2679                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2680 }
2681
2682 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2683 {
2684         struct e1000_hw *hw = &adapter->hw;
2685         u32 manc, manc2h, mdef, i, j;
2686
2687         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2688                 return;
2689
2690         manc = er32(MANC);
2691
2692         /*
2693          * enable receiving management packets to the host. this will probably
2694          * generate destination unreachable messages from the host OS, but
2695          * the packets will be handled on SMBUS
2696          */
2697         manc |= E1000_MANC_EN_MNG2HOST;
2698         manc2h = er32(MANC2H);
2699
2700         switch (hw->mac.type) {
2701         default:
2702                 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2703                 break;
2704         case e1000_82574:
2705         case e1000_82583:
2706                 /*
2707                  * Check if IPMI pass-through decision filter already exists;
2708                  * if so, enable it.
2709                  */
2710                 for (i = 0, j = 0; i < 8; i++) {
2711                         mdef = er32(MDEF(i));
2712
2713                         /* Ignore filters with anything other than IPMI ports */
2714                         if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2715                                 continue;
2716
2717                         /* Enable this decision filter in MANC2H */
2718                         if (mdef)
2719                                 manc2h |= (1 << i);
2720
2721                         j |= mdef;
2722                 }
2723
2724                 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2725                         break;
2726
2727                 /* Create new decision filter in an empty filter */
2728                 for (i = 0, j = 0; i < 8; i++)
2729                         if (er32(MDEF(i)) == 0) {
2730                                 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2731                                                E1000_MDEF_PORT_664));
2732                                 manc2h |= (1 << 1);
2733                                 j++;
2734                                 break;
2735                         }
2736
2737                 if (!j)
2738                         e_warn("Unable to create IPMI pass-through filter\n");
2739                 break;
2740         }
2741
2742         ew32(MANC2H, manc2h);
2743         ew32(MANC, manc);
2744 }
2745
2746 /**
2747  * e1000_configure_tx - Configure Transmit Unit after Reset
2748  * @adapter: board private structure
2749  *
2750  * Configure the Tx unit of the MAC after a reset.
2751  **/
2752 static void e1000_configure_tx(struct e1000_adapter *adapter)
2753 {
2754         struct e1000_hw *hw = &adapter->hw;
2755         struct e1000_ring *tx_ring = adapter->tx_ring;
2756         u64 tdba;
2757         u32 tdlen, tctl, tipg, tarc;
2758         u32 ipgr1, ipgr2;
2759
2760         /* Setup the HW Tx Head and Tail descriptor pointers */
2761         tdba = tx_ring->dma;
2762         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2763         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2764         ew32(TDBAH, (tdba >> 32));
2765         ew32(TDLEN, tdlen);
2766         ew32(TDH, 0);
2767         ew32(TDT, 0);
2768         tx_ring->head = E1000_TDH;
2769         tx_ring->tail = E1000_TDT;
2770
2771         /* Set the default values for the Tx Inter Packet Gap timer */
2772         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2773         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2774         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2775
2776         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2777                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2778
2779         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2780         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2781         ew32(TIPG, tipg);
2782
2783         /* Set the Tx Interrupt Delay register */
2784         ew32(TIDV, adapter->tx_int_delay);
2785         /* Tx irq moderation */
2786         ew32(TADV, adapter->tx_abs_int_delay);
2787
2788         if (adapter->flags2 & FLAG2_DMA_BURST) {
2789                 u32 txdctl = er32(TXDCTL(0));
2790                 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2791                             E1000_TXDCTL_WTHRESH);
2792                 /*
2793                  * set up some performance related parameters to encourage the
2794                  * hardware to use the bus more efficiently in bursts, depends
2795                  * on the tx_int_delay to be enabled,
2796                  * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
2797                  * hthresh = 1 ==> prefetch when one or more available
2798                  * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2799                  * BEWARE: this seems to work but should be considered first if
2800                  * there are Tx hangs or other Tx related bugs
2801                  */
2802                 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2803                 ew32(TXDCTL(0), txdctl);
2804                 /* erratum work around: set txdctl the same for both queues */
2805                 ew32(TXDCTL(1), txdctl);
2806         }
2807
2808         /* Program the Transmit Control Register */
2809         tctl = er32(TCTL);
2810         tctl &= ~E1000_TCTL_CT;
2811         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2812                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2813
2814         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2815                 tarc = er32(TARC(0));
2816                 /*
2817                  * set the speed mode bit, we'll clear it if we're not at
2818                  * gigabit link later
2819                  */
2820 #define SPEED_MODE_BIT (1 << 21)
2821                 tarc |= SPEED_MODE_BIT;
2822                 ew32(TARC(0), tarc);
2823         }
2824
2825         /* errata: program both queues to unweighted RR */
2826         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2827                 tarc = er32(TARC(0));
2828                 tarc |= 1;
2829                 ew32(TARC(0), tarc);
2830                 tarc = er32(TARC(1));
2831                 tarc |= 1;
2832                 ew32(TARC(1), tarc);
2833         }
2834
2835         /* Setup Transmit Descriptor Settings for eop descriptor */
2836         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2837
2838         /* only set IDE if we are delaying interrupts using the timers */
2839         if (adapter->tx_int_delay)
2840                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2841
2842         /* enable Report Status bit */
2843         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2844
2845         ew32(TCTL, tctl);
2846
2847         e1000e_config_collision_dist(hw);
2848 }
2849
2850 /**
2851  * e1000_setup_rctl - configure the receive control registers
2852  * @adapter: Board private structure
2853  **/
2854 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2855                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2856 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2857 {
2858         struct e1000_hw *hw = &adapter->hw;
2859         u32 rctl, rfctl;
2860         u32 pages = 0;
2861
2862         /* Workaround Si errata on 82579 - configure jumbo frame flow */
2863         if (hw->mac.type == e1000_pch2lan) {
2864                 s32 ret_val;
2865
2866                 if (adapter->netdev->mtu > ETH_DATA_LEN)
2867                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2868                 else
2869                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2870
2871                 if (ret_val)
2872                         e_dbg("failed to enable jumbo frame workaround mode\n");
2873         }
2874
2875         /* Program MC offset vector base */
2876         rctl = er32(RCTL);
2877         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2878         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2879                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2880                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2881
2882         /* Do not Store bad packets */
2883         rctl &= ~E1000_RCTL_SBP;
2884
2885         /* Enable Long Packet receive */
2886         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2887                 rctl &= ~E1000_RCTL_LPE;
2888         else
2889                 rctl |= E1000_RCTL_LPE;
2890
2891         /* Some systems expect that the CRC is included in SMBUS traffic. The
2892          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2893          * host memory when this is enabled
2894          */
2895         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2896                 rctl |= E1000_RCTL_SECRC;
2897
2898         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2899         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2900                 u16 phy_data;
2901
2902                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2903                 phy_data &= 0xfff8;
2904                 phy_data |= (1 << 2);
2905                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2906
2907                 e1e_rphy(hw, 22, &phy_data);
2908                 phy_data &= 0x0fff;
2909                 phy_data |= (1 << 14);
2910                 e1e_wphy(hw, 0x10, 0x2823);
2911                 e1e_wphy(hw, 0x11, 0x0003);
2912                 e1e_wphy(hw, 22, phy_data);
2913         }
2914
2915         /* Setup buffer sizes */
2916         rctl &= ~E1000_RCTL_SZ_4096;
2917         rctl |= E1000_RCTL_BSEX;
2918         switch (adapter->rx_buffer_len) {
2919         case 2048:
2920         default:
2921                 rctl |= E1000_RCTL_SZ_2048;
2922                 rctl &= ~E1000_RCTL_BSEX;
2923                 break;
2924         case 4096:
2925                 rctl |= E1000_RCTL_SZ_4096;
2926                 break;
2927         case 8192:
2928                 rctl |= E1000_RCTL_SZ_8192;
2929                 break;
2930         case 16384:
2931                 rctl |= E1000_RCTL_SZ_16384;
2932                 break;
2933         }
2934
2935         /* Enable Extended Status in all Receive Descriptors */
2936         rfctl = er32(RFCTL);
2937         rfctl |= E1000_RFCTL_EXTEN;
2938
2939         /*
2940          * 82571 and greater support packet-split where the protocol
2941          * header is placed in skb->data and the packet data is
2942          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2943          * In the case of a non-split, skb->data is linearly filled,
2944          * followed by the page buffers.  Therefore, skb->data is
2945          * sized to hold the largest protocol header.
2946          *
2947          * allocations using alloc_page take too long for regular MTU
2948          * so only enable packet split for jumbo frames
2949          *
2950          * Using pages when the page size is greater than 16k wastes
2951          * a lot of memory, since we allocate 3 pages at all times
2952          * per packet.
2953          */
2954         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2955         if (!(adapter->flags & FLAG_HAS_ERT) && (pages <= 3) &&
2956             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2957                 adapter->rx_ps_pages = pages;
2958         else
2959                 adapter->rx_ps_pages = 0;
2960
2961         if (adapter->rx_ps_pages) {
2962                 u32 psrctl = 0;
2963
2964                 /*
2965                  * disable packet split support for IPv6 extension headers,
2966                  * because some malformed IPv6 headers can hang the Rx
2967                  */
2968                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2969                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2970
2971                 /* Enable Packet split descriptors */
2972                 rctl |= E1000_RCTL_DTYP_PS;
2973
2974                 psrctl |= adapter->rx_ps_bsize0 >>
2975                         E1000_PSRCTL_BSIZE0_SHIFT;
2976
2977                 switch (adapter->rx_ps_pages) {
2978                 case 3:
2979                         psrctl |= PAGE_SIZE <<
2980                                 E1000_PSRCTL_BSIZE3_SHIFT;
2981                 case 2:
2982                         psrctl |= PAGE_SIZE <<
2983                                 E1000_PSRCTL_BSIZE2_SHIFT;
2984                 case 1:
2985                         psrctl |= PAGE_SIZE >>
2986                                 E1000_PSRCTL_BSIZE1_SHIFT;
2987                         break;
2988                 }
2989
2990                 ew32(PSRCTL, psrctl);
2991         }
2992
2993         ew32(RFCTL, rfctl);
2994         ew32(RCTL, rctl);
2995         /* just started the receive unit, no need to restart */
2996         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2997 }
2998
2999 /**
3000  * e1000_configure_rx - Configure Receive Unit after Reset
3001  * @adapter: board private structure
3002  *
3003  * Configure the Rx unit of the MAC after a reset.
3004  **/
3005 static void e1000_configure_rx(struct e1000_adapter *adapter)
3006 {
3007         struct e1000_hw *hw = &adapter->hw;
3008         struct e1000_ring *rx_ring = adapter->rx_ring;
3009         u64 rdba;
3010         u32 rdlen, rctl, rxcsum, ctrl_ext;
3011
3012         if (adapter->rx_ps_pages) {
3013                 /* this is a 32 byte descriptor */
3014                 rdlen = rx_ring->count *
3015                     sizeof(union e1000_rx_desc_packet_split);
3016                 adapter->clean_rx = e1000_clean_rx_irq_ps;
3017                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
3018         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
3019                 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3020                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
3021                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
3022         } else {
3023                 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3024                 adapter->clean_rx = e1000_clean_rx_irq;
3025                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
3026         }
3027
3028         /* disable receives while setting up the descriptors */
3029         rctl = er32(RCTL);
3030         if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3031                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3032         e1e_flush();
3033         usleep_range(10000, 20000);
3034
3035         if (adapter->flags2 & FLAG2_DMA_BURST) {
3036                 /*
3037                  * set the writeback threshold (only takes effect if the RDTR
3038                  * is set). set GRAN=1 and write back up to 0x4 worth, and
3039                  * enable prefetching of 0x20 Rx descriptors
3040                  * granularity = 01
3041                  * wthresh = 04,
3042                  * hthresh = 04,
3043                  * pthresh = 0x20
3044                  */
3045                 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
3046                 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
3047
3048                 /*
3049                  * override the delay timers for enabling bursting, only if
3050                  * the value was not set by the user via module options
3051                  */
3052                 if (adapter->rx_int_delay == DEFAULT_RDTR)
3053                         adapter->rx_int_delay = BURST_RDTR;
3054                 if (adapter->rx_abs_int_delay == DEFAULT_RADV)
3055                         adapter->rx_abs_int_delay = BURST_RADV;
3056         }
3057
3058         /* set the Receive Delay Timer Register */
3059         ew32(RDTR, adapter->rx_int_delay);
3060
3061         /* irq moderation */
3062         ew32(RADV, adapter->rx_abs_int_delay);
3063         if ((adapter->itr_setting != 0) && (adapter->itr != 0))
3064                 ew32(ITR, 1000000000 / (adapter->itr * 256));
3065
3066         ctrl_ext = er32(CTRL_EXT);
3067         /* Auto-Mask interrupts upon ICR access */
3068         ctrl_ext |= E1000_CTRL_EXT_IAME;
3069         ew32(IAM, 0xffffffff);
3070         ew32(CTRL_EXT, ctrl_ext);
3071         e1e_flush();
3072
3073         /*
3074          * Setup the HW Rx Head and Tail Descriptor Pointers and
3075          * the Base and Length of the Rx Descriptor Ring
3076          */
3077         rdba = rx_ring->dma;
3078         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
3079         ew32(RDBAH, (rdba >> 32));
3080         ew32(RDLEN, rdlen);
3081         ew32(RDH, 0);
3082         ew32(RDT, 0);
3083         rx_ring->head = E1000_RDH;
3084         rx_ring->tail = E1000_RDT;
3085
3086         /* Enable Receive Checksum Offload for TCP and UDP */
3087         rxcsum = er32(RXCSUM);
3088         if (adapter->netdev->features & NETIF_F_RXCSUM) {
3089                 rxcsum |= E1000_RXCSUM_TUOFL;
3090
3091                 /*
3092                  * IPv4 payload checksum for UDP fragments must be
3093                  * used in conjunction with packet-split.
3094                  */
3095                 if (adapter->rx_ps_pages)
3096                         rxcsum |= E1000_RXCSUM_IPPCSE;
3097         } else {
3098                 rxcsum &= ~E1000_RXCSUM_TUOFL;
3099                 /* no need to clear IPPCSE as it defaults to 0 */
3100         }
3101         ew32(RXCSUM, rxcsum);
3102
3103         /*
3104          * Enable early receives on supported devices, only takes effect when
3105          * packet size is equal or larger than the specified value (in 8 byte
3106          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
3107          */
3108         if ((adapter->flags & FLAG_HAS_ERT) ||
3109             (adapter->hw.mac.type == e1000_pch2lan)) {
3110                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3111                         u32 rxdctl = er32(RXDCTL(0));
3112                         ew32(RXDCTL(0), rxdctl | 0x3);
3113                         if (adapter->flags & FLAG_HAS_ERT)
3114                                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
3115                         /*
3116                          * With jumbo frames and early-receive enabled,
3117                          * excessive C-state transition latencies result in
3118                          * dropped transactions.
3119                          */
3120                         pm_qos_update_request(&adapter->netdev->pm_qos_req, 55);
3121                 } else {
3122                         pm_qos_update_request(&adapter->netdev->pm_qos_req,
3123                                               PM_QOS_DEFAULT_VALUE);
3124                 }
3125         }
3126
3127         /* Enable Receives */
3128         ew32(RCTL, rctl);
3129 }
3130
3131 /**
3132  *  e1000_update_mc_addr_list - Update Multicast addresses
3133  *  @hw: pointer to the HW structure
3134  *  @mc_addr_list: array of multicast addresses to program
3135  *  @mc_addr_count: number of multicast addresses to program
3136  *
3137  *  Updates the Multicast Table Array.
3138  *  The caller must have a packed mc_addr_list of multicast addresses.
3139  **/
3140 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
3141                                       u32 mc_addr_count)
3142 {
3143         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
3144 }
3145
3146 /**
3147  * e1000_set_multi - Multicast and Promiscuous mode set
3148  * @netdev: network interface device structure
3149  *
3150  * The set_multi entry point is called whenever the multicast address
3151  * list or the network interface flags are updated.  This routine is
3152  * responsible for configuring the hardware for proper multicast,
3153  * promiscuous mode, and all-multi behavior.
3154  **/
3155 static void e1000_set_multi(struct net_device *netdev)
3156 {
3157         struct e1000_adapter *adapter = netdev_priv(netdev);
3158         struct e1000_hw *hw = &adapter->hw;
3159         struct netdev_hw_addr *ha;
3160         u8  *mta_list;
3161         u32 rctl;
3162
3163         /* Check for Promiscuous and All Multicast modes */
3164
3165         rctl = er32(RCTL);
3166
3167         if (netdev->flags & IFF_PROMISC) {
3168                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3169                 rctl &= ~E1000_RCTL_VFE;
3170                 /* Do not hardware filter VLANs in promisc mode */
3171                 e1000e_vlan_filter_disable(adapter);
3172         } else {
3173                 if (netdev->flags & IFF_ALLMULTI) {
3174                         rctl |= E1000_RCTL_MPE;
3175                         rctl &= ~E1000_RCTL_UPE;
3176                 } else {
3177                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3178                 }
3179                 e1000e_vlan_filter_enable(adapter);
3180         }
3181
3182         ew32(RCTL, rctl);
3183
3184         if (!netdev_mc_empty(netdev)) {
3185                 int i = 0;
3186
3187                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3188                 if (!mta_list)
3189                         return;
3190
3191                 /* prepare a packed array of only addresses. */
3192                 netdev_for_each_mc_addr(ha, netdev)
3193                         memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3194
3195                 e1000_update_mc_addr_list(hw, mta_list, i);
3196                 kfree(mta_list);
3197         } else {
3198                 /*
3199                  * if we're called from probe, we might not have
3200                  * anything to do here, so clear out the list
3201                  */
3202                 e1000_update_mc_addr_list(hw, NULL, 0);
3203         }
3204
3205         if (netdev->features & NETIF_F_HW_VLAN_RX)
3206                 e1000e_vlan_strip_enable(adapter);
3207         else
3208                 e1000e_vlan_strip_disable(adapter);
3209 }
3210
3211 /**
3212  * e1000_configure - configure the hardware for Rx and Tx
3213  * @adapter: private board structure
3214  **/
3215 static void e1000_configure(struct e1000_adapter *adapter)
3216 {
3217         e1000_set_multi(adapter->netdev);
3218
3219         e1000_restore_vlan(adapter);
3220         e1000_init_manageability_pt(adapter);
3221
3222         e1000_configure_tx(adapter);
3223         e1000_setup_rctl(adapter);
3224         e1000_configure_rx(adapter);
3225         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring),
3226                               GFP_KERNEL);
3227 }
3228
3229 /**
3230  * e1000e_power_up_phy - restore link in case the phy was powered down
3231  * @adapter: address of board private structure
3232  *
3233  * The phy may be powered down to save power and turn off link when the
3234  * driver is unloaded and wake on lan is not enabled (among others)
3235  * *** this routine MUST be followed by a call to e1000e_reset ***
3236  **/
3237 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3238 {
3239         if (adapter->hw.phy.ops.power_up)
3240                 adapter->hw.phy.ops.power_up(&adapter->hw);
3241
3242         adapter->hw.mac.ops.setup_link(&adapter->hw);
3243 }
3244
3245 /**
3246  * e1000_power_down_phy - Power down the PHY
3247  *
3248  * Power down the PHY so no link is implied when interface is down.
3249  * The PHY cannot be powered down if management or WoL is active.
3250  */
3251 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3252 {
3253         /* WoL is enabled */
3254         if (adapter->wol)
3255                 return;
3256
3257         if (adapter->hw.phy.ops.power_down)
3258                 adapter->hw.phy.ops.power_down(&adapter->hw);
3259 }
3260
3261 /**
3262  * e1000e_reset - bring the hardware into a known good state
3263  *
3264  * This function boots the hardware and enables some settings that
3265  * require a configuration cycle of the hardware - those cannot be
3266  * set/changed during runtime. After reset the device needs to be
3267  * properly configured for Rx, Tx etc.
3268  */
3269 void e1000e_reset(struct e1000_adapter *adapter)
3270 {
3271         struct e1000_mac_info *mac = &adapter->hw.mac;
3272         struct e1000_fc_info *fc = &adapter->hw.fc;
3273         struct e1000_hw *hw = &adapter->hw;
3274         u32 tx_space, min_tx_space, min_rx_space;
3275         u32 pba = adapter->pba;
3276         u16 hwm;
3277
3278         /* reset Packet Buffer Allocation to default */
3279         ew32(PBA, pba);
3280
3281         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3282                 /*
3283                  * To maintain wire speed transmits, the Tx FIFO should be
3284                  * large enough to accommodate two full transmit packets,
3285                  * rounded up to the next 1KB and expressed in KB.  Likewise,
3286                  * the Rx FIFO should be large enough to accommodate at least
3287                  * one full receive packet and is similarly rounded up and
3288                  * expressed in KB.
3289                  */
3290                 pba = er32(PBA);
3291                 /* upper 16 bits has Tx packet buffer allocation size in KB */
3292                 tx_space = pba >> 16;
3293                 /* lower 16 bits has Rx packet buffer allocation size in KB */
3294                 pba &= 0xffff;
3295                 /*
3296                  * the Tx fifo also stores 16 bytes of information about the Tx
3297                  * but don't include ethernet FCS because hardware appends it
3298                  */
3299                 min_tx_space = (adapter->max_frame_size +
3300                                 sizeof(struct e1000_tx_desc) -
3301                                 ETH_FCS_LEN) * 2;
3302                 min_tx_space = ALIGN(min_tx_space, 1024);
3303                 min_tx_space >>= 10;
3304                 /* software strips receive CRC, so leave room for it */
3305                 min_rx_space = adapter->max_frame_size;
3306                 min_rx_space = ALIGN(min_rx_space, 1024);
3307                 min_rx_space >>= 10;
3308
3309                 /*
3310                  * If current Tx allocation is less than the min Tx FIFO size,
3311                  * and the min Tx FIFO size is less than the current Rx FIFO
3312                  * allocation, take space away from current Rx allocation
3313                  */
3314                 if ((tx_space < min_tx_space) &&
3315                     ((min_tx_space - tx_space) < pba)) {
3316                         pba -= min_tx_space - tx_space;
3317
3318                         /*
3319                          * if short on Rx space, Rx wins and must trump Tx
3320                          * adjustment or use Early Receive if available
3321                          */
3322                         if ((pba < min_rx_space) &&
3323                             (!(adapter->flags & FLAG_HAS_ERT)))
3324                                 /* ERT enabled in e1000_configure_rx */
3325                                 pba = min_rx_space;
3326                 }
3327
3328                 ew32(PBA, pba);
3329         }
3330
3331         /*
3332          * flow control settings
3333          *
3334          * The high water mark must be low enough to fit one full frame
3335          * (or the size used for early receive) above it in the Rx FIFO.
3336          * Set it to the lower of:
3337          * - 90% of the Rx FIFO size, and
3338          * - the full Rx FIFO size minus the early receive size (for parts
3339          *   with ERT support assuming ERT set to E1000_ERT_2048), or
3340          * - the full Rx FIFO size minus one full frame
3341          */
3342         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3343                 fc->pause_time = 0xFFFF;
3344         else
3345                 fc->pause_time = E1000_FC_PAUSE_TIME;
3346         fc->send_xon = 1;
3347         fc->current_mode = fc->requested_mode;
3348
3349         switch (hw->mac.type) {
3350         default:
3351                 if ((adapter->flags & FLAG_HAS_ERT) &&
3352                     (adapter->netdev->mtu > ETH_DATA_LEN))
3353                         hwm = min(((pba << 10) * 9 / 10),
3354                                   ((pba << 10) - (E1000_ERT_2048 << 3)));
3355                 else
3356                         hwm = min(((pba << 10) * 9 / 10),
3357                                   ((pba << 10) - adapter->max_frame_size));
3358
3359                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3360                 fc->low_water = fc->high_water - 8;
3361                 break;
3362         case e1000_pchlan:
3363                 /*
3364                  * Workaround PCH LOM adapter hangs with certain network
3365                  * loads.  If hangs persist, try disabling Tx flow control.
3366                  */
3367                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3368                         fc->high_water = 0x3500;
3369                         fc->low_water  = 0x1500;
3370                 } else {
3371                         fc->high_water = 0x5000;
3372                         fc->low_water  = 0x3000;
3373                 }
3374                 fc->refresh_time = 0x1000;
3375                 break;
3376         case e1000_pch2lan:
3377                 fc->high_water = 0x05C20;
3378                 fc->low_water = 0x05048;
3379                 fc->pause_time = 0x0650;
3380                 fc->refresh_time = 0x0400;
3381                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3382                         pba = 14;
3383                         ew32(PBA, pba);
3384                 }
3385                 break;
3386         }
3387
3388         /*
3389          * Alignment of Tx data is on an arbitrary byte boundary with the
3390          * maximum size per Tx descriptor limited only to the transmit
3391          * allocation of the packet buffer minus 96 bytes with an upper
3392          * limit of 24KB due to receive synchronization limitations.
3393          */
3394         adapter->tx_fifo_limit = min_t(u32, ((er32(PBA) >> 16) << 10) - 96,
3395                                        24 << 10);
3396
3397         /*
3398          * Disable Adaptive Interrupt Moderation if 2 full packets cannot
3399          * fit in receive buffer and early-receive not supported.
3400          */
3401         if (adapter->itr_setting & 0x3) {
3402                 if (((adapter->max_frame_size * 2) > (pba << 10)) &&
3403                     !(adapter->flags & FLAG_HAS_ERT)) {
3404                         if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
3405                                 dev_info(&adapter->pdev->dev,
3406                                         "Interrupt Throttle Rate turned off\n");
3407                                 adapter->flags2 |= FLAG2_DISABLE_AIM;
3408                                 ew32(ITR, 0);
3409                         }
3410                 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
3411                         dev_info(&adapter->pdev->dev,
3412                                  "Interrupt Throttle Rate turned on\n");
3413                         adapter->flags2 &= ~FLAG2_DISABLE_AIM;
3414                         adapter->itr = 20000;
3415                         ew32(ITR, 1000000000 / (adapter->itr * 256));
3416                 }
3417         }
3418
3419         /* Allow time for pending master requests to run */
3420         mac->ops.reset_hw(hw);
3421
3422         /*
3423          * For parts with AMT enabled, let the firmware know
3424          * that the network interface is in control
3425          */
3426         if (adapter->flags & FLAG_HAS_AMT)
3427                 e1000e_get_hw_control(adapter);
3428
3429         ew32(WUC, 0);
3430
3431         if (mac->ops.init_hw(hw))
3432                 e_err("Hardware Error\n");
3433
3434         e1000_update_mng_vlan(adapter);
3435
3436         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3437         ew32(VET, ETH_P_8021Q);
3438
3439         e1000e_reset_adaptive(hw);
3440
3441         if (!netif_running(adapter->netdev) &&
3442             !test_bit(__E1000_TESTING, &adapter->state)) {
3443                 e1000_power_down_phy(adapter);
3444                 return;
3445         }
3446
3447         e1000_get_phy_info(hw);
3448
3449         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3450             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3451                 u16 phy_data = 0;
3452                 /*
3453                  * speed up time to link by disabling smart power down, ignore
3454                  * the return value of this function because there is nothing
3455                  * different we would do if it failed
3456                  */
3457                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3458                 phy_data &= ~IGP02E1000_PM_SPD;
3459                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3460         }
3461 }
3462
3463 int e1000e_up(struct e1000_adapter *adapter)
3464 {
3465         struct e1000_hw *hw = &adapter->hw;
3466
3467         /* hardware has been reset, we need to reload some things */
3468         e1000_configure(adapter);
3469
3470         clear_bit(__E1000_DOWN, &adapter->state);
3471
3472         napi_enable(&adapter->napi);
3473         if (adapter->msix_entries)
3474                 e1000_configure_msix(adapter);
3475         e1000_irq_enable(adapter);
3476
3477         netif_start_queue(adapter->netdev);
3478
3479         /* fire a link change interrupt to start the watchdog */
3480         if (adapter->msix_entries)
3481                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3482         else
3483                 ew32(ICS, E1000_ICS_LSC);
3484
3485         return 0;
3486 }
3487
3488 static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
3489 {
3490         struct e1000_hw *hw = &adapter->hw;
3491
3492         if (!(adapter->flags2 & FLAG2_DMA_BURST))
3493                 return;
3494
3495         /* flush pending descriptor writebacks to memory */
3496         ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3497         ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3498
3499         /* execute the writes immediately */
3500         e1e_flush();
3501 }
3502
3503 static void e1000e_update_stats(struct e1000_adapter *adapter);
3504
3505 void e1000e_down(struct e1000_adapter *adapter)
3506 {
3507         struct net_device *netdev = adapter->netdev;
3508         struct e1000_hw *hw = &adapter->hw;
3509         u32 tctl, rctl;
3510
3511         /*
3512          * signal that we're down so the interrupt handler does not
3513          * reschedule our watchdog timer
3514          */
3515         set_bit(__E1000_DOWN, &adapter->state);
3516
3517         /* disable receives in the hardware */
3518         rctl = er32(RCTL);
3519         if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3520                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3521         /* flush and sleep below */
3522
3523         netif_stop_queue(netdev);
3524
3525         /* disable transmits in the hardware */
3526         tctl = er32(TCTL);
3527         tctl &= ~E1000_TCTL_EN;
3528         ew32(TCTL, tctl);
3529
3530         /* flush both disables and wait for them to finish */
3531         e1e_flush();
3532         usleep_range(10000, 20000);
3533
3534         napi_disable(&adapter->napi);
3535         e1000_irq_disable(adapter);
3536
3537         del_timer_sync(&adapter->watchdog_timer);
3538         del_timer_sync(&adapter->phy_info_timer);
3539
3540         netif_carrier_off(netdev);
3541
3542         spin_lock(&adapter->stats64_lock);
3543         e1000e_update_stats(adapter);
3544         spin_unlock(&adapter->stats64_lock);
3545
3546         e1000e_flush_descriptors(adapter);
3547         e1000_clean_tx_ring(adapter);
3548         e1000_clean_rx_ring(adapter);
3549
3550         adapter->link_speed = 0;
3551         adapter->link_duplex = 0;
3552
3553         if (!pci_channel_offline(adapter->pdev))
3554                 e1000e_reset(adapter);
3555
3556         /*
3557          * TODO: for power management, we could drop the link and
3558          * pci_disable_device here.
3559          */
3560 }
3561
3562 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3563 {
3564         might_sleep();
3565         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3566                 usleep_range(1000, 2000);
3567         e1000e_down(adapter);
3568         e1000e_up(adapter);
3569         clear_bit(__E1000_RESETTING, &adapter->state);
3570 }
3571
3572 /**
3573  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3574  * @adapter: board private structure to initialize
3575  *
3576  * e1000_sw_init initializes the Adapter private data structure.
3577  * Fields are initialized based on PCI device information and
3578  * OS network device settings (MTU size).
3579  **/
3580 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3581 {
3582         struct net_device *netdev = adapter->netdev;
3583
3584         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3585         adapter->rx_ps_bsize0 = 128;
3586         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3587         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3588
3589         spin_lock_init(&adapter->stats64_lock);
3590
3591         e1000e_set_interrupt_capability(adapter);
3592
3593         if (e1000_alloc_queues(adapter))
3594                 return -ENOMEM;
3595
3596         /* Explicitly disable IRQ since the NIC can be in any state. */
3597         e1000_irq_disable(adapter);
3598
3599         set_bit(__E1000_DOWN, &adapter->state);
3600         return 0;
3601 }
3602
3603 /**
3604  * e1000_intr_msi_test - Interrupt Handler
3605  * @irq: interrupt number
3606  * @data: pointer to a network interface device structure
3607  **/
3608 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3609 {
3610         struct net_device *netdev = data;
3611         struct e1000_adapter *adapter = netdev_priv(netdev);
3612         struct e1000_hw *hw = &adapter->hw;
3613         u32 icr = er32(ICR);
3614
3615         e_dbg("icr is %08X\n", icr);
3616         if (icr & E1000_ICR_RXSEQ) {
3617                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3618                 wmb();
3619         }
3620
3621         return IRQ_HANDLED;
3622 }
3623
3624 /**
3625  * e1000_test_msi_interrupt - Returns 0 for successful test
3626  * @adapter: board private struct
3627  *
3628  * code flow taken from tg3.c
3629  **/
3630 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3631 {
3632         struct net_device *netdev = adapter->netdev;
3633         struct e1000_hw *hw = &adapter->hw;
3634         int err;
3635
3636         /* poll_enable hasn't been called yet, so don't need disable */
3637         /* clear any pending events */
3638         er32(ICR);
3639
3640         /* free the real vector and request a test handler */
3641         e1000_free_irq(adapter);
3642         e1000e_reset_interrupt_capability(adapter);
3643
3644         /* Assume that the test fails, if it succeeds then the test
3645          * MSI irq handler will unset this flag */
3646         adapter->flags |= FLAG_MSI_TEST_FAILED;
3647
3648         err = pci_enable_msi(adapter->pdev);
3649         if (err)
3650                 goto msi_test_failed;
3651
3652         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3653                           netdev->name, netdev);
3654         if (err) {
3655                 pci_disable_msi(adapter->pdev);
3656                 goto msi_test_failed;
3657         }
3658
3659         wmb();
3660
3661         e1000_irq_enable(adapter);
3662
3663         /* fire an unusual interrupt on the test handler */
3664         ew32(ICS, E1000_ICS_RXSEQ);
3665         e1e_flush();
3666         msleep(50);
3667
3668         e1000_irq_disable(adapter);
3669
3670         rmb();
3671
3672         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3673                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3674                 e_info("MSI interrupt test failed, using legacy interrupt.\n");
3675         } else
3676                 e_dbg("MSI interrupt test succeeded!\n");
3677
3678         free_irq(adapter->pdev->irq, netdev);
3679         pci_disable_msi(adapter->pdev);
3680
3681 msi_test_failed:
3682         e1000e_set_interrupt_capability(adapter);
3683         return e1000_request_irq(adapter);
3684 }
3685
3686 /**
3687  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3688  * @adapter: board private struct
3689  *
3690  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3691  **/
3692 static int e1000_test_msi(struct e1000_adapter *adapter)
3693 {
3694         int err;
3695         u16 pci_cmd;
3696
3697         if (!(adapter->flags & FLAG_MSI_ENABLED))
3698                 return 0;
3699
3700         /* disable SERR in case the MSI write causes a master abort */
3701         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3702         if (pci_cmd & PCI_COMMAND_SERR)
3703                 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3704                                       pci_cmd & ~PCI_COMMAND_SERR);
3705
3706         err = e1000_test_msi_interrupt(adapter);
3707
3708         /* re-enable SERR */
3709         if (pci_cmd & PCI_COMMAND_SERR) {
3710                 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3711                 pci_cmd |= PCI_COMMAND_SERR;
3712                 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3713         }
3714
3715         return err;
3716 }
3717
3718 /**
3719  * e1000_open - Called when a network interface is made active
3720  * @netdev: network interface device structure
3721  *
3722  * Returns 0 on success, negative value on failure
3723  *
3724  * The open entry point is called when a network interface is made
3725  * active by the system (IFF_UP).  At this point all resources needed
3726  * for transmit and receive operations are allocated, the interrupt
3727  * handler is registered with the OS, the watchdog timer is started,
3728  * and the stack is notified that the interface is ready.
3729  **/
3730 static int e1000_open(struct net_device *netdev)
3731 {
3732         struct e1000_adapter *adapter = netdev_priv(netdev);
3733         struct e1000_hw *hw = &adapter->hw;
3734         struct pci_dev *pdev = adapter->pdev;
3735         int err;
3736
3737         /* disallow open during test */
3738         if (test_bit(__E1000_TESTING, &adapter->state))
3739                 return -EBUSY;
3740
3741         pm_runtime_get_sync(&pdev->dev);
3742
3743         netif_carrier_off(netdev);
3744
3745         /* allocate transmit descriptors */
3746         err = e1000e_setup_tx_resources(adapter);
3747         if (err)
3748                 goto err_setup_tx;
3749
3750         /* allocate receive descriptors */
3751         err = e1000e_setup_rx_resources(adapter);
3752         if (err)
3753                 goto err_setup_rx;
3754
3755         /*
3756          * If AMT is enabled, let the firmware know that the network
3757          * interface is now open and reset the part to a known state.
3758          */
3759         if (adapter->flags & FLAG_HAS_AMT) {
3760                 e1000e_get_hw_control(adapter);
3761                 e1000e_reset(adapter);
3762         }
3763
3764         e1000e_power_up_phy(adapter);
3765
3766         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3767         if ((adapter->hw.mng_cookie.status &
3768              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3769                 e1000_update_mng_vlan(adapter);
3770
3771         /* DMA latency requirement to workaround early-receive/jumbo issue */
3772         if ((adapter->flags & FLAG_HAS_ERT) ||
3773             (adapter->hw.mac.type == e1000_pch2lan))
3774                 pm_qos_add_request(&adapter->netdev->pm_qos_req,
3775                                    PM_QOS_CPU_DMA_LATENCY,
3776                                    PM_QOS_DEFAULT_VALUE);
3777
3778         /*
3779          * before we allocate an interrupt, we must be ready to handle it.
3780          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3781          * as soon as we call pci_request_irq, so we have to setup our
3782          * clean_rx handler before we do so.
3783          */
3784         e1000_configure(adapter);
3785
3786         err = e1000_request_irq(adapter);
3787         if (err)
3788                 goto err_req_irq;
3789
3790         /*
3791          * Work around PCIe errata with MSI interrupts causing some chipsets to
3792          * ignore e1000e MSI messages, which means we need to test our MSI
3793          * interrupt now
3794          */
3795         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3796                 err = e1000_test_msi(adapter);
3797                 if (err) {
3798                         e_err("Interrupt allocation failed\n");
3799                         goto err_req_irq;
3800                 }
3801         }
3802
3803         /* From here on the code is the same as e1000e_up() */
3804         clear_bit(__E1000_DOWN, &adapter->state);
3805
3806         napi_enable(&adapter->napi);
3807
3808         e1000_irq_enable(adapter);
3809
3810         adapter->tx_hang_recheck = false;
3811         netif_start_queue(netdev);
3812
3813         adapter->idle_check = true;
3814         pm_runtime_put(&pdev->dev);
3815
3816         /* fire a link status change interrupt to start the watchdog */
3817         if (adapter->msix_entries)
3818                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3819         else
3820                 ew32(ICS, E1000_ICS_LSC);
3821
3822         return 0;
3823
3824 err_req_irq:
3825         e1000e_release_hw_control(adapter);
3826         e1000_power_down_phy(adapter);
3827         e1000e_free_rx_resources(adapter);
3828 err_setup_rx:
3829         e1000e_free_tx_resources(adapter);
3830 err_setup_tx:
3831         e1000e_reset(adapter);
3832         pm_runtime_put_sync(&pdev->dev);
3833
3834         return err;
3835 }
3836
3837 /**
3838  * e1000_close - Disables a network interface
3839  * @netdev: network interface device structure
3840  *
3841  * Returns 0, this is not allowed to fail
3842  *
3843  * The close entry point is called when an interface is de-activated
3844  * by the OS.  The hardware is still under the drivers control, but
3845  * needs to be disabled.  A global MAC reset is issued to stop the
3846  * hardware, and all transmit and receive resources are freed.
3847  **/
3848 static int e1000_close(struct net_device *netdev)
3849 {
3850         struct e1000_adapter *adapter = netdev_priv(netdev);
3851         struct pci_dev *pdev = adapter->pdev;
3852
3853         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3854
3855         pm_runtime_get_sync(&pdev->dev);
3856
3857         if (!test_bit(__E1000_DOWN, &adapter->state)) {
3858                 e1000e_down(adapter);
3859                 e1000_free_irq(adapter);
3860         }
3861         e1000_power_down_phy(adapter);
3862
3863         e1000e_free_tx_resources(adapter);
3864         e1000e_free_rx_resources(adapter);
3865
3866         /*
3867          * kill manageability vlan ID if supported, but not if a vlan with
3868          * the same ID is registered on the host OS (let 8021q kill it)
3869          */
3870         if (adapter->hw.mng_cookie.status &
3871             E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
3872                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3873
3874         /*
3875          * If AMT is enabled, let the firmware know that the network
3876          * interface is now closed
3877          */
3878         if ((adapter->flags & FLAG_HAS_AMT) &&
3879             !test_bit(__E1000_TESTING, &adapter->state))
3880                 e1000e_release_hw_control(adapter);
3881
3882         if ((adapter->flags & FLAG_HAS_ERT) ||
3883             (adapter->hw.mac.type == e1000_pch2lan))
3884                 pm_qos_remove_request(&adapter->netdev->pm_qos_req);
3885
3886         pm_runtime_put_sync(&pdev->dev);
3887
3888         return 0;
3889 }
3890 /**
3891  * e1000_set_mac - Change the Ethernet Address of the NIC
3892  * @netdev: network interface device structure
3893  * @p: pointer to an address structure
3894  *
3895  * Returns 0 on success, negative on failure
3896  **/
3897 static int e1000_set_mac(struct net_device *netdev, void *p)
3898 {
3899         struct e1000_adapter *adapter = netdev_priv(netdev);
3900         struct sockaddr *addr = p;
3901
3902         if (!is_valid_ether_addr(addr->sa_data))
3903                 return -EADDRNOTAVAIL;
3904
3905         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3906         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3907
3908         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3909
3910         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3911                 /* activate the work around */
3912                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3913
3914                 /*
3915                  * Hold a copy of the LAA in RAR[14] This is done so that
3916                  * between the time RAR[0] gets clobbered  and the time it
3917                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3918                  * of the RARs and no incoming packets directed to this port
3919                  * are dropped. Eventually the LAA will be in RAR[0] and
3920                  * RAR[14]
3921                  */
3922                 e1000e_rar_set(&adapter->hw,
3923                               adapter->hw.mac.addr,
3924                               adapter->hw.mac.rar_entry_count - 1);
3925         }
3926
3927         return 0;
3928 }
3929
3930 /**
3931  * e1000e_update_phy_task - work thread to update phy
3932  * @work: pointer to our work struct
3933  *
3934  * this worker thread exists because we must acquire a
3935  * semaphore to read the phy, which we could msleep while
3936  * waiting for it, and we can't msleep in a timer.
3937  **/
3938 static void e1000e_update_phy_task(struct work_struct *work)
3939 {
3940         struct e1000_adapter *adapter = container_of(work,
3941                                         struct e1000_adapter, update_phy_task);
3942
3943         if (test_bit(__E1000_DOWN, &adapter->state))
3944                 return;
3945
3946         e1000_get_phy_info(&adapter->hw);
3947 }
3948
3949 /*
3950  * Need to wait a few seconds after link up to get diagnostic information from
3951  * the phy
3952  */
3953 static void e1000_update_phy_info(unsigned long data)
3954 {
3955         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3956
3957         if (test_bit(__E1000_DOWN, &adapter->state))
3958                 return;
3959
3960         schedule_work(&adapter->update_phy_task);
3961 }
3962
3963 /**
3964  * e1000e_update_phy_stats - Update the PHY statistics counters
3965  * @adapter: board private structure
3966  *
3967  * Read/clear the upper 16-bit PHY registers and read/accumulate lower
3968  **/
3969 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
3970 {
3971         struct e1000_hw *hw = &adapter->hw;
3972         s32 ret_val;
3973         u16 phy_data;
3974
3975         ret_val = hw->phy.ops.acquire(hw);
3976         if (ret_val)
3977                 return;
3978
3979         /*
3980          * A page set is expensive so check if already on desired page.
3981          * If not, set to the page with the PHY status registers.
3982          */
3983         hw->phy.addr = 1;
3984         ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
3985                                            &phy_data);
3986         if (ret_val)
3987                 goto release;
3988         if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
3989                 ret_val = hw->phy.ops.set_page(hw,
3990                                                HV_STATS_PAGE << IGP_PAGE_SHIFT);
3991                 if (ret_val)
3992                         goto release;
3993         }
3994
3995         /* Single Collision Count */
3996         hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
3997         ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
3998         if (!ret_val)
3999                 adapter->stats.scc += phy_data;
4000
4001         /* Excessive Collision Count */
4002         hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
4003         ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
4004         if (!ret_val)
4005                 adapter->stats.ecol += phy_data;
4006
4007         /* Multiple Collision Count */
4008         hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
4009         ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
4010         if (!ret_val)
4011                 adapter->stats.mcc += phy_data;
4012
4013         /* Late Collision Count */
4014         hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
4015         ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
4016         if (!ret_val)
4017                 adapter->stats.latecol += phy_data;
4018
4019         /* Collision Count - also used for adaptive IFS */
4020         hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
4021         ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
4022         if (!ret_val)
4023                 hw->mac.collision_delta = phy_data;
4024
4025         /* Defer Count */
4026         hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
4027         ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
4028         if (!ret_val)
4029                 adapter->stats.dc += phy_data;
4030
4031         /* Transmit with no CRS */
4032         hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
4033         ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
4034         if (!ret_val)
4035                 adapter->stats.tncrs += phy_data;
4036
4037 release:
4038         hw->phy.ops.release(hw);
4039 }
4040
4041 /**
4042  * e1000e_update_stats - Update the board statistics counters
4043  * @adapter: board private structure
4044  **/
4045 static void e1000e_update_stats(struct e1000_adapter *adapter)
4046 {
4047         struct net_device *netdev = adapter->netdev;
4048         struct e1000_hw *hw = &adapter->hw;
4049         struct pci_dev *pdev = adapter->pdev;
4050
4051         /*
4052          * Prevent stats update while adapter is being reset, or if the pci
4053          * connection is down.
4054          */
4055         if (adapter->link_speed == 0)
4056                 return;
4057         if (pci_channel_offline(pdev))
4058                 return;
4059
4060         adapter->stats.crcerrs += er32(CRCERRS);
4061         adapter->stats.gprc += er32(GPRC);
4062         adapter->stats.gorc += er32(GORCL);
4063         er32(GORCH); /* Clear gorc */
4064         adapter->stats.bprc += er32(BPRC);
4065         adapter->stats.mprc += er32(MPRC);
4066         adapter->stats.roc += er32(ROC);
4067
4068         adapter->stats.mpc += er32(MPC);
4069
4070         /* Half-duplex statistics */
4071         if (adapter->link_duplex == HALF_DUPLEX) {
4072                 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
4073                         e1000e_update_phy_stats(adapter);
4074                 } else {
4075                         adapter->stats.scc += er32(SCC);
4076                         adapter->stats.ecol += er32(ECOL);
4077                         adapter->stats.mcc += er32(MCC);
4078                         adapter->stats.latecol += er32(LATECOL);
4079                         adapter->stats.dc += er32(DC);
4080
4081                         hw->mac.collision_delta = er32(COLC);
4082
4083                         if ((hw->mac.type != e1000_82574) &&
4084                             (hw->mac.type != e1000_82583))
4085                                 adapter->stats.tncrs += er32(TNCRS);
4086                 }
4087                 adapter->stats.colc += hw->mac.collision_delta;
4088         }
4089
4090         adapter->stats.xonrxc += er32(XONRXC);
4091         adapter->stats.xontxc += er32(XONTXC);
4092         adapter->stats.xoffrxc += er32(XOFFRXC);
4093         adapter->stats.xofftxc += er32(XOFFTXC);
4094         adapter->stats.gptc += er32(GPTC);
4095         adapter->stats.gotc += er32(GOTCL);
4096         er32(GOTCH); /* Clear gotc */
4097         adapter->stats.rnbc += er32(RNBC);
4098         adapter->stats.ruc += er32(RUC);
4099
4100         adapter->stats.mptc += er32(MPTC);
4101         adapter->stats.bptc += er32(BPTC);
4102
4103         /* used for adaptive IFS */
4104
4105         hw->mac.tx_packet_delta = er32(TPT);
4106         adapter->stats.tpt += hw->mac.tx_packet_delta;
4107
4108         adapter->stats.algnerrc += er32(ALGNERRC);
4109         adapter->stats.rxerrc += er32(RXERRC);
4110         adapter->stats.cexterr += er32(CEXTERR);
4111         adapter->stats.tsctc += er32(TSCTC);
4112         adapter->stats.tsctfc += er32(TSCTFC);
4113
4114         /* Fill out the OS statistics structure */
4115         netdev->stats.multicast = adapter->stats.mprc;
4116         netdev->stats.collisions = adapter->stats.colc;
4117
4118         /* Rx Errors */
4119
4120         /*
4121          * RLEC on some newer hardware can be incorrect so build
4122          * our own version based on RUC and ROC
4123          */
4124         netdev->stats.rx_errors = adapter->stats.rxerrc +
4125                 adapter->stats.crcerrs + adapter->stats.algnerrc +
4126                 adapter->stats.ruc + adapter->stats.roc +
4127                 adapter->stats.cexterr;
4128         netdev->stats.rx_length_errors = adapter->stats.ruc +
4129                                               adapter->stats.roc;
4130         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4131         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4132         netdev->stats.rx_missed_errors = adapter->stats.mpc;
4133
4134         /* Tx Errors */
4135         netdev->stats.tx_errors = adapter->stats.ecol +
4136                                        adapter->stats.latecol;
4137         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4138         netdev->stats.tx_window_errors = adapter->stats.latecol;
4139         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4140
4141         /* Tx Dropped needs to be maintained elsewhere */
4142
4143         /* Management Stats */
4144         adapter->stats.mgptc += er32(MGTPTC);
4145         adapter->stats.mgprc += er32(MGTPRC);
4146         adapter->stats.mgpdc += er32(MGTPDC);
4147 }
4148
4149 /**
4150  * e1000_phy_read_status - Update the PHY register status snapshot
4151  * @adapter: board private structure
4152  **/
4153 static void e1000_phy_read_status(struct e1000_adapter *adapter)
4154 {
4155         struct e1000_hw *hw = &adapter->hw;
4156         struct e1000_phy_regs *phy = &adapter->phy_regs;
4157
4158         if ((er32(STATUS) & E1000_STATUS_LU) &&
4159             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
4160                 int ret_val;
4161
4162                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
4163                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
4164                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
4165                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
4166                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
4167                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
4168                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
4169                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
4170                 if (ret_val)
4171                         e_warn("Error reading PHY register\n");
4172         } else {
4173                 /*
4174                  * Do not read PHY registers if link is not up
4175                  * Set values to typical power-on defaults
4176                  */
4177                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4178                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4179                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4180                              BMSR_ERCAP);
4181                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4182                                   ADVERTISE_ALL | ADVERTISE_CSMA);
4183                 phy->lpa = 0;
4184                 phy->expansion = EXPANSION_ENABLENPAGE;
4185                 phy->ctrl1000 = ADVERTISE_1000FULL;
4186                 phy->stat1000 = 0;
4187                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4188         }
4189 }
4190
4191 static void e1000_print_link_info(struct e1000_adapter *adapter)
4192 {
4193         struct e1000_hw *hw = &adapter->hw;
4194         u32 ctrl = er32(CTRL);
4195
4196         /* Link status message must follow this format for user tools */
4197         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
4198                "Flow Control: %s\n",
4199                adapter->netdev->name,
4200                adapter->link_speed,
4201                (adapter->link_duplex == FULL_DUPLEX) ?
4202                "Full Duplex" : "Half Duplex",
4203                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
4204                "Rx/Tx" :
4205                ((ctrl & E1000_CTRL_RFCE) ? "Rx" :
4206                 ((ctrl & E1000_CTRL_TFCE) ? "Tx" : "None")));
4207 }
4208
4209 static bool e1000e_has_link(struct e1000_adapter *adapter)
4210 {
4211         struct e1000_hw *hw = &adapter->hw;
4212         bool link_active = 0;
4213         s32 ret_val = 0;
4214
4215         /*
4216          * get_link_status is set on LSC (link status) interrupt or
4217          * Rx sequence error interrupt.  get_link_status will stay
4218          * false until the check_for_link establishes link
4219          * for copper adapters ONLY
4220          */
4221         switch (hw->phy.media_type) {
4222         case e1000_media_type_copper:
4223                 if (hw->mac.get_link_status) {
4224                         ret_val = hw->mac.ops.check_for_link(hw);
4225                         link_active = ret_val > 0;
4226                 } else {
4227                         link_active = 1;
4228                 }
4229                 break;
4230         case e1000_media_type_fiber:
4231                 ret_val = hw->mac.ops.check_for_link(hw);
4232                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4233                 break;
4234         case e1000_media_type_internal_serdes:
4235                 ret_val = hw->mac.ops.check_for_link(hw);
4236                 link_active = adapter->hw.mac.serdes_has_link;
4237                 break;
4238         default:
4239         case e1000_media_type_unknown:
4240                 break;
4241         }
4242
4243         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4244             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4245                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4246                 e_info("Gigabit has been disabled, downgrading speed\n");
4247         }
4248
4249         return link_active;
4250 }
4251
4252 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4253 {
4254         /* make sure the receive unit is started */
4255         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4256             (adapter->flags & FLAG_RX_RESTART_NOW)) {
4257                 struct e1000_hw *hw = &adapter->hw;
4258                 u32 rctl = er32(RCTL);
4259                 ew32(RCTL, rctl | E1000_RCTL_EN);
4260                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
4261         }
4262 }
4263
4264 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
4265 {
4266         struct e1000_hw *hw = &adapter->hw;
4267
4268         /*
4269          * With 82574 controllers, PHY needs to be checked periodically
4270          * for hung state and reset, if two calls return true
4271          */
4272         if (e1000_check_phy_82574(hw))
4273                 adapter->phy_hang_count++;
4274         else
4275                 adapter->phy_hang_count = 0;
4276
4277         if (adapter->phy_hang_count > 1) {
4278                 adapter->phy_hang_count = 0;
4279                 schedule_work(&adapter->reset_task);
4280         }
4281 }
4282
4283 /**
4284  * e1000_watchdog - Timer Call-back
4285  * @data: pointer to adapter cast into an unsigned long
4286  **/
4287 static void e1000_watchdog(unsigned long data)
4288 {
4289         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4290
4291         /* Do the rest outside of interrupt context */
4292         schedule_work(&adapter->watchdog_task);
4293
4294         /* TODO: make this use queue_delayed_work() */
4295 }
4296
4297 static void e1000_watchdog_task(struct work_struct *work)
4298 {
4299         struct e1000_adapter *adapter = container_of(work,
4300                                         struct e1000_adapter, watchdog_task);
4301         struct net_device *netdev = adapter->netdev;
4302         struct e1000_mac_info *mac = &adapter->hw.mac;
4303         struct e1000_phy_info *phy = &adapter->hw.phy;
4304         struct e1000_ring *tx_ring = adapter->tx_ring;
4305         struct e1000_hw *hw = &adapter->hw;
4306         u32 link, tctl;
4307
4308         if (test_bit(__E1000_DOWN, &adapter->state))
4309                 return;
4310
4311         link = e1000e_has_link(adapter);
4312         if ((netif_carrier_ok(netdev)) && link) {
4313                 /* Cancel scheduled suspend requests. */
4314                 pm_runtime_resume(netdev->dev.parent);
4315
4316                 e1000e_enable_receives(adapter);
4317                 goto link_up;
4318         }
4319
4320         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4321             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4322                 e1000_update_mng_vlan(adapter);
4323
4324         if (link) {
4325                 if (!netif_carrier_ok(netdev)) {
4326                         bool txb2b = 1;
4327
4328                         /* Cancel scheduled suspend requests. */
4329                         pm_runtime_resume(netdev->dev.parent);
4330
4331                         /* update snapshot of PHY registers on LSC */
4332                         e1000_phy_read_status(adapter);
4333                         mac->ops.get_link_up_info(&adapter->hw,
4334                                                    &adapter->link_speed,
4335                                                    &adapter->link_duplex);
4336                         e1000_print_link_info(adapter);
4337                         /*
4338                          * On supported PHYs, check for duplex mismatch only
4339                          * if link has autonegotiated at 10/100 half
4340                          */
4341                         if ((hw->phy.type == e1000_phy_igp_3 ||
4342                              hw->phy.type == e1000_phy_bm) &&
4343                             (hw->mac.autoneg == true) &&
4344                             (adapter->link_speed == SPEED_10 ||
4345                              adapter->link_speed == SPEED_100) &&
4346                             (adapter->link_duplex == HALF_DUPLEX)) {
4347                                 u16 autoneg_exp;
4348
4349                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4350
4351                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4352                                         e_info("Autonegotiated half duplex but"
4353                                                " link partner cannot autoneg. "
4354                                                " Try forcing full duplex if "
4355                                                "link gets many collisions.\n");
4356                         }
4357
4358                         /* adjust timeout factor according to speed/duplex */
4359                         adapter->tx_timeout_factor = 1;
4360                         switch (adapter->link_speed) {
4361                         case SPEED_10:
4362                                 txb2b = 0;
4363                                 adapter->tx_timeout_factor = 16;
4364                                 break;
4365                         case SPEED_100:
4366                                 txb2b = 0;
4367                                 adapter->tx_timeout_factor = 10;
4368                                 break;
4369                         }
4370
4371                         /*
4372                          * workaround: re-program speed mode bit after
4373                          * link-up event
4374                          */
4375                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4376                             !txb2b) {
4377                                 u32 tarc0;
4378                                 tarc0 = er32(TARC(0));
4379                                 tarc0 &= ~SPEED_MODE_BIT;
4380                                 ew32(TARC(0), tarc0);
4381                         }
4382
4383                         /*
4384                          * disable TSO for pcie and 10/100 speeds, to avoid
4385                          * some hardware issues
4386                          */
4387                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
4388                                 switch (adapter->link_speed) {
4389                                 case SPEED_10:
4390                                 case SPEED_100:
4391                                         e_info("10/100 speed: disabling TSO\n");
4392                                         netdev->features &= ~NETIF_F_TSO;
4393                                         netdev->features &= ~NETIF_F_TSO6;
4394                                         break;
4395                                 case SPEED_1000:
4396                                         netdev->features |= NETIF_F_TSO;
4397                                         netdev->features |= NETIF_F_TSO6;
4398                                         break;
4399                                 default:
4400                                         /* oops */
4401                                         break;
4402                                 }
4403                         }
4404
4405                         /*
4406                          * enable transmits in the hardware, need to do this
4407                          * after setting TARC(0)
4408                          */
4409                         tctl = er32(TCTL);
4410                         tctl |= E1000_TCTL_EN;
4411                         ew32(TCTL, tctl);
4412
4413                         /*
4414                          * Perform any post-link-up configuration before
4415                          * reporting link up.
4416                          */
4417                         if (phy->ops.cfg_on_link_up)
4418                                 phy->ops.cfg_on_link_up(hw);
4419
4420                         netif_carrier_on(netdev);
4421
4422                         if (!test_bit(__E1000_DOWN, &adapter->state))
4423                                 mod_timer(&adapter->phy_info_timer,
4424                                           round_jiffies(jiffies + 2 * HZ));
4425                 }
4426         } else {
4427                 if (netif_carrier_ok(netdev)) {
4428                         adapter->link_speed = 0;
4429                         adapter->link_duplex = 0;
4430                         /* Link status message must follow this format */
4431                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4432                                adapter->netdev->name);
4433                         netif_carrier_off(netdev);
4434                         if (!test_bit(__E1000_DOWN, &adapter->state))
4435                                 mod_timer(&adapter->phy_info_timer,
4436                                           round_jiffies(jiffies + 2 * HZ));
4437
4438                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4439                                 schedule_work(&adapter->reset_task);
4440                         else
4441                                 pm_schedule_suspend(netdev->dev.parent,
4442                                                         LINK_TIMEOUT);
4443                 }
4444         }
4445
4446 link_up:
4447         spin_lock(&adapter->stats64_lock);
4448         e1000e_update_stats(adapter);
4449
4450         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4451         adapter->tpt_old = adapter->stats.tpt;
4452         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4453         adapter->colc_old = adapter->stats.colc;
4454
4455         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4456         adapter->gorc_old = adapter->stats.gorc;
4457         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4458         adapter->gotc_old = adapter->stats.gotc;
4459         spin_unlock(&adapter->stats64_lock);
4460
4461         e1000e_update_adaptive(&adapter->hw);
4462
4463         if (!netif_carrier_ok(netdev) &&
4464             (e1000_desc_unused(tx_ring) + 1 < tx_ring->count)) {
4465                 /*
4466                  * We've lost link, so the controller stops DMA,
4467                  * but we've got queued Tx work that's never going
4468                  * to get done, so reset controller to flush Tx.
4469                  * (Do the reset outside of interrupt context).
4470                  */
4471                 schedule_work(&adapter->reset_task);
4472                 /* return immediately since reset is imminent */
4473                 return;
4474         }
4475
4476         /* Simple mode for Interrupt Throttle Rate (ITR) */
4477         if (adapter->itr_setting == 4) {
4478                 /*
4479                  * Symmetric Tx/Rx gets a reduced ITR=2000;
4480                  * Total asymmetrical Tx or Rx gets ITR=8000;
4481                  * everyone else is between 2000-8000.
4482                  */
4483                 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4484                 u32 dif = (adapter->gotc > adapter->gorc ?
4485                             adapter->gotc - adapter->gorc :
4486                             adapter->gorc - adapter->gotc) / 10000;
4487                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4488
4489                 ew32(ITR, 1000000000 / (itr * 256));
4490         }
4491
4492         /* Cause software interrupt to ensure Rx ring is cleaned */
4493         if (adapter->msix_entries)
4494                 ew32(ICS, adapter->rx_ring->ims_val);
4495         else
4496                 ew32(ICS, E1000_ICS_RXDMT0);
4497
4498         /* flush pending descriptors to memory before detecting Tx hang */
4499         e1000e_flush_descriptors(adapter);
4500
4501         /* Force detection of hung controller every watchdog period */
4502         adapter->detect_tx_hung = 1;
4503
4504         /*
4505          * With 82571 controllers, LAA may be overwritten due to controller
4506          * reset from the other port. Set the appropriate LAA in RAR[0]
4507          */
4508         if (e1000e_get_laa_state_82571(hw))
4509                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4510
4511         if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
4512                 e1000e_check_82574_phy_workaround(adapter);
4513
4514         /* Reset the timer */
4515         if (!test_bit(__E1000_DOWN, &adapter->state))
4516                 mod_timer(&adapter->watchdog_timer,
4517                           round_jiffies(jiffies + 2 * HZ));
4518 }
4519
4520 #define E1000_TX_FLAGS_CSUM             0x00000001
4521 #define E1000_TX_FLAGS_VLAN             0x00000002
4522 #define E1000_TX_FLAGS_TSO              0x00000004
4523 #define E1000_TX_FLAGS_IPV4             0x00000008
4524 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
4525 #define E1000_TX_FLAGS_VLAN_SHIFT       16
4526
4527 static int e1000_tso(struct e1000_adapter *adapter,
4528                      struct sk_buff *skb)
4529 {
4530         struct e1000_ring *tx_ring = adapter->tx_ring;
4531         struct e1000_context_desc *context_desc;
4532         struct e1000_buffer *buffer_info;
4533         unsigned int i;
4534         u32 cmd_length = 0;
4535         u16 ipcse = 0, tucse, mss;
4536         u8 ipcss, ipcso, tucss, tucso, hdr_len;
4537
4538         if (!skb_is_gso(skb))
4539                 return 0;
4540
4541         if (skb_header_cloned(skb)) {
4542                 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4543
4544                 if (err)
4545                         return err;
4546         }
4547
4548         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4549         mss = skb_shinfo(skb)->gso_size;
4550         if (skb->protocol == htons(ETH_P_IP)) {
4551                 struct iphdr *iph = ip_hdr(skb);
4552                 iph->tot_len = 0;
4553                 iph->check = 0;
4554                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4555                                                          0, IPPROTO_TCP, 0);
4556                 cmd_length = E1000_TXD_CMD_IP;
4557                 ipcse = skb_transport_offset(skb) - 1;
4558         } else if (skb_is_gso_v6(skb)) {
4559                 ipv6_hdr(skb)->payload_len = 0;
4560                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4561                                                        &ipv6_hdr(skb)->daddr,
4562                                                        0, IPPROTO_TCP, 0);
4563                 ipcse = 0;
4564         }
4565         ipcss = skb_network_offset(skb);
4566         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4567         tucss = skb_transport_offset(skb);
4568         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4569         tucse = 0;
4570
4571         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4572                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4573
4574         i = tx_ring->next_to_use;
4575         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4576         buffer_info = &tx_ring->buffer_info[i];
4577
4578         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4579         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4580         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4581         context_desc->upper_setup.tcp_fields.tucss = tucss;
4582         context_desc->upper_setup.tcp_fields.tucso = tucso;
4583         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4584         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4585         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4586         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4587
4588         buffer_info->time_stamp = jiffies;
4589         buffer_info->next_to_watch = i;
4590
4591         i++;
4592         if (i == tx_ring->count)
4593                 i = 0;
4594         tx_ring->next_to_use = i;
4595
4596         return 1;
4597 }
4598
4599 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
4600 {
4601         struct e1000_ring *tx_ring = adapter->tx_ring;
4602         struct e1000_context_desc *context_desc;
4603         struct e1000_buffer *buffer_info;
4604         unsigned int i;
4605         u8 css;
4606         u32 cmd_len = E1000_TXD_CMD_DEXT;
4607         __be16 protocol;
4608
4609         if (skb->ip_summed != CHECKSUM_PARTIAL)
4610                 return 0;
4611
4612         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4613                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4614         else
4615                 protocol = skb->protocol;
4616
4617         switch (protocol) {
4618         case cpu_to_be16(ETH_P_IP):
4619                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4620                         cmd_len |= E1000_TXD_CMD_TCP;
4621                 break;
4622         case cpu_to_be16(ETH_P_IPV6):
4623                 /* XXX not handling all IPV6 headers */
4624                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4625                         cmd_len |= E1000_TXD_CMD_TCP;
4626                 break;
4627         default:
4628                 if (unlikely(net_ratelimit()))
4629                         e_warn("checksum_partial proto=%x!\n",
4630                                be16_to_cpu(protocol));
4631                 break;
4632         }
4633
4634         css = skb_checksum_start_offset(skb);
4635
4636         i = tx_ring->next_to_use;
4637         buffer_info = &tx_ring->buffer_info[i];
4638         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4639
4640         context_desc->lower_setup.ip_config = 0;
4641         context_desc->upper_setup.tcp_fields.tucss = css;
4642         context_desc->upper_setup.tcp_fields.tucso =
4643                                 css + skb->csum_offset;
4644         context_desc->upper_setup.tcp_fields.tucse = 0;
4645         context_desc->tcp_seg_setup.data = 0;
4646         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4647
4648         buffer_info->time_stamp = jiffies;
4649         buffer_info->next_to_watch = i;
4650
4651         i++;
4652         if (i == tx_ring->count)
4653                 i = 0;
4654         tx_ring->next_to_use = i;
4655
4656         return 1;
4657 }
4658
4659 static int e1000_tx_map(struct e1000_adapter *adapter,
4660                         struct sk_buff *skb, unsigned int first,
4661                         unsigned int max_per_txd, unsigned int nr_frags)
4662 {
4663         struct e1000_ring *tx_ring = adapter->tx_ring;
4664         struct pci_dev *pdev = adapter->pdev;
4665         struct e1000_buffer *buffer_info;
4666         unsigned int len = skb_headlen(skb);
4667         unsigned int offset = 0, size, count = 0, i;
4668         unsigned int f, bytecount, segs;
4669
4670         i = tx_ring->next_to_use;
4671
4672         while (len) {
4673                 buffer_info = &tx_ring->buffer_info[i];
4674                 size = min(len, max_per_txd);
4675
4676                 buffer_info->length = size;
4677                 buffer_info->time_stamp = jiffies;
4678                 buffer_info->next_to_watch = i;
4679                 buffer_info->dma = dma_map_single(&pdev->dev,
4680                                                   skb->data + offset,
4681                                                   size, DMA_TO_DEVICE);
4682                 buffer_info->mapped_as_page = false;
4683                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4684                         goto dma_error;
4685
4686                 len -= size;
4687                 offset += size;
4688                 count++;
4689
4690                 if (len) {
4691                         i++;
4692                         if (i == tx_ring->count)
4693                                 i = 0;
4694                 }
4695         }
4696
4697         for (f = 0; f < nr_frags; f++) {
4698                 const struct skb_frag_struct *frag;
4699
4700                 frag = &skb_shinfo(skb)->frags[f];
4701                 len = skb_frag_size(frag);
4702                 offset = 0;
4703
4704                 while (len) {
4705                         i++;
4706                         if (i == tx_ring->count)
4707                                 i = 0;
4708
4709                         buffer_info = &tx_ring->buffer_info[i];
4710                         size = min(len, max_per_txd);
4711
4712                         buffer_info->length = size;
4713                         buffer_info->time_stamp = jiffies;
4714                         buffer_info->next_to_watch = i;
4715                         buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
4716                                                 offset, size, DMA_TO_DEVICE);
4717                         buffer_info->mapped_as_page = true;
4718                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4719                                 goto dma_error;
4720
4721                         len -= size;
4722                         offset += size;
4723                         count++;
4724                 }
4725         }
4726
4727         segs = skb_shinfo(skb)->gso_segs ? : 1;
4728         /* multiply data chunks by size of headers */
4729         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4730
4731         tx_ring->buffer_info[i].skb = skb;
4732         tx_ring->buffer_info[i].segs = segs;
4733         tx_ring->buffer_info[i].bytecount = bytecount;
4734         tx_ring->buffer_info[first].next_to_watch = i;
4735
4736         return count;
4737
4738 dma_error:
4739         dev_err(&pdev->dev, "Tx DMA map failed\n");
4740         buffer_info->dma = 0;
4741         if (count)
4742                 count--;
4743
4744         while (count--) {
4745                 if (i == 0)
4746                         i += tx_ring->count;
4747                 i--;
4748                 buffer_info = &tx_ring->buffer_info[i];
4749                 e1000_put_txbuf(adapter, buffer_info);
4750         }
4751
4752         return 0;
4753 }
4754
4755 static void e1000_tx_queue(struct e1000_adapter *adapter,
4756                            int tx_flags, int count)
4757 {
4758         struct e1000_ring *tx_ring = adapter->tx_ring;
4759         struct e1000_tx_desc *tx_desc = NULL;
4760         struct e1000_buffer *buffer_info;
4761         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4762         unsigned int i;
4763
4764         if (tx_flags & E1000_TX_FLAGS_TSO) {
4765                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4766                              E1000_TXD_CMD_TSE;
4767                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4768
4769                 if (tx_flags & E1000_TX_FLAGS_IPV4)
4770                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4771         }
4772
4773         if (tx_flags & E1000_TX_FLAGS_CSUM) {
4774                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4775                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4776         }
4777
4778         if (tx_flags & E1000_TX_FLAGS_VLAN) {
4779                 txd_lower |= E1000_TXD_CMD_VLE;
4780                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4781         }
4782
4783         i = tx_ring->next_to_use;
4784
4785         do {
4786                 buffer_info = &tx_ring->buffer_info[i];
4787                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4788                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4789                 tx_desc->lower.data =
4790                         cpu_to_le32(txd_lower | buffer_info->length);
4791                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4792
4793                 i++;
4794                 if (i == tx_ring->count)
4795                         i = 0;
4796         } while (--count > 0);
4797
4798         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4799
4800         /*
4801          * Force memory writes to complete before letting h/w
4802          * know there are new descriptors to fetch.  (Only
4803          * applicable for weak-ordered memory model archs,
4804          * such as IA-64).
4805          */
4806         wmb();
4807
4808         tx_ring->next_to_use = i;
4809
4810         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
4811                 e1000e_update_tdt_wa(adapter, i);
4812         else
4813                 writel(i, adapter->hw.hw_addr + tx_ring->tail);
4814
4815         /*
4816          * we need this if more than one processor can write to our tail
4817          * at a time, it synchronizes IO on IA64/Altix systems
4818          */
4819         mmiowb();
4820 }
4821
4822 #define MINIMUM_DHCP_PACKET_SIZE 282
4823 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4824                                     struct sk_buff *skb)
4825 {
4826         struct e1000_hw *hw =  &adapter->hw;
4827         u16 length, offset;
4828
4829         if (vlan_tx_tag_present(skb)) {
4830                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4831                     (adapter->hw.mng_cookie.status &
4832                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4833                         return 0;
4834         }
4835
4836         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4837                 return 0;
4838
4839         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4840                 return 0;
4841
4842         {
4843                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4844                 struct udphdr *udp;
4845
4846                 if (ip->protocol != IPPROTO_UDP)
4847                         return 0;
4848
4849                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4850                 if (ntohs(udp->dest) != 67)
4851                         return 0;
4852
4853                 offset = (u8 *)udp + 8 - skb->data;
4854                 length = skb->len - offset;
4855                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4856         }
4857
4858         return 0;
4859 }
4860
4861 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4862 {
4863         struct e1000_adapter *adapter = netdev_priv(netdev);
4864
4865         netif_stop_queue(netdev);
4866         /*
4867          * Herbert's original patch had:
4868          *  smp_mb__after_netif_stop_queue();
4869          * but since that doesn't exist yet, just open code it.
4870          */
4871         smp_mb();
4872
4873         /*
4874          * We need to check again in a case another CPU has just
4875          * made room available.
4876          */
4877         if (e1000_desc_unused(adapter->tx_ring) < size)
4878                 return -EBUSY;
4879
4880         /* A reprieve! */
4881         netif_start_queue(netdev);
4882         ++adapter->restart_queue;
4883         return 0;
4884 }
4885
4886 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4887 {
4888         struct e1000_adapter *adapter = netdev_priv(netdev);
4889
4890         BUG_ON(size > adapter->tx_ring->count);
4891
4892         if (e1000_desc_unused(adapter->tx_ring) >= size)
4893                 return 0;
4894         return __e1000_maybe_stop_tx(netdev, size);
4895 }
4896
4897 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4898                                     struct net_device *netdev)
4899 {
4900         struct e1000_adapter *adapter = netdev_priv(netdev);
4901         struct e1000_ring *tx_ring = adapter->tx_ring;
4902         unsigned int first;
4903         unsigned int tx_flags = 0;
4904         unsigned int len = skb_headlen(skb);
4905         unsigned int nr_frags;
4906         unsigned int mss;
4907         int count = 0;
4908         int tso;
4909         unsigned int f;
4910
4911         if (test_bit(__E1000_DOWN, &adapter->state)) {
4912                 dev_kfree_skb_any(skb);
4913                 return NETDEV_TX_OK;
4914         }
4915
4916         if (skb->len <= 0) {
4917                 dev_kfree_skb_any(skb);
4918                 return NETDEV_TX_OK;
4919         }
4920
4921         mss = skb_shinfo(skb)->gso_size;
4922         if (mss) {
4923                 u8 hdr_len;
4924
4925                 /*
4926                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4927                  * points to just header, pull a few bytes of payload from
4928                  * frags into skb->data
4929                  */
4930                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4931                 /*
4932                  * we do this workaround for ES2LAN, but it is un-necessary,
4933                  * avoiding it could save a lot of cycles
4934                  */
4935                 if (skb->data_len && (hdr_len == len)) {
4936                         unsigned int pull_size;
4937
4938                         pull_size = min((unsigned int)4, skb->data_len);
4939                         if (!__pskb_pull_tail(skb, pull_size)) {
4940                                 e_err("__pskb_pull_tail failed.\n");
4941                                 dev_kfree_skb_any(skb);
4942                                 return NETDEV_TX_OK;
4943                         }
4944                         len = skb_headlen(skb);
4945                 }
4946         }
4947
4948         /* reserve a descriptor for the offload context */
4949         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4950                 count++;
4951         count++;
4952
4953         count += DIV_ROUND_UP(len, adapter->tx_fifo_limit);
4954
4955         nr_frags = skb_shinfo(skb)->nr_frags;
4956         for (f = 0; f < nr_frags; f++)
4957                 count += DIV_ROUND_UP(skb_frag_size(&skb_shinfo(skb)->frags[f]),
4958                                       adapter->tx_fifo_limit);
4959
4960         if (adapter->hw.mac.tx_pkt_filtering)
4961                 e1000_transfer_dhcp_info(adapter, skb);
4962
4963         /*
4964          * need: count + 2 desc gap to keep tail from touching
4965          * head, otherwise try next time
4966          */
4967         if (e1000_maybe_stop_tx(netdev, count + 2))
4968                 return NETDEV_TX_BUSY;
4969
4970         if (vlan_tx_tag_present(skb)) {
4971                 tx_flags |= E1000_TX_FLAGS_VLAN;
4972                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4973         }
4974
4975         first = tx_ring->next_to_use;
4976
4977         tso = e1000_tso(adapter, skb);
4978         if (tso < 0) {
4979                 dev_kfree_skb_any(skb);
4980                 return NETDEV_TX_OK;
4981         }
4982
4983         if (tso)
4984                 tx_flags |= E1000_TX_FLAGS_TSO;
4985         else if (e1000_tx_csum(adapter, skb))
4986                 tx_flags |= E1000_TX_FLAGS_CSUM;
4987
4988         /*
4989          * Old method was to assume IPv4 packet by default if TSO was enabled.
4990          * 82571 hardware supports TSO capabilities for IPv6 as well...
4991          * no longer assume, we must.
4992          */
4993         if (skb->protocol == htons(ETH_P_IP))
4994                 tx_flags |= E1000_TX_FLAGS_IPV4;
4995
4996         /* if count is 0 then mapping error has occurred */
4997         count = e1000_tx_map(adapter, skb, first, adapter->tx_fifo_limit,
4998                              nr_frags);
4999         if (count) {
5000                 e1000_tx_queue(adapter, tx_flags, count);
5001                 /* Make sure there is space in the ring for the next send. */
5002                 e1000_maybe_stop_tx(netdev,
5003                                     (MAX_SKB_FRAGS *
5004                                      DIV_ROUND_UP(PAGE_SIZE,
5005                                                   adapter->tx_fifo_limit) + 2));
5006         } else {
5007                 dev_kfree_skb_any(skb);
5008                 tx_ring->buffer_info[first].time_stamp = 0;
5009                 tx_ring->next_to_use = first;
5010         }
5011
5012         return NETDEV_TX_OK;
5013 }
5014
5015 /**
5016  * e1000_tx_timeout - Respond to a Tx Hang
5017  * @netdev: network interface device structure
5018  **/
5019 static void e1000_tx_timeout(struct net_device *netdev)
5020 {
5021         struct e1000_adapter *adapter = netdev_priv(netdev);
5022
5023         /* Do the reset outside of interrupt context */
5024         adapter->tx_timeout_count++;
5025         schedule_work(&adapter->reset_task);
5026 }
5027
5028 static void e1000_reset_task(struct work_struct *work)
5029 {
5030         struct e1000_adapter *adapter;
5031         adapter = container_of(work, struct e1000_adapter, reset_task);
5032
5033         /* don't run the task if already down */
5034         if (test_bit(__E1000_DOWN, &adapter->state))
5035                 return;
5036
5037         if (!((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
5038               (adapter->flags & FLAG_RX_RESTART_NOW))) {
5039                 e1000e_dump(adapter);
5040                 e_err("Reset adapter\n");
5041         }
5042         e1000e_reinit_locked(adapter);
5043 }
5044
5045 /**
5046  * e1000_get_stats64 - Get System Network Statistics
5047  * @netdev: network interface device structure
5048  * @stats: rtnl_link_stats64 pointer
5049  *
5050  * Returns the address of the device statistics structure.
5051  **/
5052 struct rtnl_link_stats64 *e1000e_get_stats64(struct net_device *netdev,
5053                                              struct rtnl_link_stats64 *stats)
5054 {
5055         struct e1000_adapter *adapter = netdev_priv(netdev);
5056
5057         memset(stats, 0, sizeof(struct rtnl_link_stats64));
5058         spin_lock(&adapter->stats64_lock);
5059         e1000e_update_stats(adapter);
5060         /* Fill out the OS statistics structure */
5061         stats->rx_bytes = adapter->stats.gorc;
5062         stats->rx_packets = adapter->stats.gprc;
5063         stats->tx_bytes = adapter->stats.gotc;
5064         stats->tx_packets = adapter->stats.gptc;
5065         stats->multicast = adapter->stats.mprc;
5066         stats->collisions = adapter->stats.colc;
5067
5068         /* Rx Errors */
5069
5070         /*
5071          * RLEC on some newer hardware can be incorrect so build
5072          * our own version based on RUC and ROC
5073          */
5074         stats->rx_errors = adapter->stats.rxerrc +
5075                 adapter->stats.crcerrs + adapter->stats.algnerrc +
5076                 adapter->stats.ruc + adapter->stats.roc +
5077                 adapter->stats.cexterr;
5078         stats->rx_length_errors = adapter->stats.ruc +
5079                                               adapter->stats.roc;
5080         stats->rx_crc_errors = adapter->stats.crcerrs;
5081         stats->rx_frame_errors = adapter->stats.algnerrc;
5082         stats->rx_missed_errors = adapter->stats.mpc;
5083
5084         /* Tx Errors */
5085         stats->tx_errors = adapter->stats.ecol +
5086                                        adapter->stats.latecol;
5087         stats->tx_aborted_errors = adapter->stats.ecol;
5088         stats->tx_window_errors = adapter->stats.latecol;
5089         stats->tx_carrier_errors = adapter->stats.tncrs;
5090
5091         /* Tx Dropped needs to be maintained elsewhere */
5092
5093         spin_unlock(&adapter->stats64_lock);
5094         return stats;
5095 }
5096
5097 /**
5098  * e1000_change_mtu - Change the Maximum Transfer Unit
5099  * @netdev: network interface device structure
5100  * @new_mtu: new value for maximum frame size
5101  *
5102  * Returns 0 on success, negative on failure
5103  **/
5104 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
5105 {
5106         struct e1000_adapter *adapter = netdev_priv(netdev);
5107         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5108
5109         /* Jumbo frame support */
5110         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
5111             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
5112                 e_err("Jumbo Frames not supported.\n");
5113                 return -EINVAL;
5114         }
5115
5116         /* Supported frame sizes */
5117         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
5118             (max_frame > adapter->max_hw_frame_size)) {
5119                 e_err("Unsupported MTU setting\n");
5120                 return -EINVAL;
5121         }
5122
5123         /* Jumbo frame workaround on 82579 requires CRC be stripped */
5124         if ((adapter->hw.mac.type == e1000_pch2lan) &&
5125             !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
5126             (new_mtu > ETH_DATA_LEN)) {
5127                 e_err("Jumbo Frames not supported on 82579 when CRC "
5128                       "stripping is disabled.\n");
5129                 return -EINVAL;
5130         }
5131
5132         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
5133                 usleep_range(1000, 2000);
5134         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
5135         adapter->max_frame_size = max_frame;
5136         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5137         netdev->mtu = new_mtu;
5138         if (netif_running(netdev))
5139                 e1000e_down(adapter);
5140
5141         /*
5142          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
5143          * means we reserve 2 more, this pushes us to allocate from the next
5144          * larger slab size.
5145          * i.e. RXBUFFER_2048 --> size-4096 slab
5146          * However with the new *_jumbo_rx* routines, jumbo receives will use
5147          * fragmented skbs
5148          */
5149
5150         if (max_frame <= 2048)
5151                 adapter->rx_buffer_len = 2048;
5152         else
5153                 adapter->rx_buffer_len = 4096;
5154
5155         /* adjust allocation if LPE protects us, and we aren't using SBP */
5156         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
5157              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
5158                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
5159                                          + ETH_FCS_LEN;
5160
5161         if (netif_running(netdev))
5162                 e1000e_up(adapter);
5163         else
5164                 e1000e_reset(adapter);
5165
5166         clear_bit(__E1000_RESETTING, &adapter->state);
5167
5168         return 0;
5169 }
5170
5171 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
5172                            int cmd)
5173 {
5174         struct e1000_adapter *adapter = netdev_priv(netdev);
5175         struct mii_ioctl_data *data = if_mii(ifr);
5176
5177         if (adapter->hw.phy.media_type != e1000_media_type_copper)
5178                 return -EOPNOTSUPP;
5179
5180         switch (cmd) {
5181         case SIOCGMIIPHY:
5182                 data->phy_id = adapter->hw.phy.addr;
5183                 break;
5184         case SIOCGMIIREG:
5185                 e1000_phy_read_status(adapter);
5186
5187                 switch (data->reg_num & 0x1F) {
5188                 case MII_BMCR:
5189                         data->val_out = adapter->phy_regs.bmcr;
5190                         break;
5191                 case MII_BMSR:
5192                         data->val_out = adapter->phy_regs.bmsr;
5193                         break;
5194                 case MII_PHYSID1:
5195                         data->val_out = (adapter->hw.phy.id >> 16);
5196                         break;
5197                 case MII_PHYSID2:
5198                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
5199                         break;
5200                 case MII_ADVERTISE:
5201                         data->val_out = adapter->phy_regs.advertise;
5202                         break;
5203                 case MII_LPA:
5204                         data->val_out = adapter->phy_regs.lpa;
5205                         break;
5206                 case MII_EXPANSION:
5207                         data->val_out = adapter->phy_regs.expansion;
5208                         break;
5209                 case MII_CTRL1000:
5210                         data->val_out = adapter->phy_regs.ctrl1000;
5211                         break;
5212                 case MII_STAT1000:
5213                         data->val_out = adapter->phy_regs.stat1000;
5214                         break;
5215                 case MII_ESTATUS:
5216                         data->val_out = adapter->phy_regs.estatus;
5217                         break;
5218                 default:
5219                         return -EIO;
5220                 }
5221                 break;
5222         case SIOCSMIIREG:
5223         default:
5224                 return -EOPNOTSUPP;
5225         }
5226         return 0;
5227 }
5228
5229 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5230 {
5231         switch (cmd) {
5232         case SIOCGMIIPHY:
5233         case SIOCGMIIREG:
5234         case SIOCSMIIREG:
5235                 return e1000_mii_ioctl(netdev, ifr, cmd);
5236         default:
5237                 return -EOPNOTSUPP;
5238         }
5239 }
5240
5241 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
5242 {
5243         struct e1000_hw *hw = &adapter->hw;
5244         u32 i, mac_reg;
5245         u16 phy_reg, wuc_enable;
5246         int retval = 0;
5247
5248         /* copy MAC RARs to PHY RARs */
5249         e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5250
5251         retval = hw->phy.ops.acquire(hw);
5252         if (retval) {
5253                 e_err("Could not acquire PHY\n");
5254                 return retval;
5255         }
5256
5257         /* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
5258         retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5259         if (retval)
5260                 goto out;
5261
5262         /* copy MAC MTA to PHY MTA - only needed for pchlan */
5263         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5264                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5265                 hw->phy.ops.write_reg_page(hw, BM_MTA(i),
5266                                            (u16)(mac_reg & 0xFFFF));
5267                 hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
5268                                            (u16)((mac_reg >> 16) & 0xFFFF));
5269         }
5270
5271         /* configure PHY Rx Control register */
5272         hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
5273         mac_reg = er32(RCTL);
5274         if (mac_reg & E1000_RCTL_UPE)
5275                 phy_reg |= BM_RCTL_UPE;
5276         if (mac_reg & E1000_RCTL_MPE)
5277                 phy_reg |= BM_RCTL_MPE;
5278         phy_reg &= ~(BM_RCTL_MO_MASK);
5279         if (mac_reg & E1000_RCTL_MO_3)
5280                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5281                                 << BM_RCTL_MO_SHIFT);
5282         if (mac_reg & E1000_RCTL_BAM)
5283                 phy_reg |= BM_RCTL_BAM;
5284         if (mac_reg & E1000_RCTL_PMCF)
5285                 phy_reg |= BM_RCTL_PMCF;
5286         mac_reg = er32(CTRL);
5287         if (mac_reg & E1000_CTRL_RFCE)
5288                 phy_reg |= BM_RCTL_RFCE;
5289         hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
5290
5291         /* enable PHY wakeup in MAC register */
5292         ew32(WUFC, wufc);
5293         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
5294
5295         /* configure and enable PHY wakeup in PHY registers */
5296         hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
5297         hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
5298
5299         /* activate PHY wakeup */
5300         wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5301         retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5302         if (retval)
5303                 e_err("Could not set PHY Host Wakeup bit\n");
5304 out:
5305         hw->phy.ops.release(hw);
5306
5307         return retval;
5308 }
5309
5310 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5311                             bool runtime)
5312 {
5313         struct net_device *netdev = pci_get_drvdata(pdev);
5314         struct e1000_adapter *adapter = netdev_priv(netdev);
5315         struct e1000_hw *hw = &adapter->hw;
5316         u32 ctrl, ctrl_ext, rctl, status;
5317         /* Runtime suspend should only enable wakeup for link changes */
5318         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5319         int retval = 0;
5320
5321         netif_device_detach(netdev);
5322
5323         if (netif_running(netdev)) {
5324                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5325                 e1000e_down(adapter);
5326                 e1000_free_irq(adapter);
5327         }
5328         e1000e_reset_interrupt_capability(adapter);
5329
5330         retval = pci_save_state(pdev);
5331         if (retval)
5332                 return retval;
5333
5334         status = er32(STATUS);
5335         if (status & E1000_STATUS_LU)
5336                 wufc &= ~E1000_WUFC_LNKC;
5337
5338         if (wufc) {
5339                 e1000_setup_rctl(adapter);
5340                 e1000_set_multi(netdev);
5341
5342                 /* turn on all-multi mode if wake on multicast is enabled */
5343                 if (wufc & E1000_WUFC_MC) {
5344                         rctl = er32(RCTL);
5345                         rctl |= E1000_RCTL_MPE;
5346                         ew32(RCTL, rctl);
5347                 }
5348
5349                 ctrl = er32(CTRL);
5350                 /* advertise wake from D3Cold */
5351                 #define E1000_CTRL_ADVD3WUC 0x00100000
5352                 /* phy power management enable */
5353                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5354                 ctrl |= E1000_CTRL_ADVD3WUC;
5355                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5356                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5357                 ew32(CTRL, ctrl);
5358
5359                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5360                     adapter->hw.phy.media_type ==
5361                     e1000_media_type_internal_serdes) {
5362                         /* keep the laser running in D3 */
5363                         ctrl_ext = er32(CTRL_EXT);
5364                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5365                         ew32(CTRL_EXT, ctrl_ext);
5366                 }
5367
5368                 if (adapter->flags & FLAG_IS_ICH)
5369                         e1000_suspend_workarounds_ich8lan(&adapter->hw);
5370
5371                 /* Allow time for pending master requests to run */
5372                 e1000e_disable_pcie_master(&adapter->hw);
5373
5374                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5375                         /* enable wakeup by the PHY */
5376                         retval = e1000_init_phy_wakeup(adapter, wufc);
5377                         if (retval)
5378                                 return retval;
5379                 } else {
5380                         /* enable wakeup by the MAC */
5381                         ew32(WUFC, wufc);
5382                         ew32(WUC, E1000_WUC_PME_EN);
5383                 }
5384         } else {
5385                 ew32(WUC, 0);
5386                 ew32(WUFC, 0);
5387         }
5388
5389         *enable_wake = !!wufc;
5390
5391         /* make sure adapter isn't asleep if manageability is enabled */
5392         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5393             (hw->mac.ops.check_mng_mode(hw)))
5394                 *enable_wake = true;
5395
5396         if (adapter->hw.phy.type == e1000_phy_igp_3)
5397                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5398
5399         /*
5400          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5401          * would have already happened in close and is redundant.
5402          */
5403         e1000e_release_hw_control(adapter);
5404
5405         pci_clear_master(pdev);
5406
5407         return 0;
5408 }
5409
5410 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5411 {
5412         if (sleep && wake) {
5413                 pci_prepare_to_sleep(pdev);
5414                 return;
5415         }
5416
5417         pci_wake_from_d3(pdev, wake);
5418         pci_set_power_state(pdev, PCI_D3hot);
5419 }
5420
5421 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5422                                     bool wake)
5423 {
5424         struct net_device *netdev = pci_get_drvdata(pdev);
5425         struct e1000_adapter *adapter = netdev_priv(netdev);
5426
5427         /*
5428          * The pci-e switch on some quad port adapters will report a
5429          * correctable error when the MAC transitions from D0 to D3.  To
5430          * prevent this we need to mask off the correctable errors on the
5431          * downstream port of the pci-e switch.
5432          */
5433         if (adapter->flags & FLAG_IS_QUAD_PORT) {
5434                 struct pci_dev *us_dev = pdev->bus->self;
5435                 int pos = pci_pcie_cap(us_dev);
5436                 u16 devctl;
5437
5438                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5439                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5440                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
5441
5442                 e1000_power_off(pdev, sleep, wake);
5443
5444                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5445         } else {
5446                 e1000_power_off(pdev, sleep, wake);
5447         }
5448 }
5449
5450 #ifdef CONFIG_PCIEASPM
5451 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5452 {
5453         pci_disable_link_state_locked(pdev, state);
5454 }
5455 #else
5456 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5457 {
5458         int pos;
5459         u16 reg16;
5460
5461         /*
5462          * Both device and parent should have the same ASPM setting.
5463          * Disable ASPM in downstream component first and then upstream.
5464          */
5465         pos = pci_pcie_cap(pdev);
5466         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5467         reg16 &= ~state;
5468         pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5469
5470         if (!pdev->bus->self)
5471                 return;
5472
5473         pos = pci_pcie_cap(pdev->bus->self);
5474         pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5475         reg16 &= ~state;
5476         pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5477 }
5478 #endif
5479 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5480 {
5481         dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5482                  (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5483                  (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5484
5485         __e1000e_disable_aspm(pdev, state);
5486 }
5487
5488 #ifdef CONFIG_PM
5489 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5490 {
5491         return !!adapter->tx_ring->buffer_info;
5492 }
5493
5494 static int __e1000_resume(struct pci_dev *pdev)
5495 {
5496         struct net_device *netdev = pci_get_drvdata(pdev);
5497         struct e1000_adapter *adapter = netdev_priv(netdev);
5498         struct e1000_hw *hw = &adapter->hw;
5499         u16 aspm_disable_flag = 0;
5500         u32 err;
5501
5502         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5503                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5504         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5505                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5506         if (aspm_disable_flag)
5507                 e1000e_disable_aspm(pdev, aspm_disable_flag);
5508
5509         pci_set_power_state(pdev, PCI_D0);
5510         pci_restore_state(pdev);
5511         pci_save_state(pdev);
5512
5513         e1000e_set_interrupt_capability(adapter);
5514         if (netif_running(netdev)) {
5515                 err = e1000_request_irq(adapter);
5516                 if (err)
5517                         return err;
5518         }
5519
5520         if (hw->mac.type == e1000_pch2lan)
5521                 e1000_resume_workarounds_pchlan(&adapter->hw);
5522
5523         e1000e_power_up_phy(adapter);
5524
5525         /* report the system wakeup cause from S3/S4 */
5526         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5527                 u16 phy_data;
5528
5529                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5530                 if (phy_data) {
5531                         e_info("PHY Wakeup cause - %s\n",
5532                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5533                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5534                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5535                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5536                                 phy_data & E1000_WUS_LNKC ? "Link Status "
5537                                 " Change" : "other");
5538                 }
5539                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5540         } else {
5541                 u32 wus = er32(WUS);
5542                 if (wus) {
5543                         e_info("MAC Wakeup cause - %s\n",
5544                                 wus & E1000_WUS_EX ? "Unicast Packet" :
5545                                 wus & E1000_WUS_MC ? "Multicast Packet" :
5546                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
5547                                 wus & E1000_WUS_MAG ? "Magic Packet" :
5548                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
5549                                 "other");
5550                 }
5551                 ew32(WUS, ~0);
5552         }
5553
5554         e1000e_reset(adapter);
5555
5556         e1000_init_manageability_pt(adapter);
5557
5558         if (netif_running(netdev))
5559                 e1000e_up(adapter);
5560
5561         netif_device_attach(netdev);
5562
5563         /*
5564          * If the controller has AMT, do not set DRV_LOAD until the interface
5565          * is up.  For all other cases, let the f/w know that the h/w is now
5566          * under the control of the driver.
5567          */
5568         if (!(adapter->flags & FLAG_HAS_AMT))
5569                 e1000e_get_hw_control(adapter);
5570
5571         return 0;
5572 }
5573
5574 #ifdef CONFIG_PM_SLEEP
5575 static int e1000_suspend(struct device *dev)
5576 {
5577         struct pci_dev *pdev = to_pci_dev(dev);
5578         int retval;
5579         bool wake;
5580
5581         retval = __e1000_shutdown(pdev, &wake, false);
5582         if (!retval)
5583                 e1000_complete_shutdown(pdev, true, wake);
5584
5585         return retval;
5586 }
5587
5588 static int e1000_resume(struct device *dev)
5589 {
5590         struct pci_dev *pdev = to_pci_dev(dev);
5591         struct net_device *netdev = pci_get_drvdata(pdev);
5592         struct e1000_adapter *adapter = netdev_priv(netdev);
5593
5594         if (e1000e_pm_ready(adapter))
5595                 adapter->idle_check = true;
5596
5597         return __e1000_resume(pdev);
5598 }
5599 #endif /* CONFIG_PM_SLEEP */
5600
5601 #ifdef CONFIG_PM_RUNTIME
5602 static int e1000_runtime_suspend(struct device *dev)
5603 {
5604         struct pci_dev *pdev = to_pci_dev(dev);
5605         struct net_device *netdev = pci_get_drvdata(pdev);
5606         struct e1000_adapter *adapter = netdev_priv(netdev);
5607
5608         if (e1000e_pm_ready(adapter)) {
5609                 bool wake;
5610
5611                 __e1000_shutdown(pdev, &wake, true);
5612         }
5613
5614         return 0;
5615 }
5616
5617 static int e1000_idle(struct device *dev)
5618 {
5619         struct pci_dev *pdev = to_pci_dev(dev);
5620         struct net_device *netdev = pci_get_drvdata(pdev);
5621         struct e1000_adapter *adapter = netdev_priv(netdev);
5622
5623         if (!e1000e_pm_ready(adapter))
5624                 return 0;
5625
5626         if (adapter->idle_check) {
5627                 adapter->idle_check = false;
5628                 if (!e1000e_has_link(adapter))
5629                         pm_schedule_suspend(dev, MSEC_PER_SEC);
5630         }
5631
5632         return -EBUSY;
5633 }
5634
5635 static int e1000_runtime_resume(struct device *dev)
5636 {
5637         struct pci_dev *pdev = to_pci_dev(dev);
5638         struct net_device *netdev = pci_get_drvdata(pdev);
5639         struct e1000_adapter *adapter = netdev_priv(netdev);
5640
5641         if (!e1000e_pm_ready(adapter))
5642                 return 0;
5643
5644         adapter->idle_check = !dev->power.runtime_auto;
5645         return __e1000_resume(pdev);
5646 }
5647 #endif /* CONFIG_PM_RUNTIME */
5648 #endif /* CONFIG_PM */
5649
5650 static void e1000_shutdown(struct pci_dev *pdev)
5651 {
5652         bool wake = false;
5653
5654         __e1000_shutdown(pdev, &wake, false);
5655
5656         if (system_state == SYSTEM_POWER_OFF)
5657                 e1000_complete_shutdown(pdev, false, wake);
5658 }
5659
5660 #ifdef CONFIG_NET_POLL_CONTROLLER
5661
5662 static irqreturn_t e1000_intr_msix(int irq, void *data)
5663 {
5664         struct net_device *netdev = data;
5665         struct e1000_adapter *adapter = netdev_priv(netdev);
5666
5667         if (adapter->msix_entries) {
5668                 int vector, msix_irq;
5669
5670                 vector = 0;
5671                 msix_irq = adapter->msix_entries[vector].vector;
5672                 disable_irq(msix_irq);
5673                 e1000_intr_msix_rx(msix_irq, netdev);
5674                 enable_irq(msix_irq);
5675
5676                 vector++;
5677                 msix_irq = adapter->msix_entries[vector].vector;
5678                 disable_irq(msix_irq);
5679                 e1000_intr_msix_tx(msix_irq, netdev);
5680                 enable_irq(msix_irq);
5681
5682                 vector++;
5683                 msix_irq = adapter->msix_entries[vector].vector;
5684                 disable_irq(msix_irq);
5685                 e1000_msix_other(msix_irq, netdev);
5686                 enable_irq(msix_irq);
5687         }
5688
5689         return IRQ_HANDLED;
5690 }
5691
5692 /*
5693  * Polling 'interrupt' - used by things like netconsole to send skbs
5694  * without having to re-enable interrupts. It's not called while
5695  * the interrupt routine is executing.
5696  */
5697 static void e1000_netpoll(struct net_device *netdev)
5698 {
5699         struct e1000_adapter *adapter = netdev_priv(netdev);
5700
5701         switch (adapter->int_mode) {
5702         case E1000E_INT_MODE_MSIX:
5703                 e1000_intr_msix(adapter->pdev->irq, netdev);
5704                 break;
5705         case E1000E_INT_MODE_MSI:
5706                 disable_irq(adapter->pdev->irq);
5707                 e1000_intr_msi(adapter->pdev->irq, netdev);
5708                 enable_irq(adapter->pdev->irq);
5709                 break;
5710         default: /* E1000E_INT_MODE_LEGACY */
5711                 disable_irq(adapter->pdev->irq);
5712                 e1000_intr(adapter->pdev->irq, netdev);
5713                 enable_irq(adapter->pdev->irq);
5714                 break;
5715         }
5716 }
5717 #endif
5718
5719 /**
5720  * e1000_io_error_detected - called when PCI error is detected
5721  * @pdev: Pointer to PCI device
5722  * @state: The current pci connection state
5723  *
5724  * This function is called after a PCI bus error affecting
5725  * this device has been detected.
5726  */
5727 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5728                                                 pci_channel_state_t state)
5729 {
5730         struct net_device *netdev = pci_get_drvdata(pdev);
5731         struct e1000_adapter *adapter = netdev_priv(netdev);
5732
5733         netif_device_detach(netdev);
5734
5735         if (state == pci_channel_io_perm_failure)
5736                 return PCI_ERS_RESULT_DISCONNECT;
5737
5738         if (netif_running(netdev))
5739                 e1000e_down(adapter);
5740         pci_disable_device(pdev);
5741
5742         /* Request a slot slot reset. */
5743         return PCI_ERS_RESULT_NEED_RESET;
5744 }
5745
5746 /**
5747  * e1000_io_slot_reset - called after the pci bus has been reset.
5748  * @pdev: Pointer to PCI device
5749  *
5750  * Restart the card from scratch, as if from a cold-boot. Implementation
5751  * resembles the first-half of the e1000_resume routine.
5752  */
5753 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5754 {
5755         struct net_device *netdev = pci_get_drvdata(pdev);
5756         struct e1000_adapter *adapter = netdev_priv(netdev);
5757         struct e1000_hw *hw = &adapter->hw;
5758         u16 aspm_disable_flag = 0;
5759         int err;
5760         pci_ers_result_t result;
5761
5762         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5763                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5764         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5765                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5766         if (aspm_disable_flag)
5767                 e1000e_disable_aspm(pdev, aspm_disable_flag);
5768
5769         err = pci_enable_device_mem(pdev);
5770         if (err) {
5771                 dev_err(&pdev->dev,
5772                         "Cannot re-enable PCI device after reset.\n");
5773                 result = PCI_ERS_RESULT_DISCONNECT;
5774         } else {
5775                 pci_set_master(pdev);
5776                 pdev->state_saved = true;
5777                 pci_restore_state(pdev);
5778
5779                 pci_enable_wake(pdev, PCI_D3hot, 0);
5780                 pci_enable_wake(pdev, PCI_D3cold, 0);
5781
5782                 e1000e_reset(adapter);
5783                 ew32(WUS, ~0);
5784                 result = PCI_ERS_RESULT_RECOVERED;
5785         }
5786
5787         pci_cleanup_aer_uncorrect_error_status(pdev);
5788
5789         return result;
5790 }
5791
5792 /**
5793  * e1000_io_resume - called when traffic can start flowing again.
5794  * @pdev: Pointer to PCI device
5795  *
5796  * This callback is called when the error recovery driver tells us that
5797  * its OK to resume normal operation. Implementation resembles the
5798  * second-half of the e1000_resume routine.
5799  */
5800 static void e1000_io_resume(struct pci_dev *pdev)
5801 {
5802         struct net_device *netdev = pci_get_drvdata(pdev);
5803         struct e1000_adapter *adapter = netdev_priv(netdev);
5804
5805         e1000_init_manageability_pt(adapter);
5806
5807         if (netif_running(netdev)) {
5808                 if (e1000e_up(adapter)) {
5809                         dev_err(&pdev->dev,
5810                                 "can't bring device back up after reset\n");
5811                         return;
5812                 }
5813         }
5814
5815         netif_device_attach(netdev);
5816
5817         /*
5818          * If the controller has AMT, do not set DRV_LOAD until the interface
5819          * is up.  For all other cases, let the f/w know that the h/w is now
5820          * under the control of the driver.
5821          */
5822         if (!(adapter->flags & FLAG_HAS_AMT))
5823                 e1000e_get_hw_control(adapter);
5824
5825 }
5826
5827 static void e1000_print_device_info(struct e1000_adapter *adapter)
5828 {
5829         struct e1000_hw *hw = &adapter->hw;
5830         struct net_device *netdev = adapter->netdev;
5831         u32 ret_val;
5832         u8 pba_str[E1000_PBANUM_LENGTH];
5833
5834         /* print bus type/speed/width info */
5835         e_info("(PCI Express:2.5GT/s:%s) %pM\n",
5836                /* bus width */
5837                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5838                 "Width x1"),
5839                /* MAC address */
5840                netdev->dev_addr);
5841         e_info("Intel(R) PRO/%s Network Connection\n",
5842                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5843         ret_val = e1000_read_pba_string_generic(hw, pba_str,
5844                                                 E1000_PBANUM_LENGTH);
5845         if (ret_val)
5846                 strncpy((char *)pba_str, "Unknown", sizeof(pba_str) - 1);
5847         e_info("MAC: %d, PHY: %d, PBA No: %s\n",
5848                hw->mac.type, hw->phy.type, pba_str);
5849 }
5850
5851 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5852 {
5853         struct e1000_hw *hw = &adapter->hw;
5854         int ret_val;
5855         u16 buf = 0;
5856
5857         if (hw->mac.type != e1000_82573)
5858                 return;
5859
5860         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
5861         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
5862                 /* Deep Smart Power Down (DSPD) */
5863                 dev_warn(&adapter->pdev->dev,
5864                          "Warning: detected DSPD enabled in EEPROM\n");
5865         }
5866 }
5867
5868 static int e1000_set_features(struct net_device *netdev, u32 features)
5869 {
5870         struct e1000_adapter *adapter = netdev_priv(netdev);
5871         u32 changed = features ^ netdev->features;
5872
5873         if (changed & (NETIF_F_TSO | NETIF_F_TSO6))
5874                 adapter->flags |= FLAG_TSO_FORCE;
5875
5876         if (!(changed & (NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX |
5877                          NETIF_F_RXCSUM)))
5878                 return 0;
5879
5880         if (netif_running(netdev))
5881                 e1000e_reinit_locked(adapter);
5882         else
5883                 e1000e_reset(adapter);
5884
5885         return 0;
5886 }
5887
5888 static const struct net_device_ops e1000e_netdev_ops = {
5889         .ndo_open               = e1000_open,
5890         .ndo_stop               = e1000_close,
5891         .ndo_start_xmit         = e1000_xmit_frame,
5892         .ndo_get_stats64        = e1000e_get_stats64,
5893         .ndo_set_rx_mode        = e1000_set_multi,
5894         .ndo_set_mac_address    = e1000_set_mac,
5895         .ndo_change_mtu         = e1000_change_mtu,
5896         .ndo_do_ioctl           = e1000_ioctl,
5897         .ndo_tx_timeout         = e1000_tx_timeout,
5898         .ndo_validate_addr      = eth_validate_addr,
5899
5900         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
5901         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
5902 #ifdef CONFIG_NET_POLL_CONTROLLER
5903         .ndo_poll_controller    = e1000_netpoll,
5904 #endif
5905         .ndo_set_features = e1000_set_features,
5906 };
5907
5908 /**
5909  * e1000_probe - Device Initialization Routine
5910  * @pdev: PCI device information struct
5911  * @ent: entry in e1000_pci_tbl
5912  *
5913  * Returns 0 on success, negative on failure
5914  *
5915  * e1000_probe initializes an adapter identified by a pci_dev structure.
5916  * The OS initialization, configuring of the adapter private structure,
5917  * and a hardware reset occur.
5918  **/
5919 static int __devinit e1000_probe(struct pci_dev *pdev,
5920                                  const struct pci_device_id *ent)
5921 {
5922         struct net_device *netdev;
5923         struct e1000_adapter *adapter;
5924         struct e1000_hw *hw;
5925         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
5926         resource_size_t mmio_start, mmio_len;
5927         resource_size_t flash_start, flash_len;
5928
5929         static int cards_found;
5930         u16 aspm_disable_flag = 0;
5931         int i, err, pci_using_dac;
5932         u16 eeprom_data = 0;
5933         u16 eeprom_apme_mask = E1000_EEPROM_APME;
5934
5935         if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
5936                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5937         if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
5938                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5939         if (aspm_disable_flag)
5940                 e1000e_disable_aspm(pdev, aspm_disable_flag);
5941
5942         err = pci_enable_device_mem(pdev);
5943         if (err)
5944                 return err;
5945
5946         pci_using_dac = 0;
5947         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
5948         if (!err) {
5949                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
5950                 if (!err)
5951                         pci_using_dac = 1;
5952         } else {
5953                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
5954                 if (err) {
5955                         err = dma_set_coherent_mask(&pdev->dev,
5956                                                     DMA_BIT_MASK(32));
5957                         if (err) {
5958                                 dev_err(&pdev->dev, "No usable DMA "
5959                                         "configuration, aborting\n");
5960                                 goto err_dma;
5961                         }
5962                 }
5963         }
5964
5965         err = pci_request_selected_regions_exclusive(pdev,
5966                                           pci_select_bars(pdev, IORESOURCE_MEM),
5967                                           e1000e_driver_name);
5968         if (err)
5969                 goto err_pci_reg;
5970
5971         /* AER (Advanced Error Reporting) hooks */
5972         pci_enable_pcie_error_reporting(pdev);
5973
5974         pci_set_master(pdev);
5975         /* PCI config space info */
5976         err = pci_save_state(pdev);
5977         if (err)
5978                 goto err_alloc_etherdev;
5979
5980         err = -ENOMEM;
5981         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
5982         if (!netdev)
5983                 goto err_alloc_etherdev;
5984
5985         SET_NETDEV_DEV(netdev, &pdev->dev);
5986
5987         netdev->irq = pdev->irq;
5988
5989         pci_set_drvdata(pdev, netdev);
5990         adapter = netdev_priv(netdev);
5991         hw = &adapter->hw;
5992         adapter->netdev = netdev;
5993         adapter->pdev = pdev;
5994         adapter->ei = ei;
5995         adapter->pba = ei->pba;
5996         adapter->flags = ei->flags;
5997         adapter->flags2 = ei->flags2;
5998         adapter->hw.adapter = adapter;
5999         adapter->hw.mac.type = ei->mac;
6000         adapter->max_hw_frame_size = ei->max_hw_frame_size;
6001         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
6002
6003         mmio_start = pci_resource_start(pdev, 0);
6004         mmio_len = pci_resource_len(pdev, 0);
6005
6006         err = -EIO;
6007         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
6008         if (!adapter->hw.hw_addr)
6009                 goto err_ioremap;
6010
6011         if ((adapter->flags & FLAG_HAS_FLASH) &&
6012             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
6013                 flash_start = pci_resource_start(pdev, 1);
6014                 flash_len = pci_resource_len(pdev, 1);
6015                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
6016                 if (!adapter->hw.flash_address)
6017                         goto err_flashmap;
6018         }
6019
6020         /* construct the net_device struct */
6021         netdev->netdev_ops              = &e1000e_netdev_ops;
6022         e1000e_set_ethtool_ops(netdev);
6023         netdev->watchdog_timeo          = 5 * HZ;
6024         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
6025         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
6026
6027         netdev->mem_start = mmio_start;
6028         netdev->mem_end = mmio_start + mmio_len;
6029
6030         adapter->bd_number = cards_found++;
6031
6032         e1000e_check_options(adapter);
6033
6034         /* setup adapter struct */
6035         err = e1000_sw_init(adapter);
6036         if (err)
6037                 goto err_sw_init;
6038
6039         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
6040         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
6041         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
6042
6043         err = ei->get_variants(adapter);
6044         if (err)
6045                 goto err_hw_init;
6046
6047         if ((adapter->flags & FLAG_IS_ICH) &&
6048             (adapter->flags & FLAG_READ_ONLY_NVM))
6049                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
6050
6051         hw->mac.ops.get_bus_info(&adapter->hw);
6052
6053         adapter->hw.phy.autoneg_wait_to_complete = 0;
6054
6055         /* Copper options */
6056         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
6057                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
6058                 adapter->hw.phy.disable_polarity_correction = 0;
6059                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
6060         }
6061
6062         if (e1000_check_reset_block(&adapter->hw))
6063                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
6064
6065         /* Set initial default active device features */
6066         netdev->features = (NETIF_F_SG |
6067                             NETIF_F_HW_VLAN_RX |
6068                             NETIF_F_HW_VLAN_TX |
6069                             NETIF_F_TSO |
6070                             NETIF_F_TSO6 |
6071                             NETIF_F_RXCSUM |
6072                             NETIF_F_HW_CSUM);
6073
6074         /* Set user-changeable features (subset of all device features) */
6075         netdev->hw_features = netdev->features;
6076
6077         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
6078                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
6079
6080         netdev->vlan_features |= (NETIF_F_SG |
6081                                   NETIF_F_TSO |
6082                                   NETIF_F_TSO6 |
6083                                   NETIF_F_HW_CSUM);
6084
6085         if (pci_using_dac) {
6086                 netdev->features |= NETIF_F_HIGHDMA;
6087                 netdev->vlan_features |= NETIF_F_HIGHDMA;
6088         }
6089
6090         if (e1000e_enable_mng_pass_thru(&adapter->hw))
6091                 adapter->flags |= FLAG_MNG_PT_ENABLED;
6092
6093         /*
6094          * before reading the NVM, reset the controller to
6095          * put the device in a known good starting state
6096          */
6097         adapter->hw.mac.ops.reset_hw(&adapter->hw);
6098
6099         /*
6100          * systems with ASPM and others may see the checksum fail on the first
6101          * attempt. Let's give it a few tries
6102          */
6103         for (i = 0;; i++) {
6104                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
6105                         break;
6106                 if (i == 2) {
6107                         e_err("The NVM Checksum Is Not Valid\n");
6108                         err = -EIO;
6109                         goto err_eeprom;
6110                 }
6111         }
6112
6113         e1000_eeprom_checks(adapter);
6114
6115         /* copy the MAC address */
6116         if (e1000e_read_mac_addr(&adapter->hw))
6117                 e_err("NVM Read Error while reading MAC address\n");
6118
6119         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
6120         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
6121
6122         if (!is_valid_ether_addr(netdev->perm_addr)) {
6123                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
6124                 err = -EIO;
6125                 goto err_eeprom;
6126         }
6127
6128         init_timer(&adapter->watchdog_timer);
6129         adapter->watchdog_timer.function = e1000_watchdog;
6130         adapter->watchdog_timer.data = (unsigned long) adapter;
6131
6132         init_timer(&adapter->phy_info_timer);
6133         adapter->phy_info_timer.function = e1000_update_phy_info;
6134         adapter->phy_info_timer.data = (unsigned long) adapter;
6135
6136         INIT_WORK(&adapter->reset_task, e1000_reset_task);
6137         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
6138         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
6139         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
6140         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
6141
6142         /* Initialize link parameters. User can change them with ethtool */
6143         adapter->hw.mac.autoneg = 1;
6144         adapter->fc_autoneg = 1;
6145         adapter->hw.fc.requested_mode = e1000_fc_default;
6146         adapter->hw.fc.current_mode = e1000_fc_default;
6147         adapter->hw.phy.autoneg_advertised = 0x2f;
6148
6149         /* ring size defaults */
6150         adapter->rx_ring->count = E1000_DEFAULT_RXD;
6151         adapter->tx_ring->count = E1000_DEFAULT_TXD;
6152
6153         /*
6154          * Initial Wake on LAN setting - If APM wake is enabled in
6155          * the EEPROM, enable the ACPI Magic Packet filter
6156          */
6157         if (adapter->flags & FLAG_APME_IN_WUC) {
6158                 /* APME bit in EEPROM is mapped to WUC.APME */
6159                 eeprom_data = er32(WUC);
6160                 eeprom_apme_mask = E1000_WUC_APME;
6161                 if ((hw->mac.type > e1000_ich10lan) &&
6162                     (eeprom_data & E1000_WUC_PHY_WAKE))
6163                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
6164         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
6165                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
6166                     (adapter->hw.bus.func == 1))
6167                         e1000_read_nvm(&adapter->hw,
6168                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
6169                 else
6170                         e1000_read_nvm(&adapter->hw,
6171                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
6172         }
6173
6174         /* fetch WoL from EEPROM */
6175         if (eeprom_data & eeprom_apme_mask)
6176                 adapter->eeprom_wol |= E1000_WUFC_MAG;
6177
6178         /*
6179          * now that we have the eeprom settings, apply the special cases
6180          * where the eeprom may be wrong or the board simply won't support
6181          * wake on lan on a particular port
6182          */
6183         if (!(adapter->flags & FLAG_HAS_WOL))
6184                 adapter->eeprom_wol = 0;
6185
6186         /* initialize the wol settings based on the eeprom settings */
6187         adapter->wol = adapter->eeprom_wol;
6188         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6189
6190         /* save off EEPROM version number */
6191         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
6192
6193         /* reset the hardware with the new settings */
6194         e1000e_reset(adapter);
6195
6196         /*
6197          * If the controller has AMT, do not set DRV_LOAD until the interface
6198          * is up.  For all other cases, let the f/w know that the h/w is now
6199          * under the control of the driver.
6200          */
6201         if (!(adapter->flags & FLAG_HAS_AMT))
6202                 e1000e_get_hw_control(adapter);
6203
6204         strncpy(netdev->name, "eth%d", sizeof(netdev->name) - 1);
6205         err = register_netdev(netdev);
6206         if (err)
6207                 goto err_register;
6208
6209         /* carrier off reporting is important to ethtool even BEFORE open */
6210         netif_carrier_off(netdev);
6211
6212         e1000_print_device_info(adapter);
6213
6214         if (pci_dev_run_wake(pdev))
6215                 pm_runtime_put_noidle(&pdev->dev);
6216
6217         return 0;
6218
6219 err_register:
6220         if (!(adapter->flags & FLAG_HAS_AMT))
6221                 e1000e_release_hw_control(adapter);
6222 err_eeprom:
6223         if (!e1000_check_reset_block(&adapter->hw))
6224                 e1000_phy_hw_reset(&adapter->hw);
6225 err_hw_init:
6226         kfree(adapter->tx_ring);
6227         kfree(adapter->rx_ring);
6228 err_sw_init:
6229         if (adapter->hw.flash_address)
6230                 iounmap(adapter->hw.flash_address);
6231         e1000e_reset_interrupt_capability(adapter);
6232 err_flashmap:
6233         iounmap(adapter->hw.hw_addr);
6234 err_ioremap:
6235         free_netdev(netdev);
6236 err_alloc_etherdev:
6237         pci_release_selected_regions(pdev,
6238                                      pci_select_bars(pdev, IORESOURCE_MEM));
6239 err_pci_reg:
6240 err_dma:
6241         pci_disable_device(pdev);
6242         return err;
6243 }
6244
6245 /**
6246  * e1000_remove - Device Removal Routine
6247  * @pdev: PCI device information struct
6248  *
6249  * e1000_remove is called by the PCI subsystem to alert the driver
6250  * that it should release a PCI device.  The could be caused by a
6251  * Hot-Plug event, or because the driver is going to be removed from
6252  * memory.
6253  **/
6254 static void __devexit e1000_remove(struct pci_dev *pdev)
6255 {
6256         struct net_device *netdev = pci_get_drvdata(pdev);
6257         struct e1000_adapter *adapter = netdev_priv(netdev);
6258         bool down = test_bit(__E1000_DOWN, &adapter->state);
6259
6260         /*
6261          * The timers may be rescheduled, so explicitly disable them
6262          * from being rescheduled.
6263          */
6264         if (!down)
6265                 set_bit(__E1000_DOWN, &adapter->state);
6266         del_timer_sync(&adapter->watchdog_timer);
6267         del_timer_sync(&adapter->phy_info_timer);
6268
6269         cancel_work_sync(&adapter->reset_task);
6270         cancel_work_sync(&adapter->watchdog_task);
6271         cancel_work_sync(&adapter->downshift_task);
6272         cancel_work_sync(&adapter->update_phy_task);
6273         cancel_work_sync(&adapter->print_hang_task);
6274
6275         if (!(netdev->flags & IFF_UP))
6276                 e1000_power_down_phy(adapter);
6277
6278         /* Don't lie to e1000_close() down the road. */
6279         if (!down)
6280                 clear_bit(__E1000_DOWN, &adapter->state);
6281         unregister_netdev(netdev);
6282
6283         if (pci_dev_run_wake(pdev))
6284                 pm_runtime_get_noresume(&pdev->dev);
6285
6286         /*
6287          * Release control of h/w to f/w.  If f/w is AMT enabled, this
6288          * would have already happened in close and is redundant.
6289          */
6290         e1000e_release_hw_control(adapter);
6291
6292         e1000e_reset_interrupt_capability(adapter);
6293         kfree(adapter->tx_ring);
6294         kfree(adapter->rx_ring);
6295
6296         iounmap(adapter->hw.hw_addr);
6297         if (adapter->hw.flash_address)
6298                 iounmap(adapter->hw.flash_address);
6299         pci_release_selected_regions(pdev,
6300                                      pci_select_bars(pdev, IORESOURCE_MEM));
6301
6302         free_netdev(netdev);
6303
6304         /* AER disable */
6305         pci_disable_pcie_error_reporting(pdev);
6306
6307         pci_disable_device(pdev);
6308 }
6309
6310 /* PCI Error Recovery (ERS) */
6311 static struct pci_error_handlers e1000_err_handler = {
6312         .error_detected = e1000_io_error_detected,
6313         .slot_reset = e1000_io_slot_reset,
6314         .resume = e1000_io_resume,
6315 };
6316
6317 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
6318         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
6319         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
6320         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
6321         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
6322         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
6323         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
6324         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
6325         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
6326         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
6327
6328         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
6329         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
6330         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
6331         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
6332
6333         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
6334         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
6335         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
6336
6337         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
6338         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
6339         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
6340
6341         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
6342           board_80003es2lan },
6343         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
6344           board_80003es2lan },
6345         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
6346           board_80003es2lan },
6347         { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
6348           board_80003es2lan },
6349
6350         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
6351         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
6352         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
6353         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
6354         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
6355         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
6356         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
6357         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
6358
6359         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
6360         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
6361         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
6362         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
6363         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
6364         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
6365         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
6366         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
6367         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
6368
6369         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
6370         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
6371         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
6372
6373         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
6374         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
6375         { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
6376
6377         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
6378         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
6379         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
6380         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
6381
6382         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
6383         { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
6384
6385         { }     /* terminate list */
6386 };
6387 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
6388
6389 #ifdef CONFIG_PM
6390 static const struct dev_pm_ops e1000_pm_ops = {
6391         SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
6392         SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
6393                                 e1000_runtime_resume, e1000_idle)
6394 };
6395 #endif
6396
6397 /* PCI Device API Driver */
6398 static struct pci_driver e1000_driver = {
6399         .name     = e1000e_driver_name,
6400         .id_table = e1000_pci_tbl,
6401         .probe    = e1000_probe,
6402         .remove   = __devexit_p(e1000_remove),
6403 #ifdef CONFIG_PM
6404         .driver.pm = &e1000_pm_ops,
6405 #endif
6406         .shutdown = e1000_shutdown,
6407         .err_handler = &e1000_err_handler
6408 };
6409
6410 /**
6411  * e1000_init_module - Driver Registration Routine
6412  *
6413  * e1000_init_module is the first routine called when the driver is
6414  * loaded. All it does is register with the PCI subsystem.
6415  **/
6416 static int __init e1000_init_module(void)
6417 {
6418         int ret;
6419         pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
6420                 e1000e_driver_version);
6421         pr_info("Copyright(c) 1999 - 2011 Intel Corporation.\n");
6422         ret = pci_register_driver(&e1000_driver);
6423
6424         return ret;
6425 }
6426 module_init(e1000_init_module);
6427
6428 /**
6429  * e1000_exit_module - Driver Exit Cleanup Routine
6430  *
6431  * e1000_exit_module is called just before the driver is removed
6432  * from memory.
6433  **/
6434 static void __exit e1000_exit_module(void)
6435 {
6436         pci_unregister_driver(&e1000_driver);
6437 }
6438 module_exit(e1000_exit_module);
6439
6440
6441 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
6442 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
6443 MODULE_LICENSE("GPL");
6444 MODULE_VERSION(DRV_VERSION);
6445
6446 /* e1000_main.c */